id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0506/quant-ph0506127.html
|
ar5iv
|
text
|
# The entanglement criterion of multiqubits
## Abstract
We present an entanglement criterion for multiqubits by using the quantum correlation tensors which rely on the expectation values of the Pauli operators for a multiqubit state. Our criterion explains not only the total entanglement of the system but also the partial entanglement in subsystems. It shows that we have to consider the subsystem entanglements in order to obtain the full description for multiqubit entanglements. Furthermore, we offer an extension of the entanglement to multiqudits.
Entanglement has been an important key word in the quantum computer and the quantum information technology. In particular, entanglements in the bipartite qubits have many applications as the superdense coding, quantum computation, teleportation, clock synchronization,and quantum cryptography. These have been clarified by a negative partial transposition ,, and quantified by concurrence, negativity, entanglement of formation, etc..
As many entangled states such as GHZ state, W state, etc., were found in multiqubits, the multiqubit entanglements have been applied to a real physical system such as quantum secret sharing and the one-way quantum computer. The investigation on the entanglement properties in the multipartite system has thus emerged as a central problem in quantum information study. However, no efficient method to clarify the status of multipartite entanglements has been introduced.
The classification of the mathematical and physical structures in multipartite states was, at first, tried by the local operation associated with classical communication(LOCC). In the multipartite systems, the investigations on entanglement measure have been proposed by the tangle which is computed by the concurrence between two intentionally divided subsystems in an effective two-dimensional Hilbert space. Recently, entanglement witnesses were suggested by another method for the classification of multipartite entanglements,. This method requires the witness operators to detect various forms of multipartite entanglements. However, it is difficult to know the witness operators before we classify the multipartite system, and additionally witness operators are defined by some a priori knowledge about the states under investigation.
In spite of trials, the entanglements in a multipartite system are complicated even in pure systems because the quantum states can share entanglements differently among possible subsystems and have the different classes of total entangled states as GHZ, W or cluster states. It is important to define an entanglement criterion that could distinguish all possible types of entanglements which exist among the constituents. In this letter, we present a general entanglement criterion that can solve the above problems for pure multiqubit systems. Furthermore we will discuss that our entanglement description can be extended to multiqubit mixed states and higher dimensional Hilbert spaces.
In general, a pure composite system with $`N`$ qubits can be represented by
$$|\mathrm{\Psi }(1,2,3,\mathrm{},N)=\underset{IJ\mathrm{}K=0}{\overset{1}{}}a_{IJ\mathrm{}K}|I_1|J_2\mathrm{}|K_N,$$
(1)
where $`_{IJ\mathrm{}K=0}^1|a_{IJ\mathrm{}K}|^2=1`$ and the state is an element of composite Hilbert space as $`_N=_2_2\mathrm{}_2`$.
Our question is whether this state is separable or entangled. In order to answer this question, we introduce the quantum correlation tensor for the given multipartite qubit system $`|\mathrm{\Psi }`$ as
$`M_{i_1i_2\mathrm{}i_n}(\alpha _1,\alpha _2,\mathrm{},\alpha _N;|\mathrm{\Psi })=\mathrm{\Psi }|(\sigma _{i_1}(\alpha _1)\lambda _i(\alpha _1))(\sigma _{i_2}(\alpha _2)\lambda _{i_2}(\alpha _2))`$
$`\mathrm{}(\sigma _{i_n}(\alpha _n)\lambda _{i_n}(\alpha _n))I(\alpha _{n+1})\mathrm{}I(\alpha _N)|\mathrm{\Psi },`$ (2)
where $`nN`$, $`\sigma _i(\alpha )`$ denotes the $`i^{th}`$-component Pauliโs operator of the $`\alpha ^{th}`$ qubit and $`\lambda _{i_j}(\alpha _j)=\mathrm{\Psi }|I(1)I(2)\mathrm{}\sigma _{i_j}(\alpha _j)\mathrm{}I(N)|\mathrm{\Psi }`$. Here, $`I(\alpha )`$ is the identity operator on the $`\alpha ^{th}`$ qubit. Obviously, $`M`$ determines whether one qubit is separated. If the state is separable such as $`|\mathrm{\Psi }(1,2,3,\mathrm{},N)=|\mathrm{\Psi }(\alpha _1,\alpha _2,\mathrm{},\alpha _{N1})|\mathrm{\Psi }(\alpha _N)`$, $`M_{i_1i_2\mathrm{}i_N}(1,2,\mathrm{},N;|\mathrm{\Psi })`$ must be zero. This fact can be proved by a simple calculation. Conversely, if $`M_{i_1i_2\mathrm{}i_N}(1,2,\mathrm{},N;|\mathrm{\Psi })=0`$ for all $`i_1,i_2,\mathrm{},i_N`$, the state is separable such as $`|\mathrm{\Psi }(1,2,3,\mathrm{},N)=|\mathrm{\Psi }(\alpha _1,\alpha _2,\mathrm{},\alpha _{N1})|\mathrm{\Psi }(\alpha _N)`$, which implies that one of the qubits is uncorrelated with the others. We can consider a $`N`$-qubit system as a bipartite system consisted of a $`(N1)`$-qubit system and a single qubit system. So the original state can be rewritten by the Schmidtโs decomposition as
$$|\mathrm{\Psi }(1,2,3,\mathrm{},N)=\alpha |a,0+\beta |b,1,$$
(3)
where $`|a`$ and $`|b`$ are orthogonal states of the $`(N1)`$-qubit system with $`\alpha ^2+\beta ^2=1`$. We denote the operators of the $`(N1)`$-qubit space as a generator, $`\widehat{L_K}`$. We can calculate $`M`$ for a bipartite system as following;
$`M_{Kx}`$ $`=`$ $`(\alpha a,0|+\beta b,1|)\widehat{L_K}\sigma _x(\alpha |a,0+\beta |b,1)L_K\lambda _x`$
$`=`$ $`2\alpha \beta Re(a|\widehat{L_K}|b),`$
$`M_{Ky}`$ $`=`$ $`(\alpha a,0|+\beta b,1|)\widehat{L_K}\sigma _y(\alpha |a,0+\beta |b,1)L_K\lambda _y`$
$`=`$ $`2\alpha \beta Im(a|\widehat{L_K}|b),`$
$`M_{Kz}`$ $`=`$ $`(\alpha a,0|+\beta b,1|)\widehat{L_K}\sigma _z(\alpha |a,0+\beta |b,1)L_K\lambda _z`$
$`=`$ $`\alpha ^2a|\widehat{L_K}|a\beta ^2b|\widehat{L_K}|b(\alpha ^2\beta ^2)(\alpha ^2a|\widehat{L_K}|a+\beta ^2b|\widehat{L_K}|b),`$
where $`L_K=(\alpha a,0|+\beta b,1|)(\widehat{L_K}I)(\alpha |a,0+\beta |b,1)`$. $`\alpha `$ or $`\beta `$ has to be vanished in order for all of $`M`$โs to be zero. This indicates that at least one of the qubits is uncorrelated with the others.
For a two-qubit system $`M_{ij}`$ is the criterion to judge whether the bipartite pure state is separated or entangled. Schlinz and Mahler suggested this scenario in a bipartite system. The three-qubit state, $`|\mathrm{\Psi }(1,2,3)`$, has three types of entanglements; a separated state as $`ABC`$, a bipartite entangled state as $`ABC,ABC`$ or $`CAB`$ and a totally entangled state as $`ABC`$. Nonzero of any $`M_{ijk}(1,2,3;|\mathrm{\Psi })`$ tells us that the state is totally entangled as the GHZ or W state. Zero of $`M_{ijk}(1,2,3;|\mathrm{\Psi })`$ for any $`i,j,k`$ means that the state can be either $`|\mathrm{\Psi }(1,2)|\mathrm{\Psi }(3)`$ or $`|\mathrm{\Psi }(1)|\mathrm{\Psi }(2)|\mathrm{\Psi }(3)`$. However, the increase of qubit numbers in the system produces the increase of the possibilities in the entanglement types. We have to differentiate all these situations in order to fully describe the entanglement structure. When $`M_{i_1i_2i_3\mathrm{}i_N}(\alpha _1,\alpha _2,\mathrm{},\alpha _N;|\mathrm{\Psi })=0`$ for all $`i_1,i_2,\mathrm{},i_N`$, there are various situations including completely separable and multi-separable cases. We can easily investigate that the state is at least one-qubit separated such as $`|\mathrm{\Psi }(\alpha _1,\alpha _2,\mathrm{},\alpha _{N1})|\mathrm{\Psi }(\alpha _N)`$, but we cannot judge directly whether the state, $`|\mathrm{\Psi }(\alpha _1,\alpha _2,\mathrm{},\alpha _{N1})`$ is entangled or separated. $`M_{i_1i_2\mathrm{}i_n}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })`$ with $`n<N`$ needs to determine the entanglement of the subsystems consisted of $`n`$ qubits. $`M_{i_1i_2\mathrm{}i_n}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })`$ is not zero for the states which have the entanglement among the $`n`$ qubits. $`M_{i_1i_2\mathrm{}i_n}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })`$ can distinguish totally entangled state from partially entangled state such as $`|\mathrm{\Psi }(\alpha _{i_1},\alpha _{i_2},\mathrm{},\alpha _{i_n})|\mathrm{\Psi }(\alpha _{i+1})|\mathrm{\Psi }(\alpha _{i+2})\mathrm{}|\mathrm{\Psi }(\alpha _N)`$.
The correlation tensor, $`M`$, is sufficient in bipartite and tripartite systems. However, the situation is different in systems consisted of more than three qubits. They cannot assure a criterion for partially entangled cases such as $`|\mathrm{\Psi }(\alpha _1,\alpha _2,\mathrm{},\alpha _l)|\mathrm{\Psi }(\beta _1,\beta _2\mathrm{}\beta _m)\mathrm{}|\mathrm{\Psi }(\gamma _1,\gamma _2\mathrm{}\gamma _n)`$. We have to modify the quantum correlation tensors to solve the problem by
$`M_{i_1i_2\mathrm{}i_n}^{}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })=M_{i_1i_2\mathrm{}i_n}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })`$
$`{\displaystyle \underset{AB\mathrm{}C=\{1,2,3\mathrm{}n\}}{}}(M_A(\alpha _1^{},\alpha _2^{},\mathrm{},\alpha _{n_1}^{};|\mathrm{\Psi })M_B(\alpha _1^{\prime \prime },\alpha _2^{\prime \prime },\mathrm{},\alpha _{n_2}^{\prime \prime };|\mathrm{\Psi })`$
$`\mathrm{}M_C(\alpha _1^{\prime \prime \prime },\alpha _2^{\prime \prime \prime },\mathrm{},\alpha _{n_m}^{\prime \prime \prime };|\mathrm{\Psi }),`$ (4)
where $`n_1+n_2+\mathrm{}+n_m=n`$ and $`A=i_1i_2\mathrm{}i_{n_1}`$ and $`i_j`$ denotes the $`i^{th}`$-component of Pauliโ operator acted on the $`j^{th}`$ qubit. The sum of the second term in the right side of eq. (4) denotes the possible disjoint partitions of indices composed of Pauliโs components of each qubit. Since $`M^{}`$ equals to $`M`$ in bipartite and tripartite systems, it is requested in the systems which consist of more than three qubits. For instance, $`M^{}`$ of the four-qubit case is written by
$`M_{i_1i_2i_3i_4}^{}(1,2,3,4;|\mathrm{\Psi })=M_{i_1i_2i_3i_4}(1,2,3,4;|\mathrm{\Psi })M_{i_1i_2}(1,2;|\mathrm{\Psi })M_{i_3i_4}(3,4;|\mathrm{\Psi })`$
$`M_{i_1i_3}(1,3;|\mathrm{\Psi })M_{i_2i_4}(2,4;|\mathrm{\Psi })M_{i_1i_4}(1,4;|\mathrm{\Psi })M_{i_2i_3}(2,3;|\mathrm{\Psi }).`$ (5)
This provides the complete criterion for multipartite entanglement which includes the multi-separable subsystems. If $`M_{i_1i_2\mathrm{}i_n}^{}(\alpha _1,\alpha _2,\mathrm{},\alpha _N;|\mathrm{\Psi })=0`$ for any $`i_1,i_2,\mathrm{},i_n`$, the given state is separable, and has two possibilities. The first is that every term on the right hand side in eq. (5) is zero. This means that the state has the from of $`|\mathrm{\Psi }(1,2,3,4)=|\psi (\alpha _1,\alpha _2,\alpha _3)|\varphi (\alpha _4)`$. The second is that the first term, $`M_{i_1i_2i_3i_4}(1,2,3,4;|\mathrm{\Psi })`$, is subtracted by three terms with a negative sign. However, we can show without difficulties that only one term out of three terms with the negative sign is nonzero; if $`M_{ij}(1,2,;|\mathrm{\Psi })M_{kl}(3,4;|\mathrm{\Psi })0`$ for the four-qubit case, $`M_{i_1i_2i_3i_4}(1,2,3,4;|\mathrm{\Psi })=M_{ij}(1,2;|\mathrm{\Psi })M_{kl}(3,4;|\mathrm{\Psi })`$. This fact lead us to find that the state is $`|\mathrm{\Psi }(1,2,3,4)=|\psi (1,2)|\varphi (3,4)`$.
If any pure state of a $`N`$-qubit system is given, we must, at first, check whether $`M_{i_1i_2\mathrm{}i_N}^{}(\alpha _1,\alpha _2,\mathrm{},\alpha _N;|\mathrm{\Psi })`$ is zero or not. In the nonzero case, the $`N`$-qubit is totally entangled but in the zero case, the state is either completely separated or partially entangled, as described earlier. In the zero case, we have to make additional checks whether $`M_{i_1,i_2,\mathrm{}ฤฑ_n}^{}(\alpha _1,\alpha _2,\mathrm{},\alpha _n;|\mathrm{\Psi })`$ for $`n<N`$ is zero or not in sequence. These sequential checks determine whether the given state is totally entangled, biseparable, triseparable, $`\mathrm{}`$ or completely separable. Then $`M^{}`$ classifies all the possible forms of entangled states.
By the tensor, $`M^{}`$, we can classify the pure multiqubit states including many different entanglement types. However, the tensor cannot distinguish the entangled states which are connected to each others under a local unitary transformation. For instance, $`M^{}`$ can distinguish the product state from the Bell states in a two-qubit system. However, the values of $`M^{}`$ of four Bell states are different from each others. One may misunderstand that the four Bell states have different degrees of entanglements. It is well known that the four Bell states are equivalent under a local unitary transformation as maximally entangled states. $`M^{}`$ just distinguishes whether the multiqubit state are entangled or multiseparated. Therefore, we need a new quantity to determine an entanglement magnitude.
Based on $`M^{}`$, we introduce a measure of an entanglement such as
$$B^{(m)}(\alpha _1,\alpha _2,\mathrm{}\alpha _m;|\mathrm{\Psi })=\underset{ijkl\mathrm{}}{}M_{ijkl\mathrm{}}^{}(\alpha _1,\alpha _2,\mathrm{}\alpha _m;|\mathrm{\Psi })M_{ijkl\mathrm{}}^{}(\alpha _1,\alpha _2,\mathrm{}\alpha _m;|\mathrm{\Psi }).$$
(6)
$`B^{(m)}(\alpha _1,\alpha _2,\mathrm{}\alpha _m;|\mathrm{\Psi })`$ calculates the entanglement magnitude among $`m`$ qubits labeled by $`\alpha _1,\alpha _2,\mathrm{}\alpha _m`$. For example, $`B^{(2)}(\alpha _1,\alpha _2;|\mathrm{\Psi })`$ describes the entanglement magnitude between the qubits $`\alpha _1`$ and $`\alpha _2`$ and $`B^{(3)}(\alpha _1,\alpha _2,\alpha _3;|\mathrm{\Psi })`$ the entanglement degree among the qubits $`\alpha _1,\alpha _2,\alpha _3`$. $`B^{(2)}(1,2;|\mathrm{\Psi })`$ in two-qubit systems is the same measure as Schlinz and Mahlerโs entanglement measure.
$`B^{(m)}`$ of eq. (6) satisfies the following properties affirming entanglement monotone, First, it is nonnegative, because $`B`$ is defined by the square of real numbers and $`M^{}=0`$ in separable states. Second, it is invariant under any local unitary transformations. This can be shown easily by using $`U^{}\sigma _iU=T_{ij}\sigma _j`$ and $`_iT_{ij}T_{ik}=\delta _{jk}`$ where $`U`$ is a unitary matrix and $`T`$ is a $`3\times 3`$ orthogonal matrix in the qubit systems. Third, it is nonincreasing under local measurements. The measurement collapses an entangled n-qubit state to biseparable state that one qubit under local measurement processes are uncorrelated with the others. Then, $`B^{(m)}`$ is vanished under local measurements and less than that of the original state. Fourth, it is invariant under the addition of an uncorrelated ancillary state. Fifth, it is not increased by tracing out a part of the system. Thus we claim that $`B^{(m)}`$ is the entanglement monotone.
For example, we consider four-qubit states which are totally entangled,
$`|GHZ_4`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|0000+|1111,`$
$`|W_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|0001+|0010+|0100+|1000),`$
$`|\varphi _6`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{6}}}(|0011+|0101+|1001+|1010+|0110+|1100)`$
$`|\varphi _4`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|0000+|0011+|1100|1111).`$ (7)
We can get $`M_{ijkl}^{}(1,2,3,4;|\mathrm{\Psi })0`$ for the above four states and know that their states are totally entangled. Then we can distinguish the entanglement difference through calculation of $`B^{(4)}`$; $`B^{(4)}(1,2,3,4;|GHZ_4)=1,B^{(4)}(1,2,3,4;|W_4)=\frac{51}{256},B^{(4)}(1,2,3,4;|\varphi _6)=\frac{7}{27}`$, and $`B^{(4)}(1,2,3,4;|\varphi _4)=\frac{1}{3}`$. Here we normalized $`B^{(4)}`$ with the value of the GHZ state. A partially three-qubit entangled state as $`|GHZ_3|0`$ and a partially two-qubit entangled state as $`|Bell_2|Bell_2`$ where $`|Bell_2=\frac{1}{2}(|00+|11)`$ have $`M^{}=0`$ for all subindexes. Then this explains these states do not have total entanglement. However, we can obtain $`M^{}(1,2,3)`$ is not zero for $`|GHZ_3|0`$ and $`M^{}(1,2)`$ and $`M^{}(3,4)`$ are not zero for $`|Bell_2|Bell_2`$. This shows that $`|GHZ_3|0`$ has a partial entanglement among qubits $`1,2`$ and $`3`$ and $`|Bell_2|Bell_2`$ has partial entanglements between qubits $`1`$ and $`2`$, and between qubits $`3`$ and $`4`$.
If the Pauliโs operators of the eq. (2) which are the generators of $`SU(2)`$ unitary group are replaced by the generators related to the higher dimensional Hilbert space, the same method can be also applied to the higher dimensional composite systems.
Here we have mainly focused on the pure systems. In the mixed case, it is not easy to apply directly as
$`M_{i_1i_2\mathrm{}i_n}(1,2,\mathrm{},n;\rho )`$ $`=`$ $`tr[\rho (\sigma _{i_1}(1)\lambda _i(1))(\sigma _{i_2}(2)\lambda _{i_2}(2))\mathrm{}`$ (8)
$`(\sigma _{i_n}(n)\lambda _{i_n}(n))],`$
where $`\rho `$ is the density operator. In the Werner state of two-qubit parameterized by fidelity with the singlet states, we get $`M^{}0`$ for the region $`F\frac{1}{2}`$. This contradicts the fact that the Werner state is separable in the region $`F\frac{1}{2}`$. It is why the density operator has many possible ensembles for the given density operator. We have to find the optimized ensemble to apply the criterion. If there exists a pure state ensemble to make $`M^{}`$ zero, the density operator is separable. Otherwise, the density operator is entangled. However, this story seems to be simple in principle but it is very complicated to find the proper ensemble practically.
We have here presented the classification and quantification scheme of a general multipartite systems. $`M^{}`$ which is defined by the expectation values of $`SU(N)`$ generators determines the entangled type of multipartite states without a priori knowledge on the quantum states investigated. However, the same entangled type has many different states which cannot distinguish with $`M^{}`$ any more. So we have introduced $`B^{(m)}`$ providing the quantification procedure. Finally, we have shown how to apply our method to the four-qubit system as an example.
Acknowledgements
Lee and Ahn were supported by the Korean Ministry of Science and Technology through the Creative Research Initiatives Program under Contact No. M10116000008-03F0000-03610. Oh was supported by KRF-2002-070-C00029.
|
warning/0506/physics0506220.html
|
ar5iv
|
text
|
# Non-destructive study of non-equilibrium states of cold, trapped atoms
(March 15, 2024)
## Abstract
Highly sensitive, non-destructive, real-time spectroscopic determination of the 2D kinetic momentum distribution of a cold-atom sample is performed with the three-beam measurement of the recoil-induced resonances. The measurements performed with an operating magneto-optical trap reveal slow velocity drifts within a stationary atomic cloud and strong anisotropy and asymmetry of the non-Maxwellian momentum distribution. The developed method can be easily extended to 3D.
Most experiments with cold, dilute atomic gases employ magneto-optical traps (MOT), which yield temperatures in a range of hundreds to a few $`\mu `$K. Further traps and cooling stages can be applied for reaching the quantum degeneracy regime. This requires matching of the momentum distributions of various traps. Knowledge of such distributions is also essential for quantum state diagnostics of the trapped sample. Below, we present reliable method of 2D momentum diagnostics based on the so called recoil-induced resonances (RIRs) and apply it to the detailed study of non-standard momentum distributions in a MOT.
The first unambiguous observation of RIRs was made in a 1D optical lattice gry94 filled with atoms much colder than in a standard MOT. RIR signals were also seen in optical molasses mea94 ; fisch01 , in a cold atomic beam dom01 and with atoms released from a MOT chen01 . The influence of the recoil effect on the probe absorption and four wave mixing spectra has been recently demonstrated in a continuously working MOT, i.e. with all light and magnetic fields on, in brz05 .
In this Letter we present evidence of three different kinds of anisotropy of the momentum distribution in an operating MOT. The measurements were conducted using our three-beam, RIR-based method developed for simultaneous probing of the momentum distribution in two perpendicular directions. The method extends the principle of 1D thermometry as suggested in Ref. mea94 . Important feature of our extension is that 2D information is acquired simultaneously in one measurement. The method can be extended to 3D.
The RIRs result from a stimulated Raman process, which couples two kinetic states of free moving atoms (Fig. 1). Two laser beams, the pump and the probe, of frequencies $`\omega `$ and $`\omega +\delta `$, respectively, drive the Raman transition after which the atoms gain kinetic energy $`\mathrm{\Delta }E_{\mathrm{kin}}=\mathrm{}\delta `$ and change momentum $`๐ฉ`$ by $`\mathrm{\Delta }๐ฉ=\mathrm{}\mathrm{\Delta }๐ค=\pm 2\mathrm{}k\widehat{๐}_i\mathrm{sin}\theta /2`$, where $`k`$ is the modulus of the light wave vector, $`\theta `$ is the angle between the beams, and $`\widehat{๐}_i`$ is the unit vector perpendicular to the bisector of $`\theta `$. The non-zero amplitude of the considered Raman resonance arises from different populations of the given kinetic states. When recorded in absorption, the RIR shape is proportional to a derivative of the momentum distribution $`\mathrm{\Pi }(p_i)/p_i`$, where $`p_i=๐ฉ\widehat{๐}_i`$ gry94 ; ver96 ; guo92 ; brz05 . This direct relation of the RIR signal to $`\mathrm{\Pi }(๐ฉ)`$ allows convenient and accurate measurement of the kinetic momentum distribution in a cold atomic sample, provided that the distribution is sufficiently narrow.
Important advantage of the RIR method is its directional selectivity. $`\mathrm{\Pi }(๐ฉ)`$ is probed in a given direction, specified by $`\mathrm{\Delta }๐ฉ`$, i.e. by angle $`\theta `$ (Fig.1a). Hence, apart from applications to standard 1D velocimetry mea94 ; dom01 , RIRs can also be used for studies of a possible momentum distribution anisotropy in non-equilibrium states of a cold atomic sample.
Our experiment (Fig. 2) employs a standard vapor-loaded MOT raab87 with <sup>85</sup>Rb atoms. Two extra beams intersect in the trap center: the pump and the probe. The probe beam is directed at a small angle $`\alpha =3^{}`$ to the MOT beams (propagating along $`z`$), and the pump is at $`\theta =5^{}`$ to the probe. The probe beam can be detected either directly or after retroreflection. The setup with retroreflected probe enables the measurement of $`\mathrm{\Pi }(๐ฉ)`$ simultaneously along two perpendicular directions: $``$, for angle $`\theta `$ between the pump and the nearly co-propagating probe, and $`||`$, for angle $`180^{}\theta `$ between the pump and the nearly counter-propagating probe. When $`\alpha `$ and $`\theta `$ are sufficiently small, $``$ and $`||`$ almost coincide with the $`y`$ and $`z`$ directions, respectively. Both pump and probe beams are derived from diode lasers synchronized by injection-locking and are blue-detuned from the trapping transition <sup>2</sup>S$`{}_{1/2}{}^{}(F=3)`$<sup>2</sup>P$`{}_{3/2}{}^{}(F^{}=4)`$ by $`\mathrm{\Delta }=2\pi `$140 MHz $`23.3\mathrm{\Gamma }`$, where $`\mathrm{\Gamma }`$ denotes the natural linewidth. Such big detuning reduces the perturbation of atoms to a very low level (scattering rate $`1/\mathrm{\Delta }^2`$) which is essential for non-destructive measurements. Moreover, non-resonant pump eliminates overlap of the RIRs and Raman-Zeeman resonances brz05 hence facilitates interpretation of the results. Despite large $`\mathrm{\Delta }`$, the pump and probe beams drive the Raman signal with a sufficiently large amplitude and signal-to-noise ratio for the pump beam intensities 5-35 mW/cm<sup>2</sup>.
The probe beam is scanned by $`\delta \pm 1`$ MHz around frequency $`\omega `$ of the pump. The probe and pump photons induce Raman transitions between atomic kinetic states separated by $`\pm \mathrm{}\delta `$. Since the polarization of the pump and probe beams is chosen to be the same, the atoms undergo Raman transitions with $`\mathrm{\Delta }m_F=0`$. Hence, the internal atomic state does not change and the only states that have to be considered are the external states associated with the kinetic energy of the atomic center-of-mass. The multi-level structure of <sup>85</sup>Rb can thus be reduced to a set of independent two-level systems which allows straightforward application of the basic RIR theory gry94 ; brz05 ; ver96 . With the assumption that $`\mathrm{\Pi }(๐ฉ)`$ is the Maxwell-Boltzmann distribution, the RIR signal $`s(\delta )`$ ver96 ; brz05 recorded with the retroreflected probe is given by two contributions. The narrow one results from the Raman process involving the pump and the probe beam making small angle $`\theta `$ and the wide one is for angle $`180^{}\theta `$. The signal is
$$s(\delta )A_{}\delta \mathrm{exp}\left(\frac{\delta ^2}{\xi _{}^2}\right)A_{||}\delta \mathrm{exp}\left(\frac{(\delta \delta _0)^2}{\xi _{||}^2}\right),$$
(1)
where, for small $`\theta `$, $`\xi _{}^22k_Bk^2m^1\theta ^2\tau _{}`$ and $`\xi _{||}^28k_Bk^2m^1\tau _{||}`$. $`\tau _{}`$ and $`\tau _{||}`$ are the distribution widths in the $``$ and $`||`$ directions in the temperature units, $`A_{||}`$ and $`A_{}`$ are the amplitudes of the corresponding contributions, $`m`$ is the atomic mass, $`k_B`$ is the Boltzmann constant, and $`\delta _0`$ is the possible frequency shift between the $``$ and $`||`$ contributions, to be discussed later.
Typical example of the retroreflected-probe transmission spectrum is shown in Fig. 3. It exhibits two distinct resonant contributions, predicted by eq. (1). The wide contribution is shifted with respect to the narrow one by 72.4 kHz, which indicates a 2.8-cm/s average velocity component in the $`||`$ direction. We thus observe an atomic drift within a cloud, which as a whole remains stationary. We understand this as a dynamic effect resulting from a small difference of the radiation pressures intrinsic to a MOT with retroreflected trapping beams. Indeed, the observed shift increases when the imbalance is purposely increased. Strong imbalance normally produces a displacement of the atomic-cloud center of mass. In our case this displacement is too weak to be detected by standard imaging technique, while the anisotropic atomic flow, even one order of magnitude slower than the mean thermal velocity, is well measurable with our method.
As the velocity distributions derived from the signal in Fig. 3 are Gaussian, one can determine the values $`\tau _{}=172\pm 6\mu `$K and $`\tau _{||}=170\pm 3\mu `$K. The equality of these $`\tau `$s implies thermodynamical equilibrium and allows their interpretation as temperature $`T`$, despite the slow drift. The equilibrium persists for various MOT-light intensities due to the fact that total intensities of each pair of the MOT beams remain the same. The observed nearly linear increase of $`T`$ with the total MOT-beam intensity agrees well with previous reports Lett89 ; Wallace94 ; Ye2002 .
The thermodynamics of the system becomes highly non-trivial when the trapping light is unevenly distributed between the MOT beam pairs. It was predicted that for such conditions the width of kinetic momentum distribution shows directional dependence gajda94 . Using the simultaneous measurement of the momentum width in two perpendicular directions, we attempted to observe such anisotropic non-equilibrium state of the cold-atom cloud. For this reason, we changed intensity balance between the longitudinal ($`I_z`$) and transverse ($`I_x`$, $`I_y`$) MOT beam pairs, while keeping the total intensity $`I_0=I_x+I_y+I_z`$ constant. We define parameter $`\kappa `$ as the relative intensity of $`I_z`$, $`I_z=\kappa I_0`$, $`I_x=I_y=(1\kappa )I_0/2`$. The results of the measurement of $`\tau _{||}`$ and $`\tau _{}`$ for different values of $`\kappa `$ are depicted in Fig 4. For equal partition of the trapping intensity ($`\kappa =1/3`$), the widths of kinetic momentum distributions are the same, as expected. However, when $`\kappa `$ increases, $`\tau _{||}`$ and $`\tau _{}`$ follow opposite trends, which is evidence of kinetic momentum anisotropy in a MOT working MOT and thereby its non-equilibrium state. A similar anisotropy was recently observed also in an optically dense sample vorozcovs . We notice that $`\tau _{}+2\tau _{}`$, which is the measure of $`v_{}^2+2v_{}^2`$, is constant within $`\pm 2\%`$ over the whole measured range of $`\kappa `$. The decrease of $`\tau _{||}`$ with the growing $`\kappa `$ is due to the fact that the heating associated with spontaneous emission is isotropic, whereas the cooling rate is higher for the direction with the increased intensity. The momentum anisotropy becomes manifest because the density of the atoms is too small to provide efficient thermalization. Indeed, simple estimation for typical conditions and Rb-Rb collision cross-section $`\sigma _{\mathrm{Rb}\mathrm{Rb}}=310^{13}`$ cm<sup>2</sup> rapol01 yields the atomic collision rate below 1 Hz in our trap, while the friction coefficient, in frequency units, is in the kHz range.
The theoretical behavior of $`\tau _{||}`$ and $`\tau _{}`$ according to Refs. gajda94 ; gajda05 is plotted in Fig. 4 along with the experimental data. They exhibit similar qualitative dependence (the decrease of $`\tau _{||}`$ and the increase of $`\tau _{}`$ with growing $`\kappa `$), but the existing theory fails to reproduce the exact shape of the experimental dependence. This discrepancy is due to additional mechanism of sub-Doppler cooling, not included in the calculations of Ref. gajda94 . Indeed, the increase of $`I_z`$ accompanied by attenuation of $`I_x`$ and $`I_y`$ results in efficient quasi-1D cooling scheme in the $`\sigma ^+\sigma ^{}`$ optical molasses dali89 . Evidence of this cooling is provided by the values of $`\tau _{||}`$ falling to 70 $`\mu `$K, well below the Doppler cooling limit of 140 $`\mu `$K. Importance of sub-Doppler cooling for anisotropy of momentum distribution in cold atomic samples has been previously noted in optical molasses Jav .
In the situation discussed above, the MOT beams were carefully aligned which resulted in high stability of the trapped-atom cloud, even for the largest departures from the equal partition of the trapping light intensity, and allowed fitting of the RIR signals by eq. (1). The sole manifestation of the non-equilibrium of the sample was the $`\mathrm{\Pi }(p_{})`$ vs. $`\mathrm{\Pi }(p_{||})`$ anisotropy.
The thermodynamical equilibrium can be altered yet in a different way, namely by enhancing imbalance between the counter-propagating MOT-beam radiation pressures. Fig. 5a depicts the RIR recorded with a standard 1D, two-beam arrangement applied to the case when the MOT beams were slightly misaligned and tuned closer to resonance. The 1D thermometry was accomplished by replacing the optional mirror in Fig. 2 by a photodiode. In this configuration, the pump and the probe make angle $`180^{}\theta `$ and the recorded signal is proportional only to $`\mathrm{\Pi }(p_{||})/p_{||}`$. Its shape deviates from a derivative of a Gaussian. By integrating the signal and scaling to the velocity units, the actual distribution of velocity component in the $`||`$ direction, $`v_{||}`$, can be retrieved. Fig. 5b shows such a distribution obtained from the experimental signal and the idealized Gaussian reference curve of the same area and of the width derived from the positions of the minimum and maximum of the RIR signal in Fig. 5a. Non-standard properties of a stable atomic gas, revealed in our experiment call for thorough theoretical analysis with proper accounting for cooling and heating mechanisms.
In conclusion, we have developed three-beam spectroscopic method of determining the momentum distributions of cold, trapped atoms, based on recoil-induced resonances. The method is non-destructive, highly sensitive and provides multi-dimensional momentum determination in a single measurement. Its potential has been demonstrated by study of three different momentum distributions of atoms in the operating magneto-optical trap: (i) thermodynamic equilibrium with well defined temperature and Gaussian momentum distribution with a slow velocity drift; (ii) non-equilibrium state characterized by Gaussian distributions with drastically different widths in the longitudinal and transverse directions; (iii) non-equilibrium state of non-Gaussian momentum distribution along one direction. The result (ii) qualitatively confirms theoretical predictions of Ref. gajda94 and indicate need for more refined MOT theory. Our method can be particularly useful for studies of anisotropy in optical molasses Jav , 2D MOTs Dieck98 , etc. The described method can be also straightforwardly applied to 3D case by introducing additional pump beam in the $`xz`$ plane in Fig. 2. To avoid overlapping of the RIR signals associated with all $`\mathrm{\Delta }๐ฉ`$ directions, the frequency of the additional pump could be shifted. Such an approach can be used for the non-destructive, on-line diagnostics of the atom dynamics in a trap carried out independently and simultaneously with other spectroscopic measurements. The method should also be applicable to quantum-degenerate gases. In fact, the widely used Bragg spectroscopy is based on the same principle of momentum and energy transfer. Measuring the Bragg-beam transmission in our three-beam geometry, rather than imaging BEC can thus become a valuable, non-destructive alternative.
This work was supported by the Polish Ministry of Science and Information Society Technologies and is part of a general program on cold-atom physics of the National Laboratory of AMO Physics in Toruล, Poland. Authors would like to thank Mariusz Gajda for his illuminating discussion and Krzysztof Sacha and Dmitry Budker for their valuable comments.
|
warning/0506/quant-ph0506010.html
|
ar5iv
|
text
|
# Coherent backscattering in nonlinear atomic media: quantum Langevin approach
## I Introduction
Over the past ten years, cold atomic gases have gradually become a widely employed and highly tunable tool for testing new ideas in many areas of quantum physics: quantum phase transitions (Bose-Einstein condensation, Fermi degenerate gases, Mott-Hubbard transition) bec ; fermi ; mott-hubbard , quantum chaos chaos , applications in metrology hsurm , disordered systems cbsat ; thierry to cite a few. In the latter case, cold atomic vapors act as dilute gases of randomly distributed atoms multiply scattering an incident monochromatic laser light. In this case, the scattered light field exhibit a speckle-like structure due to (multiple) interference between all possible scattering paths. The key point is that the disorder average is insufficient to erase all interference effects. This gives rise to weak or strong localization effects in light transport depending on the strength of disorder Houches ; AkkerMon . A hallmark of this coherent transport regime is the coherent backscattering (CBS) phenomenon: the average intensity multiply scattered off an optically thick sample is up to twice larger than the average background in a small angular range around the direction of backscattering, opposite to the incoming light cbs . This interference enhancement of the diffuse reflection off the sample is a manifestation of a two-wave interference. As such, it probes the coherence properties of the outgoing light photon . The CBS effect in cold atomic gases has been the subject of extensive studies in the weak localization regime, both from theoretical and experimental points of view cbsatoms . In particular, modifications brought by atoms, as compared to classical scatterers, for light transport properties (mean-free path, coherence length, CBS enhancement factor) have been highlighted. They are essentially due to the quantum internal atomic structure internal ; cbsB .
Another interesting feature of atoms is their ability to display a nonlinear behavior: the scattered light is no more proportional to the incident one. This leads to a wide variety of phenomena, like pattern formation, four-wave mixing, self-focusing effects, dynamical instabilities, *etc* boyd ; prl72GMP ; praDHGC ; prl85SM . For a weak nonlinearity, introducing an intensity-dependent susceptibility is enough to properly describe these effects, including quantum properties boyd ; pra70WGDM ; facteur3 , *e.g.* the Kerr effect (intensity dependence of the refractive index) can be obtained with a $`\chi ^{(3)}`$ nonlinearity. However, when the incident intensity is large enough, and this is easily achieved with atoms, perturbation theories eventually fail and a full nonlinear treatment is required. For a single two-level atom, the solution is usually given by the so-called optical Bloch (OB) equations. Together with the quantum regression theorem, they allow for a complete description of the spectral properties of the fluorescence light Cohenrouge . In particular, these equations show that the atomic nonlinear behavior is intrinsically linked to the quantum nature of the electromagnetic field. More specifically, as opposed to classical nonlinear scatterers, the radiated light exhibits quantum fluctuations characterized by peculiar time correlation properties. They define a power spectrum, known as the Mollow triplet, emphasizing inelastic scattering processes at work in the emission process pr188M ; Cohenrouge ; ZG .
However, even if all these aspects are well understood in the case of a *single* atom exposed to a strong monochromatic field Cohenrouge , the situation changes dramatically in the case of a large number of atoms where a detailed analysis including both quantum nonlinear properties and coherence effects is still lacking. Until now, the nonlinear coupling between the atoms and the quantum vacuum fluctuations is either included in a perturbative scheme facteur3 ; Wellens\_long or simply described by a classical noise pra46YMC ; pra46YMC2 ; pra51YC ; pra52DPGC ; pra56B . In the dilute regime $`\lambda R`$ where the light wavelength $`\lambda `$ is much less than the average particle separation $`R`$, one expects the quantum fluctuations to reduce the degree of coherence of the scattered light. This will alter not only propagation parameters (mean-free path, refraction index), but also weak localization corrections to transport, and the CBS enhancement factor, which is related to the coherence properties of the scattered light field thierry ; photon . We want here to stress that, even beyond interference and weak localization phenomena, any transport property which may be influenced by saturating the atomic transition deserves a special and necessary study on its own. The most striking systems falling in this category where both nonlinear and disordered descriptions are intimately interwoven are coherent random lasers cao where interference effects lead to localized light modes inside the disordered medium, comparable to resonator eigenmodes in standard lasers. Even if, in this case, one would require an active (i.e. amplifying) medium, a key point is the understanding of the mutual effects between multiple interferences and nonlinear scattering.
In the present paper, we will focus on the rather simple case of two atoms in vacuum. Our aim is threefold: (i) firstly to properly calculate quantum correlations between pairs of atoms as a crucial step towards a better understanding of the physical mechanisms at work; (ii) secondly to implement a method allowing for a simple incorporation of frequency-dependent propagation effects; (iii) finally to understand, in the CBS situation, the modifications brought by the (quantum) nonlinearity to the interference properties. We hope that these points, once mastered, can provide an efficient way to produce realistic computer models to simulate real experiments. Point (i) alone could easily be solved using the standard OB method pra45VA ; prl94SMB . But the latter almost becomes useless regarding point (ii), since frequency-dependent propagation leads to complicated time-correlation functions. From a numerical point of view, it also leads to such large linear systems of coupled equations that its practical use is limited up to only a few atoms, very far from a real experimental situation. For these reasons, we will rather use the quantum Langevin method for our purposes. This method not only solves points (i) and (ii), but also leads to a simple explanation of point (iii), through a direct evaluation of the quantum noise spectrum. Note however that, in the absence of any effective medium surrounding the two atoms, and as long as only the numerical results are concerned (but not the physical interpretation), the quantum Langevin approach is completely equivalent to solving the multi-atoms optical Bloch equations like in pra45VA ; prl94SMB .
This paper divides as follows: in section II, the notations are defined and the quantum Langevin approach is explained for the single atom case. In section III, the method is adapted to the case where two atoms are weakly coupled by the dipole interaction. The validity and relevance of the method is controlled by a comparison with a direct calculation using OB equations. Then, in the CBS configuration, numerical results for different values of the laser intensity and detuning are presented and discussed. In particular, possible reasons for the reduction of the enhancement factor are put forward.
## II Single two-level atom case
### II.1 Time-domain approach
We consider an atom with a zero angular momentum electronic ground state ($`J_g=0`$) exposed to a monochromatic light field. The light field frequency $`\omega _L`$ is near-resonant with an optical dipole transition connecting this ground state to an excited state with angular momentum $`J_e=1`$. The angular frequency separation between these two states is $`\omega _0`$ and the natural linewidth of the excited state is $`\mathrm{\Gamma }`$. We will denote hereafter by $`\delta =\omega _L\omega _0`$ the laser detuning. The ground state is denoted by $`|\mathrm{0\hspace{0.17em}0}`$ while the excited states are denoted by $`|1m_e`$, with $`m_e=1,0,1`$ the Zeeman magnetic quantum number. As we assume no magnetic field to be present throughout this paper, the excited state is triply degenerate.
In the Heisenberg picture, this two-level atom is entirely characterized by the following set of 16 time-dependent operators:
$$\mathrm{\Pi }^g=|\mathrm{0\hspace{0.17em}0}\mathrm{0\hspace{0.17em}0}|;\mathrm{\Pi }_{m_em_e^{}}^e=|1m_e1m_e^{}|;๐_{m_e}^+=|1m_e\mathrm{0\hspace{0.17em}0}|;๐_{m_e}^{}=|\mathrm{0\hspace{0.17em}0}1m_e|$$
(1)
The atomic operators obey the completeness constraint
$$๐=\mathbb{\Pi }^๐+\mathbb{\Pi }^๐$$
(2)
where $`\mathrm{\Pi }^g`$ and $`\mathrm{\Pi }^e=_{m_e}\mathrm{\Pi }_{m_em_e}^e`$ are the ground and excited state atomic population operators.
The full atom-field Hamiltonian $``$ is the sum of the free atom Hamiltonian $`_A=\mathrm{}\omega _0\mathrm{\Pi }^e`$, of the free quantized field Hamiltonian $`_F=_{\text{k},\mathit{ฯต}\text{k}}\mathrm{}\omega _\text{k}a_{\text{k}\mathit{ฯต}}^{}a_{\text{k}\mathit{ฯต}}`$ and of the dipolar interaction $`๐ฑ=๐(๐_L+๐_V)`$ between the atomic dipole $`๐`$, the classical laser field $`๐_L`$ and the quantum electromagnetic vacuum field $`๐_V`$. Performing the usual approximations of quantum optics, *i.e.* neglecting non-resonant terms (rotating wave approximation) and assuming Markov-type correlations between the atomic operators and the vacuum field, one obtains the quantum Langevin equations controlling the time evolution of any atomic observable $`๐ช`$ in the rotating frame pra46YMC ; Cohenrouge :
$$\frac{d๐ช}{dt}=i\delta _L[๐ช,\mathrm{\Pi }^e]\frac{i}{2}\underset{q}{}(1)^q[๐ช,๐_q^+]\mathrm{\Omega }_q^{L+}(\text{R})\frac{i}{2}\underset{q}{}[๐ช,๐_q^{}]\mathrm{\Omega }_q^L(\text{R})\frac{\mathrm{\Gamma }}{2}\left(๐ช\mathrm{\Pi }^e+\mathrm{\Pi }^e๐ช\right)+\mathrm{\Gamma }\underset{q}{}๐_q^+๐ช๐_q^{}+_๐ช(\text{R},t),$$
(3)
where $`\mathrm{\Omega }_q^{L+}`$ (resp. $`\mathrm{\Omega }_q^L`$) are the components of the Rabi frequency of the positive (resp. negative) frequency parts of the incident laser beam, i.e. $`\mathrm{}๐=d๐`$ where $`d`$ is the dipole strength. Finally $`_๐ช(t)`$ is the Langevin force depicting the effects of the quantum fluctuations of the vacuum electromagnetic field and reads as follows:
$$_๐ช(t)=\frac{i}{2}\underset{q}{}(1)^q[๐ช,๐_q^+]\mathrm{\Omega }_q^{0+}(\text{R},t)\frac{i}{2}\underset{q}{}\mathrm{\Omega }_q^0(\text{R},t)[๐ช,๐_q^{}],$$
(4)
where $`\mathrm{\Omega }^{0+}(\text{R},t)`$ is the vacuum Rabi field operator
$$๐^{0+}(\text{R},t)=\frac{2id}{\mathrm{}}\underset{\text{k},\mathit{ฯต}\text{k}}{}(\omega )\mathit{ฯต}a_{\text{k}\mathit{ฯต}}(t_0)e^{i\text{k}\text{R}i(\omega \omega _L)(tt_0)}$$
(5)
with $`t_0`$ an initial time far in the past. From the preceding expression, one can calculate the time correlation functions of the vacuum field Cohengris :
$$(1)^q[\mathrm{\Omega }_q^{0+}(\text{R},t),\mathrm{\Omega }_q^{}^0(\text{R},t^{})]=4\mathrm{\Gamma }\delta _{qq^{}}f(tt^{}),$$
(6)
where $`f(\tau )`$ in a function centered around $`\tau =0`$, whose width $`\tau _c`$ is much smaller than any characteristic atomic timescale (i.e. $`\tau _c\omega _0^1\mathrm{\Gamma }^1`$) and whose time integral is equal to unity. Thus, hereafter, $`f(\tau )`$ will be safely replaced by a $`\delta `$-function: $`f(\tau )\delta (\tau )`$.
The time evolution for the expectation values is obtained by averaging over the initial density matrix $`\sigma (t_0)`$, i.e., $`๐ช(t)=\mathrm{Tr}(๐ช(t)\sigma (t_0))`$. Since the atom and the vacuum field are supposed to be decoupled initially, $`\sigma (t_0)`$ is simply $`\sigma _{at}(t_0)|00|`$ ($`|0`$ being the vacuum field state). Because of the normal ordering, one immediately gets:
$$_๐ช(t)=0,$$
(7)
and the time correlation functions of the Langevin forces:
$$_๐ช(t)_๐ช^{}(t^{})=\mathrm{\Gamma }\underset{q}{}[๐ช(t),๐_q^+(t)][๐ช^{}(t^{}),๐_q^{}(t^{})]\delta (tt^{}).$$
(8)
The physical picture of the quantum Langevin approach is to represent quantum fluctuations by a fluctuating force acting on the system, in analogy with the usual Brownian motion. Not surprisingly, this leads to a diffusive-like behavior of expectation values. More precisely, because of the $`\delta `$-function in Eq. (8), we can set $`t^{}=t`$ for the atomic operators and we finally obtain in the stationary regime $`tt_0`$:
$$_๐ช(t)_๐ช^{}(t^{})=\frac{\mathrm{\Gamma }}{4}D_{๐ช๐ช^{}}\delta (tt^{}),$$
(9)
where $`D`$ is a matrix of diffusion constants depending only on the stationary values of the atomics operators. The stationary hypothesis also results from the fact that these correlation functions only depend on the time difference $`tt^{}`$.
From this, it is possible to prove that the quantum regression theorem applies CR92 ; Cohenrouge , allowing for the calculation of two-times correlation functions of the atomic operators and of their expectation values. From their Fourier transforms, one can obtain the spectrum of the radiated light. But, for the reasons mentioned in the introduction, we will explain how these properties can be obtained in a much simpler way by directly translating the Langevin equations in the Fourier domain CR92 .
### II.2 Frequency-domain approach
First, because of the constraint (2), only 15 atomic operators are actually independent. More specifically, we will use the following set, denoted by the column vector $`๐`$:
$$\text{X}\{\begin{array}{cc}\hfill \mathrm{\Pi }_{m_e}^z& =\frac{1}{2}\left[\mathrm{\Pi }_{m_em_e}^e\mathrm{\Pi }^g\right]\hfill \\ \hfill \mathrm{\Pi }_{m_em_e^{}}^e& =|1m_e1m_e^{}|m_em_e^{}\hfill \\ \hfill ๐_{m_e}^+& =|1m_e\mathrm{0\hspace{0.17em}0}|\hfill \\ \hfill ๐_{m_e}^{}& =|\mathrm{0\hspace{0.17em}0}1m_e|\hfill \end{array}.$$
(10)
The Langevin equations for $`๐`$ then formally read as follows:
$$\frac{d}{dt}๐(t)=M๐(t)+๐+๐
(t),$$
(11)
where $`M`$ is a time-independent matrix depending on the laser Rabi frequency $`\mathrm{\Omega }^{L\pm }`$, $`๐`$ is a constant vector scaling with $`\mathrm{\Gamma }`$ and $`๐
(t)`$ is a vector characterizing the Langevin forces at work on the atom (for simplicity, we have dropped the explicit position dependence). The stationary expectation values are then simply given by:
$$๐=M^1๐.$$
(12)
Using Kuboโs notations, the Fourier transforms of the different quantities are defined as follows:
$`f[\mathrm{\Delta }]`$ $`={\displaystyle ๐tf(t)e^{i\mathrm{\Delta }t}}`$ (13)
$`f(t)`$ $`={\displaystyle \frac{d\mathrm{\Delta }}{2\pi }f[\mathrm{\Delta }]e^{i\mathrm{\Delta }t}},`$
leading to the Langevin equations in the frequency domain:
$$\left(i\mathrm{\Delta }๐๐\right)๐[\mathrm{\Delta }]=2\pi \delta [\mathrm{\Delta }]๐+๐
[\mathrm{\Delta }].$$
(14)
Introducing the Greenโs function $`G[\mathrm{\Delta }]=\left(i\mathrm{\Delta }๐๐\right)^1`$, the solution of the preceding equations simply reads:
$$๐[\mathrm{\Delta }]=G[\mathrm{\Delta }]\left(2\pi \delta [\mathrm{\Delta }]๐+๐
[\mathrm{\Delta }]\right).$$
(15)
Using $`G[0]=M^1`$ and (12), this solution separates into a non-fluctuating part $`๐_L[\mathrm{\Delta }]`$ and a fluctuating (frequency-dependent) part $`๐_F[\mathrm{\Delta }]`$:
$$\{\begin{array}{cc}\hfill ๐_L[\mathrm{\Delta }]& =2\pi \delta [\mathrm{\Delta }]๐\hfill \\ \hfill ๐_F[\mathrm{\Delta }]& =G[\mathrm{\Delta }]๐
[\mathrm{\Delta }]\hfill \end{array}.$$
(16)
From the linearity of the Fourier transform, we still have $`๐
[\mathrm{\Delta }]=\mathrm{๐}`$ implying $`๐_F[\mathrm{\Delta }]=\mathrm{๐}`$. The time correlation functions for the Langevin force components, Eq. (8), become:
$$\text{F}_p[\mathrm{\Delta }^{}]\text{F}_q[\mathrm{\Delta }]=2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]D_{pq}.$$
(17)
where the $`2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]`$ function is a direct consequence of the time-translation invariance, *i.e.* that we calculate the correlation functions in the stationary regime. This implies that the correlation function for the components of $`๐_F`$ in the frequency domain are:
$$\left(๐_F[\mathrm{\Delta }^{}]\right)_p\left(๐_F[\mathrm{\Delta }]\right)_q=2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]\left(GD^tG\right)_{pq}$$
(18)
where the superscript $`t`$ means matrix transposition.
The field radiated at frequency $`\mathrm{\Delta }`$ by the atom at a distance $`r\lambda `$ (far-field regime) reads as follows:
$$\mathrm{\Omega }_q^+[\mathrm{\Delta }]=\frac{3}{2}\mathrm{\Gamma }๐ซ_{qq^{}}^\text{r}๐_q^{}^{}[\mathrm{\Delta }]\frac{e^{ikr}}{kr},$$
(19)
where we use implicit sum over repeated indices and where $`๐ซ^\text{r}`$ is the projector onto the plane perpendicular to vector r:
$$๐ซ_{qq^{}}^\text{r}=\overline{\mathit{ฯต}}_q๐ซ^\text{r}\mathit{ฯต}_q^{}=\overline{\mathit{ฯต}}_q\left(๐\frac{\text{r}{}_{}{}^{๐ฅ}\text{r}}{๐ฃ^\mathrm{๐}}\right)\mathit{ฯต}_q^{}=\delta _{qq^{}}(1)^q\frac{\text{r}_q\text{r}_q^{}}{r^2},$$
(20)
where the bar denotes complex conjugation and where $`(\text{r}{}_{}{}^{t}\text{r})`$ is a dyadic tensor.
The correlation functions $`\mathrm{\Omega }_q^{}^{}[\mathrm{\Delta }^{}]\mathrm{\Omega }_q^+[\mathrm{\Delta }]`$ of the light emitted by the atoms is then proportional to $`๐_q^{}^+[\mathrm{\Delta }^{}]๐_q^{}[\mathrm{\Delta }]`$ and read as follow:
$$\mathrm{\Omega }_q^{}^{}[\mathrm{\Delta }^{}]\mathrm{\Omega }_q^+[\mathrm{\Delta }](2\pi )^2\delta [\mathrm{\Delta }]\delta [\mathrm{\Delta }^{}]๐_q^{}^+๐_q^{}+2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]\underset{p^{}p}{}G_{i^{}p^{}}(\mathrm{\Delta }^{})G_{ip}(\mathrm{\Delta })D_{p^{}p},$$
(21)
where the index $`i`$ (resp. $`i^{}`$) corresponds to $`๐_q^{}`$ (resp. $`๐_q^{}^+`$). The non-fluctuating part gives rise to a spectral component of the emitted light at exactly the incident laser frequency and is thus naturally called the elastic part. The fluctuating part gives rise to the inelastic Mollow triplet spectrum pra5M , whose properties (position and width of the peaks) are given by the poles of $`G[\mathrm{\Delta }]`$, *i.e.* by the complex eigenvalues of $`M`$. Actually, we simply recover the results of the quantum regression theorem, which states that the atomic time correlation functions evolve with the same equations than the expectation values $`\dot{๐}=M๐+๐`$ Cohenrouge ; pr188M .
## III Two-atom case
### III.1 Optical Bloch equations
We now consider two isolated atoms, located at fixed positions $`\text{R}_1`$ and $`\text{R}_2`$. Defining $`\text{R}=\text{R}_2\text{R}_1=R\text{u}`$ (with $`R=|\text{R}|`$ and u the unit vector joining atom 1 to atom 2), we assume the far-field condition $`R\lambda `$ to hold. We also assume that $`R`$ is sufficiently small for the light propagation time $`R/c`$ to be much smaller than any typical atomic timescales $`(\mathrm{\Gamma }^1,\delta ^1,\mathrm{\Omega }_L^1`$). In this regime, all quantities involving the two atoms are to be computed at the same time $`t.`$ The contribution of the atom-atom dipole interaction in the Langevin equation for any atomic operator $`๐ช`$ reads:
$$\frac{d๐ช}{dt}|_{\text{dip.}}=i\frac{3\mathrm{\Gamma }}{4}\left\{\left([๐ช,๐_q^{1+}]๐ซ_{qq^{}}^\text{R}๐_q^{}^2+[๐ช,๐_q^{2+}]๐ซ_{qq^{}}^\text{R}๐_q^{}^1\right)\frac{e^{ikR}}{kR}+\left(๐_q^{1+}๐ซ_{qq^{}}^\text{R}[๐ช,๐_q^{}^2]+๐_q^{2+}๐ซ_{qq^{}}^\text{R}[๐ช,๐_q^{}^1]\right)\frac{e^{ikR}}{kR}\right\}.$$
(22)
In the OB equations, the two-atom system is entirely described by the set of 256 operators $`X_{ij}`$ made of all possible products $`X_i^1X_j^2`$. The stationary expectation values $`X_{ij}`$ are then obtained as solutions of a linear system resembling equation (12). This is the approach used in prl94SMB , where such optical Bloch equations are solved.
Since the two atoms are far enough from each other, the electromagnetic field radiated by one atom onto the other can be treated as a perturbation with respect to the incident laser field. More precisely, the solutions $`X_{ij}`$ can be expanded up to second order in powers of $`g`$ and $`\overline{g}`$:
$$X_{ij}=X_{ij}^{(0)}+gX_{ij}^{(g)}+\overline{g}X_{ij}^{(\overline{g})}+g\overline{g}X_{ij}^{(g\overline{g})}+g^2X_{ij}^{(gg)}+\overline{g}^2X_{ij}^{(\overline{g}\overline{g})}$$
(23)
where the complex coupling constant $`g`$ is:
$$g=i\frac{3\mathrm{\Gamma }}{2}\frac{\mathrm{exp}(ikR)}{kR}$$
(24)
In fact, it will be shown below that both terms in $`g^2`$ and $`\overline{g}^2`$ give a vanishing contribution to the coherent backscattering signal.
As explained in the introduction, this approach has two drawbacks: (i) the solutions obtained in this way are global and, thus, do not provide a simple understanding of the properties of the emitted light; (ii) when the two atoms are embedded in a medium whose susceptibility strongly depends on the frequency, the field radiated by one atom onto the other at a given time $`t`$ now depends on the atomic operators of the first atom at earlier times (since retardation effects become frequency dependent). Time correlation functions in the dipole interaction then explicitly show up.
### III.2 Langevin approach
The Langevin equations for the two sets of atomic operators $`๐^\alpha `$, with $`\alpha =1,2`$, read formally:
$$\dot{๐}^\alpha =M^\alpha ๐^\alpha +๐+๐
^\alpha +gT^{q+}๐^\alpha ๐ซ_{qq^{}}^\text{R}๐_q^{}^\beta +\overline{g}๐_q^{\beta +}๐ซ_{qq^{}}^\text{R}T^q^{}๐^\alpha ,$$
(25)
where $`\beta `$ denotes the other atom and where $`T^{q\pm }`$ are $`15\times 15`$ matrices defined by $`[X_i,๐_q^\pm ]=\pm 2T_{ij}^{q\pm }X_j`$. Taking the Fourier transform of these equations, one gets:
$$๐^\alpha [\mathrm{\Delta }]=G^\alpha [\mathrm{\Delta }]\left(2\pi \delta [\mathrm{\Delta }]๐+๐
^\alpha [\mathrm{\Delta }]\right)+gG^\alpha [\mathrm{\Delta }]T^{q+}๐ซ_{qq^{}}^\text{R}\left(๐^\alpha ๐_q^{}^\beta \right)[\mathrm{\Delta }]\overline{g}G^\alpha [\mathrm{\Delta }]๐ซ_{qq^{}}^\text{R}T^q^{}\left(๐_q^{\beta +}๐^\alpha \right)[\mathrm{\Delta }],$$
(26)
where $``$ is the convolution operator:
$$\left(AB\right)[\mathrm{\Delta }]=\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta [\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }]A[\mathrm{\Delta }_1]B[\mathrm{\Delta }_2].$$
(27)
Introducing, for simplicity, the following notations:
$$\{\begin{array}{cc}\hfill ๐^{\alpha ^{(0)}}[\mathrm{\Delta }]& =G^\alpha [\mathrm{\Delta }]\left(2\pi \delta [\mathrm{\Delta }]๐+๐
^\alpha [\mathrm{\Delta }]\right)\hfill \\ \hfill G^{\alpha _q^+}[\mathrm{\Delta }]& =G^\alpha [\mathrm{\Delta }]T^{q^{}+}๐ซ_{q^{}q}^\text{R}\hfill \\ \hfill G^{\alpha _q^{}}[\mathrm{\Delta }]& =G^\alpha [\mathrm{\Delta }]T^q^{}๐ซ_{qq^{}}^\text{R}\hfill \end{array},$$
(28)
equation (26) becomes:
$$๐^\alpha [\mathrm{\Delta }]=๐^{\alpha ^{(0)}}[\mathrm{\Delta }]+gG^{\alpha _q^+}[\mathrm{\Delta }]\left(๐^\alpha ๐_q^\beta \right)[\mathrm{\Delta }]\overline{g}G^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +}๐^\alpha \right)[\mathrm{\Delta }],$$
(29)
from which one gets the expansion in power of $`g`$ and $`\overline{g}`$ (up to $`g\overline{g}`$) for the atomic operators:
$`X_i^\alpha [\mathrm{\Delta }]`$ $`=X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]+gG_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]\overline{g}G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]`$ (30)
$`g\overline{g}\{G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}G_{๐_q^{}j^{}}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](G_{jj^{}}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})๐_q^{\beta ^{(0)}})[\mathrm{\Delta }].`$
$`.+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](๐_q^{\beta +^{(0)}}G_{jj^{}}^{\alpha _p^+}(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](G_{๐_q^{}j^{}}^{\beta _p^+}(X_j^{}^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}})X_j^{\alpha ^{(0)}})[\mathrm{\Delta }]\}.`$
Two-body terms expansions, obtained from Eq. (30), read as follows:
$`X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$ $`=X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]`$ (31)
$`+g\left\{X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\beta ^{(0)}}๐_q^{\alpha ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\right\}`$
$`\overline{g}\left\{X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]+G_{i^{}j^{}}^{\beta _q^{}}[\mathrm{\Delta }^{}]\left(๐_q^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\right\}`$
$`g\overline{g}\left\{\text{see appendix }\text{A}\right\}`$
$`X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$ $`=X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]`$
$`+g\left\{X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]+G_{i^{}j^{}}^{\alpha _q^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\right\}`$
$`\overline{g}\left\{X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]+G_{i^{}j^{}}^{\alpha _q^{}}[\mathrm{\Delta }^{}]\left(๐_q^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\right\}`$
$`g\overline{g}\left\{\text{see appendix }\text{A}\right\}.`$
Obviously, the power expansion of the expectation values can be derived from the quantum average of the preceding equations, but not as easily as it seems. Indeed, if one formally writes:
$$\{\begin{array}{cc}\hfill X_i^{}^\alpha ^{}[\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]& =\underset{ab}{}O(a,b)g^a\overline{g}^b\hfill \\ \hfill X_i^{}^\alpha ^{}[\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]& =\underset{ab}{}C(a,b)g^a\overline{g}^b\hfill \end{array},$$
(32)
then $`C(a,b)`$ is not simply equal to $`O(a,b)`$. Actually, $`C(a,b)`$ depends on all $`O(a^{},b^{})`$ for $`(a^{},b^{})(a,b)`$, and this for two reasons:
* for a given atom $`\alpha `$, the frequency correlation functions $`F_p^\alpha [\mathrm{\Delta }^{}]F_q^\alpha [\mathrm{\Delta }]`$ are given by $`2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]D_{pq}`$, where $`D_{pq}`$ depends on the stationary values. But the latter are modified by the second atom and, thus, must also be expanded in power of $`g`$ and $`\overline{g}`$. This implies, for example, that the first term $`X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]`$ in the expansion of $`X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$ (Eq. (31)) will contribute to all coefficients of $`X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$.
* the Langevin forces acting on two different atoms are correlated since they both originate from the vacuum quantum field. More precisely, their frequency correlation functions depend on their relative distance. This dependence is analogous to the correlation function of a speckle pattern (resulting from the random superposition of plane waves with the same wavelength but arbitrary directions):
$`F_i^{}^\beta [\mathrm{\Delta }^{}]F_i^\alpha [\mathrm{\Delta }]`$ $`=2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]{\displaystyle \frac{3}{2}}\mathrm{\Gamma }{\displaystyle \frac{\mathrm{sin}kR}{kR}}T_{i^{}j^{}}^{q^{}+}๐ซ_{q^{}q}^\text{R}T_{ij}^qX_j^{}^\beta X_j^\alpha `$ (33)
$`={\displaystyle \frac{1}{2}}\left(g+\overline{g}\right)2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]T_{i^{}j^{}}^{q^{}+}๐ซ_{q^{}q}^\text{R}T_{ij}^qX_j^{}^\beta X_j^\alpha `$
$`={\displaystyle \frac{1}{2}}\left(g+\overline{g}\right)2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]D_{i^{}i}^{\beta \alpha }.`$
Thus, terms like $`X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]`$ appearing in equation (31) will also contribute to higher-order coefficients in the power expansion of $`X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$. One must note that, when $`R0`$, $`๐ซ_{q^{}q}^\text{R}\frac{2}{3}\delta _{q^{}q}`$ and one recovers the single atom correlation functions given by Eq. (17), which emphasizes the consistency of the present approach.
Despite these subtleties, it is nevertheless possible to calculate power expansions of the atomic correlation functions. More precisely, in order to emphasize the validity of the present approach, we will compare the results obtain from the OB equations and from the Langevin approach. Indeed from the atomic correlation functions, the stationary solutions can be calculated by inverse Fourier transform as follows:
$$X_i^{}^\alpha X_i^\alpha ^{}=\frac{1}{(2\pi )^2}๐\mathrm{\Delta }^{}๐\mathrm{\Delta }X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha ^{}[\mathrm{\Delta }].$$
(34)
As a specific example, the coefficient proportional to $`g`$ in the perturbative expansion of $`X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]`$ is given by:
$`X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]^{(g)}`$ $`=\underset{ยฏ}{X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(g)}}+X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]^{(0)}`$ (35)
$`+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\beta ^{(0)}}๐_q^{\alpha ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(0)}`$
$`=\underset{ยฏ}{G_{i^{}j^{}}^\beta [\mathrm{\Delta }^{}]G_{ij}^\alpha [\mathrm{\Delta }]F_j^{}^\beta [\mathrm{\Delta }^{}]F_j^\alpha [\mathrm{\Delta }]^{(g)}}+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]X_j^{\alpha ^{(0)}}X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]๐_q^{\beta ^{(0)}}[\mathrm{\Delta }]^{(0)}`$
$`+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}]X_j^{}^{\beta ^{(0)}}๐_q^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(0)},`$
where we have used the fact that terms like $`X^{\alpha ^{(0)}}X^{\beta ^{(0)}}^{(0)}`$ (*i.e.* zeroth order) actually factorize into $`X^\alpha X^\beta `$ since their fluctuating parts necessarily give rise to higher orders in $`g`$ and $`\overline{g}`$, see Eq. (33). The underlined terms correspond to the non-vanishing correlations of the quantum vacuum fluctuations evaluated at the two atom positions.
Finally, separating elastic and inelastic part, one gets:
$`X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]^{(g)}`$ $`=(2\pi )^2\delta [\mathrm{\Delta }^{}]\delta [\mathrm{\Delta }]\left(G_{ij}^{\alpha _q^+}[0]X_j^{\alpha ^{(0)}}X_i^{}^{\beta ^{(0)}}๐_q^{\beta ^{(0)}}+G_{i^{}j^{}}^{\beta _q^+}[0]X_j^{}^{\beta ^{(0)}}๐_q^{\alpha ^{(0)}}X_i^{\alpha ^{(0)}}\right)`$ (36)
$`+2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]({\displaystyle \frac{1}{2}}\underset{ยฏ}{G_{i^{}j^{}}^\beta [\mathrm{\Delta }^{}]G_{ij}^\alpha [\mathrm{\Delta }]D_{j^{}j}^{\beta \alpha ^{(0)}}}+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]G_{i^{}j^{}}^\beta [\mathrm{\Delta }^{}]G_{๐_q^{}k^{}}^\beta [\mathrm{\Delta }]D_{j^{}k^{}}^{\beta \beta ^{(0)}}X_j^{\alpha ^{(0)}}.`$
$`+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}]G_{๐_q^{}k}^\alpha [\mathrm{\Delta }^{}]G_{ij}^\alpha [\mathrm{\Delta }]D_{kj}^{\alpha \alpha ^{(0)}}X_j^{}^{\beta ^{(0)}}).`$
The corresponding stationary solution then reads:
$`X_i^{}^\beta X_i^\alpha ^{(g)}`$ $`=G_{ij}^{\alpha _q^+}[0]X_j^{\alpha ^{(0)}}X_i^{}^{\beta ^{(0)}}๐_q^{\beta ^{(0)}}+G_{i^{}j^{}}^{\beta _q^+}[0]X_j^{}^{\beta ^{(0)}}๐_q^{\alpha ^{(0)}}X_i^{\alpha ^{(0)}}`$ (37)
$`+{\displaystyle \frac{1}{2\pi }}{\displaystyle }d\mathrm{\Delta }({\displaystyle \frac{1}{2}}\underset{ยฏ}{G_{i^{}j^{}}^\beta [\mathrm{\Delta }]G_{ij}^\alpha [\mathrm{\Delta }]D_{j^{}j}^{\beta \alpha ^{(0)}}}+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]G_{i^{}j^{}}^\beta [\mathrm{\Delta }]G_{๐_q^{}k^{}}^\beta [\mathrm{\Delta }]D_{j^{}k^{}}^{\beta \beta ^{(0)}}X_j^{\alpha ^{(0)}}.`$
$`+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }]G_{๐_q^{}k}^\alpha [\mathrm{\Delta }]G_{ij}^\alpha [\mathrm{\Delta }]D_{kj}^{\alpha \alpha ^{(0)}}X_j^{}^{\beta ^{(0)}}).`$
All quantities above only depend on the stationary values without coupling between the atoms and thus can be calculated from the single atom solutions. Furthermore, the integration over $`\mathrm{\Delta }`$ can be performed either numerically or analytically by the theorem of residues once the poles of $`G`$ (*i.e.* the complex eigenvalues of $`M`$) are known. Because of causality, they all lie in the lower-half of the complex plane. In practice, we have checked that we effectively recover, from the preceding expressions, the results obtained from the full OB equations. In particular, the contribution of the correlations of the quantum vacuum fluctuations evaluated at the two atom positions (the underlined term) is essential to get the correct results.
The same kind of expressions can be derived for $`g\overline{g}`$ terms, but they are slightly more complicated, since they explicitly involve three-body correlation functions, more precisely terms like:
$$\{\begin{array}{cc}& G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]^{(\overline{g})}\hfill \\ & G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]X_i^{}^\beta [\mathrm{\Delta }^{}]\left(G_{jj^{}}^{\alpha _p^{}}\left(๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}}\right)๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]^{(0)}\hfill \end{array},$$
(38)
which require the calculation of three-points Langevin force correlation functions like:
$$\{\begin{array}{cc}& G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]G_{i^{}j^{}}^\beta [\mathrm{\Delta }^{}]\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta [\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }]G_{jk}^\alpha [\mathrm{\Delta }_1]G_{๐_q^{}k^{}}^\beta [\mathrm{\Delta }_2]F_j^{}^\beta [\mathrm{\Delta }^{}]F_k^\alpha [\mathrm{\Delta }_1]F_k^{}^\beta [\mathrm{\Delta }_2]^{(\overline{g})}\hfill \\ & G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]G_{i^{}k^{}}^\beta [\mathrm{\Delta }^{}]\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta [\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }]G_{jj^{}}^{\alpha _p^{}}[\mathrm{\Delta }_1]G_{๐_p^+k}^\beta [\mathrm{\Delta }_1]G_{๐_p^+k^{\prime \prime }}^\beta [\mathrm{\Delta }_2]F_k^{}^\beta [\mathrm{\Delta }^{}]F_k^\beta [\mathrm{\Delta }_1]F_{k^{\prime \prime }}^\beta [\mathrm{\Delta }_2]^{(0)}\hfill \end{array}.$$
(39)
These correlations functions are non-zero even if they involve an odd number of Langevin forces, emphasizing that the statistical properties of the vacuum field fluctuations are far from Gaussian. Nevertheless, the explicit expressions of the above quantities can be derived (see appendix B). They lead to rather complicated and tedious formulae for the atomic correlation functions at order $`g\overline{g}`$. From that, we get the corresponding stationary expectations values. Again, we have checked that we indeed recover the OB results.
### III.3 Incorporation of an effective medium
Finally, and in sharp contrast to optical Bloch equations, it is very easy to adapt all the preceding results to the case of propagation in a medium with a frequency-dependent complex susceptibility. Indeed, propagation is controlled by the complex amplitude $`g`$ so that the field radiated by an atom at a distance $`R`$ and at frequency $`\mathrm{\Delta }`$ will be given by:
$$\mathrm{\Omega }_q^+[\mathrm{\Delta }]=ig๐ซ_{qq^{}}^\text{R}๐_q^{}^{}[\mathrm{\Delta }]\mathrm{exp}\left(\frac{1}{2}\frac{R}{\mathrm{}^+[\mathrm{\Delta }]}\right),$$
(40)
where $`\mathrm{}^+[\mathrm{\Delta }]`$ is the (complex) scattering mean-free path satisfying the dilute regime condition $`k|\mathrm{}^+[\mathrm{\Delta }]|1`$. The real part of $`1/\mathrm{}^+[\mathrm{\Delta }]`$ describes the exponential attenuation of the field during its propagation in the medium while the imaginary part describes the additional dephasing induced by the medium. More complicated formulas, accounting for possible variations of $`\mathrm{}`$ with position, birefringence effects, or even nonlinearities in propagation, can be derived in the same spirit. In all preceding equations, leading to the calculation of the correlation functions, any occurrence of the dipole operators must then simply be replaced by:
$$๐^{}๐^{}\mathrm{exp}\left(\frac{R}{2\mathrm{}^\pm [\mathrm{\Delta }]}\right)$$
(41)
while keeping the same โmedium-freeโ coupling constant $`g`$. In this way, the present approach can be easily extended to the situation where the two atoms are embedded in a medium. In the case of a nonlinear medium, this could lead to a self-consistent set of nonlinear equations.
It is important to stress that accounting for the effective medium is rather straightforward in this frequency-domain approach but is a much more difficult task in the temporal-domain approach. Indeed, one basic hypothesis for deducing OB equations from the Langevin approach โ see section III.1 โ is that the light propagation time between the two atoms is much shorter than any typical atomic timescale. When this condition is fulfilled, it is possible to evaluate expectation values at equal times for both atoms, producing the set of closed OB equations. In the presence of a surrounding medium, propagation between the two atoms is affected and this basic assumption may fail. If the refraction index of the dilute medium is smoothly varying with frequency, then the corresponding propagation term is also smoothly varying with frequency and can be factored out. Thus, except for the exponential attenuation, one may recover the OB equations where equal times must be used for atoms 1 and 2. On the contrary, if the propagation term has a complicated frequency dependence, the problem cannot be simply reduced to OB equations. It will rather involve operators evaluated at the other atom, but *at different times*, thus leading to a much more complicated structure. This difficulty may even take place in a dilute medium with refraction index close to unity. Indeed, the important parameter is the time delay induced by the medium, itself related to the *derivative* of the index of refraction with respect to frequency. If the medium is composed of atoms having sharp resonances, the effective group velocity can be reduced by several orders of magnitude, consequently increasing by the same amount the propagation time between the two atoms. Around the atomic resonance line, the typical propagation time delay induced by the medium over one mean free path depends on the laser detuning but is of the order of the atomic timescale for the internal dynamics, namely $`\mathrm{\Gamma }^1`$ Labeyrie:radiation\_trapping . In this case, only the full Langevin treatment developed in this paper can properly account for the effect of the average atomic medium. Its practical implementation calls for an investigation on its own and is thus postponed to a future paper. We must also note that, if the surrounding medium is composed of the same atoms than the scatterers, it is not completely clear that propagation in the medium can be described โclassicallyโ, *i.e.* that the correlation between the Langevin forces acting on the scatterers and the Langevin forces acting on the medium can be safely neglected.
For the rest of this paper, we will consider two isolated atoms in vacuum.
## IV Main results
### IV.1 Scattered field correlation functions in the CBS configuration
In the case of a large number of atoms and for a given configuration, the interference between all possible multiple scattering paths gives rise to a speckle pattern. When averaging the intensity scattered off the sample over all possible positions of the atoms, one recovers the CBS phenomenon: the intensity radiated in the direction opposite to the incident beam is up to twice larger than the background intensity and gradually decreases to the background value over an angular range $`\mathrm{\Delta }\theta `$ scaling essentially as $`(k\mathrm{})^1`$, with $`\mathrm{}`$ the scattering mean-free path. In the present case, the averaging procedure is performed numerically by integrating over the relative positions of the two atoms. As will be seen below, the far-field condition $`kR1`$ allows for an *a priori* selection of the dominant terms contributing to the CBS signal.
The field radiated by the two atoms in the direction n at a distance $`rR\lambda `$, in the polarization channel $`\mathit{ฯต}^{\mathrm{out}}`$ orthogonal to n ($`\mathit{ฯต}^{\mathrm{out}}\text{n}=0`$), is given by:
$$\mathrm{\Omega }_{\mathrm{out}}^+[\text{n},\mathrm{\Delta }]=\frac{3}{2}\mathrm{\Gamma }ฯต_q^{\mathrm{out}}\left(๐_q^1[\mathrm{\Delta }]e^{ik\text{n}\text{R}_1}+๐_q^2[\mathrm{\Delta }]e^{ik\text{n}\text{R}_2}\right)\frac{e^{ikr}}{kr},$$
(42)
so that the field correlation function in this channel reads:
$$\begin{array}{c}\mathrm{\Omega }_{\mathrm{out}}^{}[\text{n},\mathrm{\Delta }^{}]\mathrm{\Omega }_{\mathrm{out}}^+[\text{n},\mathrm{\Delta }]=\left(\frac{3\mathrm{\Gamma }}{2kr}\right)^2ฯต_q^{\mathrm{out}}ฯต_p^{\mathrm{out}}\{๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]+๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }].\hfill \\ \hfill .+e^{ik\text{n}\text{R}}๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]+e^{ik\text{n}\text{R}}๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]\}.\end{array}$$
(43)
The CBS effect occurs when the total phase in the interference terms in the preceding expression becomes independent of the positions of the atom. This phase accumulates during the propagation of the incident laser beam to the atoms and during the propagation of the radiated field between the two atoms. The phase factor due to the incoming laser beam (a plane wave with wave number $`\text{k}_L=k\text{n}_L`$) can be explicitly factorized out of the atomic operators as follows:
$$\stackrel{~}{๐}_q^{\alpha \pm }=๐_q^{\alpha \pm }e^{\pm i\text{k}_L\text{R}_\alpha }.$$
(44)
The other components of $`\stackrel{~}{X}`$, cf. Eq. (10), are populations and not affected by this phase factor. In the single atom case, the expectation values of the hereby defined operators $`\stackrel{~}{๐}_q^{\alpha \pm }`$ are independent of the positions of the atoms. Defining $`\varphi =\text{k}_L\text{R}`$ and
$$g_1=ge^{i\varphi }g_2=ge^{i\varphi },$$
(45)
the Langevin equations (29) then become:
$$\stackrel{~}{๐}^\alpha [\mathrm{\Delta }]=\stackrel{~}{๐}^{\alpha ^{(0)}}[\mathrm{\Delta }]+g_\alpha \stackrel{~}{G}^{\alpha _q^+}[\mathrm{\Delta }]\left(\stackrel{~}{๐}^\alpha \stackrel{~}{๐}_q^\beta \right)[\mathrm{\Delta }]+\overline{g}_\alpha \stackrel{~}{G}^{\alpha _q^{}}[\mathrm{\Delta }]\left(\stackrel{~}{๐}_q^{\beta +}\stackrel{~}{๐}^\alpha \right)[\mathrm{\Delta }],$$
(46)
In the preceding equation, the Greenโs functions $`\stackrel{~}{G}`$ are now independent of the position of the atoms, so that the phase information due to the incident laser beam is entirely contained in the coefficients $`g_\alpha `$.
Frequency correlation functions of the Langevin forces, eq. (33), must also be modified accordingly:
$$\stackrel{~}{F}_i^{}^\beta [\mathrm{\Delta }^{}]\stackrel{~}{F}_i^\alpha [\mathrm{\Delta }]=\frac{1}{2}\left(g_\beta +\overline{g}_\alpha \right)2\pi \delta [\mathrm{\Delta }^{}+\mathrm{\Delta }]\stackrel{~}{D}_{i^{}i}^{\beta \alpha }.$$
(47)
Dropping for simplicity, the tilde notation, the field correlation function (43), in the backward direction $`\text{n}=\text{n}_L`$, becomes:
$$\begin{array}{c}\mathrm{\Omega }_{\mathrm{out}}^{}[\text{n}_L,\mathrm{\Delta }^{}]\mathrm{\Omega }_{\mathrm{out}}^+[\text{n}_L,\mathrm{\Delta }]=\left(\frac{\mathrm{\Gamma }}{kr}\right)^2ฯต_q^{\mathrm{out}}ฯต_p^{\mathrm{out}}\{๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]+๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }].\hfill \\ \hfill .+e^{2i\varphi }๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]+e^{2i\varphi }๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]\}.\end{array}$$
(48)
The configuration average is then performed in two steps. Since we are working in the limit $`kR1`$, the first one is to keep only terms with a total phase independent of $`kR`$. In the power expansion with respect to the four parameters $`g_1`$, $`g_2`$, $`\overline{g}_1`$ and $`\overline{g}_2`$, this simply amounts to keep terms with even powers of $`g_\alpha \overline{g}_\alpha ^{}`$. This obviously cancels any $`\varphi `$ dependence. More precisely, the field correlation function in the backward direction, beside the trivial zeroth order (in $`g`$) term, is given by:
$`\mathrm{\Omega }_{\mathrm{out}}^{}[\text{n}_L,\mathrm{\Delta }^{}]\mathrm{\Omega }_{\mathrm{out}}^+[\text{n}_L,\mathrm{\Delta }]^{(2)}`$ $`=\left({\displaystyle \frac{\mathrm{\Gamma }}{kr}}\right)^2ฯต_q^{\mathrm{out}}ฯต_p^{\mathrm{out}}\{๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]^{(g_1\overline{g}_1)}+๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]^{(g_2\overline{g}_2)}.`$ (49)
$`.+๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]^{(g_1\overline{g}_2)}+๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]^{(g_2\overline{g}_1)}\}`$
$`=\left({\displaystyle \frac{\mathrm{\Gamma }}{kr}}\right)^2\left(L[\mathrm{\Delta }^{},\mathrm{\Delta }]+C[\mathrm{\Delta }^{},\mathrm{\Delta }]\right).`$
The preceding field correlation function still depends on the relative orientation of the atoms through the projector $`๐ซ^\text{R}`$, so that, in a second step, an additional average over R must be performed. In the preceding equation, the first two terms correspond to the usual โladderโ terms $`L[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ (they are actually independent of the direction of observation), whereas the two other terms correspond to the usual โmaximally crossedโ terms $`C[\mathrm{\Delta }^{},\mathrm{\Delta }]`$:
$`L[\mathrm{\Delta }^{},\mathrm{\Delta }]={\displaystyle \frac{9}{4}}ฯต_q^{\mathrm{out}}ฯต_p^{\mathrm{out}}\{๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]^{(g_1\overline{g}_1)}+๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]^{(g_2\overline{g}_2)}\}`$ (50)
$`C[\mathrm{\Delta }^{},\mathrm{\Delta }]={\displaystyle \frac{9}{4}}ฯต_q^{\mathrm{out}}ฯต_p^{\mathrm{out}}\{๐_p^{2+}[\mathrm{\Delta }^{}]๐_q^1[\mathrm{\Delta }]^{(g_1\overline{g}_2)}+๐_p^{1+}[\mathrm{\Delta }^{}]๐_q^2[\mathrm{\Delta }]^{(g_2\overline{g}_1)}\}`$
### IV.2 CBS enhancement factor
In the case of linear scatterers, the CBS enhancement factor achieves its maximal value 2 (recall that the CBS phenomenon is an incoherent sum of two-wave interference patterns all starting with a bright fringe at exact backscattering) if the single scattering contribution can be removed from the total signal and provided reciprocity holds. This is the case for scatterers with spherical symmetry in the so-called polarization preserving channel $`hh`$ BvTMaynard .
In this polarization channel, we have calculated the relevant quantities for an evaluation of the CBS enhancement factor *when no frequency filtering of the outgoing signal is made*. We have thus derived the elastic and inelastic ladder terms and the elastic and inelastic crossed terms, together with their corresponding frequency spectra, for different values of the on-resonance saturation parameter $`s_0=2|\mathrm{\Omega }_L|^2/\mathrm{\Gamma }^2`$. This parameter measures the intensity strength of the incident laser beam in units of the natural atomic transition line width $`\mathrm{\Gamma }`$, *i.e.* its compares the on-resonance transition rate induced by the laser to the atomic spontaneous emission rate. For a detuned laser beam, the saturation parameter is $`s(\delta )`$ and is defined as:
$$s(\delta )=\frac{s_0}{1+(2\delta /\mathrm{\Gamma })^2}$$
(51)
In the following, different values of the laser detuning have also been considered:
$$\begin{array}{cc}(a)\delta =0,s=s_0=0.02\hfill & (b)\delta =0,s=s_0=2.00\hfill \\ (c)\delta =5\mathrm{\Gamma },s_0=2.00,s=0.02\hfill & (d)\delta =0,s=s_0=50.0\hfill \end{array}.$$
The ladder and crossed terms (49) are separated into their elastic and inelastic parts according to:
$`L[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ $`=2\pi \delta (\mathrm{\Delta }+\mathrm{\Delta }^{})\left\{2\pi \delta (\mathrm{\Delta })L_{\mathrm{el}}+L_{\mathrm{inel}}(\mathrm{\Delta })\right\}`$ (52)
$`C[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ $`=2\pi \delta (\mathrm{\Delta }+\mathrm{\Delta }^{})\left\{2\pi \delta (\mathrm{\Delta })C_{\mathrm{el}}+C_{\mathrm{inel}}(\mathrm{\Delta })\right\}`$
The corresponding inelastic spectra $`L_{\mathrm{inel}}(\mathrm{\Delta })`$ and $`C_{\mathrm{inel}}(\mathrm{\Delta })`$ are displayed in figure 1. For a sufficiently low saturation parameter $`s_0`$, the inelastic contribution to the total intensity is small and the ladder intensity is almost equal to the crossed one (see graph 1$`a`$). For larger saturation parameters (see graphs 1$`b`$ and 1$`d`$), there are two effects : first, the inelastic contribution becomes comparable to the elastic one and second, the crossed term is smaller than the ladder one. For a nonzero detuning (see graph 1$`c`$), one clearly observes an asymmetry in the inelastic spectrum, which reflects that the scattering cross-section of the atomic transition is maximal for resonant light (indicated by the vertical dashed line): the symmetric inelastic spectrum emitted by a single atom is filtered out when scattered by the other one. We also observe that the crossed spectrum is much more reduced than the ladder term, highlighting the non-linear effects in the quantum correlations between the two atoms. Finally, for much larger saturation parameters (see graph 1$`d`$), the scattered light almost entirely originates from the inelastic spectrum, like for a single atom. However, contrary to the single atom case (for which the scattered intensity reaches a constant value), the total intensity scattered by the two atoms decreases when increasing the incoming intensity. Indeed, since the atomic transitions become fully saturated, the nonlinear scattering cross-section of each atom is decreasing, resulting in a smaller total intensity scattered by the two atoms compared to the one scattered by a single atom.
The CBS enhancement factor $`\eta `$ is defined as the peak to background ratio. It thus reads:
$$\eta =1+\frac{C^{\mathrm{tot}}}{L^{\mathrm{tot}}}$$
(53)
with:
$`L^{\mathrm{tot}}`$ $`=L_{\mathrm{el}}+L_{\mathrm{inel}}^{\mathrm{tot}}=L_{\mathrm{el}}+{\displaystyle \frac{d\mathrm{\Delta }}{2\pi }L_{\mathrm{inel}}(\mathrm{\Delta })}`$ (54)
$`C^{\mathrm{tot}}`$ $`=C_{\mathrm{el}}+C_{\mathrm{inel}}^{\mathrm{tot}}=C_{\mathrm{el}}+{\displaystyle \frac{d\mathrm{\Delta }}{2\pi }C_{\mathrm{inel}}(\mathrm{\Delta })}`$
If the CBS phenomenon is reducible to a two-wave interference, as it is the case here, then the enhancement factor $`\eta `$ is simply related to the degree of coherence $`\gamma `$ of the scattered light coherence . If the single scattering contribution can be removed from the detected signal, and this is the case in the $`hh`$ channel, one has simply $`\eta =1+\gamma `$ and consequently $`\gamma =C^{\mathrm{tot}}/L^{\mathrm{tot}}`$. The maximal value for $`\eta `$ is 2, meaning that full coherence $`\gamma =1`$ is maintained for the scattered field since then $`C^{\mathrm{tot}}=L^{\mathrm{tot}}`$. If all interference effects disappear, meaning $`C^{\mathrm{tot}}=0`$, $`\eta `$ reaches its minimal value 1 and correspondingly coherence is fully lost $`\gamma =0`$. Furthermore, one can show that in the $`hh`$ polarization channel, $`L_{\mathrm{el}}=C_{\mathrm{el}}`$ prl94SMB . Consequently, as soon as $`C_{\mathrm{inel}}^{\mathrm{tot}}<L_{\mathrm{inel}}^{\mathrm{tot}}`$ in this channel, the coherence of the scattered light field is partially destroyed, since then $`\eta <2`$ and $`\gamma <1`$.
Our results are summarized in table 1. At low saturation parameter $`s_0`$, $`\eta `$ reaches its maximal value 2 and $`\gamma =1`$. This is so because the ladder and crossed inelastic components are almost equal as evidenced in 1$`a`$. Increasing $`s_0`$ reduces further $`C_{\mathrm{inel}}^{\mathrm{tot}}`$ with respect to $`L_{\mathrm{inel}}^{\mathrm{tot}}`$, thus decreasing $`\eta `$ and $`\gamma `$. In the strongly saturated regime, one thus expects $`\gamma `$ to decrease. However, there is no reason for the ratio $`C_{\mathrm{inel}}^{\mathrm{tot}}/L_{\mathrm{inel}}^{\mathrm{tot}}`$ to tend to zero as $`s_0\mathrm{}.`$ It rather tends to a finite value, which depends on the detuning, in agreement with the results published in prl94SMB . Furthermore, keeping $`s_0`$ fixed and decreasing the saturation parameter $`s`$, situation $`(c)`$, $`\eta `$ increases, as expected, but to a value which strongly depends on $`s_0`$. In other words, contrary to the single atom case, the properties of the scattered light, are not only determined by the saturation parameter $`s`$ pra70WGDM . Indeed, in both situations $`(a)`$ and $`(c)`$, $`s`$ has the same (small) value, but the enhancement factor strongly differs, mainly because the inelastic ladder term has increased. This highlights the crucial role of the inelastic processes and of the rather complicated quantum correlations between the two atoms.
This is not however the full story. Depending on the $`s`$ and $`\delta `$ parameters, a rich variety of situations can be observed, with various physical interpretations. These are beyond the scope of this paper, which instead concentrates on the basic ingredients of the quantum Langevin approach and will be published elsewhere.
### IV.3 Linear response model
Some insight on the relative behavior of $`C_{\mathrm{inel}}(\mathrm{\Delta })`$ and $`L_{\mathrm{inel}}(\mathrm{\Delta })`$ can be found by comparing the respective formulae from which these quantities are extracted:
$$\begin{array}{c}X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]^{(\overline{g}_\beta g_\alpha )}=g_\alpha X_i^{}^{\beta ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]^{(\overline{g}_\beta )}\overline{g}_\beta G_{i^{}j^{}}^{\beta _q^{}}[\mathrm{\Delta }^{}]\left(๐_q^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(g_\alpha )}\hfill \\ \hfill g_\alpha \overline{g}_\beta \{X_i^{}^\beta [\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}G_{๐_q^{}j^{}}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}))[\mathrm{\Delta }]^{(0)}\\ \hfill +G_{i^{}j^{}}^{\beta _q^{}}[\mathrm{\Delta }^{}]\left(G_{๐_q^{}j}^{\alpha _p^+}\left(X_j^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}\right)X_j^{}^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(0)}\\ \hfill +.\left[G_{i^{}j^{}}^{\beta _p^{}}[\mathrm{\Delta }^{}](๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}})[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]\right]^{(0)}\}\end{array}$$
(55)
and
$$\begin{array}{c}X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]^{(\overline{g}_\alpha g_\alpha )}=X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(\overline{g}_\alpha g_\alpha )}\hfill \\ \hfill +g_\alpha \left\{X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]\left(X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }]^{(\overline{g}_\alpha )}+G_{i^{}j^{}}^{\alpha _q^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(\overline{g}_\alpha )}\right\}\\ \hfill \overline{g}_\alpha \left\{X_i^{}^{\alpha ^{(0)}}[\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]^{(g_\alpha )}+G_{i^{}j^{}}^{\alpha _q^{}}[\mathrm{\Delta }^{}]\left(๐_q^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(g_\alpha )}\right\}\\ \hfill \overline{g}_\alpha g_\alpha \{X_i^{}^\alpha [\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](G_{jj^{}}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]^{(0)}.\\ \hfill +X_i^{}^\alpha [\mathrm{\Delta }^{}]G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}G_{jj^{}}^{\alpha _p^+}\left(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}\right)\right)[\mathrm{\Delta }]^{(0)}\\ \hfill +G_{i^{}j^{}}^{\alpha _q^+}[\mathrm{\Delta }^{}]\left(G_{j^{}j}^{\alpha _p^{}}\left(๐_p^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)๐_q^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(0)}\\ \hfill +G_{i^{}j^{}}^{\alpha _q^{}}[\mathrm{\Delta }^{}]\left(๐_q^{\beta +^{(0)}}G_{j^{}j}^{\alpha _p^+}\left(X_j^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}\right)\right)[\mathrm{\Delta }^{}]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]^{(0)}\\ \hfill +\left[G_{i^{}j^{}}^{\alpha _p^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]\right]^{(0)}\\ \hfill +.\left[G_{i^{}j^{}}^{\alpha _p^{}}[\mathrm{\Delta }^{}](๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]\right]^{(0)}\}.\end{array}$$
(56)
There are twice as many terms contributing to the ladder terms as to the crossed terms. A rather simple explanation of this fact is borrowed from the usual linear response theory. Indeed, each atom is exposed to two fields : the incoming monochromatic field (angular frequency $`\omega _L`$, wave vector $`\text{k}_L`$) and the field scattered by the other atom (angular frequency $`\omega _L+\mathrm{\Delta }`$, wave vector $`\text{k}_p`$). In the far-field regime $`R\lambda `$, the incoming field is more intense than the scattered field. It thus plays the role of a pump beam with angular Rabi frequency $`\mathrm{\Omega }_L`$, while the second weaker field plays the role of a probe beam with angular Rabi frequency $`\mathrm{\Omega }_p`$. In this case, the response of each atom is simply described by its nonlinear susceptibility Cohenrouge ; boyd . More precisely, forgetting about polarization effects, we have:
$`\delta ๐^+[\mathrm{\Delta }]`$ $`=e^{i(2\text{k}_L\text{k}_p)\text{R}_\alpha }\chi _{++}[\mathrm{\Delta }]\mathrm{\Omega }_p^++e^{i\text{k}_p\text{R}_\alpha }\chi _+[\mathrm{\Delta }]\mathrm{\Omega }_p^{}`$ (57)
$`\delta ๐^{}[\mathrm{\Delta }]`$ $`=e^{i\text{k}_p\text{R}_\alpha }\chi _+[\mathrm{\Delta }]\mathrm{\Omega }_p^++e^{i(2\text{k}_L\text{k}_p)\text{R}_\alpha }\chi _{}[\mathrm{\Delta }]\mathrm{\Omega }_p^{}.`$
where the phases due to the light fields have been explicitly factorized.
As obviously seen, the two terms $`\chi _+`$ and $`\chi _+`$ generate the forward propagation of the probe whereas the two other terms $`\chi _{++}`$ and $`\chi _{}`$ can generate an additional field in the direction $`2\text{k}_L\text{k}_p`$ provided phase-matching conditions are fulfilled. This corresponds to the usual forward four-wave mixing mechanism (FFWM) boyd ; Cohenrouge . If we now replace the probe field by the field radiated by the other atom $`\beta `$, we get:
$`\delta ๐_{\beta \alpha }^+[\mathrm{\Delta }]`$ $`={\displaystyle \frac{1}{kR}}\left\{e^{i(kR+2\text{k}_L\text{R}_\alpha \text{k}_L\text{R}_\beta )}\chi _{++}[\mathrm{\Delta }]๐_\beta ^{}+e^{i(kR\text{k}_L\text{R}_\beta )}\chi _+[\mathrm{\Delta }]๐_\beta ^+\right\}`$ (58)
$`\delta ๐_{\beta \alpha }^{}[\mathrm{\Delta }]`$ $`={\displaystyle \frac{1}{kR}}\left\{e^{i(kR\text{k}_L\text{R}_\beta )}\chi _+[\mathrm{\Delta }]๐_\beta ^{}+e^{i(2\text{k}_L\text{R}_\alpha +kR\text{k}_L\text{R}_\beta )}\chi _{}[\mathrm{\Delta }]๐_\beta ^+\right\}.`$
Hence the ladder and crossed contributions are given by (dropping for sake of clarity any frequency dependence):
$`C^{(2)}`$ $`\delta ๐_{\alpha \beta }^+\delta ๐_{\beta \alpha }^{}e^{i(\text{k}_L\text{R}_\beta +\text{k}_L\text{R}_\alpha )}`$ (59)
$`e^{i(2\text{k}_L(\text{R}_\alpha \text{R}_\beta )2kR)}\chi _{++}\chi _+๐_\alpha ^{}๐_\beta ^{}+e^{4i\text{k}_L(\text{R}_\alpha \text{R}_\beta )}\chi _{++}\chi _{}๐_\alpha ^{}๐_\beta ^+`$
$`+\chi _+\chi _+๐_\alpha ^+๐_\beta ^{}+e^{i(2\text{k}_L(\text{R}_\alpha \text{R}_\beta )+2kR)}\chi _+\chi _{}๐_\alpha ^+๐_\beta ^+`$
$`L^{(2)}`$ $`\delta ๐_{\beta \alpha }^+\delta ๐_{\beta \alpha }^{}`$
$`e^{i(2\text{k}_L(\text{R}_\beta \text{R}_\alpha )2kR)}\chi _{++}\chi _+๐_\beta ^{}๐_\beta ^{}+\chi _{++}\chi _{}๐_\beta ^{}๐_\beta ^+`$
$`+\chi _+\chi _+๐_\beta ^+๐_\beta ^{}+e^{i(2\text{k}_L(\text{R}_\alpha \text{R}_\beta )+2kR)}\chi _+\chi _{}๐_\beta ^+๐_\beta ^+.`$
Averaging these expressions over the positions $`\text{R}_\alpha `$ and $`\text{R}_\beta `$ of the atoms while keeping $`R\lambda `$ fixed, only terms with position-independent phases survive, giving rise to:
$`C^{(2)}`$ $`\chi _+\chi _+๐_\alpha ^+๐_\beta ^{}`$ (60)
$`L^{(2)}`$ $`\chi _{++}\chi _{}๐_\beta ^{}๐_\beta ^++\chi _+\chi _+๐_\beta ^+๐_\beta ^{}.`$
This simple model allows to understand clearly why there are twice more terms in the ladder expression than in the crossed one. Fields generated in the FFWM process *always* interfere constructively in the case of the ladder, since they originate from the same atom. Of course, in the preceding explanation, we have discarded polarization effects and inelastic processes in the nonlinear susceptibilities. Nevertheless, even if in that case the situation becomes more involved, the differences between the ladder and crossed expressions still arise from this local four wave-mixing process. For example, in the last line of Eqs. (55) and (56), we see that the operator $`(G_{ij}^{\alpha _q^+}[\mathrm{\Delta }]X_j^{\alpha ^{(0)}})`$ plays the role of a generalized nonlinear susceptibility (actually, the standard ones are recovered from the elastic part of $`X_j^{\alpha ^{(0)}}`$). Thus we recover the same structure as previously depicted, which leads to similar conclusions.
Finally, as mentioned above, for large saturation parameters $`s_0`$, even if in that case the total scattered intensities (ladder and crossed) are dominated by the inelastic spectrum, we numerically observe that the enhancement factor does not vanish but rather goes to a finite limit $`1.096`$ (for $`\delta =0`$). Field coherence is thus not fully erased, which, at first glance, could be surprising since the inelastic spectrum is a noise spectrum at the heart of the temporal decoherence of the radiated field. But this only means that both crossed and ladder become vanishingly small relatively to the incident intensity. Nevertheless, even if it would be hard to derive it analytically from Eqs. (55) and (56), they actually decrease at the same rate, resulting in a finite (but small) enhancement factor.
## V Conclusion
In the case of two atoms, even if the quantum Langevin approach leads to calculations more tedious and involved than the direct optical Bloch method, it nevertheless gives rise to an understanding closer to the usual scattering approach developed in the linear regime. In this way, one also gets direct information about the inelastic spectrum of the radiated light. In particular, it clearly outlines the crucial roles played by the inelastic nonlinear susceptibilities and by the quantum correlations of the vacuum fluctuations. Furthermore, since the framework of the quantum Langevin approach is set in the frequency domain, frequency-dependent propagation (*i.e.* frequency-dependent mean-free paths) between the atoms can be naturally included.
The next step would be to adapt the present approach to โmacroscopicโ configurations (*i.e.* at least many atoms), allowing for a more direct comparison with existing experiments thierry . This would provide a better understanding of light transport properties in nonlinear atomic media where vacuum fluctuations play a role. In particular, for given values of the incident laser intensity and detuning, the nonlinear mean-free path becomes negative in well-defined frequency windows. This means that light amplification can be achieved in these frequency windows pra5M ; prl38WEDM . The atomic media would then constitute a very simple realization of a coherent random laser.
###### Acknowledgements.
We would like to thank Cord Mรผller, Oliver Sigwarth, Andreas Buchleitner, Vyacheslav Shatokhin, Serge Reynaud and Jean-Michel Courty for stimulating discussions. T.W. has been supported by the DFG Emmy Noether program. Laboratoire Kastler Brossel is laboratoire de lโUniversitรฉ Pierre et Marie Curie et de lโEcole Normale Supรฉrieure, UMR 8552 du CNRS.
## Appendix A
The $`g\overline{g}`$ terms in Eq. (31) read:
$$\begin{array}{c}X_i^{}^\beta [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]=\mathrm{}\hfill \\ \hfill g\overline{g}\{X_i^{}^\beta [\mathrm{\Delta }^{}][G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}G_{๐_q^{}j^{}}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](G_{jj^{}}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]..\\ \hfill .+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](๐_q^{\beta +^{(0)}}G_{jj^{}}^{\alpha _p^+}(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](G_{๐_q^{}j^{}}^{\beta _p^+}(X_j^{}^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}})X_j^{\alpha ^{(0)}})[\mathrm{\Delta }]]\\ \hfill [G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}](X_j^{}^{\beta ^{(0)}}G_{๐_q^{}j}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}))[\mathrm{\Delta }^{}]+G_{i^{}j^{}}^{\beta _q^+}[\mathrm{\Delta }^{}](G_{j^{}j}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{\beta ^{(0)}})๐_q^{\alpha ^{(0)}})[\mathrm{\Delta }^{}].\\ \hfill .+G_{i^{}j^{}}^{\beta _q^{}}[\mathrm{\Delta }^{}](๐_q^{\alpha +^{(0)}}G_{j^{}j}^{\beta _p^+}(X_j^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}}))[\mathrm{\Delta }^{}]+G_{i^{}j^{}}^{\beta _q^{}}[\mathrm{\Delta }^{}](G_{๐_q^{}j}^{\alpha _p^+}(X_j^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}})X_j^{}^{\beta ^{(0)}})[\mathrm{\Delta }^{}]]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\\ \hfill +\left[G_{i^{}j^{}}^{\beta _p^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}}\right)[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]\right]\\ \hfill +.\left[G_{i^{}j^{}}^{\beta _p^{}}[\mathrm{\Delta }^{}](๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}})[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]\right]\}\end{array}$$
(61)
$$\begin{array}{c}X_i^{}^\alpha [\mathrm{\Delta }^{}]X_i^\alpha [\mathrm{\Delta }]=\mathrm{}\hfill \\ \hfill g\overline{g}\{X_i^{}^\alpha [\mathrm{\Delta }^{}][G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}G_{๐_q^{}j^{}}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{}^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](G_{jj^{}}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]..\\ \hfill .+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](๐_q^{\beta +^{(0)}}G_{jj^{}}^{\alpha _p^+}(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}))[\mathrm{\Delta }]+G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }](G_{๐_q^{}j^{}}^{\beta _p^+}(X_j^{}^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}})X_j^{\alpha ^{(0)}})[\mathrm{\Delta }]]\\ \hfill [G_{i^{}j^{}}^{\alpha _q^+}[\mathrm{\Delta }^{}](X_j^{}^{\alpha ^{(0)}}G_{๐_q^{}j}^{\beta _p^{}}(๐_p^{\alpha +^{(0)}}X_j^{\beta ^{(0)}}))[\mathrm{\Delta }^{}]+G_{i^{}j^{}}^{\alpha _q^+}[\mathrm{\Delta }^{}](G_{j^{}j}^{\alpha _p^{}}(๐_p^{\beta +^{(0)}}X_j^{\alpha ^{(0)}})๐_q^{\beta ^{(0)}})[\mathrm{\Delta }^{}].\\ \hfill .+G_{i^{}j^{}}^{\alpha _q^{}}[\mathrm{\Delta }^{}](๐_q^{\beta +^{(0)}}G_{j^{}j}^{\alpha _p^+}(X_j^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}))[\mathrm{\Delta }^{}]+G_{i^{}j^{}}^{\alpha _q^{}}[\mathrm{\Delta }^{}](G_{๐_q^{}j}^{\beta _p^+}(X_j^{\beta ^{(0)}}๐_p^{\alpha ^{(0)}})X_j^{}^{\alpha ^{(0)}})[\mathrm{\Delta }^{}]]X_i^{\alpha ^{(0)}}[\mathrm{\Delta }]\\ \hfill +\left[G_{i^{}j^{}}^{\alpha _p^+}[\mathrm{\Delta }^{}]\left(X_j^{}^{\alpha ^{(0)}}๐_p^{\beta ^{(0)}}\right)[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^{}}[\mathrm{\Delta }]\left(๐_q^{\beta +^{(0)}}X_j^{\alpha ^{(0)}}\right)[\mathrm{\Delta }]\right]\\ \hfill +.\left[G_{i^{}j^{}}^{\alpha _p^{}}[\mathrm{\Delta }^{}](๐_p^{\beta +^{(0)}}X_j^{}^{\alpha ^{(0)}})[\mathrm{\Delta }^{}]\right]\left[G_{ij}^{\alpha _q^+}[\mathrm{\Delta }](X_j^{\alpha ^{(0)}}๐_q^{\beta ^{(0)}})[\mathrm{\Delta }]\right]\}\end{array}$$
(62)
## Appendix B Three-body correlation functions
### B.1 Single atom case
The three-body correlation function for the Langevin force reads:
$$C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]=\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta [\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }]f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]F_a^\alpha [\mathrm{\Delta }^{}]F_b^\alpha [\mathrm{\Delta }_1]F_c^\alpha [\mathrm{\Delta }_2],$$
(63)
where $`f[\mathrm{\Delta }]`$ and $`g[\mathrm{\Delta }]`$ are regular functions such that the preceding integral is well defined. Going back to the time domain, $`C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ reads as follows:
$$C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]=\frac{1}{2\pi }๐t๐t^{}e^{i\mathrm{\Delta }t}e^{i\mathrm{\Delta }^{}t^{}}๐t_1๐t_2๐t_3๐t_4\delta (t_1+t_2t)\delta (t_3+t_4t)f(t_1)g(t_3)F_a^\alpha (t^{})F_b^\alpha (t_2)F_c^\alpha (t_4).$$
(64)
Then, from the time correlation properties of the vacuum field, one can show that:
$`F_a^\alpha (t^{})F_b^\alpha (t_2)F_c^\alpha (t_4)`$ $`=4T_{aa^{}}^{q+}T_{bb^{}}^q\delta (t^{}t_2)X_a^{}^\alpha (t^{})X_b^{}^\alpha (t^{})F_c^\alpha (t_4)`$ (65)
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q\delta (t^{}t_4)X_a^{}^\alpha (t^{})F_b^\alpha (t_2)X_c^{}^\alpha (t_4)`$
$`+4T_{bb^{}}^{q+}T_{cc^{}}^q\delta (t_2t_4)F_a^\alpha (t^{})X_b^{}^\alpha (t_2)X_c^{}^\alpha (t_2).`$
where the $`T^{q\pm }`$ are $`15\times 15`$ matrices defined by $`[X_i,๐_q^\pm ]=\pm 2T_{ij}^{q\pm }X_j`$.
When taken at the same time, the atomic operators (including the identity $`๐`$) define a group entirely characterized by the group structure constants $`ฯต_{ij}^k`$, *i.e.*:
$$X_i(t)X_j(t)=\underset{k}{}ฯต_{ij}^kX_k(t),$$
(66)
so that the preceding equation becomes:
$`F_a^\alpha (t^{})F_b^\alpha (t_2)F_c^\alpha (t_4)`$ $`=4T_{aa^{}}^{q+}T_{bb^{}}^q\delta (t^{}t_2)ฯต_{a^{}b^{}}^uX_u^\alpha (t^{})F_c^\alpha (t_4)`$ (67)
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q\delta (t^{}t_4)X_a^{}^\alpha (t^{})F_b^\alpha (t_2)X_c^{}^\alpha (t_4)`$
$`+4T_{bb^{}}^{q+}T_{cc^{}}^q\delta (t_2t_4)ฯต_{a^{}b^{}}^uF_a^\alpha (t^{})X_u^\alpha (t_2).`$
Injecting the preceding relations in $`C(a,b,c)`$ and going back to the frequency domain, we get:
$`C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ $`=4T_{aa^{}}^{q+}T_{bb^{}}^qฯต_{a^{}b^{}}^u{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]X_u^\alpha [\mathrm{\Delta }^{}+\mathrm{\Delta }_1]F_c^\alpha [\mathrm{\Delta }_2]}`$ (68)
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_3g[\mathrm{\Delta }_3]f[\mathrm{\Delta }\mathrm{\Delta }_3]D_{a^{}c^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{}+\mathrm{\Delta }_3,\mathrm{\Delta }\mathrm{\Delta }_3]}`$
$`+4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{a^{}b^{}}^uF_a^\alpha [\mathrm{\Delta }^{}]X_u^\alpha [\mathrm{\Delta }]{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]}`$
$`=4T_{aa^{}}^{q+}T_{bb^{}}^qฯต_{a^{}b^{}}^u{\displaystyle \frac{1}{2\pi }}{\displaystyle }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]G_{uv}^\alpha [\mathrm{\Delta }^{}+\mathrm{\Delta }_1]F_v^\alpha [\mathrm{\Delta }^{}+\mathrm{\Delta }_1]F_c^\alpha [\mathrm{\Delta }_2][`$
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_3g[\mathrm{\Delta }_3]f[\mathrm{\Delta }\mathrm{\Delta }_3]D_{a^{}c^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{}+\mathrm{\Delta }_3,\mathrm{\Delta }\mathrm{\Delta }_3]}`$
$`+4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{a^{}b^{}}^uG_{uv}^\alpha [\mathrm{\Delta }]F_a^\alpha [\mathrm{\Delta }^{}]F_v^\alpha [\mathrm{\Delta }]{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]}`$
$`=2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]4T_{aa^{}}^{q+}T_{bb^{}}^qฯต_{a^{}b^{}}^uD_{vc}^{\alpha \alpha }{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]G_{uv}^\alpha [\mathrm{\Delta }_2]}`$
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_3g[\mathrm{\Delta }_3]f[\mathrm{\Delta }\mathrm{\Delta }_3]D_{a^{}c^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{}+\mathrm{\Delta }_3,\mathrm{\Delta }\mathrm{\Delta }_3]}`$
$`+2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{a^{}b^{}}^uD_{av}^{\alpha \alpha }G_{uv}^\alpha [\mathrm{\Delta }]{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]},`$
where we have introduced the matrix $`D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ defined by:
$$D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]=\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})X_i^\alpha [\mathrm{\Delta }_1]F_b^\alpha [\mathrm{\Delta }]X_k^\alpha [\mathrm{\Delta }_2].$$
(69)
This matrix is calculated using the same strategy (*i.e.* going back and forth to the time domain) and one finally gets:
$$\begin{array}{c}D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]=2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]\{G_{ia}^\alpha [0]L_a^\alpha G_{kc}^\alpha [\mathrm{\Delta }^{}]\stackrel{~}{D}_{bc}^{\alpha \alpha }+G_{ia}^\alpha [\mathrm{\Delta }^{}]G_{kc}^\alpha [0]L_c^\alpha \stackrel{~}{D}_{ab}^{\alpha \alpha }\hfill \\ \hfill +4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{b^{}c^{}}^v\stackrel{~}{D}_{au}^{\alpha \alpha }\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_1]G_{kc}^\alpha [\mathrm{\Delta }_2]G_{vu}^\alpha [\mathrm{\Delta }_1]\\ \hfill +4T_{aa^{}}^+T_{bb^{}}^{}ฯต_{a^{}b^{}}^v\stackrel{~}{D}_{uc}^{\alpha \alpha }\frac{1}{2\pi }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_1]G_{kc}^\alpha [\mathrm{\Delta }_2]G_{vu}^\alpha [\mathrm{\Delta }_2]\}\\ \hfill +4T_{aa^{}}^{q+}T_{cc^{}}^q\left(\frac{1}{2\pi }๐\mathrm{\Delta }_3๐\mathrm{\Delta }_4\delta (\mathrm{\Delta }_3+\mathrm{\Delta }_4\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_3]G_{kc}^\alpha [\mathrm{\Delta }_4]\right)\\ \hfill \times \left(\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})X_a^{}^\alpha [\mathrm{\Delta }_1]F_b^\alpha [\mathrm{\Delta }]X_c^{}^\alpha [\mathrm{\Delta }_2]\right).\end{array}$$
(70)
It may seem that we have taken a loop path and that we are back to square oneโฆ However, in the last line of the preceding formula, we immediately recognize the matrix $`D_{a^{}b^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]`$. Thus, the preceding equation is nothing else but a linear system for this matrix. More precisely, $`D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ is the solution of the following linear system:
$$D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]I_{ik,a^{}c^{}}^{\alpha \alpha }[\mathrm{\Delta }^{}]D_{a^{}c^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]=J_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }],$$
(71)
with
$$\{\begin{array}{cc}\hfill I_{ik,a^{}c^{}}^{\alpha \alpha }[\mathrm{\Delta }^{}]& =4T_{aa^{}}^{q+}T_{cc^{}}^q\frac{1}{2\pi }๐\mathrm{\Delta }_3๐\mathrm{\Delta }_4\delta (\mathrm{\Delta }_3+\mathrm{\Delta }_4\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_3]G_{kc}^\alpha [\mathrm{\Delta }_4]\hfill \\ \hfill J_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }^{},\mathrm{\Delta }]& =2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]\{G_{ia}^\alpha [0]L_a^\alpha G_{kc}^\alpha [\mathrm{\Delta }^{}]\stackrel{~}{D}_{bc}^{\alpha \alpha }+G_{ia}^\alpha [\mathrm{\Delta }^{}]G_{kc}^\alpha [0]L_c^\alpha \stackrel{~}{D}_{ab}^{\alpha \alpha }\hfill \\ & +4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{b^{}c^{}}^v\stackrel{~}{D}_{au}^{\alpha \alpha }\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_1]G_{kc}^\alpha [\mathrm{\Delta }_2]G_{vu}^\alpha [\mathrm{\Delta }_1]\hfill \\ & +4T_{aa^{}}^+T_{bb^{}}^{}ฯต_{a^{}b^{}}^v\stackrel{~}{D}_{uc}^{\alpha \alpha }\frac{1}{2\pi }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_1]G_{kc}^\alpha [\mathrm{\Delta }_2]G_{vu}^\alpha [\mathrm{\Delta }_2]\}\hfill \end{array}.$$
(72)
In the preceding equations, the Greenโs function $`G[\mathrm{\Delta }]`$ and the diffusion matrix $`D^{\alpha \alpha }`$ only depend on the Rabi field $`\mathrm{\Omega }_L`$ evaluated at the position of atom $`\alpha `$. Thus, for *any* value of $`\mathrm{\Delta }`$, numerical values of $`I`$ and $`J`$ can be computed, allowing for a direct calculation of $`D_{ik}^{b,\alpha \alpha \alpha }[\mathrm{\Delta },\mathrm{\Delta }]`$. Furthermore, it is not surprising that the matrix $`I`$ shows up in the linear system. Indeed, the Greenโs function $`G[\mathrm{\Delta }]`$ governs the time evolution of X through a Fourier transform. Thus the time evolution of products of operators $`\text{X}_i(t)\text{X}_j(t)`$ will be simply governed by the Fourier transform of the product of two Greenโs functions $`G(t)G(t)`$, which is precisely the convolution product found in $`I`$. Finally, from the knowledge of the matrix $`D`$, we can calculate the value of $`C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]`$:
$`C_{abc}[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ $`=2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]\{4T_{aa^{}}^{q+}T_{bb^{}}^qฯต_{a^{}b^{}}^uD_{vc}^{\alpha \alpha }{\displaystyle \frac{1}{2\pi }}{\displaystyle }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]G_{uv}^\alpha [\mathrm{\Delta }_2]`$ (73)
$`+4T_{aa^{}}^{q+}T_{cc^{}}^q{\displaystyle \frac{1}{2\pi }}{\displaystyle ๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]D_{a^{}c^{}}^{b,\alpha \alpha \alpha }[\mathrm{\Delta }_1,\mathrm{\Delta }_1]}`$
$`+4T_{bb^{}}^{q+}T_{cc^{}}^qฯต_{a^{}b^{}}^uD_{av}^{\alpha \alpha }G_{uv}^\alpha [\mathrm{\Delta }]{\displaystyle \frac{1}{2\pi }}{\displaystyle }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta })f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]\}.`$
Of course, we recover the global factor $`2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]`$, showing that the time correlation function only depends on the time difference $`t^{}t`$ (stationary condition).
### B.2 Two-atom case
The calculation of quantities like:
$$C_{abc}^{\alpha \beta }[\mathrm{\Delta }^{},\mathrm{\Delta }]=\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta [\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }]f[\mathrm{\Delta }_1]g[\mathrm{\Delta }_2]F_j^{}^\alpha [\mathrm{\Delta }^{}]F_k^\beta [\mathrm{\Delta }_1]F_k^{}^\alpha [\mathrm{\Delta }_2]^{(\overline{g})},$$
(74)
follows, more or less, the way described in the preceding section. In particular, it also involves the calculation of a matrix $`D_{ik}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }]`$ defined as follows:
$$D_{ik}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }]=\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})X_i^\alpha [\mathrm{\Delta }_1]F_b^\beta [\mathrm{\Delta }]X_k^\alpha [\mathrm{\Delta }_2]^{(\overline{g})}.$$
(75)
The latter is also found to be the solution of a linear system, resembling the preceding one (see Eq. (71)):
$$D_{ik}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }]I_{ik,a^{}c^{}}^{\alpha \alpha }[\mathrm{\Delta }^{}]D_{a^{}c^{}}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }]=J_{ik}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }],$$
(76)
with
$$\begin{array}{c}J_{ik}^{b,\alpha \beta \alpha ^{(\overline{g})}}[\mathrm{\Delta }^{},\mathrm{\Delta }]=\left(\frac{1}{2}\right)2\pi \delta [\mathrm{\Delta }+\mathrm{\Delta }^{}]\{G_{ia}^\alpha [0]L_a^\alpha G_{kc}^\alpha [\mathrm{\Delta }^{}]\stackrel{~}{D}_{bc}^{\beta \alpha ^{(0)}}+G_{ia}^\alpha [\mathrm{\Delta }^{}]G_{kc}^\alpha [0]L_c^\alpha \stackrel{~}{D}_{ab}^{\alpha \beta ^{(0)}}.\hfill \\ \hfill +4T_{bb^{}}^{q+}X_b^{}^{\beta ^{(0)}}\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^\alpha [\mathrm{\Delta }_1]G_{kc}^{\alpha _q^{}}[\mathrm{\Delta }_2]G_{cu}^\alpha [\mathrm{\Delta }_1]\stackrel{~}{D}_{au}^{\alpha \alpha ^{(0)}}\\ \hfill +4T_{bb^{}}^qX_b^{}^{\beta ^{(0)}}\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^{\alpha _q^+}[\mathrm{\Delta }_1]G_{kc}^\alpha [\mathrm{\Delta }_2]G_{au}^\alpha [\mathrm{\Delta }_2]\stackrel{~}{D}_{uc}^{\alpha \alpha ^{(0)}}\\ \hfill 2G_{๐_q^+u}^\beta [\mathrm{\Delta }^{}]\stackrel{~}{D}_{ub}^{\beta \beta ^{(0)}}\frac{1}{2\pi }๐\mathrm{\Delta }_1๐\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{ia}^{\alpha _q^{}}[\mathrm{\Delta }_1]\stackrel{~}{X}_a^{\alpha ^{(0)}}[\mathrm{\Delta }_2]\stackrel{~}{X}_k^{\alpha ^{(0)}}[\mathrm{\Delta }_2]^{(0)}\\ \hfill .2G_{๐_q^+u}^\beta [\mathrm{\Delta }^{}]\stackrel{~}{D}_{bu}^{\beta \beta ^{(0)}}\frac{1}{2\pi }d\mathrm{\Delta }_1d\mathrm{\Delta }_2\delta (\mathrm{\Delta }_1+\mathrm{\Delta }_2\mathrm{\Delta }^{})G_{kc}^{\alpha _q^{}}[\mathrm{\Delta }_2]\stackrel{~}{X}_i^{\alpha ^{(0)}}[\mathrm{\Delta }_1]\stackrel{~}{X}_c^{\alpha ^{(0)}}[\mathrm{\Delta }_1]^{(0)}\}.\end{array}$$
(77)
|
warning/0506/math0506063.html
|
ar5iv
|
text
|
# Sur la dynamique unidimensionnelle en rรฉgularitรฉ intermรฉdiaire
## Introduction
### Quelques motivations
Dans ce travail, nous nous intรฉressons ร la dynamique des groupes de diffรฉomorphismes de variรฉtรฉs unidimensionnelles dont la classe de diffรฉrentiabilitรฉ est intermรฉdiaire, cโest-ร -dire supรฉrieure ou รฉgale ร $`C^1`$ mais strictement infรฉrieure ร $`C^2`$. ร premiรจre vue, cela peut paraรฎtre un sujet dโรฉtude ร la fois artificiel et trรจs technique. Il y a nรฉanmoins plusieurs raisons qui le justifient.
La premiรจre est dโordre dynamique. On sait depuis le cรฉlรจbre travail de Denjoy que lโรฉtude de la dynamique des diffรฉomorphismes de classe $`C^2`$ du cercle dont le nombre de rotation est irrationnel se ramรจne (ร conjugaison topologique prรจs) ร celle des rotations. Ce nโest plus le cas en classe $`C^1`$, ce qui a รฉtรฉ illustrรฉ aussi dans par des $`^{_{}}`$contre-exemples $`^{_{}}`$ (voir รฉgalement ). Au ยง3 du chapitre X de , et en suivant une idรฉe de Milnor et Sergeraert, Herman a amรฉliorรฉ ces contre-exemples en les rendant de classe $`C^{1+\tau }`$ pour tout $`\tau <1`$. Au ยง5 de ce mรชme chapitre, on trouve des contre-exemples de Denjoy continus avec des centralisateurs non triviaux. Il est bien possible que Herman ait dรฉjร rรฉflรฉchi au problรจme dโamรฉliorer la classe de diffรฉrentiabilitรฉ de ces derniers. Ce problรจme est en fait au cลur de notre travail.
Une autre raison pour entreprendre cette รฉtude, de nature cohomologique, est liรฉe ร lโinvariant de Godbillon-Vey. Rappelons que pour des groupes agissant sur des variรฉtรฉs unidimensionnelles, cet invariant correspond ร une classe de 2-cohomologie (ร valeurs rรฉelles) dont le reprรฉsentant (non homogรจne) est donnรฉ par
$$๐ข๐ฑ(f,g)=\mathrm{log}(f^{})d\mathrm{log}(g^{}f).$$
Cette dรฉfinition a un sens รฉvident lorsque les diffรฉomorphismes $`f`$ et $`g`$ sont de classe $`C^2`$. Or, on sโaperรงoit facilement que le domaine de dรฉfinition le plus naturel est celui des diffรฉomorphismes dont le logarithme de la dรฉrivรฉe appartient ร lโespace de Sobolev $`H^{1/2}`$. En particulier, lโinvariant peut รชtre dรฉfini (et il est continu) dans le groupe des diffรฉomorphismes de classe $`C^{3/2+\epsilon }`$ pour tout $`\epsilon >0`$. Cette $`^{_{}}`$extension $`^{_{}}`$ proposรฉe par Hurder et Katok dans (voir aussi ) est ร comparer avec le travail remarquable de Tsuboi, oรน il est dรฉmontrรฉ (parmi bien dโautres choses) quโaucune extension nโest envisageable en classe $`C^1`$ (voir รฉgalement ). La preuve de Tsuboi sโappuie sur lโexistence dโune famille assez large de contre-exemples $`C^1`$ (dรปs ร Pixton ) ร un lemme classique de Kopell , selon lequel le centralisateur de tout diffรฉomorphisme de classe $`C^2`$ de lโintervalle $`[\mathrm{0,1}[`$ sans point fixe ร lโintรฉrieur agit librement sur $`]\mathrm{0,1}[`$. Si lโon tient compte de lโextension de Hurder et Katok, on devrait sโattendre ร ce que de tels contre-exemples ne puissent pas รชtre construits en classes intermรฉdiaires trop รฉlรฉvรฉes. Cette motivation a amenรฉ Tsuboi ร construire des contre-exemples au lemme de Kopell (et aussi au thรฉorรจme de Denjoy avec des centralisateurs non triviaux) dont la rรฉgularitรฉ est supรฉrieure ร $`C^1`$ . Il a รฉgalement conjecturรฉ des phรฉnomรจnes de rigiditรฉ en classe intermรฉdiaire, mais il nโa pas donnรฉ de rรฉsultats dans cette direction. Notre travail confirme lโintuition de Tsuboi, et de plus il montre que ses constructions sont en fait optimales en ce qui concerne la rรฉgularitรฉ atteinte.
Signalons en passant que la classe de diffรฉrentiabilitรฉ $`C^{3/2}`$ apparaรฎt de maniรจre naturelle dans dโautres situations. Par exemple, dans elle se trouve รชtre la classe critique pour la dรฉfinition dโun autre invariant cohomologique : le $`^{_{}}`$cocycle de Liouville $`^{_{}}`$. Ce cocycle est un outil important dans lโรฉtude de phรฉnomรจnes de rigiditรฉ (dans lโesprit des thรฉorรจmes de Margulis et Zimmer) pour les actions de groupes $`^{_{}}`$de rang supรฉrieur $`^{_{}}`$ par diffรฉomorphismes du cercle (voir pour un lien entre le cocycle de Liouville et la classe de Godbillon-Vey ; voir aussi ). Dโautre part, la rรฉsolution dโรฉquations diffรฉrentielles stochastiques dans le groupe des diffรฉomorphismes du cercle donne lieu ร des analogues de la mesure de Wiener sur $`\mathrm{Diff}_+^{\mathrm{}}(\mathrm{S}^1)`$ dont le support sโavรจre รชtre prรฉcisรฉment lโespace des diffรฉomorphismes de classe $`C^{3/2}`$ (voir ). ร notre connaissance, le cas $`d=2`$ des thรฉorรจmes A et B que nous verrons plus loin correspondent aux premiers rรฉsultats de nature essentiellement dynamique et liรฉs ร la classe $`C^{3/2}`$.
La derniรจre justification de notre รฉtude concerne la thรฉorie les feuilletages de codimension 1. En effet, en sโinspirant du thรฉorรจme classique de Denjoy, cette thรฉorie sโest developpรฉe en admettant la plupart des cas une hypothรจse de rรฉgularitรฉ transverse $`C^2`$. Or, si lโon tient compte de sa nature dynamique, et du fait que de nombreux rรฉsultats fondamentaux en systรจmes dynamiques, notamment la thรฉorie de Pesin, sont encore valables en classe $`C^{1+\tau }`$ (et parfois mรชme en classe $`C^1`$), on peut espรฉrer que plusieurs propriรฉtรฉs des feuilletages de classe $`C^2`$ soient encore valables dans ce contexte (quelques progrรจs importants dans cette direction sont dรฉjร connus ). Si tel est le cas, on devrait sโattendre รฉgalement ร voir apparaรฎtre des obstructions en rรฉgularitรฉ intermรฉdiaire.
### Prรฉsentation des rรฉsultats
Suivant un principe exprimรฉ par Herman dans lโintroduction de , $`^{_{}}`$tout chercheur dรฉsireux de travailler sur les diffรฉomorphismes du cercle doit sโhabituer ร construire et รฉtudier des exemples $`^{_{}}`$. Ce point de vue est illustrรฉ par le fait que Denjoy lui-mรชme ait abouti ร son cรฉlรจbre thรฉorรจme parce que $`^{_{}}`$il ne pouvait pas en construire des contre-exemples de classe $`C^2`$$`^{_{}}`$ (des contre-exemples qui, nรฉanmoins, รฉtaient vraisemblables ร Poincarรฉ). Dans ce mรชme esprit, si lโon essaie de construire des contre-exemples en rรฉgularitรฉ infรฉrieure ร $`C^2`$ ร plusieurs rรฉsultats de dynamique unidimensionnelle, on voit apparaรฎtre des obstructions pour des classes de diffรฉrentiabilitรฉ intermรฉdiaires trรจs prรฉcises. La premiรจre est liรฉe ร des actions de groupes abรฉliens sur le cercle.
Thรฉorรจme A. Si $`d`$ est un entier supรฉrieur ou รฉgal ร $`2`$ et $`\epsilon >0`$, alors toute action libre de $`^d`$ par diffรฉomorphismes de classe $`C^{1+1/d+\epsilon }`$ du cercle est minimale.
Ce rรฉsultat peut รชtre considรฉrรฉ comme une gรฉnรฉralisation du thรฉorรจme de Denjoy (le cas $`d=1`$) avec une petite hypothรจse de rรฉgularitรฉ supplรฉmentaire (donnรฉe par le $`\epsilon >0`$). Il est bien possible quโil soit encore valable en classe $`C^{1+1/d}`$. Nous en donnons une preuve simple dans ce cas sous une hypothรจse dynamique assez naturelle (voir la proposition 1.1). Signalons que le thรฉorรจme A a รฉtรฉ conjecturรฉ dans par Tsuboi, qui lโa รฉgalement illustrรฉ par des contre-exemples de classe $`C^{1+1/d\epsilon }`$ pour tout $`ฯต>0`$ (nous discuterons de telles constructions au ยง1). Il serait sans doute intรฉressant dโobtenir, sous les hypothรจses du thรฉorรจme, un rรฉsultat dโergodicitรฉ (par rapport ร la mesure de Lebesgue) similaire ร celui valable pour les diffรฉomorphismes de classe $`C^2`$ du cercle dont le nombre de rotation est irrationnel (voir ). Quant ร la rรฉgularitรฉ de la linรฉarisation, le lecteur trouvera dans le ยง3 de des rรฉsultats optimaux (qui gรฉnรฉralisent ceux de Moser ) pour des groupes commutatifs engendrรฉs par des petites perturbations de rotations<sup>1</sup><sup>1</sup>1Malheureusement, nous devons signaler que les exemples du ยง2 de sont erronรฉs. Ceci est une consรฉquence directe de la proposition 1.1, et rรฉsulte รฉgalement du thรฉorรจme A..
Dรป ร lโabsence de rรฉcurrence pour la dynamique, le cas oรน il existe des points fixes globaux, i.e. le cas de lโintervalle, est lรฉgรจrement diffรฉrent. Dans ce contexte, on a un analogue du thรฉorรจme de Denjoy ; cโest le thรฉorรจme de Kopell-Szekeres concernant la rigiditรฉ des centralisateurs de diffรฉomorphismes de classe $`C^2`$ (voir par exemple ). Le rรฉsultat suivant peut รชtre vu comme une gรฉnรฉralisation du fameux lemme de Kopell (le cas $`d=1`$) pour les actions de $`^{d+1}`$ (sous une lรฉgรจre hypothรจse supplรฉmentaire de rรฉgularitรฉ). En effet, lโhypothรจse combinatoire (1) ci-dessus est exactement celle que lโon trouve lors de la dรฉmonstration du lemme de Kopell (voir le ยง1.2). De plus, cโest en utilisant des diffรฉomorphismes (de petite rรฉgularitรฉ) qui vรฉrifient cette propriรฉtรฉ que Tsuboi a dรฉmontrรฉ son thรฉorรจme dโacyclicitรฉ cohomologique dans .
Thรฉorรจme B. Soient $`d2`$ un entier et $`\epsilon >0`$. Soient $`f_1,\mathrm{},f_{d+1}`$ des diffรฉomorphismes de classe $`C^1`$ de lโintervalle $`[\mathrm{0,1}]`$ qui commutent entre eux. Supposons quโil existe des intervalles ouverts disjoints $`I_{n_1,\mathrm{},n_d}`$ qui sont disposรฉs dans $`]\mathrm{0,1}[`$ en respectant lโordre lexicographique et tels que, pour tout $`(n_1,\mathrm{},n_d)^d`$,
$$f_i(I_{n_1,\mathrm{},n_i,\mathrm{},n_d})=I_{n_1,\mathrm{},n_i1,\mathrm{},n_d}\text{pour tout}i\{1,\mathrm{},d\}\text{et}f_{d+1}(I_{n_1,\mathrm{},n_d})=I_{n_1,\mathrm{},n_d}.$$
(1)
Si $`f_1,\mathrm{},f_d`$ sont de classe $`C^{1+1/d+\epsilon }`$, alors la restriction de $`f_{d+1}`$ ร la rรฉunion des $`I_{n_1,\mathrm{},n_d}`$ est lโidentitรฉ.
Une nouvelle fois, la classe de diffรฉrentiabilitรฉ $`C^{1/d}`$ pour la dรฉrivรฉe est optimale. La construction de contre-exemples, bien plus dรฉlicate que dans le cas du cercle, apparaรฎt aussi dans .
Dans les deux rรฉsultats prรฉcรฉdents, la classe intermรฉdiaire optimale est liรฉe ร la $`^{_{}}`$croissance $`^{_{}}`$ du groupe (ou plutรดt des orbites des actions respectives). Il serait intรฉressant (et pas trรจs difficile) dโobtenir des rรฉsultats de ce type pour des actions de groupes nilpotents, ร croissance sous-exponentielle ou rรฉsolubles, qui soient des extensions en classe intermรฉdiaire de ceux obtenus dans , et respectivement. En prรฉsence dโune dynamique $`^{_{}}`$ร croissance exponentielle $`^{_{}}`$, on voit apparaรฎtre des phรฉnomรจnes de rigiditรฉ analogues en classe $`C^{1+\tau }`$ pour tout $`\tau >0`$. De plus, en utilisant des mรฉthodes un peu techniques mais tout ร fait standard en systรจmes dynamiques, on peut dรฉmontrer que certains de ces phรฉnomรจnes ont encore lieu en classe $`C^1`$. Un exemple est donnรฉ par le rรฉsultat ci-dessous, lequel peut รชtre pensรฉ comme une version gรฉnรฉralisรฉe du thรฉorรจme de Sacksteder.
Thรฉorรจme C. Soit $``$ un feuilletage de codimension 1 et transversalement de classe $`C^1`$. Si $``$ nโadmet pas de mesure transverse invariante (au sens de Plante ), alors il possรจde des feuilles ressort hyperboliques.
Ce rรฉsultat nโest absolument pas nouveau. Dans toute sa gรฉnรฉralitรฉ, il doit รชtre atribuรฉ ร Hurder , si bien que des rรฉsultats reliรฉs (mais plus faibles) se trouvent dans . La dรฉmonstration que nous proposons est รฉlรฉmentaire et sโapplique en gรฉneral ร des pseudo-groupes de diffรฉomorphismes en dimension 1. En effet, dโaprรจs la thรฉorie bien connue des feuilletages de codimension 1, la preuve du thรฉorรจme C se rรฉduit ร dรฉmontrer que, en prรฉsence dโune feuille ressort topologique, il existe des feuilles ressort hyperboliques (voir lโappendice 5.1). Or, cโest exactement cette affirmation que nous dรฉmontrons ; la mรฉthode est simple et directe, et elle pourrait รชtre utile dans dโautres circonstances.
Le problรจme de la validitรฉ du thรฉorรจme de Sacksteder en classe infรฉrieure ร $`C^2`$ a รฉtรฉ dโabord abordรฉ par Hurder pour les groupes de diffรฉomorphismes du cercle . En prรฉsence dโun minimal exceptionnel avec une dynamique $`^{_{}}`$suffisamment riche $`^{_{}}`$, il dรฉmontre lโexistence dโรฉlรฉments avec des points fixes hyperboliques. Nous proposons ici une version $`^{_{}}`$globale$`^{_{}}`$ de ce fait.
Thรฉorรจme D. Si $`\mathrm{\Gamma }`$ est un sous-groupe dรฉnombrable de $`\mathrm{Diff}_+^1(\mathrm{S}^1)`$ qui ne prรฉserve aucune mesure de probabilitรฉ du cercle, alors $`\mathrm{\Gamma }`$ contient des รฉlรฉments nโayant que des points fixes hyperboliques.
Remarquons que lโhypothรจse de non existence dโune mesure de probabilitรฉ invariante ci-dessus รฉquivaut ร ce que le groupe $`\mathrm{\Gamma }`$ ne soit pas semi-conjuguรฉ ร un groupe de rotations et quโil nโait pas dโorbite finie.
Tel quโil est รฉnoncรฉ, le thรฉorรจme D gรฉnรฉralise les rรฉsultats obtenus par Hurder dans dans trois directions : lโhypothรจse faite sur la dynamique (nous admettons des groupes non abรฉliens avec toutes ses orbites denses), le nombre de points fixes hyperboliques, et la non existence dโautres points fixes. La premiรจre direction nโest pas nouvelle, car il avait รฉtรฉ dรฉjร remarquรฉ par Ghys que, en classe $`C^2`$, le thรฉorรจme de Sacksteder (qui en gรฉnรฉral est prรฉsentรฉ comme un rรฉsultat qui nโest valable quโen prรฉsence dโun minimal exceptionnel) reste valide pour des groupes non commutatifs qui agissent de maniรจre minimale (voir la page 11 de ). De plus, Hurder a rรฉcemment dรฉmontrรฉ que tout sous-groupe de $`\mathrm{Diff}_+^1(\mathrm{S}^1)`$ sans probabilitรฉ invariante possรจde des รฉlรฉments avec des points fixes hyperboliques . La deuxiรจme direction, suivant laquelle il existe des รฉlรฉments avec des points fixes hyperboliquement dilatants et contractants, est une amรฉlioration non banale dรฉjร en classe $`C^2`$. Nous en donnons une preuve simple dans ce contexte au ยง4.3. Finalement, lโobtention dโun รฉlรฉment nโayant que des points fixes hyperboliques est un raffinement bien plus subtil. En effet, mรชme dans le cas analytique rรฉel, on ne dispose dโaucune dรฉmonstration de ce fait par des mรฉthodes $`^{_{}}`$standard $`^{_{}}`$. Pour la preuve nous nous appuyons fortement sur des rรฉsultats concernant la mesure stationnaire et les exposants de Lyapunov dโune action, lesquels portent un interรชt en soi et sont inclus dans lโappendice de ce travail.
### Sur la mรฉthode de dรฉmonstration
Lโun des aspects techniques les plus importants de la dynamique unidimensionnelle diffรฉrentiable est liรฉ ร la possibilitรฉ de contrรดler la distorsion (linรฉaire ou projective) des itรฉrรฉes dโune application (ou de composรฉes de diffรฉrentes applications). Dans le cas dโune dynamique $`^{_{}}`$inversible $`^{_{}}`$, i.e. lorsquโil nโy a pas de point critique, deux principes sont plus au moins canoniques pour contrรดler la distorsion linรฉaire. Le premier vient du thรฉorรจme de Denjoy : en classe $`C^2`$, on peut contrรดler la distorsion des itรฉrรฉes sur des intervalles disjoints en la comparant avec la somme des longueurs de ces intervalles. Le deuxiรจme principe, dit du $`^{_{}}`$folklore $`^{_{}}`$ dans , est valable en classe $`C^{1+\tau }`$ lorsquโon sait a priori quโil y a suffisamment dโhyperbolicitรฉ.
Dans ce travail nous introduisons des nouvelles mรฉthodes pour contrรดler la distorsion en classe intermรฉdiaire lorsque la dynamique a une structure topologique et combinatoire bien prรฉcise et qui apparaรฎt de maniรจre naturelle dans de nombreuses situations. Ces mรฉthodes sโappuient sur des arguments de nature probabiliste (inspirรฉs en partie de ) : nous cherchons ร assurer un contrรดle uniforme pour la distorsion dโune suite alรฉatoire $`^{_{}}`$typique $`^{_{}}`$ de compositions (si bien quโen gรฉnรฉral il est impossible de dรฉterminer prรฉcisรฉment quelle suite conviendra !). Nous croyons que, au delร mรชme des thรฉorรจmes prรฉsentรฉs, ces mรฉthodes soient celles qui donnent le plus dโinterรชt ร cet article. Dโailleurs, il est trรจs raisonnable dโessayer de les utiliser afin de raffiner dโautres rรฉsultats classiques de la thรฉorie des feuilletages de codimension 1, comme par exemple ceux de Duminy (voir ร ce propos ). Il pourrait sโavรฉrer รฉgalement intรฉressant de repenser en rรฉgularitรฉ intermรฉdiaire la thรฉorie des niveaux , tout en sโappuyant sur les techniques introduites dans ce travail.
Quelques recommandations pour la lecture. La structure de cet article nโest pas toujours linรฉaire. Le lecteur pressรฉ pour la dรฉmonstration du thรฉorรจme de Denjoy gรฉnรฉralisรฉ peut aller directement au ยง2 (si bien que la lecture des exemples du ยง1.1 peut aider ร รฉclaircir un peu le panorama). Celui qui est plutรดt intรฉressรฉ par le thรฉorรจme de Sacksteder peut passer immรฉdiatement au ยง2.1 et puis aux ยง4.1 et ยง4.2 (oรน lโon dรฉmontre les versions gรฉnรฉralisรฉes pour des pseudo-groupes). Le cas des diffรฉomorphismes du cercle nรฉcessite cependant les rรฉsultats probabilistes de lโappendice (qui peuvent รชtre รฉtudiรฉs sans la nรฉcessitรฉ du reste de lโarticle). Nous proposons ci-dessous un schรฉma dโinterdรฉpendance logique entre les diffรฉrents paragraphes. Une flรจche continue reprรฉsente une lecture indispensable, alors quโune flรจche pointillรฉe indique une lecture non indispensable mais tout ร fait convenable pour la comprรฉhension.
Introduction$`\mathrm{\S }1.1`$$`\mathrm{\S }1.2`$$`\mathrm{\S }2.1`$$`\mathrm{\S }2.2`$$`\mathrm{\S }3`$$`\mathrm{\S }4.1`$$`\mathrm{\S }4.2`$$`\mathrm{\S }4.3`$$`\mathrm{\S }4.4`$$`\mathrm{\S }5.1`$$`\mathrm{\S }5.2`$$`\mathrm{\S }5.3`$$`\mathrm{\S }5.4`$ ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... ....................................
Remerciemments. Nous sommes trรจs reconnaissants envers รtienne Ghys pour avoir partagรฉ avec nous ses idรฉes et connaissances sur les groupes de diffรฉomorphismes du cercle, ainsi quโenvers Sylvain Crovisier, sans lโaide de qui nous nโaurions jamais rรฉalisรฉ la validitรฉ des versions $`C^1`$ (et non seulement $`C^{1+\tau }`$) du thรฉorรจme de Sacksteder. Nous exprimons aussi notre gratitude ร Jean Christophe Yoccoz, Danijela Damjanovic et Anna Erschler pour lโintรชret quโils ont portรฉ ร cet article. Ce travail sโest dรฉroulรฉ ร lโIHรS et ร lโUMPA de lโENS-Lyon, et nous voudrions remercier ces deux institutions par leur hospitalitรฉ.
## 1 Quelques exemples
Tout au long de ce travail, nous ne considรฉrerons que des transformations qui respectent lโorientation. En nous inspirant du travail de Denjoy , oรน la construction des contre-exemples $`C^1`$ prรฉcรจde au thรฉorรจme (et en constitue la partie la plus longue), nous ferons une rรฉvision rapide des constructions dโexemples de diffรฉomorphismes du cercle et de lโintervalle ร des centralisateurs non triviaux. La prรฉsentation est inspirรฉe de , bien que des constructions similaires, basรฉes sur les exemples de Pixton , apparaissent dรฉjร dans le travail de Tsuboi .
Rappelons quโรฉtant donnรฉ un homรฉomorphisme $`\eta :[0,\mathrm{}[[0,\mathrm{}[`$, on dit quโune fonction $`\psi `$ dรฉfinie sur le cercle ou un intervalle et ร valeurs rรฉelles est $`\eta `$-continue sโil existe $`C<\mathrm{}`$ tel que pour tout $`x,y`$ on ait
$$|\psi (x)\psi (y)|C\eta \left(|xy|\right).$$
Lorsque $`\tau `$ appartient ร $`]\mathrm{0,1}[`$, un diffรฉomorphisme $`f`$ est dit de classe $`C^{1+\tau }`$ si sa dรฉrivรฉe est $`\eta `$-continue par rapport ร $`\eta (s)=s^\tau `$, cโest-ร -dire sโil existe une constante $`C<\mathrm{}`$ telle que pour tout $`x,y`$ on ait $`|f^{}(x)f^{}(y)|C|xy|^\tau `$ (on dit รฉgalement que $`f^{}`$ est $`\tau `$-Hรถlder continue). Nous dirons que $`f`$ est de classe $`C^{1+lip}`$ si sa dรฉrivรฉe est lipschitzienne.
### 1.1 Les contre-exemples de Denjoy-Herman
Pour construire nos exemples dโactions libres et non minimales de $`^d`$ sur le cercle, nous utiliserons la fa- mille dโapplications $`\phi _{a,b}:[0,a][0,b]`$ introduite par Yoccoz et donnรฉe sur $`]0,a[`$ par $`\phi _{a,b}(x)=\phi _b(\phi _a)^1(x)`$, oรน $`\phi _a:]0,a[`$ est dรฉfini par
$$\phi _a(u)=\frac{1}{\pi }_{\mathrm{}}^u\frac{ds}{s^2+(1/a)^2}=\frac{a}{2}+\frac{a}{\pi }\mathrm{arctan}(au).$$
En faisant $`u=\phi _a^1(x)`$ on voit que
$$\phi _{a,b}^{}(x)=\phi _b^{}(u)(\phi _a^1)^{}(x)=\frac{\phi _b^{}(u)}{\phi _a^{}(u)}=\frac{u^2+1/a^2}{u^2+1/b^2},$$
ce qui montre que $`\phi _{a,b}`$ est un diffรฉomorphisme de classe $`C^1`$ tangent ร lโidentitรฉ aux extrรฉmitรฉs. De plus, pour la deuxiรจme dรฉrivรฉe on trouve aisรฉment la majoration
$$\left|\phi _{a,b}^{\prime \prime }(x)\right|\frac{6\pi }{a}\left|\frac{b}{a}1\right|.$$
(2)
Si pour des intervalles non dรฉgรฉnรฉrรฉs $`I=[x_0,y_0]`$ et $`J=[x_1,y_1]`$ on dรฉsigne par $`\phi (I,J):IJ`$ le diffรฉomorphisme donnรฉ par
$$\phi (I,J)(x)=x_1+\phi _{y_0x_0,y_1x_1}(xx_0),$$
alors on constate immรฉdiatemment que la famille des $`\phi (I,J)`$ est รฉquivariante, dans le sens que
$$\phi (J,K)\phi (I,J)=\phi (I,K).$$
Grรขce aux propriรฉtรฉs prรฉcรฉdentes, cette famille permet de construire des contre-exemples de Denjoy de classe $`C^{1+1/d\epsilon }`$ dont le centralisateur contient un sous-groupe isomorphe ร $`^d`$ qui agit librement. Plus gรฉnรฉralement, nous construirons de tels contre-exemples en classe $`C^{1+\eta }`$ par rapport au module de continuitรฉ $`\eta (s)=s^{1/d}[\mathrm{log}(1/s)]^{1/d+\epsilon }`$. Pour cela, on commence avec $`d`$ rotations $`R_{\theta _1},\mathrm{},R_{\theta _d}`$ dโangles linรฉairรฉment indรฉpendants sur les rationnels. On fixe un entier $`m2`$ et un point $`p\mathrm{S}^1`$, et pour chaque $`(i_1,\mathrm{},i_d)^d`$ on remplace le point $`p_{i_1,\mathrm{},i_d}=R_{\theta _1}^{i_1}\mathrm{}R_{\theta _d}^{i_d}(p)`$ par un intervalle $`I_{i_1,\mathrm{},i_d}`$ de longueur
$$\mathrm{}_{(i_1,\mathrm{},i_d)}=\frac{1}{\left(|i_1|+\mathrm{}+|i_d|+m\right)^d\left[\mathrm{log}(|i_1|+\mathrm{}+|i_d|+m)\right]^{1+\epsilon }}.$$
On obtient ainsi un nouveau cercle (de longueur finie $`T_m2^d\epsilon /[\mathrm{log}(m)]^\epsilon (d1)!`$), sur lequel les $`R_{\theta _j}`$ induisent de maniรจre unique des homรฉomorphismes $`f_j`$ vรฉrifiant, pour tout $`xI_{i_1,\mathrm{},i_j,\mathrm{},i_d}`$,
$$f_j(x)=\phi (I_{i_1,\mathrm{},i_j,\mathrm{},i_d},I_{i_1,\mathrm{}\mathrm{,1}+i_j,\mathrm{},i_d})(x).$$
En vertu des propriรฉtรฉs dโรฉquivariance des $`\phi (I,J)`$, ces homรฉomorphismes $`f_j`$ commutent entre eux. Vรฉrifions maintenant que $`f_1`$ est de classe $`C^{1+\eta }`$, le cas des autres $`f_i`$ รฉtant analogue. Pour cela, on remarque dโabord que si $`x`$ et $`y`$ appartiennent au mรชme intervalle $`I_{i_1,\mathrm{},i_d}`$, alors $`|f_1^{}(x)f_1^{}(y)|=f_1^{\prime \prime }(p)|xy|`$ pour un point $`pI_{i_1,\mathrm{},i_d}`$. Ceci donne, grรขce ร (2) et au fait que la fonction $`ss/\eta (s)`$ est croissante,
$$\frac{|f_1^{}(x)f_1^{}(y)|}{\eta \left(|xy|\right)}\frac{|xy|}{\eta \left(|xy|\right)}\frac{6\pi }{|I_{i_1,\mathrm{},i_d}|}\left|\frac{|I_{1+i_1,\mathrm{},i_d}|}{|I_{i_1,\mathrm{},i_d}|}1\right|\frac{6\pi }{\eta \left(|I_{i_1,\mathrm{},i_d}|\right)}\left|\frac{|I_{1+i_1,\mathrm{},i_d}|}{|I_{i_1,\mathrm{},i_d}|}1\right|.$$
Or, lโexpression ร droite est รฉgale ร
$$\left|\frac{\left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)^d\left[\mathrm{log}\left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)\right]^{1+\epsilon }}{\left(\left|1+i_1\right|+\mathrm{}+\left|i_d\right|+m\right)^d\left[\mathrm{log}\left(\left|1+i_1\right|+\mathrm{}+\left|i_d\right|+m\right)\right]^{1+\epsilon }}1\right|\frac{6\pi \left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)\left[\mathrm{log}\left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)\right]^{\left(1+\epsilon \right)/d}}{\left[d\mathrm{log}\left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)+\left(1+\epsilon \right)\mathrm{log}\left(\mathrm{log}\left(\left|i_1\right|+\mathrm{}+\left|i_d\right|+m\right)\right)\right]^{1/d+\epsilon }}.$$
En appliquant le thรฉorรจme des accroissements finis ร la fonction $`ss^d[\mathrm{log}(s)]^{1+\epsilon }`$, on constate aisรฉment que cette expression est majorable par $`(d+2)/[\mathrm{log}(m)]^{\epsilon (11/d)}`$. Lorsque $`x`$ et $`y`$ appartiennent ร des intervalles $`|I_{i_1,\mathrm{},i_d}|`$ distincts, la mรชme majoration modulo un facteur $`2`$ est valable, car les diffรฉomorphismes sont tangents ร lโidentitรฉ aux extrรฉmitรฉs de ces intervalles. Cela montre quโen faisant tendre $`m`$ vers lโinfini, les diffรฉomorphismes $`f_j`$ deviennent aussi proches que lโon veut de lโidentitรฉ en topologie $`C^{1+\eta }`$. Puisque $`T_m1`$ pour tout $`m`$ assez grand, ceci reste encore valable aprรจs renormalisation du cercle (de faรงon ร ce que sa longueur soit รฉgale ร $`1`$).
Les exemples que nous venons de construire vรฉrifient $`f_j^{}(x)=1`$ pour tout point $`x`$ qui est une extrรฉmitรฉ de lโun des intervalles $`I_{i_1,\mathrm{},i_d}`$. Nous verrons ci-dessous que, sous une telle hypothรจse, il est impossible de fabriquer des exemples analogues de classe $`C^{1+1/d}`$. Lโargument de la preuve prรฉsentรฉe ร continuation est ร rapprocher avec .
###### Proposition 1.1.
Considรฉrons une action de $`^d`$ par diffรฉomorphismes du cercle qui est libre et semiconjuguรฉe ร une action par des rotations sans y รชtre conjuguรฉe. Supposons que pour toute composante connexe $`I`$ du complรฉmentaire du Cantor invariant minimal, et pour tout รฉlรฉment $`h^d`$, la dรฉrivรฉe de (lโimage de) $`h`$ aux extrรฉmitรฉs de $`I`$ soit รฉgale ร $`1`$. Alors la rรฉgularitรฉ de lโaction est strictement infรฉrieure ร $`C^{1+1/d}`$.
Dรฉmonstration. Supposons le contraire et soient $`f_1,\mathrm{},f_d`$ les gรฉnรฉrateurs de $`^d`$ en tant que sous-groupe de $`\mathrm{Diff}_+^{1+1/d}(\mathrm{S}^1)`$. Notons $`C_i`$ la constante de Hรถlder dโexposant $`1/d`$ pour $`f_i^{}`$, et posons $`C=\mathrm{max}\{C_1,\mathrm{},C_d\}`$. Fixons une composante connexe $`I`$ du complรฉmentaire du Cantor invariant, et pour chaque entier $`n0`$ notons $`\mathrm{}_n`$ la longueur minimale dโun intervalle de la forme $`f_1^{i_1}\mathrm{}f_d^{i_d}(I)`$, avec $`i_j0`$ et $`_ji_j=n`$. Sans perdre en gรฉnรฉralitรฉ, nous pouvons supposer que tous ces intervalles ont une longueur infรฉrieure ou รฉgale ร $`(\frac{1}{C(1+1/d)})^{1/d}`$. Nous affirmons que
$$\mathrm{}_{n+1}\mathrm{}_n(1C\mathrm{}_n^{1/d}).$$
(3)
En effet, si $`\mathrm{}_{n+1}`$ est rรฉalisรฉ comme la longueur correspondante ร un intervalle $`J`$, alors $`J`$ est lโimage par lโun des gรฉnรฉrateurs $`f_i`$ dโun intervalle $`K=[a,b]`$ dont la longueur est par dรฉfinition supรฉrieure ou รฉgale ร $`\mathrm{}_n`$. Puisquโil existe $`qK`$ tel que $`f_i^{}(q)=|J|/|K|`$ et que $`f_i^{}(a)=f_i^{}(b)=1`$, on a
$$\left|\frac{|J|}{|K|}1\right|C_i|qa|^{1/d}C|K|^{1/d},$$
dโoรน on obtient
$$\left|\mathrm{}_{n+1}|K|\right|=\left||J||K|\right|C|K|^{1+1/d}.$$
Donc
$$\mathrm{}_{n+1}|K|\left(1C|K|^{1/d}\right),$$
et ceci implique (3), car $`|K|\mathrm{}_n`$ et la fonction $`ss(1Cs^{1/d})`$ est croissante sur lโintervalle $`[0,(\frac{1}{C(1+1/d)})^{1/d}]`$.
Nous affirmons maintenant que pour $`A=\mathrm{min}\{\mathrm{}_1,d^d/2^{d^2}C^d\}`$ et pour tout $`n`$,
$$\mathrm{}_n\frac{A}{n^d}.$$
(4)
Cette inรฉgalitรฉ est vรฉrifiable aisรฉment par rรฉcurrence. Pour $`n=1`$ elle a lieu ร cause de la condition $`A\mathrm{}_1`$. Dโautre part, si elle est valable pour un entier $`n1`$ alors, la fonction $`ss(1Cs^{1/d})`$ รฉtant croissante sur lโintervalle $`[0,(\frac{1}{C(1+1/d)})^{1/d}]`$, ร partir de lโinรฉgalitรฉ (3) et de la condition $`A^{1/d}d/2^dC`$ on obtient
$$\mathrm{}_{n+1}\frac{A}{n^d}\left(1C\frac{A^{1/d}}{n}\right)\frac{A}{(n+1)^d}.$$
Nous sommes maintenant en mesure dโachever la preuve de la proposition. Pour cela, remarquons que la quantitรฉ dโintervalles de la forme $`I_{i_1,\mathrm{},i_d}=f_1^{i_1}\mathrm{}f_d^{i_d}(I)`$, avec $`i_j0`$ et $`_ji_j=n`$, est supรฉrieure ou รฉgale ร $`Bn^{d1}`$ pour une certaine constante universelle $`B>0`$. Ceci donne, grรขce ร (4),
$$\underset{(i_1,\mathrm{},i_d)_0^d}{}|I_{i_1,\mathrm{},i_d}|=|I|+\underset{n1}{}\underset{{\scriptscriptstyle i_j}=n}{}|I_{i_1,\mathrm{},i_d}|B\underset{n1}{}n^{d1}\mathrm{}_nAB\underset{n1}{}\frac{1}{n}=\mathrm{},$$
ce qui est absurde, car la somme des longueurs des intervalles (disjoints) $`I_{i_1,\mathrm{},i_d}`$ est finie. $`\mathrm{}`$
### 1.2 Les contre-exemples de Pixton-Tsuboi
Dans le contexte des groupes abรฉliens de diffรฉomorphismes de lโintervalle on dispose du fameux lemme de Kopell , dont la version la plus connue nโest dโhabitude รฉnoncรฉe quโen classe $`C^2`$. Pour la commoditรฉ du lecteur, nous donnons ci-dessous la version gรฉnรฉrale avec une preuve simple dont lโidรฉe sous-jacente sera utilisรฉe ร plusieurs reprises. Rappelons quโun diffรฉomorphisme $`f`$ de $`[\mathrm{0,1}[`$ est de classe $`C^{1+vb}`$ si sa dรฉrivรฉe est ร variation bornรฉe sur tout intervalle compact contenu dans $`[\mathrm{0,1}[`$.
Lemme \[Kopell\]. Soient $`f`$ et $`g`$ deux diffรฉomorphismes de lโintervalle $`[\mathrm{0,1}[`$ qui commutent entre eux. Supposons que $`f`$ soit de classe $`C^{1+vb}`$ et $`g`$ de classe $`C^1`$. Si $`f`$ nโa pas de point fixe sur $`]\mathrm{0,1}[`$ et $`g`$ possรจde de tels points, alors $`g`$ est lโidentitรฉ.
Dรฉmonstration. Quitte ร changer $`f`$ par son inverse, on peut supposer que $`f(x)<x`$ pour tout $`x]\mathrm{0,1}[`$. Soit $`b]\mathrm{0,1}[`$ lโun des points fixes de $`g`$. Pour chaque $`n`$ notons $`b_n=f^n(b)`$, et soit $`a=b_1=f(b)`$<sup>2</sup><sup>2</sup>2Remarquons que pour $`f_1=f,f_2=g`$ et $`I_n=f^n(]a,b[)`$, nous sommes exactement sous une hypothรจse combinatoire du type (1), avec $`d=1`$.. Puisque $`g`$ fixe lโintervalle $`[b_{n+1},b_n]`$, pour chaque $`n`$ il existe $`c_n[b_n,b_{n+1}]`$ tel que $`g^{}(c_n)=1`$. รtant donnรฉ que la suite $`(c_n)`$ tend vers lโorigine et que $`g`$ est de classe $`C^1`$, on a nรฉcessairement $`g^{}(0)=1`$. Notons $`M=M(f)`$ la variation du logarithme de la dรฉrivรฉe de $`f`$ sur lโintervalle $`[0,b]`$. Si $`u`$ et $`v`$ appartiennent ร $`[a,b]`$ alors
$$\left|\mathrm{log}\left(\frac{(f^n)^{}(v)}{(f^n)^{}(u)}\right)\right|\underset{i=1}{\overset{n}{}}\left|\mathrm{log}\left(f^{}(f^{i1}(v))\right)\mathrm{log}\left(f^{}(f^{i1}(u))\right)\right|M.$$
(5)
En posant $`u=x[a,b]`$ et $`v=f^ngf^n(x)=g(x)[a,b]`$, en utilisant lโรฉgalitรฉ
$$g^{}(x)=\frac{(f^n)^{}(x)}{(f^n)^{}(f^ngf^n(x))}g^{}(f^n(x))=\frac{(f^n)^{}(x)}{(f^n)^{}(g(x))}g^{}(f^n(x)),$$
et en passant ร la limite lorsque $`n`$ tend vers lโinfini, on obtient lโinรฉgalitรฉ $`sup_{x[a,b]}g^{}(x)\mathrm{exp}(M)`$. Or, ceci reste valable lorsquโon remplace $`g`$ par $`g^j`$ pour nโimporte quel $`j`$ (car $`M`$ ne dรฉpend que de $`f`$). On en dรฉduit que $`sup_{x[a,b]}(g^j)^{}(x)\mathrm{exp}(M)`$. Comme $`g`$ fixe $`a`$ et $`b`$, ceci implique que la restriction de $`g`$ ร lโintervalle $`[a,b]`$ est lโidentitรฉ. Finalement, en conjugant successivement par $`f`$, on conclut que $`g`$ est lโidentitรฉ sur tout lโintervalle $`[\mathrm{0,1}[`$. $`\mathrm{}`$
###### Remarque 1.2.
Notons que dans la dรฉmonstration prรฉcรฉdente, tout le contrรดle de distorsion est rรฉalisรฉ par le diffรฉomorphisme $`f`$, qui doit donc รชtre de classe au moins $`C^{1+vb}`$, tandis que lโautre diffรฉomorphisme $`g`$ nโa besoin que dโรชtre de classe $`C^1`$. Le lemme nโest cependant plus valable lorsque $`g`$ est seulement lipschitzien. En effet, si lโon fixe nโimporte quel homรฉomorphisme lipschitzien $`h`$ de $`[a,b]`$ et on lโรฉtend ร $`[\mathrm{0,1}[`$ en commutant avec $`f`$, alors pour tout $`c]\mathrm{0,1}[`$ lโapplication quโon obtient est encore lipschitzienne sur $`[0,c]`$, et sa constante de Lipschitz y diffรจre de celle de $`h`$ par un facteur au plus รฉgal ร $`\mathrm{exp}(M_c)`$, oรน $`M_c`$ dรฉsigne la variation du logarithme de la dรฉrivรฉe de $`f`$ sur $`[0,c]`$. Lโinterรชt de cette observation vient du fait que, dans , il a รฉtรฉ remarquรฉ que si $`f:[\mathrm{0,1}[[\mathrm{0,1}[`$ est un diffรฉomorphisme de classe $`C^2`$ sans point fixe ร lโintรฉrieur et $`g`$ est un homรฉomorphisme de $`[\mathrm{0,1}[`$ qui commute avec $`f`$, alors la restriction de $`g`$ ร $`]\mathrm{0,1}[`$ est automatiquement de classe $`C^2`$ lorsque $`g`$ est un diffรฉomorphisme de classe $`C^1`$. De plus, si $`f`$ est hyperbolique, i.e. si $`f^{}(0)1`$, alors $`g`$ est de classe $`C^2`$ sur $`[\mathrm{0,1}[`$ dรจs quโil est diffรฉrentiable ร lโorigine .
On peut donner des contre-exemples au lemme de Kopell analogues ร ceux du thรฉorรจme de Denjoy, tout en respectant la propriรฉtรฉ combinatoire (1). Pour cela, fixons un entier $`m2`$, et pour $`(i_1,\mathrm{},i_d)^d`$ posons une nouvelle fois
$$\mathrm{}_{(i_1,\mathrm{},i_d)}=\frac{1}{\left(|i_1|+\mathrm{}+|i_d|+m\right)^d\left[\mathrm{log}(|i_1|+\mathrm{}+|i_d|+m)\right]^{1+\epsilon }}.$$
Par rรฉcurrence sur $`j`$ dรฉfinisons $`\mathrm{}_{(i_1,\mathrm{},i_{j1})}=_{i_j}\mathrm{}_{(i_1,\mathrm{},i_{j1},i_j)}`$. Notons $`[x_{(i_1,\mathrm{},i_j,\mathrm{},i_d)},y_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}]`$ lโintervalle dโextrรฉmitรฉs
$$x_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}=\underset{i_1^{}<i_1}{}\mathrm{}_{(i_1^{})}+\underset{i_2^{}<i_2}{}\mathrm{}_{(i_1,i_2^{})}+\mathrm{}+\underset{i_d^{}<i_d}{}\mathrm{}_{(i_1,\mathrm{},i_{d1},i_d^{})},y_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}=x_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}+\mathrm{}_{(i_1,\mathrm{},i_d)}.$$
Considรฉrons le diffรฉomorphisme $`f_j`$ de $`[\mathrm{0,1}]`$ dont la restriction aux intervalles $`[x_{(i_1,\mathrm{},i_j,\mathrm{},i_d)},y_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}]`$ coรฏncide avec
$$\phi ([x_{(i_1,\mathrm{},i_j,\mathrm{},i_d)},y_{(i_1,\mathrm{},i_j,\mathrm{},i_d)}],[x_{(i_1,\mathrm{},i_j1,\mathrm{},i_d)},y_{(i_1,\mathrm{},i_j1,\mathrm{},i_d)}]).$$
Les estimรฉes qui prรฉcรจdent la proposition (1.1) permettent de dรฉmontrer une nouvelle fois que les $`f_i`$ ainsi obtenus sont de classe $`C^{1+\eta }`$ par rapport au module de continuitรฉ $`\eta (s)=s^{1/d}[\mathrm{log}(1/s)]^{1/d+\epsilon }`$. On obtient donc des contre-exemples au lemme de Kopell donnรฉs par des actions de $`^d`$ par diffรฉomorphismes de classe $`C^{1+1/d\epsilon }`$ de lโintervalle. Cependant, nous devons impรฉrativement souligner que la rรฉgularitรฉ atteinte par cette mรฉthode nโest pas optimale. Pour aboutir ร la classe optimale $`C^{1+1/(d1)\epsilon }`$, il est nรฉcรฉssaire de modifier la construction prรฉcรฉdente afin de suprimer les tangences ร lโidentitรฉ $`^{_{}}`$excessives $`^{_{}}`$. En effet, une modification simple de la proposition 1.1 montre que par cette mรฉthode on nโatteindra mรชme pas la classe $`C^{1+1/d}`$. Le lecteur intรฉressรฉ trouvera dans la construction des contre-exemples optimaux au lemme de Kopell ; nous reviendrons sur ce point dans le ยง3.
Pour finir ce paragraphe, nous voudrions prรฉsenter une preuve simple dโune version du lemme de Kopell pour des actions de $`^3`$ avec une hypothรจse de rรฉgularitรฉ intermรฉdiaire (qui nรฉanmoins nโest pas optimale). Pour cela, commenรงons par un lemme รฉlรฉmentaire mais trรจs utile.
###### Lemme 1.3.
Soit $`g:[a,b][a,b]`$ un diffรฉomorphisme de classe $`C^{1+\tau }`$. Si $`C=C(g)`$ dรฉsigne la constante de $`\tau `$-hรถlderianitรฉ de $`g^{}`$, alors pour tout $`x]a,b[`$ on a
$$|xg(x)|C|ba|^{1+\tau }.$$
(6)
Dรฉmonstration. Fixons un point arbitraire $`x`$ dans $`]a,b[`$, et prenons deux points $`p[a,x]`$ et $`q[a,b]`$ tels que $`g^{}(p)=\left(g(x)g(a)\right)/(xa)`$ and $`g^{}(q)=\left(g(b)g(a)\right)/(ba)=1`$. Alors on a
$$|xg(x)|=\left|x\left[(xa)g^{}(p)+g(a)\right]\right|=\left|(xa)\left(1g^{}(p)\right)\right|.$$
(7)
Puisque $`g^{}`$ est $`\tau `$-Hรถlder continue,
$$|1g^{}(p)|=|g^{}(q)g^{}(p)|C|qp|^\tau C|ba|^\tau .$$
(8)
De (7) et (8) on dรฉduit
$$|xg(x)||xa|C|ba|^\tau C|ba|^{1+\tau },$$
ce qui conclut la preuve du lemme. $`\mathrm{}`$
La proposition suivante nโest quโune consรฉquence du lemme de Kopell gรฉnรฉralisรฉ (thรฉorรจme B). Cependant, lโidรฉe de la preuve que nous prรฉsentons, dans un esprit dynamique $`^{_{}}`$classique $`^{_{}}`$, pourrait รชtre utile dans dโautres situations.
###### Proposition 1.4.
Soient $`f`$, $`g`$ et $`h`$ des diffรฉomorphismes de lโintervalle $`[\mathrm{0,1}]`$ qui commutent entre eux, avec $`f`$ et $`g`$ de classe $`C^{1+\tau }`$ et $`h`$ de classe $`C^1`$. Supposons quโil existe des intervalles ouverts disjoints de la forme $`I_{i,j}`$ qui soient disposรฉs en respectant lโordre lexicografique sur $`]\mathrm{0,1}[`$, et de sorte que pour tout $`(i,j)^2`$ on ait
$$f(I_{i,j})=I_{i1,j},g(I_{i,j})=I_{i,j1}h(I_{i,j})=I_{i,j}.$$
Si $`\tau (\sqrt{5}1)/2`$ alors la restriction de $`h`$ ร la rรฉunion des $`I_{i,j}`$ est lโidentitรฉ.
Dรฉmonstration. Reprenons lโargument de la preuve du lemme de Kopell. Pour tout $`xI=I_{\mathrm{0,0}}`$ et tout $`(n,m)^2`$ on a
$$\mathrm{log}(h^n)^{}(x)=\mathrm{log}\left((h^n)^{}(f^m(x))\right)+\mathrm{log}\left((f^m)^{}(x)\right)\mathrm{log}\left((f^m)^{}(h^n(x))\right).$$
Puisque $`f^m(x)`$ converge vers un points fixe (parabolique) de $`h`$, en prenant la limite lorsque $`m`$ tend vers lโinfini on obtient
$`|\mathrm{log}\left((h^n)^{}(x)\right)|`$ $``$ $`{\displaystyle \underset{k0}{}}\left|\mathrm{log}\left(f^{}(f^k(x))\right)\mathrm{log}\left(f^{}(h^n(f^k(x)))\right)\right|`$
$``$ $`\overline{C}{\displaystyle \underset{k0}{}}\left|f^k(x)h^n(f^k(x))\right|^\tau `$
$``$ $`\overline{C}{\displaystyle \underset{k0}{}}|f^k(I)|^\tau ,`$
$`\overline{C}=\overline{C}(f)`$ dรฉsigne la constante de $`\tau `$-hรถlderianitรฉ de $`\mathrm{log}(f^{})`$. Pour obtenir la convergence de la derniรจre sรฉrie, lโidรฉe consiste ร remarquer que $`|f^k(I)|`$ est trรจs petit par rapport ร $`|f^k(J)|`$, oรน $`J`$ dรฉsigne le plus petit intervalle contenant tous les $`g^m(I)`$, avec $`m`$ (notons que $`g`$ fixe lโintervalle $`J`$). En effet, si $`C=C(g)`$ est la constante de $`\tau `$-hรถlderianitรฉ de $`g^{}`$, alors lโinรฉgalitรฉ (6) donne $`|f^k(I)|C|f^k(J)|^{1+\tau }`$. Par suite,
$$\left|\mathrm{log}\left((h^n)^{}(x)\right)\right|\overline{C}C\underset{k0}{}|f^k(J)|^{\tau (1+\tau )}.$$
Si $`\tau (\sqrt{5}1)/2`$ alors $`\tau (1+\tau )1`$, et lโexpression prรฉcรฉdente est majorรฉe par $`\overline{C}C_{k0}|f^k(J)|\overline{C}C`$. Autrement dit, pour tout $`xI`$ et tout $`n`$ on a lโinรฉgalitรฉ $`(h^n)^{}(x)\mathrm{exp}(\overline{C}C)`$. Comme dans la fin de la preuve du lemme (classique) de Kopell, ceci implique que la restriction de $`h`$ ร $`I`$ (et donc ร tous les $`I_{i,j}`$) est lโidentitรฉ. $`\mathrm{}`$
Si pour chaque entier $`d3`$ on note $`\tau _d`$ lโunique rรฉel positif qui vรฉrifie lโรฉgalitรฉ $`\tau _d(1+\tau _d)^{d2}=1`$, alors la mรฉthode prรฉcรฉdente permet de dรฉmontrer une proposition analogue pour des actions de $`^d`$ par diffรฉomorphismes de classe $`C^{1+\tau _d}`$ (et qui vรฉrifient une condition combinatoire du type (1)). Or, cet exposant $`\tau _d`$ nโest pas du tout optimal. En effet, bien que la suite $`(\tau _d)`$ converge vers zรฉro lorsque $`d`$ tend vers lโinfini, le nombre $`\tau _d`$ est supรฉrieur ร $`1/(d1)`$ (par exemple, $`\tau _3`$ est รฉgal au nombre dโor $`(\sqrt{5}1)/2>1/2`$).
## 2 Le thรฉorรจme de Denjoy gรฉnรฉralisรฉ
### 2.1 Le principe gรฉnรฉral
Nous rappelons dans la suite lโidรฉe de la preuve du thรฉorรจme de Denjoy donnรฉe par Schwartz dans . Cโest un principe qui est devenu classique grรขce ร la formulation et les applications aux feuilletages de codimension 1 donnรฉes par Sacksteder dans (voir pour une discussion plus dรฉtaillรฉe).
###### Lemme 2.1.
Soit $`\mathrm{\Gamma }`$ un groupe de diffรฉomorphismes de classe $`C^{1+lip}`$ dโune variรฉtรฉ unidimensionnelle compacte. Supposons quโil existe un sous-ensemble fini $`๐ข`$ de $`\mathrm{\Gamma }`$ et un intervalle ouvert $`I`$ satisfaisant la propriรฉtรฉ suivante : pour chaque $`n`$ il existe un รฉlรฉment $`h_n=g_{i_{n,n}}g_{i_{n1,n}}\mathrm{}g_{i_{1,n}}`$ dans $`\mathrm{\Gamma }`$ tel que tous les $`g_{i_{m,n}}`$ appartiennent ร $`๐ข`$, les intervalles $`I,g_{i_{1,n}}(I),g_{i_{2,n}}g_{i_{1,n}}(I),\mathrm{},g_{i_{n1,n}}\mathrm{}g_{i_{1,n}}(I)`$ sont deux ร deux disjoints, et $`h_n(I)`$ sโaccumule sur lโune des extrรฉmitรฉs de $`I`$. Alors pour $`n`$ assez grand, lโapplication $`h_n`$ possรจde un point fixe hyperbolique (qui est proche de lโextrรฉmitรฉ correspondante de $`I`$).
La dรฉmonstration de ce lemme utilise le fait que les intervalles $`I,\mathrm{},g_{i_{n1,n}}\mathrm{}g_{i_{1,n}}(I)`$ sont disjoints, ainsi que lโhypothรจse de rรฉgularitรฉ $`C^{1+lip}`$, pour contrรดler la distorsion de $`h_n`$ sur $`I`$ grรขce ร lโargument classique de Denjoy. Ensuite, lโidรฉe consiste ร contrรดler la distorsion de $`h_n`$ sur un intervalle plus large $`J`$ contenant $`I`$ et indรฉpendant de $`n`$. On aboutit ร ceci par un argument de rรฉcurrence assez subtil (remarquons que, en gรฉnรฉral, les intervalles $`J,\mathrm{},g_{i_{n1,n}}\mathrm{}g_{i_{1,n}}(J)`$ ne sont pas disjoints!). La contraction (topologique) devient alors รฉvidente (voir la figure 1), et cette contraction doit รชtre hyperbolique ร cause du contrรดle (uniforme) de la distorsion.
.....................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................$`|`$$`|`$$`|`$$`|`$$`(`$$`)`$$`(`$$`)`$$`h_n`$$`h_n`$$`h_n`$$`a`$$`b`$..........................................................................$`J=[a,b]`$Figure 1$`h_n(I)`$$`I`$$``$point fixehyperbolique
Si lโon veut contrรดler les distorsions en classe $`C^{1+\tau }`$, on est amenรฉ ร estimer des sommes du type
$$|I|^\tau +\underset{k=1}{\overset{n1}{}}|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I)|^\tau $$
(9)
Or, mรชme si les intervalles $`I,\mathrm{},g_{i_{n1,n}}\mathrm{}g_{i_{1,n}}(I)`$ sont deux ร deux disjoints, cette somme peut devenir trop grande avec $`n`$. Pour rรฉsoudre ce problรจme, notre idรฉe consiste ร penser les compositions de maniรจre alรฉatoire (contrairement au cas de la preuve de Schwartz et Sacksteder, oรน la suite des compositions est fixรฉe de maniรจre dรฉterministe). Plus prรฉcisรฉment, nous considรฉrerons des suites dโapplications $`h_n`$ de la forme $`h_n=g_nh_{n1}`$ de faรงon ร ce que les intervalles $`h_0(I)=I,h_1(I),\mathrm{}`$ soient deux ร deux disjoints, et par des arguments dโordre probabiliste nous chercherons ร montrer que pour un $`^{_{}}`$chemin typique $`^{_{}}`$ la valeur de la somme (9) est uniformรฉment bornรฉe. Cela nous permettra de trouver des รฉlรฉments avec des points fixes hyperboliques grรขce au lemme gรฉnรฉral suivant.
###### Lemme 2.2.
Soit $`\mathrm{\Gamma }`$ un pseudo-groupe de diffรฉomorphismes de classe $`C^{1+\tau }`$ dโune variรฉtรฉ unidimensionnelle compacte. Supposons quโil existe un sous-ensemble fini $`๐ข`$ de $`\mathrm{\Gamma }`$, un intervalle ouvert $`I`$ et une constante $`M<\mathrm{}`$ tels que la propriรฉtรฉ suivante soit satisfaite : ร chaque รฉlรฉment $`g๐ข`$ on peut associer un intervalle compact $`\mathrm{C}_g`$ contenu dans un domaine ouvert de dรฉfinition de $`g`$ de sorte que, pour tout $`n`$, il existe un รฉlรฉment $`h_n=g_{i_{n,n}}\mathrm{}g_{i_{1,n}}`$ dans $`\mathrm{\Gamma }`$ vรฉrifiant les propriรฉtรฉs suivantes :
โ tous les $`g_{i_{m,n}}`$ appartiennent ร $`๐ข`$ ;
โ si $`g_{i_{k,n}}=g`$ alors lโintervalle $`g_{i_{k1,n}}\mathrm{}g_{i_{1,n}}(I)`$ est contenu dans $`\mathrm{C}_g`$ (oรน nous convenons que $`g_{i_{k1,n}}\mathrm{}g_{i_{1,n}}`$ est lโidentitรฉ lorsque $`k=1`$) ;
โ on a lโinรฉgalitรฉ
$$\underset{k=0}{\overset{n1}{}}|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I)|^\tau M.$$
Alors il existe une constante strictement positive $`L=L(\tau ,M,|I|;๐ข)`$ telle que si $`n`$ est tel que $`h_n(I)`$ est contenu dans un $`L`$-voisinage de lโintervalle $`I`$, alors lโapplication $`h_n`$ possรจde un point fixe hyperbolique (qui est proche de lโextrรฉmitรฉ correspondante de $`I`$).
Dรฉmonstration. La preuve du lemme รฉtant bien connue, nous ne la rรฉpรฉtons que pour le cas dโun groupe de diffรฉomorphismes. Fixons une constante $`C>0`$ telle que pour tout $`g๐ข`$ et tout $`x,y`$ on ait
$$\left|\mathrm{log}(g^{}(x))\mathrm{log}(g^{}(y))\right|C|xy|^\tau .$$
Nous montrerons alors que pour
$$L=L(\tau ,M,|I|;๐ข)=\frac{|I|}{2\mathrm{exp}(2^\tau CM)},$$
lโaffirmation du lemme est satisfaite.
Dรฉsignons par $`J`$ le $`2L`$-voisinage de $`I`$, et notons $`I^{}`$ (resp. $`I^{\prime \prime }`$) la composante connexe de $`JI`$ ร droite (resp. ร gauche) de $`I`$. Pour $`n`$ fixรฉ nous montrerons par rรฉcurrence sur $`k\{0,\mathrm{},n\}`$ que les deux propriรฉtรฉs suivantes sont simultanรฉment vรฉrifiรฉes :
$`(i)_k|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I^{})||g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I)|`$ ;
$`(ii)_ksup_{x,yII^{}}\frac{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(x)}{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(y)}\mathrm{exp}(2^\tau CM)`$.
La condition $`(ii)_0`$ est trivialement vรฉrifiรฉe, tandis que $`(i)_0`$ est satisfaite grรขce ร lโhypothรจse $`|I^{}|=2L|I|`$. Supposons que $`(i)_j`$ et $`(ii)_j`$ soient valables pour tout $`j\{0,\mathrm{},k1\}`$. Dans ce cas, pour tout $`x,y`$ dans $`II^{}`$ nous avons
$`\left|\mathrm{log}\left({\displaystyle \frac{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(x)}{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(y)}}\right)\right|`$ $``$ $`{\displaystyle \underset{j=0}{\overset{k1}{}}}\left|\mathrm{log}(g_{i_{j+1,n}}^{}(g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(x)))\mathrm{log}(g_{i_{j+1,n}}^{}(g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(y)))\right|`$
$``$ $`C{\displaystyle \underset{j=0}{\overset{k1}{}}}\left|g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(x)g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(y)\right|^\tau `$
$``$ $`C{\displaystyle \underset{j=0}{\overset{k1}{}}}\left(|g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(I)|+|g_{i_{j,n}}\mathrm{}g_{i_{1,n}}(I^{})|\right)^\tau `$
$``$ $`C\mathrm{\hspace{0.17em}2}^\tau M.`$
Ceci montre $`(ii)_k`$. Quant ร $`(i)_k`$, remarquons quโil existe $`xI`$ et $`yI^{}`$ tels que
$$|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I)|=|I|(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(x)\text{ et }|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I^{})|=|I^{}|(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(y).$$
Donc, par $`(ii)_k`$,
$$\frac{|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I^{})|}{|g_{i_{k,n}}\mathrm{}g_{i_{1,n}}(I)|}=\frac{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(x)}{(g_{i_{k,n}}\mathrm{}g_{i_{1,n}})^{}(y)}\frac{|I^{}|}{|I|}\mathrm{exp}(2^\tau CM)\frac{|I^{}|}{|I|}1,$$
ce qui montre $`(i)_k`$. Bien sรปr, un argument analogue montre que $`(i)_k`$ et $`(ii)_k`$ sont vรฉrifiรฉes pour tout $`k`$ dans $`\{0,\mathrm{},n\}`$ lorsquโon remplace $`I^{}`$ par $`I^{\prime \prime }`$.
Supposons maintenant que $`h_n(I)`$ soit contenu dans le $`L`$-voisinage de lโintervalle $`I`$. La propriรฉtรฉ $`(i)_n`$ donne alors $`h_n(J)J`$. De plus, si $`h_n(I)J`$ se trouve ร droite (resp. ร gauche) de $`I`$, alors $`h_n(II^{})I^{}`$ (resp. $`h_n(I^{\prime \prime }I)I^{\prime \prime }`$). Les deux cas รฉtant analogues, considรฉrons seulement le premier. Bien รฉvidemment, $`h_n`$ possรจde au moins un point fixe $`x`$ dans $`I^{}`$. Il nous reste donc ร vรฉrifier quโil sโagit dโun point fixe hyperbolique contractant. Or, il existe $`yI`$ tel que
$$h_n^{}(y)=\frac{|h_n(I)|}{|I|}\frac{L/2}{|I|}.$$
Par consรฉquent, si $`h_n^{}(x)1`$ alors on aurait $`h_n^{}(x)/h_n^{}(y)2|I|/L`$, et donc, dโaprรจs $`(ii)_n`$,
$$\mathrm{exp}(2^\tau CM)\frac{2|I|}{L},$$
ce qui contredirait la dรฉfinition de $`L`$. $`\mathrm{}`$
### 2.2 Preuve du thรฉorรจme de Denjoy gรฉnรฉralisรฉ
Avant de rentrer dans les dรฉtails de la dรฉmonstration du thรฉorรจme A, nous en donnons lโidรฉe essentielle. Supposons que $`I`$ soit un intervalle ouvert errant pour la dynamique de deux diffรฉomorphismes $`g_1`$ et $`g_2`$ du cercle qui sont de classe $`C^{1+\tau }`$ et qui commutent entre eux. Remarquons que lโensemble
$$\{g_1^mg_2^n(I):(m,n)_0\times _0,m+nk\}$$
contient exactemment $`(k+1)(k+2)/2`$ intervalles. Puisquโils sont deux ร deux disjoints, leur $`^{_{}}`$longueur typique $`^{_{}}`$ est de lโordre de $`1/k^2`$. Donc, pour une $`^{_{}}`$suite alรฉatoire typique $`^{_{}}`$ $`I,h_1(I),h_2(I)\mathrm{}`$, oรน $`h_{n+1}=g_1h_n`$ ou $`h_{n+1}=g_2h_n`$, on sโattend ร ce que, pour $`\tau >1/2`$,
$$\underset{k1}{}|h_k(I)|^\tau C\underset{k1}{}\frac{1}{k^{2\tau }}<\mathrm{}.$$
Or, la sรฉrie ร gauche est exactement celle dont la convergence permet de contrรดler les distorsions, et donc de trouver des รฉlรฉments avec des points fixes (hyperboliques), contredisant ainsi la libertรฉ de lโaction.
Afin de $`^{_{}}`$modeler $`^{_{}}`$ une preuve dans lโesprit de lโidรฉe ci-dessus, nous devons prรฉciser quelles sont nos $`^{_{}}`$suites alรฉatoires typiques $`^{_{}}`$. Pour cela, considรฉrons le processus de Markov sur $`_0\times _0`$ dont les probabilitรฉs de transition sont
$$p\left((m,n)(m+1,n)\right)=\frac{m+1}{m+n+2}\text{et}p\left((m,n)(m,n+1)\right)=\frac{n+1}{m+n+2}.$$
(10)
Ce processus markovien induit une mesure de probabilitรฉ $``$ sur lโespace de chemins correspondant $`\mathrm{\Omega }`$. On vรฉrifie aisรฉment que, en partant de lโorigine, la probabilitรฉ dโarriver en $`k`$ pas au point $`(m,n)`$ est รฉgale ร $`1/(k+1)`$ (resp. 0) si $`m+n=k`$ (resp. $`m+nk`$).
###### Remarque 2.3.
Il est intรฉressant de constater que les probabilitรฉs de passage (10) ci-dessus coรฏncident avec celles qui apparaissent dans le modรจle dโurne de Polya.
###### Remarque 2.4.
On peut canoniquement identifier lโespace $`(\mathrm{\Omega },)`$ ร lโintervalle unitรฉ (muni de la mesure de Lebesgue). Pour cela, ร chaque $`x[\mathrm{0,1}]`$ on associe le chemin dont la position au $`k`$-iรจme pas est รฉgale ร $`([kx],k[kx])`$. Avec cette identification, la propriรฉtรฉ dโรฉquidistribution prรฉcรฉdente devient complรจtement naturelle.
................................................Figure 2.................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................$`(\mathrm{0,0})`$$`p\left((m,n)(m,n+1)\right)\frac{1}{2}`$$`nm`$$`p\left((m,n)(m+1,n)\right)\frac{1}{2}`$$`mn`$$`1/2`$$`1/3`$$`1/4`$$`2/3`$$`2/4`$$`3/4`$$`1/2`$$`1/3`$$`1/4`$$`2/3`$$`2/4`$$`3/4`$$`g_1`$$`g_2`$$``$
Pour dรฉmontrer le thรฉorรจme A dans le cas $`d=2`$, nous procรฉdons par contradiction. Soient $`g_1`$ et $`g_2`$ deux diffรฉomorphismes du cercle de classe $`C^{1+\tau }`$ qui commutent. Le semigroupe $`\mathrm{\Gamma }^+`$ engendrรฉ par $`g_1`$ et $`g_2`$ sโidentifie ร $`_0\times _0`$. Par suite, le processus markovien dรฉcrit prรฉcรฉdemment induit une $`^{_{}}`$promenade alรฉatoire $`^{_{}}`$ sur $`\mathrm{\Gamma }^+`$. Dans ce qui suit nous identifierons $`\mathrm{\Omega }`$ ร lโespace des chemins correspondants sur $`\mathrm{\Gamma }^+`$. Pour tout $`\omega \mathrm{\Omega }`$ et tout $`n`$, dรฉsignons par $`h_n(\omega )\mathrm{\Gamma }^+`$ le produit des $`n`$ premiรจres cordonnรฉes de $`\omega `$. Autrement dit, pour $`\omega =(g_{i_1},g_{i_2},\mathrm{})\mathrm{\Omega }`$ notons $`h_n(\omega )=g_{i_n}\mathrm{}g_{i_1}`$ (oรน les $`g_{i_j}`$ ce sont des $`g_1`$ ou des $`g_2`$), et convenons que $`h_0(\omega )=id`$.
Si lโaction du groupe $`\mathrm{\Gamma }=g_1,g_2^2`$ est libre, alors les nombres de rotation $`\rho (g_1)`$ et $`\rho (g_2)`$ de $`g_1`$ et $`g_2`$ respectivement sont indรฉpendants sur les rationnels, dans le sens que pour tout $`(r_0,r_1,r_2)^3`$ distinct de $`(\mathrm{0,0,0})`$ on a $`r_1\rho (g_1)+r_2\rho (g_2)r_0`$. En effet, dans le cas contraire on pourrait trouver des รฉlรฉments non triviaux (et donc dโordre infini) avec un nombre de rotation rationnel ; de tels รฉlรฉments possรฉdant des points pรฉriodiques, ceci contredirait la libertรฉ de lโaction.
Supposons maintenant que lโaction de $`\mathrm{\Gamma }`$ soit (libre et) non minimale. Dans ce cas, il existe un ensemble de Cantor invariant et minimal pour lโaction. De plus, toute composante connexe $`I`$ du complรฉmentaire de cet ensemble est errante pour la dynamique, dans le sens que ses images par des รฉlรฉments distincts du groupe sont disjointes.
###### Lemme 2.5.
Si $`\tau >1/2`$ alors la valeur de la sรฉrie $`_{n0}|h_n(\omega )(I)|^\tau `$ est finie pour $``$-presque tout $`\omega \mathrm{\Omega }`$.
Dรฉmonstration. Pour tout chemin $`\omega \mathrm{\Omega }`$ dรฉfinissons
$$\mathrm{}_\tau (\omega )=\underset{k0}{}|h_k(\omega )(I)|^\tau .$$
Nous allons vรฉrifier que, si $`\tau >1/2`$, alors lโespรฉrance (par rapport ร $``$) de la fonction $`\mathrm{}_\tau `$ est finie, ce qui implique รฉvidemment lโaffirmation du lemme. Remarquons dโabord que
$$๐ผ(\mathrm{}_\tau )=๐ผ\left(\underset{k0}{}|h_k(\omega )(I)|^\tau \right)=\underset{k0}{}๐ผ\left(|h_k(\omega )(I)|^\tau \right)=\underset{k0}{}\underset{m+n=k}{}\frac{|g_1^mg_2^n(I)|^\tau }{k+1}.$$
Or, lโinรฉgalitรฉ de Hรถlder montre aisรฉment que
$$\underset{m+n=k}{}\frac{|g_1^mg_2^n(I)|^\tau }{k+1}\left(\underset{m+n=k}{}|g_1^mg_2^n(I)|\right)^\tau \left((k+1)\frac{1}{(k+1)^{1/(1\tau )}}\right)^{1\tau },$$
et donc,
$$๐ผ(\mathrm{}_\tau )\underset{k0}{}\frac{\left(_{m+n=k}|g_1^mg_2^n(I)|\right)^\tau }{(k+1)^\tau }.$$
En appliquant une nouvelle fois lโinรฉgalitรฉ de Hรถlder on obtient
$$๐ผ(\mathrm{}_\tau )\left[\underset{(m,n)_0\times _0}{}|g_1^mg_2^n(I)|\right]^\tau \left[\underset{k1}{}\left(\frac{1}{k^\tau }\right)^{\frac{1}{1\tau }}\right]^{1\tau }.$$
รtant donnรฉ que $`\tau >1/2`$, la sรฉrie
$$\underset{k1}{}\left(\frac{1}{k^\tau }\right)^{\frac{1}{1\tau }}=\underset{k1}{}\frac{1}{k^{\tau /(1\tau )}}$$
converge, et puisque les intervalles $`g_1^mg_2^n(I)`$ sont deux ร deux disjoints, ceci montre la finitude de $`๐ผ(\mathrm{}_\tau )`$. $`\mathrm{}`$
###### Remarque 2.6.
Dans la dรฉmonstration prรฉcรฉdente, la seule propriรฉtรฉ du processus de diffusion sur $`_0\times _0`$ que lโon a utilisรฉ est le fait que les probabilitรฉs dโarrivรฉe en $`k`$ pas sont รฉquidistribuรฉes sur lโensemble des points ร distance (simpliciale) $`k`$ de lโorigine.
Dโaprรจs le lemme prรฉcedent, si $`M`$ est suffisamment grand alors lโensemble $`\mathrm{\Omega }(M)=\{\omega \mathrm{\Omega }:\mathrm{}_\tau (\omega )M\}`$ possรจde une probabilitรฉ strictement positive (en fait, $`[\mathrm{\Omega }(M)]`$ converge vers $`1`$ lorsque $`M`$ tend vers lโin- fini). Fixons un tel $`M`$, et soit $`L=L(\tau ,M,|I|;\{g_1,g_2\})`$ la constante du lemme 2.2. Considรฉrons finalement lโintervalle ouvert $`K^{}`$ de taille $`|K^{}|=L`$ et adjacent ร droite ร $`I`$. Nous affirmons que
$$[\omega \mathrm{\Omega }:h_n(\omega )(I)K^{}\text{ pour tout }n]=0.$$
(11)
Pour dรฉmontrer (11) remarquons dโabord que lโaction du groupe engendrรฉ par les diffรฉomorphismes $`g_1`$ et $`g_2`$ est semiconjuguรฉe ร une action par des rotations. Par suite, si lโon $`^{_{}}`$รฉcrase $`^{_{}}`$ chaque composante connexe du complรฉmentaire du Cantor invariant minimal $`\mathrm{\Lambda }`$, alors on obtient un cercle topologique $`\mathrm{S}_\mathrm{\Lambda }^1`$ sur lequel $`g_1`$ et $`g_2`$ induisent des homรฉomorphismes dont toutes les orbites sont denses. De plus, lโintervalle $`K^{}`$ devient un intervalle $`U`$ dโintรฉrieur non vide dans $`\mathrm{S}_\mathrm{\Lambda }^1`$. Les nombres de rotation de $`g_1`$ et $`g_2`$ รฉtant irrationnels, il existe $`N`$ tel que, aprรจs รฉcrasement, $`g_1^1(U),\mathrm{},g_1^N(U)`$ recouvrent le cercle topologique $`\mathrm{S}_\mathrm{\Lambda }^1`$, et de mรชme pour $`g_2^1(U),\mathrm{},g_2^N(U)`$. Sur le cercle original $`\mathrm{S}^1`$ cela se traduit par le fait que, pour toute composante connexe $`I_0`$ de $`\mathrm{S}^1\mathrm{\Lambda }`$, il existe $`n_1`$ et $`n_2`$ dans $`\{1,\mathrm{},N\}`$ tels que $`g_1^{n_1}(I_0)K^{}`$ et $`g_2^{n_2}(I_0)K^{}`$.
Soulignons maintenant la propriรฉtรฉ รฉlรฉmentaire suivante et qui dรฉcoule directement de (10) : les probabilitรฉs de passage ร droite (resp. vers le haut) du processus markovien considรฉrรฉ sont $`1/2`$ au dessous (resp. au dessus) de la diagonale (voir la figure 2). Dโaprรจs la dรฉfinition de $`N`$, cette propriรฉtรฉ donne, pour tout entier $`k0`$,
$$\left[g_1^ih_k(\omega )(I)K^{}\text{ et }g_2^ih_k(\omega )(I)K^{}\text{ pour tout }iN|h_j(\omega )(I)K^{}\text{ pour tout }jk\right]1\frac{1}{2^N}.$$
Cette derniรจre inรฉgalitรฉ implique รฉvidemment que
$$\left[h_{k+i}(\omega )(I)K^{}\text{ pour tout }iN|h_j(\omega )(I)K^{}\text{ pour tout }jk\right]1\frac{1}{2^N}.$$
(12)
Par consรฉquent, pour tout $`r`$,
$$\left[h_n(\omega )(I)K^{}\text{ pour tout }n\right]\left[h_n(\omega )(I)K^{}\text{ pour tout }n\{1,\mathrm{},rN\}\right]\left(1\frac{1}{2^N}\right)^r,$$
dโoรน lโon obtient (11) en faisant tendre $`r`$ vers lโinfini.
Pour finir la preuve du thรฉorรจme A (toujours dans le cas $`d=2`$), remarquons que si $`\omega \mathrm{\Omega }(M)`$ et $`n`$ sont tels que $`h_n(\omega )(I)K^{}`$, alors le lemme 2.2 permet de trouver un point fixe hyperbolique pour $`h_n(\omega )\mathrm{\Gamma }^+`$, contredisant ainsi lโhypothรจse de libertรฉ de lโaction.
Modifications pour le cas $`๐>\mathrm{๐}`$. La dรฉmonstration du thรฉorรจme A pour $`d>2`$ est tout ร fait analogue ร celle donnรฉe pour le cas $`d=2`$. Elle se fait aussi par contradiction : on suppose lโexistence dโun intervalle errant et on considรจre le processus markovien sur $`_0^d`$ ร probabilitรฉs de transition
$$p\left((n_1,\mathrm{},n_i,\mathrm{},n_d)(n_1,\mathrm{}\mathrm{,1}+n_i,\mathrm{},n_d)\right)=\frac{1+n_i}{n_1+\mathrm{}+n_d+d}.$$
Les probabilitรฉs dโarrivรฉe en $`k`$ pas pour ce processus sont aussi รฉquidistribuรฉes sur lโensemble des points ร distance (simpliciale) $`k`$ de lโorigine. Cela permet ร nouveau de contrรดler les distortions pour presque toute suite alรฉatoire, cโest-ร -dire de dรฉmontrer un analogue du lemme 2.5 lorsque $`\tau >1/d`$. On remarque ensuite que chaque point $`(n_1,\mathrm{},n_d)`$ de $`_0^d`$ est le point de dรฉpart dโau moins une ligne droite telle les probabilitรฉs de passage entre deux sommets consรฉcutifs est supรฉrieure ou รฉgale ร $`1/d`$ (il suffit de suivre la direction de la coordonnรฉe $`i`$-รจme pour laquelle $`n_i`$ prend la valeur la plus grande). Cela permet dโobtenir une inรฉgalitรฉ analogue ร (12) (dont le membre ร droite sera รฉgal ร $`11/d^N`$ pour un certain entier $`N`$ assez grand). Une telle inรฉgalitรฉ entraรฎne la propriรฉtรฉ (11), qui grรขce au contrรดle de distorsion permet dโutiliser le lemme 2.2. On trouve ainsi des รฉlรฉments avec des points fixes hyperboliques, contredisant une nouvelle fois la libertรฉ de lโaction.
###### Remarque 2.7.
Dans la dรฉmonstration prรฉcรฉdente nous nโavons eu besoin de la finitude de la fonction $`\mathrm{}_\tau `$ que pour un ensemble de suites de mesure strictement positive. Nรฉanmoins, la mรฉthode du lemme 2.5 mรจne ร une conclusion beaucoup plus forte : lโespรฉrance de la fonction $`\mathrm{}_\tau `$ est finie dรจs que $`\tau >1/d`$. On pourrait donc imaginer quโen affaiblisant cette derniรจre affirmation on puisse attaquer le cas critique $`\tau =1/d`$ par des mรฉthodes analogues. On constatera cependant que pour la preuve du lemme 2.5 nous nous sommes appuyรฉs uniquemment sur le fait que $`I`$ รฉtait un intervalle errant. Or, cโest aussi le cas de lโintervalle $`I_{0,\mathrm{}\mathrm{,0}}`$ de lโexemple du ยง1.1, alors que pour cet exemple la valeur de la fonction $`\mathrm{}_{1/d}`$ est infinie pour toute suite $`\omega `$ (lorsque $`\epsilon d1`$). Ceci indique que pour dรฉmontrer le thรฉorรจme dans le cas critique, il est nรฉcessaire dโintroduire des mรฉthodes plus fines et qui prennent en compte la nature dynamique des intervalles $`g_1^mg_2^n(I)`$ (et non seulement le fait quโil soient deux ร deux disjoints). En effet, la preuve prรฉcรฉdente ne permet pas de conclure que (pour $`\epsilon d1`$) lโexemple du ยง1.1 nโest pas de classe $`C^{1+1/d}`$, alors que ceci est aisรฉment vรฉrifiable dโaprรจs les dรฉfinitions (et rรฉsulte รฉgalement de la proposition 1.1).
Nous voudrions conclure ce paragraphe en donnant le schรฉma dโune autre preuve du thรฉorรจme A dont lโidรฉe sera essentielle ร la fin du ยง3. Pour simplifier, nous ne considรฉrons que le cas $`d=2`$, et nous gardons les notations introduites tout au long de ce paragraphe. En raffinant lรฉgรจrement les arguments donnรฉs prรฉcรจ- demment, on dรฉmontre que pour presque toute suite $`\omega \mathrm{\Omega }`$ et pour tout point $`p`$ appartenant au cercle topolo- gique $`\mathrm{S}_\mathrm{\Lambda }^1`$, lโensemble des points de la forme $`h_n(\omega )(p)`$ est dense dans $`\mathrm{S}_\mathrm{\Lambda }^1`$. Fixons $`M>0`$ suffisamment grand de faรงon ร ce que $`[\mathrm{\Omega }(M)]>0`$, et comme dans la preuve du lemme 2.2 notons $`J=I^{}II^{\prime \prime }`$ le $`2L`$-voisinage de $`I`$. Choisissons un รฉlรฉment $`h\mathrm{\Gamma }`$ tel que $`h(I)`$ et $`h^1(I)`$ soient des intervalles de taille strictement infรฉrieure ร $`|I|\mathrm{exp}(2^\tau CM)`$ contenus dans $`I^{}`$ et $`I^{\prime \prime }`$ respectivement. Fixons un point arbitraire $`a\mathrm{\Lambda }`$, et supposons que $`h^{}(a)1`$ (le cas oรน $`h^{}(a)1`$ est rรฉglรฉ en appliquant les arguments qui suivent ร $`h^1`$ au lieu de $`h`$, tout en remarquant que dans ce cas $`(h^1)^{}(h(a))=1/h^{}(a)1`$ et $`h(a)\mathrm{\Lambda }`$). La premiรจre partie de la dรฉmonstration du lemme 2.2 montre que pour tout $`x`$ et $`y`$ appartenant ร $`\overline{I}^{}\overline{I}`$, tout $`\omega \mathrm{\Omega }(M)`$ et tout $`n`$,
$$\frac{h_n(\omega )^{}(x)}{h_n(\omega )^{}(y)}\mathrm{exp}(2^\tau CM).$$
ร partir de lโรฉgalitรฉ $`h=h_n(\omega )^1hh_n(\omega )`$ on en dรฉduit que, pour tout $`yI`$,
$$h^{}(y)=\frac{h_n(\omega )^{}(y)}{h_n(\omega )^{}(h(y))}h^{}\left(h_n(\omega )(y)\right)\mathrm{exp}(2^\tau CM)\underset{n}{lim\; sup}h^{}\left(h_n(\omega )(y)\right).$$
Or, puisque la suite $`\left(h_n(\omega )(p)\right)`$ est dense sur $`\mathrm{S}_\mathrm{\Lambda }^1`$ pour tout $`p\mathrm{S}_\mathrm{\Lambda }^1`$, il existe une suite croissante et infinie dโen- tiers $`n_k`$ tels que lโintervalle $`h_{n_k}(\omega )(I)`$ tend vers le point $`a`$. Par suite, $`lim\; sup_nh^{}\left(h_n(\omega )(y)\right)h^{}(a)1`$, et donc $`h^{}(y)\mathrm{exp}(2^\tau CM)`$ pour tout $`yI`$. Ceci implique que la taille de lโintervalle $`h(I)`$ est au moins รฉgale ร $`|I|\mathrm{exp}(2^\tau CM)`$, ce qui contredit notre choix de $`h`$.
## 3 Le lemme de Kopell gรฉnรฉralisรฉ
De maniรจre analogue ร ce que nous avons fait pour la gรฉnรฉralisation du thรฉorรจme de Denjoy, pour la preuve du thรฉorรจme B nous ne considรฉrerons que le cas $`d=2`$, et nous laisserons au lecteur le soin dโadapter nos arguments au cas $`d3`$. Tout en gardant les notations de lโรฉnoncรฉ du thรฉorรจme (avec $`d=2`$), identifions le semigroupe $`\mathrm{\Gamma }^+`$ engendrรฉ par les รฉlรฉments $`f_1`$ et $`f_2`$ de $`\mathrm{Diff}_+^{1+\tau }\left([\mathrm{0,1}]\right)`$ avec $`_0\times _0`$, et considรฉrons le processus markovien du ยง2.2. Si lโon fixe lโintervalle $`I=I_{\mathrm{0,0}}`$ et pour chaque $`\omega \mathrm{\Omega }`$ on dรฉfinit
$$\mathrm{}_\tau (\omega )=\underset{i0}{}|h_i(\omega )(I)|^\tau ,$$
alors lโargument de la preuve du lemme 2.5 montre que, lorsque $`\tau >1/2`$, la fonction $`\mathrm{}_\tau :\mathrm{\Omega }`$ est presque sรปrement finie (et quโen fait, son espรฉrance est finie).
Soit $`C`$ une constante de $`\tau `$-hรถlderianitรฉ pour $`\mathrm{log}(f_1^{})`$ et $`\mathrm{log}(f_2^{})`$. Pour chaque $`\omega =(f_{j_1},f_{j_2},\mathrm{})\mathrm{\Omega }`$, tout $`n,k`$ dans $``$ et tout $`xI`$, lโรฉgalitรฉ $`f_3^n=h_k(\omega )^1f_3^nh_k(\omega )`$ donne
$$\mathrm{log}\left((f_3^n)^{}(x)\right)=\mathrm{log}\left((f_3^n)^{}(h_k(\omega )(x))\right)+\underset{i=1}{\overset{k}{}}\left[\mathrm{log}\left(f_{j_i}^{}(h_{i1}(\omega )(x))\right)\mathrm{log}\left(f_{j_i}^{}(f_3^nh_{i1}(\omega )(x))\right)\right],$$
et donc
$`\left|\mathrm{log}\left((f_3^n)^{}(x)\right)\right|`$ $``$ $`\left|\mathrm{log}\left((f_3^n)^{}(h_k(\omega )(x))\right)\right|+C{\displaystyle \underset{i=1}{\overset{k}{}}}\left|h_{i1}(\omega )(x)f_3^nh_{i1}(\omega )(x)\right|^\tau `$
$``$ $`\left|\mathrm{log}\left((f_3^n)^{}(h_k(\omega )(x))\right)\right|+C{\displaystyle \underset{i=1}{\overset{k}{}}}|h_{i1}(\omega )(I)|^\tau .`$
En prenant $`\omega \mathrm{\Omega }`$ tel que $`\mathrm{}_\tau (\omega )=M`$ soit fini, lโinรฉgalitรฉ prรฉcรฉdente implique que
$$\left|\mathrm{log}\left((f_3^n)^{}(x)\right)\right|\left|\mathrm{log}\left((f_3^n)^{}(h_k(\omega )(x))\right)\right|+CM.$$
Or, le point $`h_k(\omega )(x)`$ converge nรฉcessairement vers un point fixe (parabolique) de $`f_3`$. En faisant tendre $`k`$ vers lโinfini on en dรฉduit que $`\left|\mathrm{log}\left((f_3^n)^{}(x)\right)\right|CM.`$ Par suite, $`(f_3^n)^{}(x)\mathrm{exp}(CM)`$ pour tout $`xI`$ et tout $`n`$, ce qui entraรฎne รฉvidemment que la restriction de $`f_3`$ ร lโintervalle $`I`$ est lโidentitรฉ. Par commutativitรฉ, ceci est aussi vrai sur tous les intervalles $`I_{n_1,n_2}`$, ce qui conclut la dรฉmonstration.
###### Remarque 3.1.
Une lecture attentive de la preuve prรฉcรฉdente montre que le thรฉorรจme B est encore valable pour des diffรฉomorphismes de classe $`C^{1+\tau }`$ de lโintervalle $`[\mathrm{0,1}[`$, i.e. pour des diffรฉomorphismes dont on ne dispose dโune borne uniforme pour la constante de $`\tau `$-hรถlderianitรฉ que sur chaque intervalle compact contenu dans $`[\mathrm{0,1}[`$.
Le lecteur pourrait รชtre incommodรฉ ร cause de lโhypothรจse combinatoire (1). Nรฉanmoins, nous avons dรฉjร expliquรฉ que les contre-exemples au thรฉorรจme B qui vรฉrifient cette condition correspondent aux actions de $`^{d+1}`$ sur $`[\mathrm{0,1}]`$ $`^{_{}}`$les plus intรฉressantes $`^{_{}}`$ du point de vue cohomologique. Ils portent aussi un interรชt dynamique, car il nโest pas difficile de construire des actions de $`^{d+1}`$ par diffรฉomorphismes de classe $`C^{2\epsilon }`$ de lโintervalle sans point global ร lโintรฉrieur. En effet, considรฉrons par exemple des diffรฉomorphismes $`f_2,\mathrm{},f_{d+1}`$ dโun intervalle $`[a,b]]\mathrm{0,1}[`$ dont les supports soient disjoints, et soit $`f_1`$ un diffรฉomorphisme de lโintervalle $`[\mathrm{0,1}]`$ dans lui mรชme sans point fixe ร lโintรฉrieur et qui envoie $`]a,b[`$ sur un intervalle disjoint. En รฉtendant $`f_2,\mathrm{},f_{d+1}`$ ร tout lโintervalle $`[\mathrm{0,1}]`$ de faรงon ร ce quโils commutent avec $`f_1`$, on obtient une action (fidรจle) de $`^{d+1}`$ par homรฉomorphismes de lโintervalle et sans point fixe global ร lโintรฉrieur. Bien sรปr, les mรฉthodes du ยง1.2 (resp. de ) permettent de rendre cette action de classe $`C^{3/2\epsilon }`$ (resp. $`C^{2\epsilon }`$) pour tout $`\epsilon >0`$.
Lโexemple naรฏf ci-dessus rend naturel le problรจme dโobtenir un rรฉsultat qui gรฉnรฉralise simultanรฉment les thรฉorรจmes A et B, et qui permette de dรฉcrire (toutes) les actions de $`^{d+1}`$ par diffรฉomorphismes du cercle de rรฉgularitรฉ intermรฉdiaire ; pour cela il faudrait รฉtudier les actions non libres mais sans point fixe global. Remarquons ร ce propos lโexistence dโun exemple intรฉressant : il sโagit dโune action de $`^2`$ par diffรฉomorphismes de classe $`C^{2\epsilon }`$ du cercle. Dans cet exemple, lโun des gรฉnรฉrateurs correspond ร un contre-exemple de Denjoy, alors que lโautre gรฉnรฉrateur laisse invariant le Cantor minimal du premier et agit non trivialement sur les composantes connexes de son complรฉmentaire. Pour rendre cette action de classe $`C^{2\epsilon }`$, on doit utiliser les mรฉthodes de .
Un autre aspect intรฉressant et important ร remarquer est lโexistence dโactions sur lโintervalle qui ร lโintรฉrieur sont libres mais non conjuguรฉes ร des actions par des translations. Ce phรฉnomรจne ne peut pas se produire en classe $`C^{1+vb}`$ (voir le lemme 3.2 de ). Cependant, en classe $`C^{3/2\epsilon }`$ on peut construire des actions de $`^2`$ sur $`[\mathrm{0,1}]`$ qui sont libres sur $`]\mathrm{0,1}[`$ mais qui admettent des intervalles errants, tout en utilisant les techniques du ยง1. Une nouvelle fois, la classe de diffรฉrentiabilitรฉ $`3/2\epsilon `$ est optimale pour de tels exemples. Le rรฉsultat suivant est ร rapprocher avec celui qui stipule lโexistence de champs de vecteurs associรฉs aux diffรฉomorphismes de classe $`C^2`$ de lโintervalle sans point fixe ร lโintรฉrieur .
###### Proposition 3.2.
Soit $`\mathrm{\Gamma }`$ un sous-groupe de $`\mathrm{Diff}_+^{1+\tau }([\mathrm{0,1}])`$ isomorphe ร $`^d`$, avec $`\tau >1/d`$ et $`d2`$. Si la restriction ร $`]\mathrm{0,1}[`$ de lโaction de $`\mathrm{\Gamma }`$ est libre, alors elle est minimale et topologiquement conjuguรฉe ร lโaction dโun groupe de translations.
Avant de passer ร la dรฉmonstration, rappelons quโun thรฉorรจme classique dรป ร Hรถlder stipule que tout groupe dโhomรฉomorphismes de lโintervalle $`]\mathrm{0,1}[`$ qui agit librement est topologiquement semi-conjuguรฉ ร un groupe de translations (voir pour une preuve complรจte de ce rรฉsultat). Lorsque lโaction est minimale, cette semi-conjugaison est forcรฉment une conjugaison ; dans le cas contraire, on voit apparaรฎtre des intervalles errants pour la dynamique.
Supposons maintenant que $`f_1,\mathrm{},f_d`$ soient les gรฉnรฉrateurs dโun groupe $`\mathrm{\Gamma }^d`$ qui agit librement sur $`]\mathrm{0,1}[`$ mais qui nโest pas conjuguรฉ ร un groupe de translations. Si lโon identifie les points des orbites par $`f_1`$, alors $`f_2,\mathrm{},f_d`$ deviennent les gรฉnรฉrateurs dโun groupe agissant librement sur le cercle et qui nโest pas conjuguรฉ ร un groupe de rotations ; bien sรปr, le thรฉorรจme A implique que les $`f_i`$ ne peuvent pas รชtre tous de classe $`C^{1/(d1)+\epsilon }`$. Notons que cet argument nโutilise que la diffรฉrentiabilitรฉ des applications ร lโintรฉrieur ; dans ce contexte la rรฉgularitรฉ de lโobstruction prรฉcรฉdente est en fait optimale. Nรฉanmoins, ce nโest pas le cas dans le cadre des groupes de diffรฉomorphismes de lโintervalle fermรฉ $`[\mathrm{0,1}]`$ (et mรชme de $`[\mathrm{0,1}[`$). Pour chercher ร dรฉmontrer cela nos devrons donc tenir compte de la rรฉgularitรฉ des applications aux extrรฉmitรฉs.
Dรฉmonstration de la proposition 3.2. Encore une fois, nous ne donnons la preuve que pour le cas $`d=2`$. Soient $`f_1`$ et $`f_2`$ les gรฉnรฉrateurs dโun groupe $`\mathrm{\Gamma }^2`$ de diffรฉomorphismes de classe $`C^{1+\tau }`$ de $`[\mathrm{0,1}]`$ qui agit librement ร lโintรฉrieur. Quitte ร les รฉchanger par leurs inverses, nous pouvons supposer quโils contractent topologiquement vers lโorigine. Supposons que lโaction de $`\mathrm{\Gamma }`$ sur $`]\mathrm{0,1}[`$ ne soit pas conjuguรฉe ร une action par des translations. Dans ce cas, un argument simple de contrรดle de distorsion hyperbolique montre que tous les รฉlรฉments de $`\mathrm{\Gamma }`$ doivent รชtre tangents ร lโidentitรฉ ร lโorigine. En effet, supposons par contradiction quโil existe des intervalles errants ainsi quโun รฉlรฉment $`f\mathrm{\Gamma }`$ tel que $`f^{}(0)<1`$. Fixons $`\lambda <1`$ et $`c]\mathrm{0,1}[`$ tels que $`f^{}(x)\lambda `$ pour tout $`x[0,c[`$, et fixons un intervalle errant ouvert et maximal $`I=]a,b[`$ contenu dans $`]0,c[`$. Si lโon dรฉsigne par $`K`$ lโintervalle $`[f(b),b]`$ alors
$$\underset{n0}{}|f^n(K)|^\tau |K|^\tau \underset{n0}{}\lambda ^{n\tau }=\frac{|K|^\tau }{1\lambda ^\tau }=\overline{M}.$$
Par consรฉquent, $`(g^n)^{}(x)/(g^n)^{}(y)\mathrm{exp}(C\overline{M})`$ pour tout $`x,y`$ dans $`K`$, oรน $`C>0`$ est une constante de $`\tau `$-hรถlderianitรฉ pour $`\mathrm{log}(f_1^{})`$ et $`\mathrm{log}(f_2^{})`$. Or, cette derniรจre estimรฉe permet dโappliquer les arguments de la preuve du lemme 3.2 de , en parvenant ainsi ร une contradiction<sup>3</sup><sup>3</sup>3Remarquons que dans cette partie de la preuve nous nโavons utilisรฉ que le fait que $`\tau >0`$. Une maniรจre plus conceptuelle dโexpliquer le phรฉnomรจne sous-jacent consiste ร rappeler que le thรฉorรจme de linรฉarisation de Sternberg est encore valable pour des germes hyperboliques et de classe $`C^{1+\tau }`$ de lโintervalle : la preuve donnรฉe dans en classe $`C^2`$ sโรฉtend aisรฉment dans ce contexte (voir aussi ). Or, le centralisateur dโun germe linรฉaire non trivial est le groupe ร un paramรจtre des germes linรฉaires ; en particulier, lโexistence dโintervalles errants pour la dynamique de sous-groupes denses de ce centralisateur est interdite..
Identifions maintenant le semigroupe $`\mathrm{\Gamma }^+`$ engendrรฉ par $`f_1`$ et $`f_2`$ ร $`_0\times _0`$, et considรฉrons le processus markovien du ยง2.2. Lorsque $`\tau >1/2`$, la preuve du lemme 2.5 montre la finitude de lโespรฉrance de la fonction
$$\omega \mathrm{}_\tau (\omega )=\underset{k0}{}|h_k(\omega )(I)|^\tau .$$
Prenons $`M>0`$ suffisamment grand de faรงon ร ce que lโensemble $`\mathrm{\Omega }(M)=\{\omega \mathrm{\Omega }:\mathrm{}_\tau (\omega )M\}`$ ait une probabilitรฉ strictement positive, et notons $`\overline{L}=|I|/\mathrm{exp}(2^\tau CM)`$. Dโaprรจs la premiรจre partie de la preuve du lemme 2.2, si $`I^{\prime \prime }`$ dรฉsigne lโintervalle adjacent ร gauche ร $`I`$ et de longueur $`\overline{L}`$, alors pour tout $`x`$ et $`y`$ appartenant ร $`J=\overline{I}^{\prime \prime }\overline{I}`$, tout $`\omega \mathrm{\Omega }(M)`$ et tout $`n`$,
$$\frac{h_n(\omega )^{}(x)}{h_n(\omega )^{}(y)}\mathrm{exp}(2^\tau CM).$$
(13)
Lโintervalle $`I`$ nโรฉtant pas contenu dans aucun autre intervalle errant et ouvert, il existe $`h\mathrm{\Gamma }`$ tel que $`h(I)I^{\prime \prime }`$ et $`|h(I)|<|I|\mathrm{exp}(2^\tau CM)`$. Fixons un point arbitraire $`y`$ dans $`I`$. Puisque $`h^{}(0)=1`$ et que $`h_n(\omega )(y)`$ tend vers lโorigine pour tout $`\omega \mathrm{\Omega }(M)`$, ร partir de lโรฉgalitรฉ
$$h^{}(y)=\frac{h_n(\omega )^{}(y)}{h_n(\omega )^{}(h(y))}h^{}\left(h_n(\omega )(y)\right)$$
et de (13) on conclut que $`h^{}(y)\mathrm{exp}(2^\tau CM).`$ Par suite,
$$|h(I)||I|\mathrm{exp}(2^\tau CM),$$
ce qui est en contradiction avec le choix de $`h`$. $`\mathrm{}`$
## 4 Le thรฉorรจme de Sacksteder gรฉnรฉralisรฉ
### 4.1 Pseudo-groupes de diffรฉomorphismes et feuilletages : le cas $`C^{1+\tau }`$
Dans ce paragraphe nous donnons la preuve du thรฉorรจme C en classe $`C^{1+\tau }`$. La cas $`C^1`$ nรฉcessite lโintroduction dโautres techniques, et il sera traitรฉ au paragraphe suivant.
Comme nous lโavons dรฉjร signalรฉ, la preuve du thรฉorรจme C se rรฉduit ร dรฉmontrer que, en prรฉsence dโune feuille ressort, il existe des feuilles ressort hyperboliques (voir lโappendice 5.1). En considรฉrant une transversale au feuilletage qui $`^{_{}}`$capture $`^{_{}}`$ cette feuille ressort topologique, nous sommes amenรฉs ร considรฉrer la dynamique sur un intervalle $`[a,a^{}]`$ de deux diffรฉomorphismes locaux $`f`$ et $`h`$ de classe $`C^{1+\tau }`$ qui vรฉrifient :
โ $`f`$ est defini sur tout lโintervalle $`[a,a^{}[`$ et il contracte topologiquement vers le point fixe $`a`$ ;
โ $`h`$ est defini sur un voisinage de ce point fixe et $`h(a)]a,a^{}[`$.
Posons $`c=h(a)`$ et fixons $`d^{}]c,a^{}[`$. Quitte ร remplacer $`f`$ par $`f^n`$ pour $`n`$ assez grand, nous pouvons supposer que $`f(d^{})<c`$, que $`f(d^{})`$ appartient au domaine de dรฉfinition de $`h`$, et que $`hf(d^{})]c,d^{}[`$. Cette derniรจre condition implique en particulier que $`hf`$ possรจde des points fixes dans $`]c,d^{}[`$. Soit $`d`$ le premier point fixe de $`hf`$ ร droite de $`c`$, et soit $`b=f(d)`$. Lโintervalle ouvert $`I=]b,c[`$ correspond au premier gap dโun ensemble de Cantor $`\mathrm{\Lambda }`$ qui est invariant par $`f`$ et $`g=hf`$ (voir la figure 3).
........................ ........................ ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ .........................................Figure 3$`a`$$`f(c)=b`$$`c=h(b)`$$`d`$$`f`$$`h`$$`I`$$`a`$$`b`$$`c`$$`d`$$`b`$$`c`$$`d`$$`I`$$`f`$$`g`$
Puisque les orbites du pseudo-groupe engendrรฉ par $`f`$ et $`g`$ sont denses sur $`\mathrm{\Lambda }`$, la preuve du thรฉorรจme se rรฉduit ร dรฉmontrer la proposition suivante.
###### Proposition 4.1.
Sous les hypothรจses prรฉcedentes, le pseudo-groupe engendrรฉ par $`f`$ et $`g`$ contient des รฉlรฉments avec des points fixes hyperboliques appartenant ร lโensemble $`\mathrm{\Lambda }`$.
Dรฉmonstration. Considรจrons lโespace $`\mathrm{\Omega }=\{f,g\}^{}`$ muni de la mesure de Bernoulli $``$ de poids $`(1/\mathrm{2,1}/2)`$. Pour chaque $`\omega =(g_1,g_2,\mathrm{})\mathrm{\Omega }`$ et chaque $`n`$ notons $`h_n(\omega )=g_n\mathrm{}g_1`$, et posons $`h_0(\omega )=id`$. Puisque les intervalles de lโensemble $`\{g_n\mathrm{}g_1(I):n0,(g_1,\mathrm{},g_n)\{f,g\}^n\}`$ sont deux ร deux disjoints,
$$\underset{n0}{}\underset{(g_1,\mathrm{},g_n)\{f,g\}^n}{}|g_n\mathrm{}g_1(I)|da<\mathrm{}.$$
Or, par le thรฉorรจme de Fubini,
$$\underset{n0}{}\underset{(g_1,\mathrm{},g_n)\{f,g\}^n}{}|g_n\mathrm{}g_1(I)|=\underset{n0}{}2^n\left(_\mathrm{\Omega }|h_n(\omega )(I)|๐(\omega )\right)=_\mathrm{\Omega }\left(\underset{n0}{}2^n|h_n(\omega )(I)|\right)๐(\omega ).$$
Donc, pour $``$-presque toute suite $`\omega \mathrm{\Omega }`$, on a la convergence de la sรฉrie
$$\underset{n0}{}2^n|h_n(\omega )(I)|.$$
Fixons $`\epsilon ]\mathrm{0,1}[`$, et pour chaque $`B>0`$ considรฉrons lโensemble $`\mathrm{\Omega }(B,\epsilon )`$ dรฉfini par
$$\mathrm{\Omega }(B,\epsilon )=\{\omega \mathrm{\Omega }:|h_n(\omega )(I)|B/(2\epsilon )^n\text{ pour tout }n0\}.$$
Dโaprรจs ce qui prรฉcรจde, $`[\mathrm{\Omega }(B,\epsilon )]`$ converge vers $`1`$ lorsque $`B`$ tend vers lโinfini. Nous pouvons en particulier fixer $`B`$ suffisamment grand de faรงon ร ce que $`[\mathrm{\Omega }(B,\epsilon )]>0`$. Remarquons que si $`\omega `$ appartient ร $`\mathrm{\Omega }(B,\epsilon )`$ alors
$$\underset{n0}{}|h_n(\omega )(I)|^\tau B^\tau \underset{n0}{}\frac{1}{(2\epsilon )^{n\tau }}=M<\mathrm{}.$$
(14)
Considรฉrons lโintervalle $`J^{}=[bL,c+L]`$ contenant lโintervalle errant $`I`$, oรน $`L=L(\tau ,M,|I|;\{f,g\})`$ est la constante qui apparaรฎt dans le lemme 2.2. Si $`N`$ est suffisamment grand alors $`f^Ng`$ et $`g^Nf`$ envoient tout lโintervalle $`[\mathrm{0,1}]`$ sur $`J^{}I`$. La loi $`01`$ de Borel donne ainsi
$$[h_n(\omega )(I)J^{}I\text{ une infinitรฉ de fois}]=1.$$
Si $`\omega \mathrm{\Omega }(B,\epsilon )`$ et $`n`$ satisfont $`h_n(\omega )(I)J^{}I`$, alors le lemme 2.2 montre que $`h_n(\omega )`$ possรจde un point fixe hyperbolique. Finalement, puisque lโensemble $`\mathrm{\Lambda }`$ est invariant par le pseudo-groupe et que le point fixe que lโon a trouvรฉ attire une partie de cet ensemble, ce point fixe appartient nรฉcessairement ร $`\mathrm{\Lambda }`$. $`\mathrm{}`$
###### Remarque 4.2.
Si lโon examine la preuve prรฉcรฉdente de plus prรจs, on sโaperรงoit que la mรชme technique permet de dรฉmontrer que pour presque tout $`\omega \mathrm{\Omega }`$ il existe une infinitรฉ dโentiers $`n`$ tels que $`h_n(\omega )`$ possรจde un point fixe hyperbolique $`p_n\mathrm{\Lambda }`$ de telle sorte que, asymptotiquement, la contraction $`h_n(\omega )^{}(p_n)`$ est au moins de lโordre de $`1/(2\epsilon )^n`$ pour tout $`\epsilon >0`$.
###### Remarque 4.3.
En classe $`C^{1+lip}`$, la proposition 4.1 dรฉcoule directement du thรฉorรจme de Sacksteder classique. Soulignons cependant que, dans ce contexte, on dispose dโune version beaucoup plus fine. En effet, considรฉrons un intervalle $`]\overline{a},\overline{d}[`$ contenant strictement $`[a,d]`$, et fixons une application $`T:[\overline{a},\overline{d}][\overline{a},\overline{d}]`$ de classe $`C^{1+lip}`$, nโayant que deux points critiques, et dont la restriction ร $`[a,b]`$ (resp. $`[c,d]`$) coรฏncide avec $`f^1`$ (resp. $`g^1`$). Il rรฉsulte alors du thรฉorรจme dโhyperbolicitรฉ de Maรฑรฉ que tous les points pรฉriodiques de $`T`$ appartenant ร lโensemble de Cantor $`\mathrm{\Lambda }`$ et dont la pรฉriode est suffisamment grande sont hyperboliques (dilatants). Or, pour tout $`n`$ la restriction de $`T^n`$ ร $`\mathrm{\Lambda }`$ coรฏncide localement avec des รฉlรฉments du pseudo-groupe engendrรฉ par $`f`$ et $`g`$.
Nous finissons ce paragraphe en donnant le schรฉma de la preuve dโune version raffinรฉe de lโun des rรฉsultats contenus dans .
###### Proposition 4.4.
Soient $`\mathrm{\Gamma }_1`$ et $`\mathrm{\Gamma }_2`$ deux groupes de diffรฉomorphismes du cercle dont la classe de diffรฉrentiabi- litรฉ $`C^\alpha `$ est strictement supรฉrieure ร $`C^1`$. Supposons quโils ne possรจdent pas dโorbite finie et quโils ne soient pas semi-conjuguรฉs ร des groupes de rotations. Alors tout diffรฉomorphisme de classe $`C^1`$ qui conjugue $`\mathrm{\Gamma }_1`$ avec $`\mathrm{\Gamma }_2`$ est nรฉcessairement de classe $`C^\alpha `$.
Dรฉmonstration. Supposons que $`\phi `$ soit un diffeomorphisme de classe $`C^1`$ qui conjugue $`\mathrm{\Gamma }_1`$ avec $`\mathrm{\Gamma }_2`$. Les hypothรจses faites sur ces groupes impliquent, par des arguments bien connus (voir par exemple les appendices 5.1 et 5.2), quโils contiennent des รฉlรฉments vรฉrifiant les hypothรจses de la proposition 4.1. Par suite, ces groupes possรจdent des รฉlรฉments avec des points fixes hyperboliques. De plus, lorsquโil existe un minimal exceptionnel, de tels points fixes peuvent รชtre pris appartenant ร cet ensemble minimal. Le diffรฉomorphisme $`\phi `$ รฉtant de classe $`C^1`$, il envoie des points fixes hyperboliques sur des points fixes hyperboliques. En particulier, il conjugue les germes correspondants. Dโaprรจs le thรฉorรจme de linรฉarisation de Sternberg, lโapplication $`\phi `$ est localement un diffรฉomorphisme de classe $`C^\alpha `$ autour de ces points fixes (voir la note au bas de la page 17). Puisque lโensemble des points autour desquels $`\phi `$ est un diffรฉomorphisme local de classe $`C^\alpha `$ est invariant par $`\mathrm{\Gamma }_1`$, lorsque les orbites sont denses cela suffit pour dรฉmontrer que $`\phi `$ est un diffรฉomorphisme de classe $`C^\alpha `$ sur tout le cercle. Cโest le cas aussi lorsquโil y a un minimal exceptionnel, car toute orbite sโaccumule sur ce minimal. Ceci termine la dรฉmonstration. $`\mathrm{}`$
### 4.2 Pseudo-groupes de diffรฉomorphismes et feuilletages : le cas $`C^1`$
Mรชme en prรฉsence dโune dynamique dilatante, il est en gรฉnรฉral impossible de contrรดler les distorsions en classe $`C^1`$ (une illustration classique de ce fait apparaรฎt dans ). Nรฉanmoins, lorsquโon sait a priori quโil y a de lโhyperbolicitรฉ quelque part, des arguments รฉlรฉmentaires permettent dโen dรฉduire quโil y a de lโhyperbolicitรฉ partout : cโest lโidรฉe sous-jacente de la preuve que nous donnons dans la suite <sup>4</sup><sup>4</sup>4Il est possible que par une mรฉthode similaire on puisse traiter le cas critique (i.e. en classe $`C^{1+1/d}`$) de nos rรฉsultats pour des groupes abรฉliens de diffรฉomorphismes.. Une formulation plus conceptuelle de cette idรฉe (en termes dโexposants de Lyapunov et de mesures stationnaires) sera esentielle au ยง4.4.
Tout en gardant les notations du ยง4.1, fixons une fois pour toutes une constante $`\epsilon ]\mathrm{0,1}/3[`$. Nous savons que pour $``$-presque tout $`\omega \mathrm{\Omega }`$ il existe $`B=B(\omega )1`$ tel que
$$|h_n(\omega )(I)|\frac{B}{(2\epsilon )^n}\text{ pour tout}n0.$$
(15)
###### Lemme 4.5.
Il existe une constante $`\overline{C}`$ ne dรฉpendant que de $`f`$ et $`g`$ telle que, si $`\omega =(g_1,g_2,\mathrm{})\mathrm{\Omega }`$ satisfait (15), alors pour tout $`xI`$ et tout entier $`n0`$ on a
$$h_n(\omega )^{}(x)\frac{B\overline{C}}{(22\epsilon )^n}.$$
(16)
Dรฉmonstration. Fixons $`\epsilon _0>0`$ suffisamment petit de faรงon ร ce que pour tout $`y,z`$ dans $`[a,d]`$ ร distance infรฉrieure ou รฉgale ร $`\epsilon _0`$ on ait
$$\frac{f^{}(y)}{f^{}(z)}\frac{2\epsilon }{22\epsilon }\text{et}\frac{g^{}(y)}{g^{}(z)}\frac{2\epsilon }{22\epsilon }.$$
(17)
Il est facile de voir quโil existe $`N`$ tel que, pour tout $`\omega \mathrm{\Omega }`$ et tout $`i0`$, la taille de lโintervalle $`h_{N+i}(I)`$ est infรฉrieure ou รฉgale ร $`\epsilon _0`$. Nous affirmons alors que (16) a lieu pour $`\overline{C}=\mathrm{max}\{A,\overline{A}\}`$, oรน
$$A=\underset{xI,nN,\omega \mathrm{\Omega }}{sup}\frac{h_n(\omega )^{}(x)(22\epsilon )^n}{B},\overline{A}=\underset{x,yI,\omega \mathrm{\Omega }}{sup}\frac{h_N(\omega )^{}(x)}{h_N(\omega )^{}(y)|I|}\left(\frac{22\epsilon }{2\epsilon }\right)^N.$$
En effet, si $`nN`$ alors (16) a lieu ร cause de lโinรฉgalitรฉ $`\overline{C}A`$. Supposons donc $`n>N`$, et fixons $`y=y(n)I`$ tel que $`|h_n(\omega )(I)|=h_n(\omega )^{}(y)|I|`$. Si $`x`$ appartient ร $`I`$ alors la distance entre les points $`h_{N+i}(\omega )(x)`$ et $`h_{N+i}(\omega )(y)`$ est infรฉrieure ou รฉgale ร $`\epsilon _0`$ pour tout $`i0`$. Par (17),
$$\frac{h_n(\omega )^{}(x)}{h_n(\omega )^{}(y)}=\frac{h_N(\omega )^{}(x)}{h_N(\omega )^{}(y)}\frac{g_{N+1}^{}(h_N(\omega )(x))}{g_{N+1}^{}(h_N(\omega )(y))}\mathrm{}\frac{g_n^{}(h_{n1}(\omega )(x))}{g_n^{}(h_{n1}(\omega )(y))}\frac{h_N(\omega )^{}(x)}{h_N(\omega )^{}(y)}\left(\frac{2\epsilon }{22\epsilon }\right)^{nN},$$
et donc
$$h_n(\omega )^{}(x)\frac{h_N(\omega )^{}(x)}{h_N(\omega )^{}(y)}\frac{|h_n(\omega )(I)|}{|I|}\left(\frac{2\epsilon }{22\epsilon }\right)^{nN}\frac{h_N(\omega )^{}(x)}{h_N(\omega )^{}(y)}\frac{B}{|I|(2\epsilon )^n}\left(\frac{2\epsilon }{22\epsilon }\right)^{nN}\frac{B\overline{C}}{(22\epsilon )^n},$$
oรน la derniรจre inรฉgalitรฉ dรฉcoule de la condition $`\overline{C}\overline{A}`$. $`\mathrm{}`$
Pour obtenir de lโhyperbolicitรฉ au delร de lโintervalle $`I`$ on doit utiliser un argument $`^{_{}}`$dual $`^{_{}}`$ mais lรฉgรจrement plus รฉlaborรฉ que celui du lemme prรฉcรฉdent. Pour le formuler fixons une constante $`\epsilon _1>0`$ suffisamment petite de sorte que pour tout $`y,z`$ dans $`[a,d]`$ ร distance infรฉrieure ou รฉgale ร $`\epsilon _1`$ on ait
$$\frac{f^{}(y)}{f^{}(z)}\frac{22\epsilon }{23\epsilon }\text{et}\frac{g^{}(y)}{g^{}(z)}\frac{22\epsilon }{23\epsilon }.$$
(18)
###### Lemme 4.6.
Soient $`C1`$, $`\omega =(g_1,g_2,\mathrm{})\mathrm{\Omega }`$ et $`x[a,d]`$ tels que
$$h_n(\omega )^{}(x)\frac{C}{(22\epsilon )^n}\text{ pour tout}n0.$$
(19)
Alors pour tout point $`y[a,d]`$ ร distance infรฉrieure ou รฉgale ร $`\epsilon _1/C`$ de $`x`$ et tout $`n0`$ on a
$$h_n(\omega )^{}(y)\frac{C}{(23\epsilon )^n}.$$
(20)
Dรฉmonstration. La vรฉrification de lโinรฉgalitรฉ (20) se fait par rรฉcurrence. Pour $`n=0`$ elle a lieu ร cause de lโhypothรจse $`C1`$. Admettons quโelle soit valable pour tout $`j\{0,\mathrm{},n\}`$, et notons $`y_j=h_j(\omega )(y)`$ et $`x_j=h_j(\omega )(y)`$. Supposons que $`yx`$, lโautre cas รฉtant analogue. Chaque point $`y_j`$ appartient alors ร lโintervalle $`h_j(\omega )([x\epsilon _1/C,x])`$. Or, par hypothรจse de rรฉcurrence,
$$\left|h_j(\omega )([x\epsilon _1/C,x])\right|\frac{C}{(23\epsilon )^j}\left|[x\epsilon _1/C,x]\right|C\frac{\epsilon _1}{C}=\epsilon _1.$$
Par la dรฉfinition de $`\epsilon _1`$ on conclut que, pour tout $`jn`$ on a $`g_{j+1}^{}(y_j)g_{j+1}^{}(x_j)\left(\frac{22\epsilon }{23\epsilon }\right)`$. Donc, dโaprรจs lโhypothรจse (19),
$`h_{n+1}(\omega )^{}(y)`$ $`=`$ $`g_1^{}(y_0)\mathrm{}g_{n+1}^{}(y_n)g_1^{}(x_0)\mathrm{}g_{n+1}^{}(x_n)\left({\displaystyle \frac{22\epsilon }{23\epsilon }}\right)^{n+1}`$
$``$ $`{\displaystyle \frac{C}{(22\epsilon )^{n+1}}}\left({\displaystyle \frac{22\epsilon }{23\epsilon }}\right)^{n+1}{\displaystyle \frac{C}{(23\epsilon )^{n+1}}}.`$
Ceci achรจve la vรฉrification par rรฉcurrence de (20). $`\mathrm{}`$
Nous sommes maintenant en mesure de terminer la preuve du thรฉorรจme C en classe $`C^1`$. Pour cela remarquons que, dโaprรจs le lemme 16, si $`C`$ est suffisamment grand alors la probabilitรฉ de lโensemble
$$\mathrm{\Omega }(C)=\{\omega \mathrm{\Omega }:h_n(\omega )^{}(x)\frac{C}{(22\epsilon )^n}\text{pour tout }n0\text{ et tout }xI\}$$
est strictement positive. Fixons un tel $`C1`$ et notons $`L=\mathrm{min}\{\epsilon _1/2C,|I|/2\}`$. Si lโon dรฉsigne par $`J`$ le $`2L`$-voisinage de $`I`$, alors le lemme 20 entraรฎne que pour tout $`\omega \mathrm{\Omega }(C)`$, tout $`n0`$ et tout $`yJ`$,
$$h_n(\omega )^{}(y)\frac{C}{(23\epsilon )^n}.$$
(21)
Si lโon dรฉsigne par $`J^{}`$ le $`L`$-voisinage de $`I`$, alors $`\left[h_n(\omega )(I)J^{}I\text{ une infinitรฉ de fois}\right]=1.`$ Par consรฉquent, et รฉtant donnรฉ que $`\epsilon <1/3`$, il existe $`\omega \mathrm{\Omega }(C)`$ et $`m`$ tels que $`h_m(\omega )(I)J^{}I`$ et $`(23\epsilon )^m>C`$. Dโaprรจs (21), et puisque $`L|I|/2`$, cela implique que $`h_m(\omega )`$ envoie $`J`$ sur lโune des deux composantes connexes de $`JI`$ de telle maniรจre que $`h_m(\omega )`$ possรจde un point fixe dans cette composante connexe, lequel est nรฉcessairement hyperbolique (et appartient ร $`\mathrm{\Lambda }`$).
### 4.3 Groupes de diffรฉomorphismes du cercle : mรฉthode dรฉterministe
Avant de nous plonger dans la preuve du thรฉorรจme D, nous voudrions montrer que par des mรฉthodes $`^{_{}}`$dรฉterministes $`^{_{}}`$ on peut dรฉjร obtenir une amรฉlioration du thรฉorรจme de Sacksteder concernant le nombre de points fixes hyperboliques.
###### Proposition 4.7.
Soit $`\mathrm{\Gamma }`$ un sous-groupe de $`\mathrm{Diff}_+^{1+lip}(\mathrm{S}^1)`$. Si $`\mathrm{\Gamma }`$ ne prรฉserve aucune mesure de probabilitรฉ du cercle, alors il contient des รฉlรฉments avec (au moins) deux points fixes hyperboliques, lโun contractant et lโautre dilatant.
Remarquons que ce rรฉsultat est trรจs naturel, car le thรฉorรจme C fournit dรฉjร un รฉlรฉment avec un point fixe hyperbolique, et tout diffรฉomorphisme du cercle avec un tel point possรจde nรฉcessairement dโautres points fixes. La proposition ci-dessus stipule donc lโexistence dโรฉlรฉments pour lesquels au moins lโun de ces autres points fixes est hyperbolique. Pour aboutir ร une dรฉmonstration, nous chercherons ร vรฉrifier lโexistence dโune suite $`(h_n)`$ dโรฉlรฉments dans $`\mathrm{\Gamma }`$ telle que $`h_n`$ et $`h_n^1`$ satisfassent les hypothรจses du lemme 2.1.
.........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................$``$$``$$`a`$$`b`$$`I`$$`f^1`$$`f`$$`a^{}`$$`b^{}`$$`g`$$`g`$$`K`$...........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................Figure 4
###### Lemme 4.8.
Soit $`\mathrm{\Gamma }`$ un groupe de diffรฉomorphismes de classe $`C^{1+lip}`$ du cercle qui admet un ensemble invariant fermรฉ et non vide $`\mathrm{\Lambda }\mathrm{S}^1`$. Supposons quโil existe un รฉlรฉment $`f\mathrm{\Gamma }`$ et une composante connexe $`I=]a,b[`$ de $`\mathrm{S}^1\mathrm{\Lambda }`$ tels que pour certains points $`a^{}`$ et $`b^{}`$ dans $`\mathrm{S}^1[a,b]`$ on ait $`lim_n\mathrm{}f^n(x)=a`$ pour tout $`x]a^{},a[`$ et $`lim_n\mathrm{}f^n(y)=b`$ pour tout $`y]b,b^{}[`$. Supposons aussi que $`\mathrm{\Gamma }`$ contienne un รฉlรฉment $`g`$ tel que $`g(I)]a^{},a[`$ et $`g^1(I)]b,b^{}[`$. Alors pour $`n`$ assez grand, lโรฉlรฉment $`f^ng\mathrm{\Gamma }`$ possรจde un point fixe hyperbolique contractant et un autre point fixe hyperbolique dilatant.
Dรฉmonstration. Par hypothรจse, pour $`\epsilon >0`$ assez petit les intervalles $`]a,a+\epsilon [=J,g(J),fg(J),\mathrm{},f^ng(J)`$ sont deux ร deux disjoints et convergent vers lโextrรฉmitรฉ gauche de $`I`$ (voir la figure 4). Dโaprรจs le lemme 2.1, pour $`n`$ assez grand lโapplication $`h_{n+1}=f^ng`$ possรจde un point fixe hyperbolique contractant proche de $`a`$. Considรฉrons maintenant lโintervalle $`K=g^1(]b\epsilon ,b[)`$. Encore par hypothรจse, si $`\epsilon >0`$ est assez petit alors les intervalles $`K,f^1(K),\mathrm{},f^n(K)`$ et $`g^1f^n(K)`$ sont deux ร deux disjoints et convergent vers lโextrรฉmitรฉ droite de $`K`$. Comme prรฉcรฉdemment, $`h_{n+1}^1=g^1f^n`$ possรจde un point fixe hyperbolique proche de (lโextrรฉmitรฉ droite de) $`K`$ dรจs que $`n`$ est assez grand. $`\mathrm{}`$
Si lโon est capable de sรฉparer les ensembles de points fixes de deux รฉlรฉments dโun sous-groupe $`\mathrm{\Gamma }`$ de $`\mathrm{Diff}_+^{1+lip}(\mathrm{S}^1)`$, alors le lemme prรฉcรฉdent permet dโobtenir des รฉlรฉments dans $`\mathrm{\Gamma }`$ avec deux points fixes hyperboliques.
###### Lemme 4.9.
Soit $`\mathrm{\Gamma }`$ un sous-groupe de $`\mathrm{Diff}_+^{1+lip}(\mathrm{S}^1)`$. Supposons que $`\mathrm{\Gamma }`$ contienne deux รฉlรฉments $`g_1`$ et $`g_2`$ qui possรจdent au moins deux points fixes et tels quโil existe deux intervalles ouverts $`U`$ et $`V`$ dans $`\mathrm{S}^1`$ contenant $`Fix(g_1)`$ et $`Fix(g_2)`$ respectivement et dont les fermetures sont disjointes. Alors $`\mathrm{\Gamma }`$ contient un รฉlรฉment avec un point fixe hyperbolique contractant et un autre point fixe hyperbolique dilatant.
Dรฉmonstration. Soient $`p_1,q_1`$ (resp. $`p_2,q_2`$) les points fixes de $`g_1`$ (resp. $`g_2`$) qui determinent un intervalle fermรฉ contenant $`Fix(g_1)`$ (resp. $`Fix(g_2)`$), et soient $`U=]u,u^{}[`$ et $`V=]v,v^{}[`$ les intervalles donnรฉs par lโhypothรจse. Quitte ร changer $`g_1`$ et $`g_2`$ par leurs inverses, et quitte ร prendre des itรฉrรฉs suffisamment grands, nous pouvons supposer que $`g_2([q_2,u^{}])]q_2,v^{}]`$ et $`g_1([v,p_1])]u,p_1]`$ (voir la figure 5). Soit $`\mathrm{\Lambda }`$ lโensemble fermรฉ non vide invariant et minimal pour le groupe engendrรฉ par $`g_1`$ et $`g_2`$. Un argument facile de type ping-pong montre que $`\mathrm{\Lambda }`$ nโest pas tout le cercle ; il est en fait homรฉomorphe ร lโensemble de Cantor.
Considรฉrons maintenant les composantes connexes $`I=]a,b[`$ et $`K`$ du complรฉmentaire de $`\mathrm{\Lambda }`$ qui contiennent $`]v^{},u[`$ et $`]u^{},v[`$ respectivement. Notons que $`g_1(K)=g_2(K)=I`$. En particulier, lโapplication $`f=g_2g_1^1`$ fixe $`a`$ et $`b`$. Nous affirmons que ce sont les seuls points fixes de $`f`$ sur $`[b,a]`$. En effet, on vรฉrifie aisรฉment que $`x<f(x)<a`$ pour tout $`x]b,a[`$. Pour finir la preuve de la proposition, il suffit de remarquer que pour $`g=g_2`$ et pour $`f`$, $`I`$ et $`K`$ comme prรฉcรฉdemment, les hypothรจses du lemme 4.8 sont satisfaites. $`\mathrm{}`$
..............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................$``$$``$$``$$``$$``$$``$$``$$``$...................................................................................$`I`$$`K`$$`f`$$`a`$$`b`$$`u`$$`v`$$`u^{}`$$`v^{}`$$`p_1`$$`q_1`$$`p_2`$$`q_2`$$`g_1`$$`g_2`$......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................Figure 5
Pour finir la preuve de la proposition 4.7 on pourrait essayer de trouver deux รฉlรฉments vรฉrifiant les hypothรจses du lemme prรฉcรฉdent. Nรฉanmoins, mรชme pour des groupes qui ne prรฉservent pas de mesure de probabilitรฉ du cercle, de tels รฉlรฉments peuvent ne pas exister. Cโest le cas par exemple des revรชtements finis de sous-groupes non mรฉtabรฉliens de $`\mathrm{PSL}(2,)`$. Lโidรฉe consiste alors ร montrer que ces $`^{_{}}`$extensions finies $`^{_{}}`$ sont les seules obstructions ร la sรฉparation dโensembles de points fixes. Pour cela nous utilisons quelques arguments de la preuve de Ghys de lโalternative de Tits faible pour les groupes dโhomรฉomorphismes du cercle (voir lโappendice 5.2).
Dรฉmonstration de la proposition 4.7. Puisque $`\mathrm{\Gamma }`$ ne prรฉserve pas de mesure de probabilitรฉ, il nโa pas dโorbite finie. Donc, soit toutes ses orbites sont denses, soit il prรฉserve un ensemble fermรฉ et minimal homรฉomorphe ร lโensemble de Cantor. Dans la suite nous ne considรฉrerons que le cas minimal (i.e. lorsque toutes les orbites sont denses), et nous laisserons au lecteur le soin dโadapter les arguments ci-dessous au cas oรน il existe un ensemble minimal exceptionnel (pour cela on utilise la technique classique qui consiste ร $`^{_{}}`$รฉcraser $`^{_{}}`$ les composantes connexes du complรฉmentaire de cet ensemble).
Dโaprรจs le ยง5.2, il existe un revรชtement (topologique) fini $`\pi :\mathrm{S}^1\mathrm{S}^1/`$ tel que lโaction de $`\mathrm{\Gamma }`$ sur $`\mathrm{S}^1`$ induit une action de $`\mathrm{\Gamma }`$ par homรฉomorphismes du cercle topologique $`\mathrm{S}^1/`$. De plus, si lโon dรฉsigne par $`\widehat{\mathrm{\Gamma }}`$ le sous-groupe correspondant de $`\mathrm{Homeo}_+(\mathrm{S}^1/)`$, alors la propriรฉtรฉ dโexpansivitรฉ forte suivante est vรฉrifiรฉe : pour chaque sous-intervalle fermรฉ $`[a,b]`$ de $`\mathrm{S}^1/`$ il existe une suite $`(\widehat{h}_n)`$ dans $`\widehat{\mathrm{\Gamma }}`$ telle que $`\widehat{h}_n([a,b])`$ converge vers un seul point. On conclut en particulier quโil existe un รฉlรฉment non trivial $`\widehat{g}_1\widehat{\mathrm{\Gamma }}`$ possรฉdant au moins deux points fixes. En effet, en prenant $`x\mathrm{S}^1/`$, un petit intervalle $`]c,d[`$ contenant $`x`$, et $`y[c,d]`$, il existe une suite $`(\widehat{h}_n)`$ dans $`\widehat{\mathrm{\Gamma }}`$ telle que les complรฉmentaires de $`\widehat{h}_n([c,d])`$ convergent vers le point $`y`$, et ceci implique que pour $`n`$ assez grand lโรฉlรฉment $`\widehat{h}_n`$ a un point fixe dans $`]c,d[`$ et un autre point fixe proche de $`y`$.
Soit $`\widehat{U}`$ un intervalle ouvert contenant lโensemble des points fixes de $`\widehat{g}_1`$ et dont la fermeture ne soit pas tout le cercle $`\mathrm{S}^1/`$. En vertu des propriรฉtรฉs de minimalitรฉ et dโexpansivitรฉ forte de $`\widehat{\mathrm{\Gamma }}`$, il existe $`\widehat{h}\widehat{\mathrm{\Gamma }}`$ tel que $`\widehat{h}(\widehat{U})\widehat{U}=\mathrm{}`$. Soient $`\widehat{V}=\widehat{h}(\widehat{U})`$ et $`\widehat{g}_2=\widehat{h}\widehat{g}_1\widehat{h}^1`$. Tous les arguments topologiques de la preuve du lemme 4.9 peuvent รชtre appliquรฉs au sous-groupe de $`\mathrm{Homeo}_+(\mathrm{S}^1/)`$ engendrรฉ par (les itรฉrรฉs correspondants de) $`\widehat{g}_1`$ et $`\widehat{g}_2`$. Ceci donne en particulier un intervalle ouvert $`\widehat{I}`$ tel que les seuls points fixes de lโรฉlรฉment $`\widehat{g}_2\widehat{g}_1^1\widehat{\mathrm{\Gamma }}`$ dans le complรฉmentaire de $`\widehat{I}`$ ce sont ses extrรฉmitรฉsโฆ
Dรฉsignons par $`\kappa `$ le degrรฉ du revรชtement $`\pi :\mathrm{S}^1\mathrm{S}^1/`$, et notons $`\varphi `$ le morphisme correspondant de $`\mathrm{\Gamma }`$ vers $`\widehat{\mathrm{\Gamma }}`$. Considรฉrons deux รฉlรฉments $`g_1`$ et $`g_2`$ de $`\mathrm{\Gamma }`$ tels que $`\varphi (g_1)=\widehat{g}_1`$ et $`\varphi (g_2)=\widehat{g}_2`$, et soit $`f=(g_2g_1^1)^\kappa `$. Cet รฉlรฉment $`f\mathrm{\Gamma }`$ fixe toutes les prรฉimages de $`\widehat{I}`$ par $`\pi `$. Notons $`I=I_1,I_2,\mathrm{},I_\kappa `$ ces prรฉimages, tout en respectant leur ordre cyclique sur le cercle. Soit $`\overline{g}\mathrm{\Gamma }`$ tel que $`\varphi (\overline{g})=\widehat{g}_2`$, et soient $`g=\overline{g}^\kappa `$ et $`K=g^1(I)`$. Remarquons que $`K`$ coรฏncide avec la prรฉimage par $`\varphi `$ de $`\widehat{g}_2^\kappa (\widehat{I})`$ qui est situรฉe entre $`I=I_1`$ et $`I_2`$ (voir la figure 6 pour une illustration du cas $`\kappa =3`$). On vรฉrifie rapidement que les รฉlรฉments $`f`$ et $`g`$ de $`\mathrm{\Gamma }`$, ainsi que les intervalles $`I`$ et $`K`$, satisfont les hypothรจses du lemme 4.8, ce qui permet dโachever la dรฉmonstration. $`\mathrm{}`$
..................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................$``$$``$$``$$``$$``$$``$$`I=I_1`$$`I_2`$$`I_\kappa `$$`f^1`$$`f`$$`g`$$`g`$$`K`$......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................Figure 6
Pour finir ce paragraphe, remarquons que la preuve prรฉcรฉdente montre en fait que dans $`\mathrm{\Gamma }`$ il existe des รฉlรฉments qui possรจdent (au moins) $`\kappa (\mathrm{\Gamma })`$ points fixes hyperboliques contractants (resp. dilatants), oรน $`\kappa (\mathrm{\Gamma })`$ dรฉsigne le degrรฉ de $`\mathrm{\Gamma }`$ (voir le ยง5.2). Dans le paragraphe suivant, et en utilisant des mรฉthodes probabilistes, nous montrerons que, mรชme en classe $`C^1`$, il existe des รฉlรฉments avec exactement $`2\kappa (\mathrm{\Gamma })`$ points fixes hyperboliques (toujours sous lโhypothรจse de non existence de mesure de probabilitรฉ invariante).
### 4.4 Groupes de diffรฉomorphismes du cercle : mรฉthode probabiliste
Les techniques des paragraphes 4.1 et 4.3 (resp. des ยง4.2 et ยง4.3) permettent dรฉjร de dรฉmontrer que si $`\mathrm{\Gamma }`$ est un sous-groupe de $`\mathrm{Diff}_+^{1+\tau }(\mathrm{S}^1)`$ (resp. de $`\mathrm{Diff}_+^1(\mathrm{S}^1)`$) qui ne prรฉserve aucune mesure de probabilitรฉ du cercle, alors il contient des รฉlรฉments avec (au moins) $`2\kappa (\mathrm{\Gamma })`$ points fixes hyperboliques. Nous en donnons un schรฉma de preuve, car ceci permettra dโillustrer la difficultรฉ de la dรฉmonstration du thรฉorรจme D.
Au cours de la preuve de la proposition 4.7, on sโest ramenรฉ ร รฉtudier une situation qui (ร indice fini prรจs) est illustrรฉe par la figure 4. Or, si lโon regarde lโaction de $`f`$ et $`g`$ sur lโintervalle $`[a^{},a]`$, on sโaperรงoit que leur dynamique est analogue ร celle qui est illustrรฉe par la figure 3. Les mรฉthodes du ยง4.1 (resp. du ยง4.2) permettent ainsi de trouver des รฉlรฉments avec des points fixes hyperboliques contractants. Pour trouver des points fixes hyperboliques dilatants, lโidรฉe consiste ร examiner lโaction de $`f^1`$ et $`g^1`$ sur lโintervalle $`[b^{},b]`$, laquelle aussi est analogue ร celle illustrรฉe par la figure 3. Nous sommes donc confrontรฉs ร la difficultรฉ dโรฉtudier simultanรฉment deux processus alรฉatoires, dont lโun va dans le $`^{_{}}`$sens inverse $`^{_{}}`$ de lโautre. Celui-ci nโest quโun problรจme de nature probabiliste et relativement simple du point de vue technique.
Le problรจme qui consiste ร trouver des รฉlรฉments avec exactement $`2\kappa (\mathrm{\Gamma })`$ points fixes hyperboliques est bien plus dรฉlicat. La stratรฉgie que nous proposons pour le rรฉsoudre diffรจre radicalement des mรฉthodes employรฉes dans les paragraphes prรฉcรฉdents. Pour sa mise en ลuvre nous aurons fortement besoin des rรฉsultats contenus dans lโappendice de ce travail, dont la lecture prรฉliminaire est fondamentale pour la comprรฉhension de ce qui suit.
Si $`\mathrm{\Gamma }`$ est un sous-groupe dรฉnombrable de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$ qui ne prรฉserve aucune mesure de probabilitรฉ du cercle, alors un argument simple montre lโexistence dโun sous-groupe de type fini de $`\mathrm{\Gamma }`$ qui ne fixe aucune probabilitรฉ sur $`\mathrm{S}^1`$. Donc, dans lโรฉnoncรฉ du thรฉorรจme D, nous pouvons supposer (sans perdre en gรฉnรฉralitรฉ) que $`\mathrm{\Gamma }`$ est de type fini. Considรฉrons une mesure de probabilitรฉ $`\mu `$ sur $`\mathrm{\Gamma }`$ qui soit non dรฉgรฉnรฉrรฉe, symรฉtrique et de support fini, et dรฉsignons par $`\mathrm{\Omega }`$ lโespace $`\mathrm{\Gamma }^{}`$ (muni de la mesure $`=\mu ^{}`$). Notons $`\sigma :\mathrm{\Omega }\mathrm{\Omega }`$ le dรฉcalage, et considรฉrons la transformation $`T:\mathrm{\Omega }\times \mathrm{S}^1\mathrm{\Omega }\times \mathrm{S}^1`$ dรฉfinie par $`T(\omega ,x)=(\sigma (\omega ),h_1(\omega )(x))`$. Soit $`\nu `$ une probabilitรฉ sur $`\mathrm{S}^1`$ stationnaire par rapport ร $`\mu `$, i.e. telle que la transformation $`T`$ prรฉserve la mesure de probabilitรฉ $`\times \nu `$. Le thรฉorรจme ergodique de Birkhoff donne lโexistence (presque partout) de la limite
$$\lambda _{(\omega ,x)}(\nu )=\underset{n\mathrm{}}{lim}\frac{\mathrm{log}\left(h_n(\omega )^{}(x)\right)}{n}.$$
Ce nombre est par dรฉfinition lโexposant de Lyapunov du point $`(\omega ,x)\mathrm{\Omega }\times \mathrm{S}^1`$.
Lorsque $`\mathrm{\Gamma }`$ ne prรฉserve pas de mesure de probabilitรฉ du cercle, la mesure stationnaire (par rapport ร $`\mu `$) est unique et ne possรจde pas dโatome ; par ailleurs, la transformation $`T`$ est ergodique. Lโexposant de Lyapunov est donc $`\times \nu `$-presque partout constant ; de plus, la valeur $`\lambda =\lambda (\nu )`$ de cette constante est strictement nรฉgative (voir les appendices 5.2, 5.3 et 5.4).
Pour dรฉmontrer le thรฉorรจme D nous examinons dโabord le cas le plus simple, ร savoir celui dโune action minimale qui vรฉrifie la propriรฉtรฉ dโexpansivitรฉ forte. Dans ce cas, nous savons dโaprรจs le ยง5.2 que pour tout $`\omega `$ appartenant ร un sous-ensemble $`\mathrm{\Omega }^{}`$ de probabilitรฉ totale de $`\mathrm{\Omega }`$, le coefficient de contraction $`c(h_n(\omega ))`$ converge vers zรฉro lorsque $`n`$ tend vers lโinfini. Cela se traduit par le fait que pour tout $`\omega \mathrm{\Omega }^{}`$ il existe des intervalles fermรฉs $`I_n(\omega )`$ et $`J_n(\omega )`$ dont la taille converge vers zรฉro et tels que $`h_n(\omega )\left(\overline{\mathrm{S}^1I_n(\omega )}\right)=J_n(\omega )`$. De plus, lโintervalle $`I_n(\omega )`$ converge vers un point $`\varsigma _+(\omega )`$ (et lโapplication $`\varsigma _+:\mathrm{\Omega }^{}\mathrm{S}^1`$ ainsi dรฉfinie est mesurable). Si $`I_n(\omega )`$ et $`J_n(\omega )`$ sont disjoints alors les points fixes de $`h_n(\omega )`$ sont inclus ร lโintรฉrieur de ces deux intervalles. Pour dรฉmontrer lโunicitรฉ et lโhyperbolicitรฉ dโau moins le point fixe contractant (i.e. celui situรฉ dans $`J_n(\omega )`$), nous aimerions nous servir du fait que lโexposant de Lyapunov de lโaction est strictement nรฉgatif (ce qui correspond ร une propriรฉtรฉ de contraction diffรฉrentielle au niveau local). Nรฉanmoins, on voit apparaรฎtre immรฉdiatement deux difficultรฉs techniques : les intervalles $`I_n(\omega )`$ et $`J_n(\omega )`$ ne sont pas toujours disjoints, et la rapiditรฉ de contraction diffรฉrentielle locale dรฉpend du point initial $`(\omega ,x)`$. Pour rรฉsoudre le premier problรจme il suffit de remarquer que les instants $`n`$ pour lesquels $`I_n(\omega )J_n(\omega )\mathrm{}`$ sont assez rares (en effet, la densitรฉ de cet ensemble dโentiers est gรฉnรฉriquement รฉgale ร zรฉro). La deuxiรจme difficultรฉe est surmontรฉe en utilisant le fait que la transformation $`T`$ est ergodique, ce qui implique que presque tout point initial $`(\omega ,x)`$ tombera au bout dโun certain moment sur un point oรน la rapiditรฉ de contraction est bien contrรดlรฉe. Soulignons finalement que, pour obtenir lโunicitรฉ et lโhyperbolicitรฉ du point fixe dilatant, il est nรฉcessaire dโutiliser un argument analogue ร celui qui prรฉcรจde mais pour les compositions des inverses et dans lโordre opposรฉ. Pour cela nous passons aux distributions en temps finis ; en effet, puisque la mesure de dรฉpart $`\mu `$ est symรฉtrique, en temps fini les distributions pour ces deux processus coรฏncident.
Pour mettre en ลuvre les idรฉes ci-dessus, commenรงons par fixer une constante $`C1`$ suffisamment grande de faรงon ร ce que lโensemble
$$E(C)=\{(\omega ,x)\mathrm{\Omega }\times \mathrm{S}^1:h_n(\omega )^{}(x)C\mathrm{exp}\left(\frac{2\lambda n}{3}\right)\text{ pour tout }n0\}$$
ait une mesure strictement positive <sup>5</sup><sup>5</sup>5ร lโaide du lemme classique de Pliss , on peut dรฉmontrer que la mesure de lโensemble $`E(1)`$ est dรฉjร strictement positive. Cependant, nous nโavons pas besoin de cela pour poursuivre notre argumentation.. Considรฉrons $`\overline{\epsilon }>0`$ tel que pour tout $`g`$ dans le support de $`\mu `$ et tout $`x,y`$ ร distance au plus $`\overline{\epsilon }`$ on ait
$$\frac{g^{}(x)}{g^{}(y)}\mathrm{exp}\left(\frac{\lambda }{6}\right).$$
La preuve du lemme suivant est analogue ร celle du lemme 20, et nous la laissons au lecteur.
###### Lemme 4.10.
Si $`(\omega ,y)`$ appartient ร $`E(C)`$ et $`x\mathrm{S}^1`$ est un point ร distance infรฉrieure ou รฉgale ร $`\epsilon =\overline{\epsilon }/C`$ de $`y`$, alors
$$h_n(\omega )^{}(x)C\mathrm{exp}\left(\frac{\lambda n}{2}\right)\text{ pour tout}n0.$$
Pour $`\delta ]\mathrm{0,1}[`$ posons
$$\alpha _\nu (\delta )=inf\{\nu (I):I\text{ intervalle de taille supรฉrieure ou รฉgale ร }\mathrm{\hspace{0.17em}1}\delta \}.$$
Puisque $`\nu `$ est sans atome, $`\alpha _\nu (\delta )`$ tend vers $`1`$ lorsque $`\delta `$ tend vers zรฉro. Pour tout $`\omega \mathrm{\Omega }`$ et chaque entier $`k>10`$ notons<sup>6</sup><sup>6</sup>6La condition $`k>10`$ est imposรฉe afin que les arguments qui suivent ne tombent pas en dรฉfaut par des raisons $`^{_{}}`$stupides $`^{_{}}`$ (impossibilitรฉ dโexistence dโintervalles avec les propriรฉtรฉs dรฉmandรฉes, etc).
$$(\omega ,k)=\{n:dist(\varsigma _+(\omega ),J_n(\omega ))4/k\},$$
et posons
$$\mathrm{\Omega }_{1/k}^{}=\{\omega \mathrm{\Omega }^{}:dens\left((\omega ,k)\right)\alpha _\nu (10/k)\},$$
$`dens`$ dรฉsigne la densitรฉ de lโensemble dโentiers strictement positifs correspondant<sup>7</sup><sup>7</sup>7Nous entendons par densitรฉ dโun sous-ensemble $`X`$ de $``$ la valeur
$$dens(X)=\underset{N\mathrm{}}{lim\; inf}\frac{card(X\{1,\mathrm{},N\})}{N}.$$
.
###### Lemme 4.11.
Pour chaque entier $`k>10`$ la probabilitรฉ de lโensemble $`\mathrm{\Omega }_{1/k}^{}`$ est totale.
Dรฉmonstration. Lorsque lโaction est minimale, le support de toute probabilitรฉ stationnaire est total. Par consรฉquent, dโaprรจs le thรฉorรจme ergodique de Birkhoff, pour chaque $`k>10`$ on peut fixer un point $`y_k`$ dans $`\mathrm{S}^1[\varsigma _+(\omega )5/k,\varsigma _+(\omega )+5/k]`$ de sorte que pour un sous-ensemble gรฉnรฉrique $`\overline{\mathrm{\Omega }}_{1/k}^{}`$ de $`\mathrm{\Omega }^{}`$ on ait
$$\underset{N\mathrm{}}{lim}\frac{card\{nN:h_n(p_k)[\varsigma _+(\omega )5/k,\varsigma _+(\omega )+5/k]\}}{N}=\nu \left(\mathrm{S}^1[\varsigma _+(\omega )5/k,\varsigma _+(\omega )+5/k]\right).$$
(22)
Pour chaque $`\omega \overline{\mathrm{\Omega }}_{1/k}^{}`$ fixons $`n(\omega )`$ de sorte que $`c\left(h_n(\omega )\right)1/k`$ et $`I_n(\omega )[\varsigma _+(\omega )1/k,\varsigma _+(\omega )+1/k]`$ pour tout $`nn(\omega )`$. Si $`nn(\omega )`$ alors $`y_k`$ est รฉvidemment dans $`\mathrm{S}^1I_n(\omega )`$, et donc $`h_n(\omega )(y_k)`$ appartient ร $`J_n(\omega )`$. Par suite, si $`nn(\omega )`$ est tel que $`h_n(\omega )(y_k)`$ nโest pas dans $`[\varsigma _+(\omega )5/k,\varsigma _+(\omega )+5/k]`$, alors $`dist(\varsigma _+(\omega ),J_n(\omega ))4/k`$. Dโaprรจs (22), ceci implique que lโensemble $`\overline{\mathrm{\Omega }}_{1/k}^{}`$ est contenu dans $`\mathrm{\Omega }_{1/k}^{}`$, ce qui permet de conclure la preuve du lemme. $`\mathrm{}`$
Puisque lโapplication $`T`$ est ergodique, si lโon dรฉsigne par $`\mathrm{\Omega }(C)`$ lโensemble des $`\omega \mathrm{\Omega }`$ tels que pour $`\nu `$-presque tout $`y\mathrm{S}^1`$ on a $`(\sigma ^n(\omega ),h_n(\omega )(y))E(C)`$ pour une infinitรฉ dโentiers $`n`$, alors la probabilitรฉ de $`\mathrm{\Omega }(C)`$ est totale. Par suite, pour tout entier $`k>10`$ lโensemble $`\mathrm{\Omega }_{1/k}^{}=\mathrm{\Omega }_{1/k}^{}\mathrm{\Omega }(C)`$ possรจde lui aussi une probabilitรฉ รฉgale ร $`1`$. Pour chaque $`\eta >0`$ dรฉsignons par $`D_\eta (\mathrm{S}^1)`$ lโensemble des diffรฉomorphismes $`g`$ de classe $`C^1`$ du cercle qui vรฉrifient la propriรฉtรฉ suivante : il existe deux intervalles fermรฉs $`I^{}`$ et $`J^{}`$ de taille au plus $`\eta `$ et ร distance au moins $`2\eta `$ entre eux tels que $`g`$ envoie $`\mathrm{S}^1I^{}`$ dans $`J^{}`$ et la contraction $`g:\mathrm{S}^1I^{}J^{}`$ est uniformรฉment hyperbolique (i.e. la dรฉrivรฉe de $`g`$ sur $`\overline{\mathrm{S}^1I^{}}`$ est strictement infรฉrieure ร $`1`$).
###### Lemme 4.12.
Si $`\omega `$ appartient ร $`\mathrm{\Omega }_{1/k}^{}`$ alors lโensemble $`^{}(\omega ,k)`$ des entiers $`n`$ tels que $`h_n(\omega )`$ appartient ร $`D_{1/k}(\mathrm{S}^1)`$ a une densitรฉ au moins รฉgale ร $`\alpha _\nu (10/k)`$.
Dรฉmonstration. Posons $`\epsilon _k=\mathrm{min}\{1/2k,\epsilon \}`$, et pour chaque $`\omega \mathrm{\Omega }_{1/k}^{}`$ prenons :
โ un entier positif $`n_0=n_0(\omega )`$ tel que $`c\left(h_n(\omega )\right)\epsilon _k`$ et $`I_n(\omega )[\varsigma _+(\omega )1/k,\varsigma _+(\omega )+1/k]`$ pour tout $`nn_0`$ ;
โ un point $`y`$ nโappartenant pas ร $`[\varsigma _+(\omega )1/k,\varsigma _+(\omega )+1/k]`$ et un entier positif $`n_1=n_1(\omega )n_0`$ tels que $`(\sigma ^{n_1}(\omega ),h_{n_1}(\omega )(y))E(C)`$ ;
โ un entier positif $`n_2=n_2(\omega )`$ tel que si lโon dรฉsigne par $`M`$ le supremum des valeurs de $`g^{}(z)`$ avec $`z\mathrm{S}^1`$ et $`g`$ dans le support de $`\mu `$, alors
$$C\mathrm{exp}\left(\frac{\lambda n_2}{2}\right)M^{n_1}<1.$$
Supposons que $`n`$ soit supรฉrieur ou รฉgal ร $`n_1+n_2`$ et quโil appartienne ร $`(\omega ,k)`$. Le point $`h_{n_1}(\omega )(y)`$ appartient ร lโintervalle $`J_{n_1}(\omega )`$, dont la taille nโest pas plus grande que $`\epsilon `$. Par le lemme 4.10, pour tout $`xJ_{n_1}(\omega )`$ et tout $`m0`$,
$$h_m\left(\sigma ^{n_1}(\omega )\right)^{}(x)C\mathrm{exp}\left(\frac{\lambda m}{2}\right).$$
En particulier, si $`mn_2`$ alors pour tout $`zI_{n_1}(\omega )`$ on a
$$h_{n_1+m}(\omega )^{}(z)=h_{n_1}(\omega )^{}(z)h_m\left(\sigma ^{n_1}(\omega )\right)^{}\left(h_{n_1}(\omega )(z)\right)M^{n_1}C\mathrm{exp}\left(\frac{\lambda m}{2}\right)<1.$$
Donc, si $`mn_2`$ est tel que $`n=n_1+m`$ appartient ร $`(\omega ,k)`$, alors :
โ la dรฉrivรฉe de $`h_n(\omega )`$ sur $`\overline{\mathrm{S}^1I_{n_1}(\omega )}`$ est bornรฉe par une constante strictement infรฉrieure ร $`1`$ ;
โ la taille de lโintervalle $`h_n(\omega )(I_{n_1}(\omega ))=h_m(\sigma ^{n_1}(\omega ))(J_{n_1}(\omega ))`$ est infรฉrieure ou รฉgale ร $`|J_{n_1}(\omega )|\epsilon _k`$ ;
โ on a $`h_n(\omega )\left(\overline{\mathrm{S}^1I_n(\omega )}\right)=J_n(\omega )`$.
Puisque les quatre intervalles concernรฉs ci-dessus ont une taille au plus รฉgale ร $`\epsilon _k1/2k`$, des arguments รฉlรฉmentaires montrent que $`I_{n_1}(\omega )`$ et $`I_n(\omega )`$ sโintersectent, et de mรชme pour $`J_n(\omega )`$ et $`h_n(\omega )(I_{n_1}(\omega ))`$ ; de plus si lโon dรฉsigne par $`I_n^{}(\omega )`$ (resp. $`J_n^{}(\omega )`$) lโintervalle $`I_{n_1}(\omega )I_n(\omega )`$ (resp. $`J_n(\omega )h_n(\omega )(I_{n_1}(\omega ))`$), alors $`I_n^{}(\omega )`$ et $`J_n^{}(\omega )`$ ont une taille au plus รฉgale ร $`1/k`$, ils sont ร une distance supรฉrieure ou รฉgale ร $`2/k`$ et $`h_n(\omega )\left(\mathrm{S}^1I_n^{}(\omega )\right)J_n^{}(\omega )`$.
Nous avons donc montrรฉ que tout entier $`nn_1(\omega )+n_2(\omega )`$ dans $`(\omega ,k)`$ appartient ร $`^{}(\omega ,k)`$. Puisque la densitรฉ de $`(\omega ,k)`$ est supรฉrieure ou รฉgale ร $`\alpha _\nu (10/k)`$, ceci conclut la preuve du lemme. $`\mathrm{}`$
Fixons maintenant une constante $`\rho >1/2`$ et un entier $`k_0>10`$ tels que $`\alpha _\nu (10/k_0)>\rho `$. Prenons une autre constante $`\varrho ]\mathrm{0,1}[`$ telle que $`\rho \varrho >1/2`$, et pour chaque $`N`$ dรฉsignons par $`\mathrm{\Omega }(k_0,N)`$ lโensemble des $`\omega \mathrm{\Omega }_{1/k_0}^{}`$ tels que
$$\frac{card\{n\{1,\mathrm{},N\}:h_n(\omega )D_{1/k_0}(\mathrm{S}^1)\}}{N}\rho .$$
Bien รฉvidemment, pour $`N`$ suffisamment grand nous avons $`\left[\mathrm{\Omega }(k_0,N)\right]\varrho `$. Pour $`n`$ notons $`_n`$ la probabilitรฉ $`\mu ^n`$ dรฉfinie sur $`\mathrm{\Gamma }^n`$.
###### Lemme 4.13.
Si $`\left[\mathrm{\Omega }(k_0,N)\right]\varrho `$ alors il existe $`n\{1,\mathrm{},N\}`$ tel que
$$_n[(g_1,\mathrm{},g_n):g_n\mathrm{}g_1D_{1/k_0}(\mathrm{S}^1)]\rho \varrho .$$
(23)
Dรฉmonstration. Pour chaque $`n`$ notons $`\xi _n`$ la variable alรฉatoire dรฉfinie par
$$\xi _n(\omega )=๐ณ_{D_{1/k_0}(\mathrm{S}^1)}\left(h_n(\omega )\right),$$
$`๐ณ`$ dรฉsigne la fonction caractรฉristique. Un รฉlรฉment $`\omega \mathrm{\Omega }_{1/k_0}^{}`$ appartient ร $`\mathrm{\Omega }(k_0,N)`$ si et seulement si
$$\frac{1}{N}\underset{n=1}{\overset{N}{}}\xi _n(\omega )\rho .$$
Donc, si $`\left[\mathrm{\Omega }(k_0,N)\right]\varrho `$ alors
$$๐ผ\left(\frac{1}{N}\underset{n=1}{\overset{N}{}}\xi _n\right)\rho \varrho .$$
Par suite, il existe $`n\{1,\mathrm{},N\}`$ tel que $`๐ผ(\xi _n)\rho \varrho `$. Puisque
$$๐ผ(\xi _n)=_n[(g_1,\mathrm{},g_n):g_n\mathrm{}g_1D_{1/k_0}(\mathrm{S}^1)],$$
ceci achรจve la preuve du lemme. $`\mathrm{}`$
Rappelons que la mesure $`\mu `$ sur $`\mathrm{\Gamma }`$ est supposรฉe symรฉtrique. Par suite, la distribution dans $`\mathrm{\Gamma }`$ des รฉlรฉments de la forme $`g_n\mathrm{}g_1`$ obtenus ร partir dโรฉlรฉments $`(g_1,\mathrm{},g_n)\mathrm{\Gamma }^n`$ est exactement la mรชme que celle des $`g_1^1\mathrm{}g_n^1`$. Donc, si $`n`$ vรฉrifie (23) alors
$$_n[(g_1,\mathrm{},g_n):g_n\mathrm{}g_1\text{ et }(g_n\mathrm{}g_1)^1\text{ appartiennent ร }D_{1/k_0}(\mathrm{S}^1)]2\rho \varrho 1>0,$$
La preuve du thรฉorรจme D (dans le cas minimal et fortement expansif) est alors conclue par le lemme suivant.
###### Lemme 4.14.
Soit $`g`$ un diffรฉomorphisme de classe $`C^1`$ du cercle. Si $`g`$ et $`g^1`$ appartiennent ร $`D_{1/k_0}(\mathrm{S}^1)`$ alors $`g`$ ne possรจde que deux points fixes, lโun hyperboliquement contractant et lโautre hyperboliquement dilatant.
Dรฉmonstration. Il est facile de voir que lโintersection des intervalles $`I^{}(g)`$ et $`J^{}(g^1)`$ est dโintรฉrieur non vide, et de mรชme pour $`J^{}(g)`$ et $`I^{}(g^1)`$. Puisque tous ces intervalles ont une taille au plus รฉgale ร $`1/k_0`$ et $`dist(I^{}(g),J^{}(g))2/k_0`$, les intervalles $`K(g)=I^{}(g)J^{}(g^1)`$ et $`K^{}(g)=J^{}(g)I^{}(g^1)`$ sont disjoints. De plus, $`g`$ envoie $`\mathrm{S}^1K(g)`$ dans $`K^{}(g)`$, et $`g^1`$ envoie $`\mathrm{S}^1K^{}(g)`$ dans $`K(g)`$. Puisque la dรฉrivรฉe de $`g`$ (resp. de $`g^1`$) dans $`\mathrm{S}^1K(g)`$ (resp. dans $`\mathrm{S}^1K^{}(g)`$) est strictement infรฉrieure ร $`1`$, ceci dรฉmontre le lemme. $`\mathrm{}`$
###### Remarque 4.15.
Une bonne lecture de la preuve prรฉcรฉdente permet de conclure que, lorsquโon fait des compositions alรฉatoires, trรจs probablement ร partir dโun certain moment on tombera assez souvent sur un diffรฉomorphisme avec deux points fixes hyperboliques.
Lโextension de la dรฉmonstration prรฉcรฉdente au cas minimal et expansif (mais non fortement expansif) ne prรฉsente pas de difficultรฉ majeure. En effet, nous savons dโaprรจs le ยง5.2 quโil existe un homรฉomorphisme $`\theta :\mathrm{S}^1\mathrm{S}^1`$ dโordre fini $`\kappa =\kappa (\mathrm{\Gamma })`$ et qui commute avec tous les รฉlรฉments de $`\mathrm{\Gamma }`$. De plus, si lโon dรฉsigne par $`\mathrm{S}^1/`$ le cercle topologique obtenu en identifiant les points des orbites de $`\theta `$ et par $`\widehat{\mathrm{\Gamma }}`$ le sous-groupe de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1/)`$ induit, alors lโaction de $`\widehat{\mathrm{\Gamma }}`$ est minimale et fortement expansive. Remarquons que $`\theta `$ nโest pas nรฉcessairement diffรฉrentiable. Cependant, tous les arguments topologiques de la preuve donnรฉe pour le cas minimal et fortement expansif sโappliquent ร lโaction de $`\widehat{\mathrm{\Gamma }}`$ sur $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1/)`$. Dโautre part, les arguments de nature diffรฉrentielle doivent รชtre lรปs dans lโaction originale de $`\mathrm{\Gamma }`$ sur $`\mathrm{S}^1`$. On dรฉmontre ainsi lโexistence dโun รฉlรฉment $`g=g_n\mathrm{}g_1`$ dans $`\mathrm{\Gamma }`$ tel quโil existe des intervalles disjoints $`K(g)`$ et $`K^{}(g)`$ dans $`\mathrm{S}^1/`$ qui se rรฉlรจvent respectivement en des intervalles $`K_1(g),\mathrm{},K_\kappa (g)`$ et $`K_1^{}(g),\mathrm{},K_\kappa ^{}(g)`$ qui sont deux ร deux disjoints et qui vรฉrifient :
โ $`g`$ envoie $`\mathrm{S}^1_{i=1}^\kappa K_i(g)`$ sur $`_{i=1}^\kappa K_i^{}(g)`$ avec dรฉrivรฉe strictement infรฉrieure ร $`1`$ ;
โ $`g^1`$ envoie $`\mathrm{S}^1_{i=1}^\kappa K_i^{}(g)`$ sur $`_{i=1}^\kappa K_i(g)`$ avec dรฉrivรฉe strictement infรฉrieure ร $`1`$.
Il est alors facile de sโapercevoir que lโรฉlรฉment $`g^\kappa \mathrm{\Gamma }`$ ne contient que $`2\kappa `$ points fixes, dont la moitiรฉ sont hyperboliquement contractants et lโautre moitiรฉ sont hyperboliquement dilatants.
###### Remarque 4.16.
Le dernier argument de la preuve prรฉcรฉdente (i.e. celui qui consiste ร remplacer $`g`$ par $`g^\kappa `$) nโest pas de nature probabiliste, dans le sens que si bien $`g`$ correspond ร un produit alรฉatoire gรฉnรฉrique dโรฉlรฉments de $`\mathrm{\Gamma }`$, lโรฉlรฉment $`g^\kappa \mathrm{\Gamma }`$ ne jouit pas de cette propriรฉtรฉ de gรฉnรฉricitรฉ. Il est donc naturel dโessayer de dรฉterminer la frรฉquence avec laquelle on voit apparaรฎtre des diffรฉomorphismes nโayant que des points fixes hyperboliques dans une suite alรฉatoire typique. Ce problรจme a une rรฉponse รฉvidente -cette frรฉquence est (gรฉnรฉriquement) รฉgale ร $`1/\kappa `$\- mais la dรฉmonstration de ce fait nโest pas complรฉtement รฉlรฉmentaire (elle utilise une loi de type $`02`$).
Le cas oรน il existe un minimal exceptionnel (et il nโy a pas de probabilitรฉ invariante) est plus compliquรฉ. En effet, dans cette situation le support de lโunique mesure stationnaire coรฏncide avec cet ensemble minimal $`\mathrm{\Lambda }`$ ; le fait que lโexposant de Lyapunov soit strictement nรฉgatif ne donne donc aucune information dโhyperbolicitรฉ loin de lโensemble $`\mathrm{\Lambda }`$.
En รฉcrasant les composantes connexes de $`\mathrm{S}^1\mathrm{\Lambda }`$ on obtient un cercle topologique $`\mathrm{S}_\mathrm{\Lambda }^1`$ muni dโune action (non nรฉcessairement fidรจle) de $`\mathrm{\Gamma }`$ par homรฉomorphismes (pour chaque $`g\mathrm{\Gamma }`$ nous noterons encore $`g`$ lโhomรฉomorphisme induit sur $`\mathrm{S}_\mathrm{\Lambda }^1`$). Si lโon dรฉsigne par $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ le sous-groupe correspondant de $`\mathrm{Homeo}_+(\mathrm{S}_\mathrm{\Lambda }^1)`$, alors lโaction de $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ est minimale et expansive. Pour simplifier, nous supposerons dans la suite que cette action est fortement expansive, en laissant le traitement du cas gรฉnรฉral au lecteur. Lโidรฉe de la preuve consiste alors ร appliquer les arguments topologiques du dรฉbut de ce paragraphe ร lโaction de $`\mathrm{\Gamma }_\mathrm{\Lambda }`$, et puis de lire les propriรฉtรฉs de nature diffรฉrentielle dans lโaction originale sur $`\mathrm{S}^1`$. Bien รฉvidemment, pour aboutir ร lโhyperbolicitรฉ nous devrons travailler $`^{_{}}`$suffisamment loin $`^{_{}}`$ des composantes connexes de $`\mathrm{S}^1\mathrm{\Lambda }`$ qui sont $`^{_{}}`$trop grandes $`^{_{}}`$.
Remarquons que le cercle topologique $`\mathrm{S}_\mathrm{\Lambda }^1`$ peut รชtre muni dโune structure mรฉtrique naturelle : on peut le paramรฉtrer en utilisant la mesure $`\nu `$. Nous entendons alors par $`^{_{}}`$taille $`^{_{}}`$ dโun intervalle $`V`$ de $`\mathrm{S}_\mathrm{\Lambda }^1`$ la valeur de $`\nu (V)`$, et par $`^{_{}}`$distance $`^{_{}}`$ entre deux points $`p`$ et $`q`$ de $`\mathrm{S}_\mathrm{\Lambda }^1`$ la valeur $`dist_\mathrm{\Lambda }(p,q)=\mathrm{min}\{\nu \left([p,q]\right),\nu \left([q,p]\right)\}`$.
Notons encore une fois
$$E(C)=\{(\omega ,x)\mathrm{\Omega }\times \mathrm{S}^1:h_n(\omega )^{}(x)C\mathrm{exp}\left(\frac{2\lambda n}{3}\right)\text{ pour tout }n0\}$$
et fixons une constante $`C1`$ telle que $`\times \nu \left[E(C)\right]>0`$. Considรฉrons $`\overline{\epsilon }>0`$ tel que pour tout $`g`$ dans le support de $`\mu `$ et tout $`x,y`$ ร distance au plus $`\overline{\epsilon }`$ on ait $`g^{}(x)/g^{}(y)\mathrm{exp}(\lambda /6)`$, et posons $`\epsilon =\overline{\epsilon }/C`$. Pour chaque $`\eta >0`$ les composantes connexes de $`\mathrm{S}^1\mathrm{\Lambda }`$ de taille supรฉrieure ou รฉgale ร $`\eta /3`$ se projettent en des points distincts $`p_1,\mathrm{},p_{m(\eta )}`$ de $`\mathrm{S}_\mathrm{\Lambda }^1`$. Il nโest pas difficile de voir que si $`\eta \epsilon `$ est suffisamment petit (et strictement positif), alors il existe des constantes strictement positives $`\rho `$, $`\overline{\rho }`$, $`\varrho `$, $`\overline{\varrho }`$, $`\delta `$ et $`\overline{\delta }`$ telles que :
$`(i)`$$`1>\overline{\varrho }>\varrho >\overline{\rho }>\rho >1/2`$, $`\rho \varrho >1/2`$ ;
$`(ii)`$ si $`V`$ est un intervalle quelconque de $`\mathrm{S}_\mathrm{\Lambda }^1`$ de taille au plus $`\delta `$ et dont le centre est ร distance au moins $`(1\overline{\varrho })/2m(\eta )`$ de tous les $`p_i`$, alors la taille de la prรฉimage de $`V`$ dans $`\mathrm{S}^1`$ est infรฉrieure ou รฉgale ร $`\eta /2`$ ;
$`(iii)`$ si $`V`$ est un intervalle quelconque de $`\mathrm{S}_\mathrm{\Lambda }^1`$ de taille au moins รฉgale ร $`\overline{\delta }`$, alors la taille de la prรฉimage de $`V`$ dans $`\mathrm{S}^1`$ est supรฉrieure ou รฉgale ร $`2\eta `$ ;
$`(iv)`$ on a lโinรฉgalitรฉ $`\overline{\varrho }2(\delta +\overline{\delta })\overline{\rho }`$.
Fixons de telles constantes et pour chaque $`i\{1,\mathrm{},m(\eta )\}`$ dรฉsignons par $`U_i`$ (resp. $`U_i^+`$) lโintervalle fermรฉ dans $`\mathrm{S}_\mathrm{\Lambda }^1`$ de centre $`p_i`$ et de taille $`(1\overline{\varrho })/m(\eta )`$ (resp. de taille $`(1\overline{\varrho }+\delta )/m(\eta )`$). Bien รฉvidement,
$$\nu (\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i)=\overline{\varrho },\nu (\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i^+)=\overline{\varrho }\delta .$$
(24)
Pour chaque $`\omega \mathrm{\Omega }^{}`$ notons $`I_n(\omega )`$ et $`J_n(\omega )`$ les intervalles de $`\mathrm{S}_\mathrm{\Lambda }^1`$ donnรฉs par la propriรฉtรฉ dโexpansivitรฉ forte de lโaction de $`\mathrm{\Gamma }_\mathrm{\Lambda }`$ sur $`\mathrm{S}_\mathrm{\Lambda }^1`$, et dรฉsignons par $`\stackrel{~}{I}_n(\omega )`$ et $`\stackrel{~}{J}_n(\omega )`$ ses prรฉimages dans $`\mathrm{S}^1`$. Posons
$$_{}(\omega ,\eta )=\{n:\varsigma _+(\omega )\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i,|\stackrel{~}{I}_n(\omega )|\eta /2,|\stackrel{~}{J}_n(\omega )|\eta /2,dist(\stackrel{~}{I}_n(\omega ),\stackrel{~}{J}_n(\omega ))2\eta \},$$
$$\mathrm{\Omega }_{}(\eta )=\{\omega \mathrm{\Omega }^{}:dens\left(_{}(\omega ,\eta )\right)\overline{\varrho }\}.$$
###### Lemme 4.17.
La probabilitรฉ de lโensemble $`\mathrm{\Omega }_{}(\eta )`$ est supรฉrieure ou รฉgale ร $`\overline{\varrho }`$.
Dรฉmonstration. La distribution des points $`\varsigma _+(\omega )`$ coรฏncide avec $`\nu `$ (voir la preuve de la proposition 5.11 plus loin). Par suite,
$$[\omega \mathrm{\Omega }^{}:\varsigma _+(\omega )\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i]=\nu (\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i)=\overline{\varrho }.$$
Nous affirmons que presque tout $`\omega `$ dans lโensemble
$$\overline{\mathrm{\Omega }}_{}(\eta )=\{\omega \mathrm{\Omega }^{}:\varsigma _+(\omega )\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i\}$$
appartient ร $`\mathrm{\Omega }_{}(\eta )`$, ce qui implique รฉvidemment lโaffirmation du lemme. Pour vรฉrifier cela, notons dโabord que si $`\varsigma _+(\omega )`$ appartient ร $`\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i`$ alors $`I_n(\omega )`$ est contenu dans $`\mathrm{S}_\mathrm{\Lambda }^1_{i=1}^{m(\eta )}U_i`$ pour tout $`n`$ assez grand. Donc, dโaprรจs $`(ii)`$, on a $`|\stackrel{~}{I}_n(\omega )|\eta /2`$ pour tout $`n`$ suffisamment grand. Pour chaque $`\omega \mathrm{\Omega }^{}`$ dรฉsignons par $`V(\omega )`$ lโintervalle de centre $`\varsigma _+(\omega )`$ et de taille $`2\overline{\delta }+\delta `$. Dโaprรจs le thรฉorรจme ergodique de Birkhoff et (24), pour presque tout $`\omega \overline{\mathrm{\Omega }}_{}(\eta )`$ il existe $`p\mathrm{S}_\mathrm{\Lambda }^1[\varsigma _+(\omega )1/3,\varsigma _+(\omega )+1/3]`$ tel que
$$\underset{N\mathrm{}}{lim}\frac{card\{n\{1,\mathrm{},N\}:h_n(\omega )_{i=1}^{m(\eta )}U_i^+V(\omega )\}}{N}\overline{\varrho }2(\delta +\overline{\delta }).$$
(25)
Pour $`n`$ assez grand on a $`h_n(\omega )(p)J_n(p)`$ et $`\nu \left(J_n(\omega )\right)\delta `$, et donc la condition $`h_n(\omega )(p)_{i=1}^{m(\eta )}U_i^+`$ (resp. $`h_n(\omega )(p)V(\omega )`$) implique dโaprรจs $`(ii)`$ que $`|\stackrel{~}{J}_n(\omega )|\eta /2`$ (resp. implique dโaprรจs $`(iii)`$ que la valeur de $`dist(\stackrel{~}{I}_n(\omega ),\stackrel{~}{J}_n(\omega ))`$ est supรฉrieure ou รฉgale ร $`2\eta `$). Le lemme sโen suit alors de (25) et de la condition $`\overline{\varrho }2(\delta +\overline{\delta })\overline{\rho }`$. $`\mathrm{}`$
La dรฉmonstration continue de maniรจre analogue ร celle donnรฉe pour le cas minimal (et fortement expansif). On note $`\mathrm{\Omega }(C)`$ lโensemble des $`\omega \mathrm{\Omega }`$ tels que pour $`\nu `$-presque tout $`y\mathrm{S}^1`$ on ait $`(\sigma ^n(\omega ),h_n(\omega )(y))E(C)`$ pour une infinitรฉ dโentiers $`n`$, et on considรจre lโensemble $`\mathrm{\Omega }_{}(\eta )=\mathrm{\Omega }_{}(\eta )\mathrm{\Omega }(C)`$ (dont la probabilitรฉ est encore supรฉrieure ou รฉgale ร $`\overline{\varrho }`$). Si lโon tient compte du fait que $`\eta \epsilon `$ (et que la conclusion du lemme 4.10 est donc valable), alors par des arguments analogues ร ceux du lemme 4.12 on constate aisรฉment que pour tout $`\omega \mathrm{\Omega }_{}(\eta )`$ la densitรฉ des entiers $`n`$ tels que $`h_n(\omega )`$ appartient ร $`D_\eta (\mathrm{S}^1)`$ est supรฉrieure ou รฉgale ร $`\overline{\rho }`$. Par suite, pour $`N`$ suffisamment grand on a
$$[\omega \mathrm{\Omega }_{}(\eta ):\frac{card\{n\{1,\mathrm{},N\}:h_n(\omega )D_\eta (\mathrm{S}^1)\}}{N}\rho ]\varrho .$$
La preuve du lemme 23 montre alors quโil existe $`n\{1,\mathrm{},N\}`$ tel que
$$_n[(g_1,\mathrm{},g_n):g_n\mathrm{}g_1D_\eta (\mathrm{S}^1)]\rho \varrho >1/2.$$
En passant aux inverses on conclut que
$$_n[(g_1,\mathrm{},g_n):g_n\mathrm{}g_1\text{ et }(g_n\mathrm{}g_1)^1\text{ appartiennent ร }D_\eta (\mathrm{S}^1)]2\rho \varrho 1>0,$$
ce qui permet de conclure la dรฉmonstration par un argument analogue ร celui du lemme 4.14.
## 5 Appendice
### 5.1 Une alternative topologique pour des pseudo-groupes dโhomรฉomorphismes
Soient $`X`$ une variรฉtรฉ unidimensionnelle compacte, $`\mathrm{\Gamma }`$ un pseudo-groupe dโhomรฉomorphismes de $`X`$ et $`\mathrm{\Lambda }`$ un sous-ensemble de $`X`$ qui est compact, invariant par $`\mathrm{\Gamma }`$, non vide et minimal. Lโorbite $`O(x)`$ dโun point $`xX`$ est dite de type ressort si dโune part il existe un รฉlรฉment $`g\mathrm{\Gamma }`$ dont le domaine de dรฉfinition $`dom(g)`$ contient un demi-intervalle $`[x,p[`$ ou $`]p,x]`$ pour lequel $`x`$ est un point fixe attractif topologique (i.e. $`g^n(p)`$ converge vers $`x`$ quand $`n`$ tend vers lโinfini), et dโautre part $`O(x)`$ intersecte ce demi-intervalle. La proposition suivante sโappuie fortement sur la partie moins connue (et pourtant trรจs intรฉressante) de lโarticle de Sacksteder .
###### Proposition 5.1.
Avec les notations prรฉcรฉdentes, soit il existe une mesure de probabilitรฉ supportรฉe sur $`\mathrm{\Lambda }`$ et invariante par tous les รฉlรฉments de $`\mathrm{\Gamma }`$, soit il existe dans $`\mathrm{\Lambda }`$ une orbite de type ressort.
Pour dรฉmontrer cette proposition, commenรงons par remarquer quโil y a trois possibilitรฉs pour lโensemble minimal $`\mathrm{\Lambda }`$ : soit il est fini, soit cโest un ensemble de Cantor, soit $`\mathrm{\Lambda }=X`$. Ceci se montre en remarquant que la frontiรจre $`Fr(\mathrm{\Lambda })`$ et lโensemble $`\mathrm{\Lambda }^{ac}`$ des points dโaccumulation de $`\mathrm{\Lambda }`$ sont des fermรฉs invariants contenus dans $`\mathrm{\Lambda }`$. Donc soit ils sont vides, soit il sont รฉgaux ร $`\mathrm{\Lambda }`$. En รฉtudiant toutes les possibilitรฉs on dรฉmontre le fait escomptรฉ.
Dans le cas oรน $`\mathrm{\Lambda }`$ est fini, la moyenne des masses de Dirac aux points de $`\mathrm{\Lambda }`$ est une mesure invariante par $`\mathrm{\Gamma }`$. Dans le cas oรน $`\mathrm{\Lambda }=X`$, la proposition 5.1 a รฉtรฉ dรฉmontrรฉe par Sacksteder (voir le lemme 9.1 de ). Notre dรฉmonstration pour le cas oรน $`\mathrm{\Lambda }`$ est exceptionnel (i.e. cโest un ensemble de Cantor) suit sa stratรฉgie : nous dรฉfinissons une notion dโรฉquicontinuitรฉ pour la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ et puis nous prouvons que si cette restriction nโest pas รฉquicontinue, alors il existe une orbite de type ressort dans $`\mathrm{\Lambda }`$. De plus, dans le cas รฉquicontinu nous dรฉmontrons lโexistence dโune mesure de probabilitรฉ invariante par $`\mathrm{\Gamma }`$ et supportรฉe sur $`\mathrm{\Lambda }`$. La subtilitรฉ de la preuve vient de la dรฉfinition suivante.
###### Dรฉfinition 5.2.
Fixons une mesure de probabilitรฉ $`\nu `$ sur $`\mathrm{\Lambda }`$ qui soit sans atome et dont le support coรฏncide avec tout lโensemble $`\mathrm{\Lambda }`$. Nous dirons que la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est รฉquicontinue si pour tout $`\epsilon >0`$ il existe $`\delta >0`$ tel que pour tout intervalle $`I`$ de $`X`$ de mesure $`\nu (I)`$ infรฉrieure ร $`\delta `$ et pour tout รฉlรฉment $`g`$ de $`\mathrm{\Gamma }`$ dont le domaine de dรฉfinition contient $`I`$, on a $`\nu \left(g(I)\right)<\epsilon `$.
Il est facile de vรฉrifier que cette notion dโรฉquicontinuitรฉ ne dรฉpend pas du choix de la mesure $`\nu `$.
###### Lemme 5.3.
Si la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est non รฉquicontinue alors $`\mathrm{\Lambda }`$ contient une orbite de type ressort.
Dรฉmonstration. Supposons que la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ ne soit pas รฉquicontinue. Il existe donc $`\epsilon _0>0`$, une suite dโintervalle $`I_n`$ dont les mesures tendent vers $`0`$, et une suite dโรฉlรฉments $`g_n\mathrm{\Gamma }`$ dont les domaines de dรฉfinition contiennent $`I_n`$, tels que $`\nu \left(g_n(I_n)\right)\epsilon _0`$. Supposons que les longueurs des intervalles $`I_n`$ soient minorรฉes par une constante strictement positive indรฉpendante de $`n`$. Dans ce cas, nous pouvons extraire de la suite des $`I_n`$ une sous-suite dโintervalles qui converge vers un intervalle $`I`$. Puisque $`\nu `$ nโa pas dโatome, la mesure de $`I`$ est nulle, ce qui signifie que $`I`$ est contenu dans une composante de $`X\mathrm{\Lambda }`$. Pour tout $`n`$ lโextรฉrieur de $`I`$ dans $`I_n`$ est formรฉ dโau plus deux intervalles $`J_n^1`$ et $`J_n^2`$ dont les longueurs tendent vers $`0`$. De plus, nous avons
$$\nu \left(g_n(J_n^1J_n^2)\right)\epsilon _0.$$
Ainsi, lโun dโentre eux, par exemple $`J_n^1`$, vรฉrifie $`\nu \left(g_n(J_n^1)\right)\epsilon _0/2`$, et sa longueur tend vers $`0`$ lorsque $`n`$ tend vers lโinfini. Si les longueurs des $`I_n`$ ne sont pas minorรฉes par une constante strictement positive, on peut en extraire une sous-suite dโintervalles dont les longueurs tendent vers $`0`$ et tels que leurs images respectives sont de mesure supรฉrieure ou รฉgale ร $`\epsilon _0`$. Dans tous les cas, nous avons trouvรฉ une suite dโintervalles $`J_m`$ et des รฉlรฉments $`g_m`$ de $`\mathrm{\Gamma }`$ dรฉfinis sur $`J_m`$ tels que
$$\underset{m\mathrm{}}{lim}|J_m|=0\text{ et }\nu \left(g_m(J_m)\right)\epsilon _0/2\text{pour tout }m.$$
De plus, ces intervalles $`J_m`$ peuvent รชtre pris fermรฉs et avec leurs extrรฉmitรฉs appartenant ร $`\mathrm{\Lambda }`$. Quitte ร extraire une sous-suite de la suite dโintervalles $`g_m(J_m)`$, nous pouvons supposer quโils contiennent tous dans leur intรฉrieur un intervalle $`J`$ de mesure supรฉrieure ou รฉgale ร $`\epsilon _0/2`$. Puisque toutes les orbites sont denses dans $`\mathrm{\Lambda }`$ et que $`\mathrm{\Lambda }`$ est compact, il existe des intervalles $`U_1,\mathrm{},U_k`$ de $`X`$ recouvrant $`\mathrm{\Lambda }`$, ainsi quโune famille $`\{h_1,\mathrm{},h_k\}`$ dโรฉlรฉments de $`\mathrm{\Gamma }`$, tels que chaque $`h_i`$ est dรฉfini sur $`U_i`$ et $`h_i(U_i)J`$ pour tout $`i\{1,\mathrm{},k\}`$. Lorsque $`m`$ est assez grand, les intervalles $`J_m`$ sont de longueurs assez petites pour quโils soient entiรจrement contenus dans lโun des intervalles $`U_i`$, disons $`U_{i_m}`$. La transformation $`g=h_{i_m}g_m^1`$ est alors dรฉfinie sur lโintervalle fermรฉ $`g_m(J_m)`$ et son image est contenue dans lโintรฉrieur de $`g_m(J_m)`$. รcrivons $`g_m(J_m)=[p,p^{}]`$, avec $`p`$ et $`p^{}`$ dans $`\mathrm{\Lambda }`$. Nous avons donc $`p<g(p)<g(p^{})<p^{}`$. Lโorbite du point $`x=limg^n(p)`$ est alors de type ressort, puisque dโune part les itรฉrรฉs de tout point $`q`$ de $`[p,x[`$ par $`g`$ convergent vers $`x`$, et dโautre part lโensemble $`]p,x[\mathrm{\Lambda }`$ est non vide (ร cause de la minimalitรฉ de $`\mathrm{\Lambda }`$). $`\mathrm{}`$
###### Lemme 5.4.
Si la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est รฉquicontinue, alors il existe une mesure de probabilitรฉ supportรฉe sur $`\mathrm{\Lambda }`$ qui est invariante par tous les รฉlรฉments de $`\mathrm{\Gamma }`$.
Dรฉmonstration. La preuve repose sur les idรฉes de Weil pour construire la mesure de Haar dโun groupe localement compact . Notons $`C_\mathrm{\Lambda }(X)`$ lโensemble des fonctions continues sur $`X`$ qui sont constantes sur les composantes connexes de $`X\mathrm{\Lambda }`$, et dรฉsignons par $`C_\mathrm{\Lambda }^+(X)`$ le sous-ensemble des fonctions positives et non identiquement nulles<sup>8</sup><sup>8</sup>8Rappelons que, en franรงais, un nombre $`a`$ est dit positif (resp. nรฉgatif) lorsque $`a0`$ (resp. $`a0`$). Si $`g`$ est un รฉlรฉment de $`\mathrm{\Gamma }`$ et $`\psi `$ un รฉlรฉment de $`C_\mathrm{\Lambda }^+(X)`$, notons $`\psi g^1`$ la fonction qui vaut $`\psi g^1`$ sur lโimage de $`g`$ et $`0`$ en dehors. Si le domaine de dรฉfinition de $`g`$ contient le support de $`\psi `$, alors la fonction $`\psi g^1`$ est ร nouveau continue. Lorsque la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est รฉquicontinue, nous allons construire une fonctionnelle $`L:C_\mathrm{\Lambda }^+(X)]0,\mathrm{}[`$ qui est linรฉaire et homogรจne, et qui vรฉrifie
$$L(\zeta g^1)=L(\zeta )$$
pour toute fonction $`\zeta C_\mathrm{\Lambda }^+(X)`$ et tout $`g\mathrm{\Gamma }`$ dont le domaine de dรฉfinition contient le support de $`\zeta `$. Le lecteur pourra sโassurer que cela dรฉfinit une mesure de probabilitรฉ supportรฉe sur $`\mathrm{\Lambda }`$ qui est invariante par $`\mathrm{\Gamma }`$.
Soient $`\psi ,\zeta `$ des รฉlรฉments de $`C_\mathrm{\Lambda }^+(X)`$. Puisque $`\psi `$ est non nulle sur $`\mathrm{\Lambda }`$, il est aisรฉ de trouver un nombre fini de rรฉels positifs $`c_i`$, ainsi que des รฉlรฉments $`g_i`$ de $`\mathrm{\Gamma }`$ dont le domaine de dรฉfinition soit connexe, de telle sorte que
$$\zeta \underset{i}{}c_i\psi g_i^1.$$
Nous dรฉfinissons $`(\zeta :\psi )`$ comme รฉtant lโinfimum des valeurs des $`_ic_i`$ parmi tous choix possibles des $`c_i`$ pour lesquels une telle inรฉgalitรฉ a lieu. Nous laissons au lecteur la vรฉrification des propriรฉtรฉs suivantes :
โ $`(\zeta :\psi )\frac{\zeta _{\mathrm{}}}{\psi _{\mathrm{}}}`$ ;
โ $`(c\zeta :d\psi )=\frac{c}{d}(\zeta :\psi )`$ pour toute paire de rรฉels strictement positifs $`c,d`$ ;
โ $`(\zeta _1+\zeta _2:\psi )(\zeta _1:\psi )+(\zeta _2:\psi )`$ ;
โ $`(\zeta :\psi )(\zeta :\xi )(\xi :\psi )`$ ;
โ $`(\zeta g^1:\psi )(\zeta :\psi )`$, avec รฉgalitรฉ si le domaine de $`g`$ contient le support de $`\psi `$.
Pour chaque $`\psi C_\mathrm{\Lambda }^+(X)`$ considรฉrons la fonctionnelle $`L_\psi :C_\mathrm{\Lambda }^+(X)`$ donnรฉe par
$$L_\psi (\zeta )=\frac{(\zeta :\psi )}{(1:\psi )}.$$
Ces fonctionnelles sont positives, homogรจnes et sous-additives. Nous allons vรฉrifier que lorsque le diamรจtre du support de $`\psi `$ tend vers $`0`$, et que la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est รฉquicontinue, ces fonctionnelles deviennent de plus en plus linรฉaires. Choisissons deux fonctions $`\xi _1`$ et $`\xi _1`$ dans $`C_\mathrm{\Lambda }^+(X)`$ telles que $`\xi _1+\xi _2=1`$. Nous allons dรฉmontrer que
$$\left|L_\psi (\xi _1\zeta )+L_\psi (\xi _2\zeta )L_\psi (\zeta )\right|$$
tend vers $`0`$ lorsque le diamรจtre du support de $`\psi `$ tend vers $`0`$. Pour cela considรฉrons un rรฉel $`\eta >0`$. Puisque $`\xi _1`$ et $`\xi _2`$ sont continues et que $`X`$ est compact, il existe $`\epsilon >0`$ tel que si $`dist(x,x^{})\epsilon `$ alors $`|\xi _1(x)\xi _1(x^{})|<\eta `$ et $`|\xi _2(x)\xi _2(x^{})|<\eta `$. De plus, puisque la restriction de $`\mathrm{\Gamma }`$ ร $`\mathrm{\Lambda }`$ est รฉquicontinue, il existe $`\delta >0`$ tel que si $`I`$ est un intervalle vรฉrifiant $`\nu (I)\delta `$, alors pour tout $`g\mathrm{\Gamma }`$ dont le domaine de dรฉfinition est connexe on a $`\nu \left(g(Idom(g))\right)<\epsilon `$. Supposons que le support de $`\psi `$ soit contenu dans un intervalle de longueur plus petite que $`\delta `$. Considรฉrons une famille de rรฉels positifs $`c_i`$ et des $`g_i\mathrm{\Gamma }`$ dont le domaine de dรฉfinition soit connexe et tels que
$$\zeta \underset{i}{}c_i\psi g_i^1.$$
Si le domaine de dรฉfinition de $`g_i`$ nโintersecte pas le support de $`\psi `$, alors la fonction $`\psi g_i^1`$ est identiquement nulle, et on peut lโenlever de la somme en conservant lโinรฉgalitรฉ. Nous supposerons donc que le domaine de $`g_i`$ intersecte le support de $`\psi `$. Choisissons un point $`y_i`$ dans $`g_i\left(supp(\psi )dom(g_i)\right)`$. Pour tout $`x`$ dans $`X`$ nous avons
$$\zeta (x)\xi _1(x)\underset{i}{}c_i\xi _1(x)\psi g_i^1(x)=\underset{i}{}{}_{}{}^{}c_{i}^{}\xi _1(x)\psi \left(g_i^1(x)\right),$$
oรน la deuxiรจme somme ne porte que sur les termes pour lesquels $`g_i^1(x)`$ appartient au support de $`\psi `$, cโest-ร -dire tels que $`xg_i\left(supp(\psi )dom(g_i)\right)`$. Dโaprรจs lโhypothรจse dโรฉquicontinuitรฉ nous avons $`dist(g_i^1(x),y_i)<\epsilon `$, et donc
$$\left|\xi _1(y_i)\xi _1(g_i^1(x))\right|\eta .$$
Par suite,
$$\xi _1\zeta \underset{i}{}c_i\left(\xi (y_i)+\eta \right)\psi g_i^1,$$
ce qui implique lโinรฉgalitรฉ
$$(\xi _1\zeta :\psi )\underset{i}{}c_i(\xi _1(y_i)+\eta ).$$
Puisque nous avons les mรชmes inรฉgalitรฉs pour $`\xi _2`$, nous obtenons
$$(\xi _1\zeta :\psi )+(\xi _2\zeta :\psi )\underset{i}{}c_i(\xi _1(y_i)+\xi _2(y_i)+2\eta )=(1+2\eta )\underset{i}{}c_i.$$
En passant ร lโinfimum sur les $`_ic_i`$, et puis en divisant par $`(1:\psi )`$, nous en dรฉduisons que
$$L_\psi (\xi _1\zeta )+L_\psi (\xi _2\zeta )(1+2\eta )L_\psi (\zeta )$$
lorsque le diamรจtre du support de $`\psi `$ est contenu dans un intervalle de taille infรฉrieure ร $`\delta `$. Comme nous savons dรฉjร que $`L_\psi `$ est sous-additive, ceci montre que
$$\left|L_\psi (\xi _1\zeta )+L_\psi (\xi _2\zeta )L_\psi (\zeta )\right|$$
tend vers $`0`$ lorsque le diamรจtre du support de $`\psi `$ tend vers $`0`$.
Pour conclure la dรฉmonstration, il nous suffit de trouver une suite de fonctions $`\psi _n`$ dans $`C_\mathrm{\Lambda }^+(X)`$ telle que les fonctionnelles $`L_{\psi _n}`$ convergent simplement vers une fonctionnelle $`L`$. Pour cela, remarquons que pour tout $`\psi ,\zeta `$ dans $`C_\mathrm{\Lambda }^+(X)`$ on a
$$\frac{1}{(1:\zeta )}L_\psi (\zeta )(\zeta :1).$$
Ainsi, les fonctionnelles $`L_\psi `$ dรฉfinissent un point $`(L_\psi (\zeta ))_{\zeta C_\mathrm{\Lambda }^+(X)}`$ dans lโespace produit
$$\mathrm{\Pi }=\underset{\zeta C_\mathrm{\Lambda }^+(X)}{}[\frac{1}{(1:\zeta )},(\zeta :1)].$$
Cet espace $`\mathrm{\Pi }`$ muni de la topologie produit est compact. Donc, si $`\psi _n`$ est une suite dโรฉlรฉments de $`C_\mathrm{\Lambda }^+(X)`$ dont les supports tendent vers un point, de la suite dโรฉlรฉments $`(L_{\psi _n}(\zeta ))_{\zeta C_\mathrm{\Lambda }^+(X)}`$ de $`\mathrm{\Pi }`$ on peut extraire une sous-suite qui converge vers un รฉlรฉment $`\left(L(\zeta )\right)_{\zeta C_\mathrm{\Lambda }^+(X)}`$. Dโaprรจs ce qui prรฉcรจde, $`L`$ dรฉfinit une fonctionnelle sur $`C_\mathrm{\Lambda }^+(X)`$ qui vรฉrife les conditions suivantes :
โ $`L`$ est strictement positive ;
โ pour tout $`\zeta C_\mathrm{\Lambda }^+(X)`$ et tout $`g\mathrm{\Gamma }`$ tel que $`dom(g)`$ contient le support de $`\zeta `$ on a $`L(\zeta g^1)=L(\zeta )`$ ;
โ $`L`$ est homogรจne ;
โ pour toutes fonctions $`\zeta ,\xi _1,\xi _2`$ de $`C_\mathrm{\Lambda }^+(X)`$ qui vรฉrifient $`\xi _1+\xi _2=1`$ on a $`L(\xi _1\zeta )+L(\xi _2\zeta )=L(\zeta )`$.
Il nโest pas difficile de sโassurer que les deux derniรจres conditions impliquent que $`L`$ est en fait linรฉaire. Ceci achรจve la preuve du lemme, et donc celle de la proposition 5.1. $`\mathrm{}`$
La terminologie $`^{_{}}`$ressort $`^{_{}}`$ vient de la thรฉorie des feuilletages (voir par exemple pour les dรฉtails ainsi quโun joli dessin dโun ressort). Dans ce contexte ce qui prรฉcรจde se traduit par le rรฉsultat suivant, ร comparer avec .
###### Proposition 5.5.
Soit $``$ un feuilletage de codimension 1 dโune variรฉtรฉ compacte. Si $``$ nโadmet pas de mesure (de probabilitรฉ) transverse invariante, alors il contient une feuille ressort (topologique).
Dรฉmonstration. Quitte ร passer au revรชtement double orientable, nous pouvons supposer que $``$ est transversalement orientable. Il suffit alors dโappliquer la conclusion de la proposition 5.1 au pseudo-groupe dโholonomie de $``$. $`\mathrm{}`$
### 5.2 Contraction presque sรปre pour les groupes dโhomรฉomorphismes du cercle
Une รฉlaboration plus simple et conceptuelle des idรฉes du paragraphe prรฉcรฉdent peut รชtre faite pour les groupes dโhomรฉomorphismes du cercle. En suivant Ghys (pages 360-362 ; voir aussi ), considรฉrons un tel groupe $`\mathrm{\Gamma }`$ et supposons dโabord que ses orbites soient denses. Si son action sur le cercle est รฉquicontinue (dans le sens que pour tout $`\epsilon >0`$ il existe $`\delta >0`$ tel que si $`dist(x,y)\delta `$ alors $`dist(g(x),g(y))\epsilon `$ pour tout $`g\mathrm{\Gamma }`$), alors $`\mathrm{\Gamma }`$ est topologiquement conjuguรฉ ร un groupe de rotations. Sinon, cโest que son action est expansive, dans le sens que pour tout $`x\mathrm{S}^1`$ il existe un intervalle ouvert $`I`$ contenant $`x`$ et une suite $`(h_n)`$ dโรฉlรฉments de $`\mathrm{\Gamma }`$ tels que la taille de $`h_n(I)`$ tend vers zรฉro. En particulier, il existe des points $`y\mathrm{S}^1`$ tels que lโintervalle $`[x,y[`$ est $`^{_{}}`$contractable $`^{_{}}`$. Si lโon denote par $`\theta (x)`$ le $`^{_{}}`$supremum $`^{_{}}`$ de ces points, alors lโapplication $`x\theta (x)`$ sโavรจre รชtre un homรฉomorphisme dโordre fini $`\kappa (\mathrm{\Gamma })`$ qui commute avec tous les รฉlรฉments de $`\mathrm{\Gamma }`$ (le nombre $`\kappa (\mathrm{\Gamma })`$ est appelรฉ le degrรฉ de $`\mathrm{\Gamma }`$). Par suite, si lโon identifie les points des orbites par $`\theta `$, alors on obtient un cercle topologique $`\mathrm{S}^1/`$ sur lequel le groupe $`\mathrm{\Gamma }`$ agit par homรฉomorphismes. Remarquons que cette derniรจre action nโest pas nรฉcessairement fidรจle. Cependant, elle vรฉrifie une propriรฉtรฉ dโexpansivitรฉ forte, ร savoir tout intervalle dont le complรฉmentaire nโest ni vide ni rรฉduit ร un seul point est contractable par une suite dโรฉlรฉments du groupe.
Si les orbites de notre groupe original ne sont pas denses, alors il peut se prรฉsenter deux cas : soit il possรจde une orbite finie, soit il existe un (unique) ensemble de Cantor invariant minimal (et sur lequel les orbites de tous les points du cercle sโaccumulent). En รฉcrasant les composantes connexes du complรฉmentaire de cet ensemble, ce dernier cas peut รชtre ramenรฉ ร celui dโorbites denses.
Cherchons maintenant ร donner des versions probabilistes de ce qui prรฉcรจde. Pour simplifier, supposons que $`\mathrm{\Gamma }`$ soit dรฉnombrable, et fixons une mesure de probabilitรฉ $`\mu `$ sur $`\mathrm{\Gamma }`$ qui soit non dรฉgรฉnรฉrรฉe, cโest-ร -dire telle que son support engendre $`\mathrm{\Gamma }`$ en tant que semigroupe. Une mesure de probabilitรฉ $`\nu `$ sur le cercle est dite stationnaire (par rapport ร $`\mu `$) si $`\mu \nu =\nu `$, cโest-ร -dire si pour toute fonction continue $`\psi `$ dรฉfinie sur le cercle on a
$$_{\mathrm{S}^1}\psi (x)๐\nu (x)=_\mathrm{\Gamma }_{\mathrm{S}^1}\psi \left(g(x)\right)๐\nu (x)๐\mu (g).$$
Pour mieux comprendre cette dรฉfinition, considรฉrons lโopรฉrateur de diffusion agissant sur les fonctions continues du cercle par la formule
$$D\psi (x)=_\mathrm{\Gamma }\psi \left(g(x)\right)๐\mu (g).$$
(26)
Cet opรฉrateur agit de maniรจre duale sur les mesures de probabilitรฉ en prรฉservant le compact convexe des mesures de probabilitรฉ : cette action duale correspond prรฉcisรฉment ร celle donnรฉe par $`\upsilon \mu \upsilon `$. Lโexistence de (au moins) une mesure stationnaire $`\nu `$ sur le cercle est ainsi assurรฉe par le thรฉorรจme du point fixe de Kakutani (bien sรปr, on peut aussi dรฉmontrer cela par un argument de moyennes de Birkhoff).
###### Lemme 5.6.
Si les orbites de $`\mathrm{\Gamma }`$ sont denses, alors $`\nu `$ est ร support total et sans atome. Si $`\mathrm{\Gamma }`$ admet un ensemble de Cantor invariant et minimal, alors cet ensemble coรฏncide avec le support de $`\nu `$, et $`\nu `$ est encore sans atome.
Dรฉmonstration. Montrons dโabord que si $`\nu `$ possรจde des atomes, alors $`\mathrm{\Gamma }`$ a des orbites finies (signalons en passant que, contrairemment ร ce que lโon peut croire, ce dernier cas peut รชtre trรจs compliquรฉ ; voir par exemple la proposition 5.13 et la remarque 5.14 plus loin). En effet, si $`p`$ est un point de mesure (positive et) maximale, alors ร partir de lโรฉgalitรฉ
$$\nu (p)=_\mathrm{\Gamma }\nu \left(g^1(p)\right)๐\mu (g),$$
on conclut que $`\nu \left(g^1(p)\right)=\nu (p)`$ pour tout $`g`$ dans le support de $`\nu `$. Ceci reste vrai pour tout รฉlรฉment de $`\mathrm{\Gamma }`$, car $`\mu `$ est une mesure non dรฉgรฉnรฉrรฉe. La masse totale de $`\nu `$ รฉtant finie, lโorbite de $`p`$ par $`\mathrm{\Gamma }`$ doit nรฉcessairement รชtre finie.
Si lโaction de $`\mathrm{\Gamma }`$ est minimale, alors le support de $`\nu `$ est tout le cercle puisque cโest un ensemble fermรฉ invariant. Si $`\mathrm{\Gamma }`$ admet un minimal exceptionnel $`\mathrm{\Lambda }`$ alors cet ensemble รฉtant unique, il doit รชtre contenu dans le support de $`\nu `$. Donc, pour montrer que $`\mathrm{\Lambda }`$ et $`supp(\nu )`$ coรฏncident, nous devons montrer que $`\nu (I)=0`$ pour toute composante connexe de $`\mathrm{S}^1\mathrm{\Lambda }`$. Or, si ce nโest pas le cas, en prenant une telle composante de masse maximale, on conclut (par un argument analogue ร celui donnรฉ dans le cas dโexistence dโatomes) que lโorbite de $`I`$ est finie. Nรฉanmoins, ceci est absurde, car les orbites des extrรฉmitรฉs de $`I`$ sont denses dans $`\mathrm{\Lambda }`$. $`\mathrm{}`$
Dรฉsignons par $`\mathrm{\Omega }`$ lโespace des suites $`(g_1,g_2,\mathrm{})\mathrm{\Gamma }^{}`$ (muni de la mesure $`=\mu ^{}`$). Si lโon dรฉsigne par $`\sigma `$ le dรฉcalage sur $`\mathrm{\Omega }`$, alors on vรฉrifie aisรฉment quโune mesure de probabilitรฉ $`\nu `$ sur le cercle est stationnaire par rapport ร $`\mu `$ si et seulement si la mesure $`\times \nu `$ est invariante par le produit croisรฉ $`T:\mathrm{\Omega }\times \mathrm{S}^1\mathrm{\Omega }\times \mathrm{S}^1`$ dรฉfini par
$$T(\omega ,x)=(\sigma (\omega ),h_1(\omega )(x))=(\sigma (\omega ),g_1(x)),\omega =(g_1,g_2,\mathrm{}).$$
En suivant Furstenberg , pour รฉtudier lโรฉvolution des compositions alรฉatoires on considรจre le processus inverse$`\overline{h}_n(\omega )=g_1\mathrm{}g_n`$. Si $`\psi `$ est une fonction continue dรฉfinie sur le cercle et $`\nu `$ est une mesure stationnaire, alors la suite de variables alรฉatoires
$$\xi _n(\omega )=_{\mathrm{S}^1}\psi d\left(g_1\mathrm{}g_n(\nu )\right)$$
est une martingale. Le thรฉorรจme de convergence de martingales (plus un argument de densitรฉ) montre ainsi que, pour un sous-ensemble de probabilitรฉ total $`\mathrm{\Omega }_0`$ de $`\mathrm{\Omega }`$, la limite suivante (par rapport ร la topologie faible) existe :
$$\underset{n\mathrm{}}{lim}g_1g_2\mathrm{}g_n(\nu )=\underset{n\mathrm{}}{lim}\overline{h}_n(\omega )(\nu )=\omega (\nu ).$$
De plus, lโapplication (bien dรฉfinie presque partout) $`\omega \omega (\nu )`$ est mesurable (voir , page 199). Une proposition รฉquivalente ร celle prรฉsentรฉe ร continuation a รฉtรฉ originalement dรฉmontrรฉe par Antonov dans (voir aussi ) ; elle est ร rapprocher avec des rรฉsultats contenus dans et le ยง2 du chapitre VI de .
###### Proposition 5.7.
Supposons que $`\mathrm{\Gamma }`$ soit dรฉnombrable, que ses orbites soient denses et que la propriรฉtรฉ dโexpansivitรฉ forte soit vรฉrifiรฉe. Alors pour presque toute suite $`\omega `$ dans $`\mathrm{\Omega }_0`$, la mesure $`\omega (\nu )`$ est une mesure de Dirac.
Dรฉmonstration. Nous montrerons que pour tout $`\epsilon ]\mathrm{0,1}]`$ il existe un sous-ensemble de probabilitรฉ total $`\mathrm{\Omega }_\epsilon `$ de $`\mathrm{\Omega }_0`$ tel que pour tout $`\omega \mathrm{\Omega }_\epsilon `$ il existe un intervalle $`I`$ de longueur $`\epsilon `$ tel que $`\omega (\nu )(I)1\epsilon `$. Ceci permet de conclure de maniรจre รฉvidente que, pour tout $`\omega `$ dans lโensemble $`\mathrm{\Omega }^{}=_n\mathrm{\Omega }_{1/n}`$ (dont la probabilitรฉ est totale), la mesure $`\omega (\nu )`$ est une mesure de Dirac.
Fixons donc $`\epsilon >0`$. Pour chaque $`n`$ dรฉsignons par $`\mathrm{\Omega }^{n,\epsilon }`$ lโensemble des suites $`\omega \mathrm{\Omega }_0`$ telles que, pour tout $`m0`$ et tout intervalle $`I`$ de $`\mathrm{S}^1`$ de longueur $`\epsilon `$, on a $`\overline{h}_{n+m}(\omega )(\nu )(I)<1\epsilon `$. Nous devons dรฉmontrer que la mesure de $`\mathrm{\Omega }^{n,\epsilon }`$ est nulle. Pour cela, nous allons commencer par exhiber un sous-ensemble fini $`๐ข_\epsilon `$ de $`supp(\mu )`$ ainsi quโun entier $`l`$ tels que, pour tout $`r`$ et tout $`(g_1,\mathrm{},g_r)\mathrm{\Gamma }^r`$, il existe un intervalle $`I`$ de longueur $`\epsilon `$, un entier $`\mathrm{}2l`$, et des รฉlรฉments $`f_1,\mathrm{},f_{\mathrm{}}`$ dans $`๐ข_\epsilon `$ vรฉrifiant
$$g_1\mathrm{}g_rf_1\mathrm{}f_{\mathrm{}}(\nu )(I)1\epsilon .$$
(27)
Fixons deux points distincts $`a`$ et $`b`$ du cercle, ainsi quโun entier $`q>1/\epsilon `$, et prenons $`q`$ points distincts $`a_1,\mathrm{},a_q`$ de lโorbite de $`a`$ par $`\mathrm{\Gamma }`$. Pour chaque $`i\{1,\mathrm{},q\}`$ fixons un รฉlรฉment $`h_i\mathrm{\Gamma }`$ et un intervalle ouvert $`U_i`$ contenant $`a_i`$ de faรงon ร ce que les $`U_i`$ soient deux ร deux disjoints, $`h_i(a)=a_i`$ et $`h_i(U)=U_i`$ pour un certain voisinage $`U`$ de $`a`$ qui ne contient pas $`b`$. Prenons maintenant un voisinage $`V`$ de $`b`$ disjoint de $`U`$ et tel que $`\nu (\mathrm{S}^1V)1\epsilon `$. Par la minimalitรฉ et la propriรฉtรฉ dโexpansivitรฉ forte, il existe $`h\mathrm{\Gamma }`$ tel que $`h(\mathrm{S}^1V)U`$. Chaque รฉlรฉment dans $`\{h_1,\mathrm{},h_q,h\}`$ peut รชtre รฉcrit comme un produit dโรฉlรฉments du support de $`\mu `$. Cela peut รชtre fait de plusieurs maniรจres diffรฉrentes, mais si lโon fixe une fois pour toutes une รฉcriture pour chaque รฉlรฉment, alors lโensemble $`๐ข_\epsilon `$ des รฉlรฉments de $`supp(\mu )`$ qui sont utilisรฉs est fini. Soit $`l`$ le nombre maximal dโรฉlรฉments qui apparaissent dans lโune des รฉcritures prรฉcรฉdentes. Pour vรฉrifier (27), notons que pour $`g=g_1\mathrm{}g_r`$ les intervalles $`g(U_i)`$ sont deux ร deux disjoints, et donc la longueur dโau moins lโun dโentre eux doit รชtre majorรฉe par $`\epsilon `$. Si lโon fixe un tel intervalle $`I=g(U_i)`$ alors on obtient
$$g_1\mathrm{}g_rh_ih(\nu )(I)=\nu \left(h^1(U)\right)\nu (\mathrm{S}^1V)1\epsilon ,$$
ce qui achรจve la vรฉrification de (27).
Notons maintenant $`\rho =\mathrm{min}\{\mu (f):f๐ข_\epsilon \}`$ et posons
$$\mathrm{\Omega }_{n+m}^\epsilon =\{\omega \mathrm{\Omega }_0:\text{ pour tout intervalle }I\text{ de longueur }|I|\epsilon \text{ et tout }km\text{ on a }\overline{h}_{n+k}(\omega )(\nu )(I)<1\epsilon \}.$$
Dโaprรจs (27) nous avons $`(\mathrm{\Omega }_{n+2lt}^\epsilon )(1\rho ^{2l})^t`$. Donc, en passant ร la limite lorsque $`t`$ tend vers lโinfini, on conclut que $`(\mathrm{\Omega }^{n,\epsilon })=0`$, ce qui permet de finir la dรฉmonstration. $`\mathrm{}`$
###### Remarque 5.8.
Il est trรจs intรฉressant de remarquer que les seules propriรฉtรฉs de $`\nu `$ que lโon a utilisรฉ dans la preuve prรฉcรฉdente sont le fait que son support est total et que, pour presque tout $`\omega \mathrm{\Omega }`$, la suite de mesures de probabilitรฉ $`g_1\mathrm{}g_n(\nu )`$ converge faiblement.
On dรฉfinit le coefficient de contraction $`c(h)`$ dโun homรฉomorphisme $`h`$ du cercle comme รฉtant lโinfimum parmi tous les $`\epsilon >0`$ tels quโil existe deux intervalles fermรฉs $`I`$ et $`J`$ du cercle de taille au plus $`\epsilon `$ et tels que $`h(\overline{\mathrm{S}^1I})=J`$. Cette dรฉfinition permet de donner une $`^{_{}}`$version topologique $`^{_{}}`$ de la proposition prรฉcรฉdente pour les compositions dans $`^{_{}}`$lโordre naturel $`^{_{}}`$.
###### Proposition 5.9.
Sous les hypothรจses de la proposition 5.7, pour tout $`\omega =(g_1,g_2,\mathrm{})\mathrm{\Omega }^{}`$ le coefficient de contraction de $`h_n(\omega )=g_n\mathrm{}g_1`$ converge vers zรฉro lorsque $`n`$ tend vers lโinfini.
Dรฉmonstration. Comme $`\nu `$ est sans atome et son support est total, il existe un homรฉomorphisme $`\phi `$ du cercle qui envoit $`\nu `$ sur la mesure de Lebesgue. Par suite, et puisque lโaffirmation ร dรฉmontrer est de nature purement topologique, nous pouvons supposer que $`\nu `$ coรฏncide avec la mesure de Lebesgue.
Dรฉsignons par $`\overline{\mu }`$ la mesure sur $`\mathrm{\Gamma }`$ dรฉfinie par $`\overline{\mu }(g)=\mu (g^1)`$, et notons $`\overline{\mathrm{\Omega }}`$ lโespace de probabilitรฉ $`\mathrm{\Gamma }^{}`$ muni de la mesure $`\overline{\mu }^{}`$. Sur cet espace considรฉrons le processus $`\overline{h}_n(\overline{\omega })=g_1\mathrm{}g_n`$, oรน $`\overline{\omega }=(g_1,g_2,\mathrm{})`$. Dโaprรจs la proposition 5.7, pour tout $`\epsilon >0`$ il existe $`n(\epsilon )`$ tel que si $`nn(\epsilon )`$ alors il existe un intervalle (fermรฉ) $`I`$ tel que $`\nu (I)\epsilon `$ et $`\overline{h}_n(\overline{\omega })(\nu )(I)1\epsilon `$. Si lโon note $`J`$ la fermeture de $`\mathrm{S}^1g_n^1\mathrm{}g_1^1(I)`$, alors on voit que $`|I|=\nu (I)\epsilon `$,
$$|J|=1|g_n^1\mathrm{}g_1^1(I)|=1\nu \left(\overline{h}_n(\overline{\omega })^1(I)\right)=1\overline{h}_n(\overline{\omega })(\nu )(I)\epsilon $$
et $`g_n^1\mathrm{}g_1^1(\overline{\mathrm{S}^1I})=J`$. Par suite, $`c\left(g_n^1\mathrm{}g_1^1\right)\epsilon `$ pour tout $`nn(\epsilon )`$. La preuve est conclue en remar- quant que la transformation $`(g_1,g_2,\mathrm{})(g_1^1,g_2^1,\mathrm{})`$ est un isomorphisme entre $`(\overline{\mathrm{\Omega }},\overline{\mu }^{})`$ et $`(\mathrm{\Omega },\mu ^{})`$. $`\mathrm{}`$
Soulignons que le coefficient de contraction est toujours rรฉalisรฉ, dans le sens que pour tout homรฉomor- phisme $`h`$ du cercle il existe des intervalles fermรฉs $`I`$ et $`J`$ tels que $`\mathrm{max}\{|I|,|J|\}=c(h)`$ et $`h(\overline{\mathrm{S}^1I})=J`$ (ces intervalles ne sont pas nรฉcessairement uniques). En consรฉquence, pour tout $`\omega \mathrm{\Omega }^{}`$ on peut choisir deux suites dโintervalles fermรฉs $`I_n(\omega )`$ et $`J_n(\omega )`$ dont la taille tend vers zรฉro et tels que $`h_n(\omega )\left(\overline{\mathrm{S}^1I_n(\omega )}\right)=J_n(\omega )`$ pour tout $`n`$. Pour nโimporte quel choix, les intervalles $`I_n(\omega )`$ convergent vers le point $`\varsigma _+(\omega )`$, tandis que (gรฉnรฉriquement) les intervalles $`J_n(\omega )`$ se promรจnent un peu partout sur le cercle.
###### Remarque 5.10.
Les propositions 5.7 et 5.9 suggรจrent quโune version faible du thรฉorรจme dโOseledets devrait รชtre valable pour la plupart des transformations de fibrรฉs en cercles. Pour รชtre plus prรฉcis, considรฉrons un espace mรฉtrique quelconque $`\overline{\mathrm{\Omega }}`$ muni dโune mesure de probabilitรฉ $`\overline{}`$, et soit $`T:\overline{\mathrm{\Omega }}\overline{\mathrm{\Omega }}`$ un homรฉomorphisme qui prรฉserve $`\mu `$ de maniรจre ergodique. รtant donnรฉe une application borรฉlienne $`L:\overline{\mathrm{\Omega }}\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$, considรฉrons la transformation $`\mathrm{\Phi }:\overline{\mathrm{\Omega }}\times \mathrm{S}^1\overline{\mathrm{\Omega }}\times \mathrm{S}^1`$ dรฉfinie par
$$\mathrm{\Phi }(\omega ,x)=(T(\omega ),L(\omega )(x)),$$
et notons
$$L_n(\omega )(x)=L(T^n(\omega ))\mathrm{}L(T(\omega ))L(\omega )(x),n0,$$
$$L_n(\omega )(x)=L(T^n(\omega ))^1\mathrm{}L(T^2(\omega ))^1L(T^1(\omega ))^1(x),n<0,$$
Supposons que la transformation $`\mathrm{\Phi }`$ ne prรฉserve aucune mesure de probabilitรฉ, et admettons pour simplifier quโelle soit continue et que ses orbites soient denses. Sous ces hypothรจses, il devrait exister un homรฉomorphisme dโordre fini $`\theta :\mathrm{S}^1\mathrm{S}^1`$ qui commute avec presque tous les $`L(\omega )`$ et tel que lโaction de $`\mathrm{\Phi }`$ sur lโespace $`\overline{\mathrm{\Omega }}\times \mathrm{S}^1/`$ des orbites par lโapplication $`(\omega ,x)(\omega ,\theta (x))`$ admette deux sections mesurables $`\varsigma _+`$ et $`\varsigma _{}`$ telles que pour $`\overline{}`$-presque tout point $`\omega \overline{\mathrm{\Omega }}`$ et tout $`x\mathrm{S}^1`$ on ait
$$\underset{n+\mathrm{}}{lim}dist(L_n(\omega )(x),\varsigma _{}(T^n(\omega ))=0\underset{n\mathrm{}}{lim}dist(L_n(\omega )(x),\varsigma _+(T^n(\omega ))=0.$$
### 5.3 Quelques remarques autour de la mesure stationnaire
La proposition 5.7 permet (par un argument bien connu) de dรฉmontrer lโunicitรฉ de la mesure stationnaire. Signalons que ce rรฉsultat est valable dans le cadre beaucoup plus gรฉnรฉral des feuilletages de codimension 1 sans mesure transverse invariante (la notion de mesure stationnaire dans ce contexte est celle de Garnett ; voir ).
###### Proposition 5.11.
Soit $`\mathrm{\Gamma }`$ un groupe dรฉnombrable dโhomรฉomorphismes du cercle muni dโune mesure de probabilitรฉ non dรฉgรฉnรฉrรฉe $`\mu `$. Si $`\mathrm{\Gamma }`$ ne prรฉserve aucune probabilitรฉ du cercle, alors la mesure stationnaire (par rapport ร $`\mu `$) est unique.
Dรฉmonstration. Supposons dโabord que lโaction de $`\mathrm{\Gamma }`$ est minimale et satisfait la propriรฉtรฉ dโexpansivitรฉ forte, et fixons une mesure $`\nu `$ sur $`\mathrm{S}^1`$ qui soit stationnaire par rapport ร $`\mu `$. Pour chaque $`\omega \mathrm{\Omega }`$ telle que la limite $`lim_n\mathrm{}\overline{h}_n(\omega )(\nu )`$ existe et soit une mesure de Dirac, dรฉsignons par $`\varsigma _\nu (\omega )`$ lโatome de la mesure $`\omega (\nu )`$, i.e. le point de $`\mathrm{S}^1`$ tel que $`\omega (\nu )=\delta _{\varsigma _\nu (\omega )}`$. Lโapplication $`\varsigma _\nu :\mathrm{\Omega }\mathrm{S}^1`$ est presque partout bien dรฉfinie et mesurable. Nous affirmons que les mesures $`\nu `$ et $`\varsigma _\nu ()`$ coรฏncident. En effet, dโaprรจs la stationnaritรฉ de $`\nu `$,
$$\nu =\mu ^n\nu =\underset{g\mathrm{\Gamma }}{}\mu ^n(g)g(\nu )=_\mathrm{\Omega }\overline{h}_n(\omega )(\nu )๐(\omega ).$$
Donc, en passant ร la limite lorsque $`n`$ tend vers lโinfini (ce qui peut facilement รชtre justifiรฉ par le thรฉorรจme de convergence dominรฉe),
$$\nu =_\mathrm{\Omega }\underset{n\mathrm{}}{lim}h_n(\omega )(\nu )d(\omega )=_\mathrm{\Omega }\delta _{\varsigma _\nu (\omega )}๐(\omega ),$$
cโest-ร -dire $`\nu =\varsigma _\nu ()`$.
Considรฉrons maintenant deux mesures stationnaires $`\nu _1`$ et $`\nu _2`$. La mesure $`\nu =(\nu _1+\nu _2)/2`$ est elle aussi stationnaire, et la fonction $`\varsigma _\nu `$ vรฉrifie, pour $``$-presque tout $`\omega \mathrm{\Omega }`$,
$$\frac{\delta _{\varsigma _{\nu _1}(\omega )}+\delta _{\varsigma _{\nu _2}(\omega )}}{2}=\delta _{\varsigma _\nu (\omega )}.$$
Bien รฉvidemment, ceci nโest possible que si $`\varsigma _{\nu _1}`$ et $`\varsigma _{\nu _2}`$ coรฏncident presque partout. Par consรฉquent, dโaprรจs lโaffirmation de la premiรจre partie de la preuve,
$$\nu _1=\varsigma _{\nu _1}()=\varsigma _{\nu _2}()=\nu _2.$$
Supposons maintenant que $`\mathrm{\Gamma }`$ agisse de faรงon minimale et expansive (mais non fortement expansive), et fixons une probabilitรฉ $`\nu `$ qui soit stationnaire (par rapport ร $`\mu `$). Dโaprรจs le ยง5.2, il existe un homรฉomorphisme $`\theta :\mathrm{S}^1\mathrm{S}^1`$ dโordre fini, commutant avec (tous les รฉlรฉments de) $`\mathrm{\Gamma }`$, et tel que lโaction induite sur le cercle topologique $`\mathrm{S}^1/`$ obtenu en tant quโespace dโorbites de $`\theta `$ est (minimale et) fortement expansive. Pour chaque $`x\mathrm{S}^1`$ notons $`\psi (x)=\nu ([x,\theta (x)[)`$. Comme $`\nu `$ est sans atome, la fonction $`\psi `$ est continue, et puisque $`\theta `$ commute avec tous les รฉlรฉments de $`\mathrm{\Gamma }`$, elle est harmonique. Par suite, lโensemble des points oรน $`\psi `$ prend sa valeur maximale est invariant par $`\mathrm{\Gamma }`$. Les orbites de $`\mathrm{\Gamma }`$ รฉtant denses, $`\psi `$ est une fonction constante ; en dโautres termes, $`\nu `$ est invariante par $`\theta `$. Par ailleurs, $`\nu `$ se projette sur une mesure de probabilitรฉ stationnaire pour lโaction de $`\mathrm{\Gamma }`$ sur $`\mathrm{S}^1/`$. Dโaprรจs la premiรจre partie de la preuve, cette derniรจre mesure est unique, ce qui montre alors lโunicitรฉ de $`\nu `$.
Si $`\mathrm{\Gamma }`$ admet un ensemble de Cantor invariant et minimal, alors cet ensemble coรฏncide avec le support de $`\nu `$. Si lโon รฉcrase les composantes connexes du complรฉmentaire de cet ensemble on obtient une action avec toutes ses orbites denses ; lorsque cette action est expansive, on peut appliquer les arguments prรฉcรฉdents afin de conclure lโunicitรฉ de la mesure stationnaire. Pour complรฉter la dรฉmonstration on constate aisรฉment que, dans tous les cas qui nโont pas encore รฉtรฉ considรฉrรฉs, le groupe $`\mathrm{\Gamma }`$ laisse invariante une mesure de probabilitรฉ du cercle. $`\mathrm{}`$
###### Remarque 5.12.
Sous les hypothรจses de la proposition prรฉcรฉdente, il est intรฉressant de chercher des conditions sous lesquelles le cercle $`\mathrm{S}^1`$ (lโespace $`\mathrm{S}^1/`$ le cas รฉchรฉant), muni de la mesure stationnaire, est un bord stochastique maximal pour $`(\mathrm{\Gamma },\mu )`$ (i.e. il coรฏncide avec son bord de Poisson-Furstenberg).
Comme nous lโavons dรฉjร signalรฉ, lorsquโil existe une mesure de probabilitรฉ invariante on est ramenรฉ ร lโรฉtude dโun groupe de rotations ou ร celle dโun groupe avec des orbites finies. Dans le premier cas la mesure stationnaire $`\nu `$ est encore unique et coรฏncide avec lโunique mesure invariante. En effet, pour chaque borรฉlien $`X\mathrm{S}^1`$ la fonction $`g\nu \left(g^1(X)\right)`$ est harmonique (par rapport ร $`\mu `$), et donc constante dโaprรจs . Le cas oรน il existe des orbites finies est plus intรฉressant. ร indice fini prรจs, il se ramรจne ร celui dโun groupe agissant sur lโintervalle. Or, dans ce contexte la mesure stationnaire ne donne aucune information dynamique (tout au moins lorsque la mesure de dรฉpart $`\mu `$ est symรฉtrique).
###### Proposition 5.13.
Soit $`\mathrm{\Gamma }`$ un sous-groupe de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+([\mathrm{0,1}])`$ sans points fixe global ร lโintรฉrieur. Si $`\mu `$ est une mesure de probabilitรฉ non dรฉgรฉnรฉrรฉe et symรฉtrique sur $`\mathrm{\Gamma }`$, alors toute mesure de probabilitรฉ sur $`[\mathrm{0,1}]`$ qui est stationnaire par rapport ร $`\mu `$ est supportรฉe sur les extrรฉmitรฉs de $`[\mathrm{0,1}]`$.
Premiรจre dรฉmonstration. Supposons par contradiction que $`\overline{\nu }`$ soit une mesure stationnaire telle que $`\overline{\nu }(\{\mathrm{0,1}\})<1`$. Notons $`\nu `$ la partie de $`\overline{\nu }`$ supportรฉe dans $`]\mathrm{0,1}[`$ convenablement normalisรฉe : cโest encore une mesure stationnaire, et elle vรฉrifie $`\nu (]\mathrm{0,1}[)=1`$. Comme dans la preuve du lemme 5.6, on montre aisรฉment que $`\nu `$ est sans atome. Donc, quitte ร รฉcraser les composantes connexes du complรฉmentaire du support de $`\nu `$, nous pouvons supposer que le support de $`\nu `$ coรฏncide avec $`]\mathrm{0,1}[`$. De plus, en reparamรฉtrant lโintervalle, nous pouvons supposer que $`\nu `$ est la mesure de Lebesgue.
Puisque $`\nu `$ est une mesure stationnaire par rapport ร la probabilitรฉ symรฉtrique $`\mu `$, pour tout $`s]\mathrm{0,1}[`$ on a
$$s=\nu \left([0,s]\right)=_\mathrm{\Gamma }\nu \left(g^1([0,s])\right)๐\mu (g)=_\mathrm{\Gamma }\frac{\nu \left(g([0,s])\right)+\nu \left(g^1([0,s])\right)}{2}๐\mu (g),$$
et donc
$$s=_\mathrm{\Gamma }\frac{g(s)+g^1(s)}{2}๐\mu (g).$$
En intรฉgrant entre 0 et un point arbitraire $`t]\mathrm{0,1}[`$ on obtient
$$t^2=_\mathrm{\Gamma }_0^t\left(g(s)+g^1(s)\right)๐s๐\mu (g).$$
(28)
Or, pour tout homรฉomorphisme $`f`$ de lโintervalle et tout $`t[\mathrm{0,1}]`$ on a
$$_0^t\left(f(s)+f^1(s)\right)๐st^2,$$
avec lโรฉgalitรฉ si et seulement si $`f(t)=t`$ (voir la figure 7). Dโaprรจs (28), ceci implique que $`g(t)=t`$ pour tout $`g`$ appartenant au support de $`\mu `$. Puisque $`\mu `$ est une mesure non dรฉgรฉnรฉrรฉe, on conclut que $`t]\mathrm{0,1}[`$ est un point fixe global pour lโaction de $`\mathrm{\Gamma }`$, ce qui contredit notre hypothรจse. $`\mathrm{}`$
............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... $`f`$$`f^1`$$`A`$$`A`$$`B`$$`\mathrm{\Delta }`$$`0`$$`t`$$`1`$$`t`$$`f^1(t)`$$`A=_0^tf(s)๐s`$$`B=_0^tf^1(s)๐s`$ $`A+B=t^2+\mathrm{\Delta }`$ $``$$``$ Figure 7 $``$
Deuxiรจme dรฉmonstration. Pour une mesure stationnaire $`\nu `$ considรฉrons la fonction $`x\nu \left([0,x]\right)`$. Si $`\mu `$ est symรฉtrique, alors cette fonction est harmonique par rapport ร $`\mu `$ le long de toute orbite. Dโaprรจs le thรฉorรจme ergodique de Garnett (Theorem 1. (b) dans ), cette fonction est constante pour $`\nu `$ presque toute orbite. Autrement dit, pour $`\nu `$ presque tout point $`x`$ lโรฉgalitรฉ $`\nu \left([0,g(x)]\right)=\nu \left([0,x]\right)`$ est vรฉrifiรฉe pour tout $`g\mathrm{\Gamma }`$. Lโhypothรจse de non existence de point fixe global pour lโaction de $`\mathrm{\Gamma }`$ sur $`]\mathrm{0,1}[`$ implique alors que le support de $`\nu `$ est contenu dans $`\{\mathrm{0,1}\}`$. $`\mathrm{}`$
###### Remarque 5.14.
Comme nous lโavons dรฉjร suggรฉrรฉ, lโhypothรจse de symรฉtrie pour la mesure $`\mu `$ est essentielle pour la validitรฉ de la proposition prรฉcรฉdente, comme le montre par exemple .
Voici une proposition gรฉnรฉrale de rรฉgularitรฉ dont la preuve sโappuie sur la mesure stationnaire. Signalons en passant que, dโaprรจs , un tel rรฉsultat ne peut รชtre valable quโen dimension 1.
###### Proposition 5.15.
Tout sous-groupe dรฉnombrable de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$ (resp. de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+([\mathrm{0,1}])`$) est topologiquement conjuguรฉ ร un groupe dโhomรฉomorphismes lipschitziens.
Dรฉmonstration. Considรฉrons dโabord le cas dโun sous-groupe dรฉnombrable $`\mathrm{\Gamma }`$ de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$ dont les orbites sont denses. Munissons ce groupe dโune mesure de probabilitรฉ non dรฉgรฉnรฉrรฉe et symรฉtrique $`\mu `$, et considรฉrons une mesure de probabilitรฉ $`\nu `$ sur le cercle qui soit stationnaire par rapport ร $`\mu `$ . Pour chaque intervalle $`I\mathrm{S}^1`$ et chaque รฉlรฉment $`gsupp(\mu )`$ on a
$$\nu (I)=\underset{gsupp(\mu )}{}\nu \left(g^1(I)\right)\mu (g)\nu \left(g(I)\right)\mu (g^1),$$
et donc
$$\nu \left(g(I)\right)\frac{1}{\mu (g)}\nu (I).$$
(29)
Prenons maintenant un homรฉomorphisme $`\phi `$ du cercle dans lui-mรชme qui envoie $`\nu `$ sur la mesure de Lebesgue. Si $`J`$ est un intervalle quelconque du cercle alors, dโaprรจs (29), pour tout $`gsupp(\mu )`$ on a
$$|\phi g\phi ^1(J)|=\nu \left(g\phi ^1(J)\right)\frac{1}{\mu (g)}\nu \left(\phi ^1(J)\right)=\frac{1}{\mu (g)}|J|.$$
Ainsi, pour tout $`g`$ dans $`supp(\mu )`$, la transformation $`\phi g\phi ^1`$ est lipschitzienne de rapport $`1/\mu (g)`$. Puisque la mesure $`\mu `$ est non dรฉgรฉnรฉrรฉe, cela dรฉmontre la proposition dans le cas minimal.
Si $`\mathrm{\Gamma }`$ est un sous-groupe dรฉnombrable quelconque de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$, alors en rajoutant une rotation dโangle irrationnel et en considรฉrant le groupe engendrรฉ, on se ramรจne au cas minimal. Les arguments plus haut montrent que ce nouveau groupe (et donc le groupe originel $`\mathrm{\Gamma }`$) est topologiquement conjuguรฉ ร un groupe dโhomรฉomorphismes lipschitziens. Finalement, en identifiant les extrรฉmitรฉs de lโintervalle $`[\mathrm{0,1}]`$, chaque sous-groupe $`\mathrm{\Gamma }`$ de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+([\mathrm{0,1}])`$ induit un groupe dโhomรฉomorphismes du cercle (avec un point fixe global marquรฉ). Dโaprรจs ce qui prรจcรจde, si $`\mathrm{\Gamma }`$ est dรฉnombrable alors ce nouveau groupe est conjuguรฉ par un รฉlรฉment $`\phi `$ de $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+(\mathrm{S}^1)`$ ร un groupe dโhomรฉomorphismes lipschitziens du cercle. Pour obtenir une vraie conjugaison dans $`\mathrm{Hom}\stackrel{ยด}{\mathrm{e}}\mathrm{o}_+([\mathrm{0,1}])`$, il suffit de composer $`\phi `$ avec une rotation de faรงon ร ramener le point marquรฉ du cercle sur lui-mรชme. $`\mathrm{}`$
La proposition prรฉcรฉdente indique que lโon ne doit pas sโattendre ร des rรฉsultats de rigiditรฉ pour des groupes dโhomรฉomorphismes du cercle qui soient spรฉcifiques ร une rรฉgularitรฉ strictement comprise entre $`C^0`$ et $`C^{lip}`$. Par contre, la comprรฉhension des passages $`C^{lip}C^1`$ et $`C^1C^{1+\tau }`$ semble รชtre un problรจme trรจs intรฉressant. Une obstruction pour le premier est donnรฉe par le thรฉorรจme de stabilitรฉ de Thurston . Par ailleurs, la question de savoir si la proposition 4.4 est encore valable pour des conjugaisons lipschitziennes reste ouverte. Quant au deuxiรจme passage, les rรฉsultats de cet article suggรจrent que le groupe de Grigorchuk-Maki considรฉrรฉ dans ne devrait pas รชtre isomorphe ร un sous-groupe de $`\mathrm{Diff}_+^{1+\tau }([\mathrm{0,1}])`$ pour aucun $`\tau >0`$ (mรชme sโil agit fidรจlement par diffรฉomorphismes de classe $`C^1`$ de lโintervalle !).
### 5.4 Exposants de Lyapunov et mesures invariantes
Considรฉrons un sous-groupe dรฉnombrable $`\mathrm{\Gamma }`$ de $`\mathrm{Diff}_+^1(\mathrm{S}^1)`$. Pour simplifier, supposons que $`\mathrm{\Gamma }`$ soit de type fini, et munissons-le dโune mesure de probabilitรฉ $`\mu `$ qui soit non dรฉgรฉnรฉrรฉe et ร support fini. Soit $`\nu `$ une probabilitรฉ stationnaire (par rapport ร $`\mu `$) sur le cercle, et considรฉrons la transformation $`T`$ de $`\mathrm{\Omega }\times \mathrm{S}^1`$ donnรฉe par $`T(\omega ,x)=(\sigma (\omega ),h_1(\omega )(x))`$. Puisque $`T`$ prรฉserve $`\times \nu `$, en appliquant le thรฉorรจme ergodique de Birkhoff ร la fonction $`(\omega ,x)\mathrm{log}\left(h_1(\omega )^{}(x)\right)`$ on conclut que pour $`\nu `$-presque tout point $`x`$ du cercle et pour $``$-presque tout chemin alรฉatoire $`\omega `$ dans $`\mathrm{\Omega }`$, la limite
$$\lambda _{(\omega ,x)}(\nu )=\underset{n\mathrm{}}{lim}\frac{\mathrm{log}\left(h_n(\omega )^{}(x)\right)}{n}$$
existe : cโest lโexposant de Lyapunov correspondant au point $`(\omega ,x)`$. Si la mesure de probabilitรฉ stationnaire $`\nu `$ est ergodique, cโest-ร -dire si elle ne peut pas รชtre exprimรฉe comme une combinaison convexe de deux probabilitรฉs stationnaires distinctes, alors la transformation $`T`$ est ergodique (au sens classique) par rapport ร $`\times \nu `$ (cโest le thรฉorรจme ergodique alรฉatoire de Kakutani : voir ). Dans ce cas, lโexposant de Lyapunov est $`\times \nu `$-presque partout constant et รฉgal ร
$$\lambda (\nu )=_\mathrm{\Omega }_{\mathrm{S}^1}\mathrm{log}\left(h_1(\omega )^{}(x)\right)๐\nu (x)๐(\omega )=_\mathrm{\Gamma }_{\mathrm{S}^1}\mathrm{log}\left(g^{}(x)\right)๐\nu (x)๐\mu (g).$$
Lorsque $`\mu `$ est symรฉtrique et la mesure $`\nu `$ est invariante par lโaction de $`\mathrm{\Gamma }`$, lโexposant de Lyapunov $`\lambda _{(\omega ,x)}(\nu )`$ est presque partout nul. En effet, si $`\nu `$ est invariante alors la transformation $`S`$ de $`\mathrm{\Omega }\times \mathrm{S}^1`$ sur lui mรชme dรฉfinie par $`S(\omega ,x)=(\omega ,h_1(\omega )^1(x))`$ prรฉserve $`\times \nu `$. Ceci implique que
$`\lambda (\nu )`$ $`=`$ $`{\displaystyle _\mathrm{\Omega }}{\displaystyle _{\mathrm{S}^1}}\mathrm{log}\left(h_1(\omega )^{}(x)\right)๐\nu (x)๐(\omega )={\displaystyle _\mathrm{\Omega }}{\displaystyle _{\mathrm{S}^1}}\mathrm{log}\left(h_1(\omega )^{}(h_1(\omega )^1(x))\right)๐\nu (x)๐(\omega )`$
$`=`$ $`{\displaystyle _\mathrm{\Omega }}{\displaystyle _{\mathrm{S}^1}}\mathrm{log}\left((h_1(\omega )^1)^{}(x)\right)๐\nu (x)๐(\omega )={\displaystyle _\mathrm{\Gamma }}{\displaystyle _{\mathrm{S}^1}}\mathrm{log}\left((g^1)^{}(x)\right)๐\nu (x)๐\mu (g)=\lambda (\nu ),`$
oรน la derniรจre รฉgalitรฉ est une consรฉquence de la symรฉtrie de $`\mu `$. Par suite, si $`\nu `$ est invariante et ergodique, alors son exposant de Lyapunov est nul. Le cas gรฉnรฉral sโen dรฉduit par un argument de dรฉcomposition ergodique.
Lโobjectif de cet appendice est de donner une dรฉmonstration courte et autocontenue dโune affirmation rรฉciproque. Le rรฉsultat suivant a รฉtรฉ originalement dรฉmontrรฉ par Baxendale dans un contexte bien plus gรฉnรฉral ; ici nous donnons une preuve qui est inspirรฉe de .
###### Proposition 5.16.
Si la mesure $`\mu `$ est (non dรฉgรฉnรฉrรฉe, ร support fini et) symรฉtrique et $`\mathrm{\Gamma }`$ ne prรฉserve aucune mesure de probabilitรฉ du cercle, alors lโexposant de Lyapunov de lโunique mesure stationnaire est strictement nรฉgatif.
Pour la dรฉmonstration notons
$$\psi (x)=_\mathrm{\Gamma }\mathrm{log}\left(g^{}(x)\right)๐\mu (g),$$
et supposons par contradiction que $`\lambda (\nu )0`$, cโest-ร -dire
$$_{\mathrm{S}^1}\psi (x)๐\nu (x)0.$$
(30)
Nous allons dรฉmontrer dans ce cas que $`\mathrm{\Gamma }`$ prรฉserve une mesure de probabilitรฉ, contredisant ainsi notre hypothรจse. Pour cela nous nous appuyons sur un lemme inspirรฉ des travaux de Sullivan sur les cycles feuilletรฉs (voir รฉgalement ). Rappelons que le laplacien $`\mathrm{\Delta }\zeta `$ dโune fonction rรฉelle et continue $`\zeta `$ est dรฉfini par $`\mathrm{\Delta }\zeta =D\zeta \zeta `$, oรน $`D`$ dรฉsigne lโopรฉrateur de diffusion (26).
###### Lemme 5.17.
Sous lโhypothรจse (30), il existe une suite de fonctions continues $`\zeta _n`$ dรฉfinies sur le cercle telles que, pour tout entier $`n`$ et tout point $`x`$ du cercle,
$$\psi (x)+\mathrm{\Delta }\zeta _n(x)\frac{1}{n}.$$
(31)
Dรฉmonstration. Dรฉsignons par $`C(\mathrm{S}^1)`$ lโespace des fonctions continues sur le cercle. Notons $`E`$ le sous-espace constituรฉ des fonctions qui sont des laplaciens de fonctions dans $`C(\mathrm{S}^1)`$, et soit $`C_+`$ le cรดne convexe des fonctions partout positives. Nous devons dรฉmontrer que si $`\psi `$ satisfait (30), alors son image par la projection $`\pi :C(\mathrm{S}^1)C(\mathrm{S}^1)/\overline{E}`$ est contenue dans $`\pi (C_+)`$. Supposons que ce ne soit pas le cas. Le thรฉorรจme de sรฉparation de Hahn-Banach donne alors lโexistence dโune fonctionnelle continue $`\overline{L}:C(\mathrm{S}^1)/\overline{E}`$ telle que $`\overline{L}(\pi (\psi ))<0\overline{L}(\pi (\mathrm{\Phi }))`$ pour tout $`\mathrm{\Phi }C_+`$. Bien sรปr, $`\overline{L}`$ induit une fonctionnelle continue $`L:C(\mathrm{S}^1)`$ qui est identiquement nulle sur $`E`$ et telle que $`L(\psi )<0L(\mathrm{\Phi })`$ pour tout $`\mathrm{\Phi }C_+`$. Nous affirmons que $`L=c\nu `$ pour certain $`c`$ (nous identifions les mesures de probabilitรฉ aux fonctionnelles linรฉaires quโelles dรฉfinissent sur lโespace des fonctions continues). Pour montrer cela, commenรงons par remarquer que, puisque $`L`$ est nul sur $`E`$, pour tout $`\zeta C(\mathrm{S}^1)`$ on a
$$DL,\zeta =L,D\zeta =L,\mathrm{\Delta }\zeta +\zeta =L,\zeta ,$$
cโest-ร -dire que $`L`$ est invariant par la diffusion. Supposons que la dรฉcomposition de Hahn de $`L`$ sโexprime sous la forme
$$L=\alpha \nu _1\beta \nu _2,$$
$`\nu _1`$ et $`\nu _2`$ sont des mesures de probabilitรฉ de supports disjoints du cercle, $`\alpha >0`$ et $`\beta >0`$. Dans ce cas, lโรฉgalitรฉ $`DL=L`$ et lโunicitรฉ de la dรฉcomposition de Hahn pour $`DL`$ montrent que $`\nu _1`$ et $`\nu _2`$ sont elles aussi invariantes par la diffusion. Par suite, $`\nu _1=\nu _2=\nu `$, ce qui contredit le fait que les supports de $`\nu _1`$ et $`\nu _2`$ sont disjoints. La fonctionnelle $`L`$ sโexprime donc sous la forme $`L=c\upsilon `$ pour certaine mesure de probabilitรฉ $`\upsilon `$ du cercle ; lโรฉgalitรฉ $`L=DL`$ donne รฉvidemment $`\upsilon =\nu `$.
Notons maintenant que, puisque
$$0>L(\psi )=c\nu (\psi )=c_{\mathrm{S}^1}\psi (x)๐\nu (x),$$
lโhypothรจse (30) entraรฎne que $`\nu (\psi )>0`$ et $`c<0`$. Or, comme la fonction constante รฉgale ร $`1`$ appartient ร $`C_+`$, nous avons $`c=L(1)0`$. Cette contradiction conclut la dรฉmonstration. $`\mathrm{}`$
Revenons ร la preuve de la proposition 5.16. Quitte ร rajouter une constante ร chaque $`\zeta _n`$, nous pouvons supposer que lโintรฉgrale de $`\mathrm{exp}(\zeta _n)`$ est รฉgale ร $`1`$ pour tout $`n`$. Considรฉrons les mesures de probabilitรฉ $`\nu _n`$ sur le cercle dรฉfinies par
$$\frac{d\nu _n(s)}{ds}=\mathrm{exp}\left(\zeta _n(s)\right).$$
Prenons une sous-suite $`\nu _{n_i}`$ qui converge vers une mesure de probabilitรฉ $`\overline{\nu }`$ sur le cercle. Nous allons montrer que $`\overline{\nu }`$ est invariante par $`\mathrm{\Gamma }`$.
Commenรงons par dรฉmontrer que $`\overline{\nu }`$ est une mesure harmonique. Pour cela, remarquons dโabord que si lโon dรฉsigne par $`Jac_n(g)`$ le jacobien de $`g\mathrm{\Gamma }`$ par rapport ร $`\nu _n`$, alors la relation (31) donne, pour tout point $`x`$ du cercle,
$`{\displaystyle _\mathrm{\Gamma }}\mathrm{log}\left(Jac_n(g)(x)\right)๐\mu (g)`$ $`=`$ $`{\displaystyle _\mathrm{\Gamma }}\mathrm{log}\left(g^{}(x)\right)๐\mu (g)+{\displaystyle _\mathrm{\Gamma }}\left[\zeta _n(g(x))\zeta _n(x)\right]๐\mu (g)`$
$`=`$ $`\psi (x)+\mathrm{\Delta }\zeta _n(x){\displaystyle \frac{1}{n}}.`$
Remarquons maintenant que puisque la diffusion agit continรปment sur lโespace des mesures de probabilitรฉ sur le cercle, la suite de mesures $`D\nu _{n_i}`$ converge faiblement vers la mesure $`D\overline{\nu }`$. Or, la diffusion de $`\nu _n`$ est une mesure absolument continue par rapport ร $`\nu _n`$ dont la densitรฉ sโexprime par la formule
$$\frac{dD\nu _n(x)}{d\nu _n(x)}=_\mathrm{\Gamma }Jac_n(g^1)(x)๐\mu (g)=_\mathrm{\Gamma }Jac_n(g)(x)๐\mu (g).$$
Par la concavitรฉ de la fonction logarithme nous avons
$$\frac{dD\nu _{n_i}(x)}{d\nu _{n_i}(x)}\mathrm{exp}\left(_\mathrm{\Gamma }\mathrm{log}\left(Jac_{n_i}(g)(x)\right)๐\mu (g)\right)\mathrm{exp}(1/n_i),$$
cโest-ร -dire $`D\nu _{n_i}\mathrm{exp}(1/n_i)\nu _{n_i}`$. ร la limite nous obtenons $`D\overline{\nu }\overline{\nu }`$, et puisque $`\overline{\nu }`$ et $`D\overline{\nu }`$ sont des mesures de probabilitรฉ, elles sont รฉgales ; $`\overline{\nu }`$ est donc harmonique.
Nous sommes maintenant en mesure de dรฉmontrer que $`\overline{\nu }`$ est invariante par nโimporte quel รฉlรฉment de $`\mathrm{\Gamma }`$. Pour cela, donnons nous un intervalle $`I`$ tel que $`\overline{\nu }(I)>0`$, et considรฉrons les fonctions $`\psi _{n,I}:\mathrm{\Gamma }]\mathrm{0,1}]`$ dรฉfinies par $`\psi _{n,I}(g)=\nu _n\left(g(I)\right)`$. Les inรฉgalitรฉs suivantes montrent que le laplacien de $`\mathrm{log}(\psi _{n,I})`$ est minorรฉ par $`1/n`$ en lโรฉlรฉment neutre $`e`$ de $`\mathrm{\Gamma }`$ :
$`\mathrm{\Delta }\mathrm{log}(\psi _{n,I})(e)`$ $`=`$ $`{\displaystyle _\mathrm{\Gamma }}\mathrm{log}\left({\displaystyle \frac{\nu _n(g(I))}{\nu _n(I)}}\right)๐\mu (g)={\displaystyle _\mathrm{\Gamma }}\mathrm{log}\left({\displaystyle _I}Jac_n(g)(x){\displaystyle \frac{d\nu _n(x)}{\nu _n(I)}}\right)๐\mu (g)`$
$``$ $`{\displaystyle _\mathrm{\Gamma }}\left({\displaystyle _I}\mathrm{log}\left(Jac_n(g)(x)\right){\displaystyle \frac{d\nu _n(x)}{\nu _n(I)}}\right)๐\mu (g)={\displaystyle _I}\left({\displaystyle _\mathrm{\Gamma }}\mathrm{log}\left(Jac_n(g)(x)\right)๐\mu (g)\right){\displaystyle \frac{d\nu _n(x)}{\nu _n(I)}}{\displaystyle \frac{1}{n}}.`$
Ceci est valable pour tout intervalle $`I`$ vรฉrifiant $`\overline{\nu }(I)>0`$. Par consรฉquent, et ร cause des relations
$$\psi _{n,I}(gf)=\psi _{n,f(I)}(g),$$
le laplacien de $`\mathrm{log}(\psi _{n,I})`$ est partout minorรฉ par $`1/n`$. Donc, si lโon dรฉsigne par $`\psi _I`$ la limite simple de la suite des $`\psi _{n_i,I}`$, i.e.$`\psi _I(g)=\overline{\nu }\left(g(I)\right)`$, alors la fonction $`\mathrm{log}(\psi _I)`$ est surharmonique, dans le sens que son laplacien est positif. Par ailleurs, puisque $`\mu `$ est symรฉtrique, $`\psi _I`$ est harmonique. Par consรฉquent, pour tout $`f\mathrm{\Gamma }`$ nous avons partout รฉgalitรฉ dans les inรฉgalitรฉs suivantes :
$$\mathrm{log}(\psi _I)(f)_\mathrm{\Gamma }\mathrm{log}(\psi _I)(gf)๐\mu (g)\mathrm{log}\left(_\mathrm{\Gamma }\psi _I(gf)๐\mu (g)\right)=\mathrm{log}(\psi _I)(f).$$
La fonction $`\psi _I`$ est donc constante. Or, comme lโintervalle $`I`$ vรฉrifiant $`\overline{\nu }(I)>0`$ est arbitraire, nous en dรฉduisons que la mesure $`\overline{\nu }`$ est invariante par tous les รฉlรฉments de $`\mathrm{\Gamma }`$. La proposition 5.16 est dรฉmontrรฉe.
Bertrand Deroin
IHรS, 35 route de Chartres, 91440 Bures sur Yvette, France (bderoin@ihes.fr)
Victor Kleptsyn
UMPA, ENS-Lyon, 46 allรฉe dโItalie, 69007 Lyon, France (victor.kleptsyn@umpa.ens-lyon.fr)
Andrรฉs Navas
IHรS, 35 route de Chartres, 91440 Bures sur Yvette, France (anavas@ihes.fr)
Univ. de Chile, Las Palmeras 3425, รuรฑoa, Santiago, Chile (andnavas@uchile.cl)
|
warning/0506/math-ph0506067.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Investigation of nonlinear heat (or diffusion if $`u`$ represents mass concentration) equations by means of symmetry methods was started as early as in 1959 with Ovsiannikovโs work where the author performed the group classification of the class of equations of the form
$$u_t=(f(u)u_x)_x.$$
(1)
Nonclassical symmetries of equations from class (1) were investigated in . In particular, the authors of obtained the determining equations for the coefficients of conditional symmetry operators for the wider class of nonlinear reactionโdiffusion equations of the form $`u_t=(f(u)u_x)_x+g(u)`$ and constructed a number of their exact solutions. Review of results on symmetries, exact solutions and conservation laws of such equations is given e.g. in .
The diffusion processes described by (1) are known to arise in different fields of physics such as plasma physics, kinetic theory of gases, solid state and transport in porous medium. In many metals and ceramic materials the diffusion coefficient $`f(u)`$ can, over a wide range of temperatures, be approximated as $`u^\alpha `$, where $`0<\alpha <2`$ . So, one of the mathematical model of the diffusion processes is
$$u_t=(u^\alpha u_x)_x.$$
(2)
Equations (2) are called fast diffusion equations in the case $`0<\alpha <2`$ since these values of $`\alpha `$ correspond to a much faster spread of mass than in the linear case ($`\alpha =0`$).
In this Letter we restrict ourselves with the special case $`\alpha =1`$, i.e. with the equation
$$u_t=(u^1u_x)_x.$$
(3)
It emerges in plasma physics as a model of the cross-field convective diffusion of plasma including mirror effects and in the central limit approximation to Calermanโs model of the Boltzmann equation. Equation (3) governs the expansion of a thermalized electron cloud described by isothermal Maxwell distribution. It is also the one-dimensional Ricci flow equation. (See and references therein.)
Equation (3) has a number of remarkable mathematical properties which distinguish it from class (2). Thus, (3) can be rewritten in the form $`u_t=(\mathrm{ln}u)_{xx}`$ whereas for the other values of $`\alpha `$ the function under $`_{xx}`$ is a power one. It admits a discrete potential invariance transformation. For this equation wide classes of exact solutions were constructed in a closed form while reduction of (2) in the general case results in ordinary differential equations which usually cannot be integrated explicitly. Its potential form admits two kinds of variable separation.
The fact that (3) is written in a conserved form allows us, following Bluman et al. , to consider the corresponding potential system
$$v_x=u,v_t=u^1u_x$$
(4)
and to find potential symmetries of equation (3). Namely, any local symmetry of system (4) induces a symmetry of the initial equation (3). If transformations of some of the โnon-potentialโ variables $`t`$, $`x`$ and $`u`$ explicitly depend on the potential $`v`$, this symmetry is a nonlocal (potential) symmetry of equation (3).
It follows from (4) that the potential $`v`$ satisfies the nonlinear filtration equation
$$v_t=v_x{}_{}{}^{1}v_{xx}^{}$$
(5)
with the special value $`v_x^1`$ of the filtration coefficient. We will also call equation (5) the potential fast diffusion equation. Akhatov, Gazizov and Ibragimov carried out group classification of the nonlinear filtration equations of the general form
$$v_t=f(v_x)v_{xx}$$
(6)
and investigated their contact and quasi-local symmetries .
Lie symmetries of (3) are well known (see Section 2). All its exact solutions constructed in closed form by reduction with Lie symmetries are listed e.g. in .
Some non-Lie exact solutions of (3) were obtained in . Thus, Rosenau found that equation (5) admits, in addition to the usual variable separation $`v=T(t)X(x)`$, the additive one $`v=Y(x+\lambda t)+Z(x\lambda t)`$ which is a potential additive variable separation for equation (3). To construct nonclassical solutions of (3), Qu made use of generalized conditional symmetry method, looking for the conditional symmetry operators in the special form $`Q=(u_{xx}+H(u)u_{x}^{}{}_{}{}^{2}+F(u)u_x+G(u))_u`$. Gandarias investigated some families of usual and potential nonclassical symmetries of (2). In particular, using an ansatz for the coefficient $`\eta `$, she found nontrivial reduction operators in the so-called โno-goโ case when the coefficient of $`_t`$ vanishes, i.e. operators can be reduced to the form $`Q=_x+\eta (t,x,u)_u`$.
In the recent paper a preliminary analysis of nonclassical symmetries of equations from class (6) was performed. A more detailed consideration was carried out for the case $`f=(v_x^2+v_x)^1`$, and only some examples of reduction operators and corresponding exact solutions were constructed. Let us note that equation (6) with $`f=(v_x^2+v_x)^1`$ is reduced by the point transformation $`\stackrel{~}{t}=t`$, $`\stackrel{~}{x}=x+v`$, $`\stackrel{~}{v}=v`$ to equation (5) which corresponds to the value $`\stackrel{~}{f}=\stackrel{~}{v}_{\stackrel{~}{x}}^1`$ and is simpler and more convenient for investigation. All results on symmetries and exact solutions of the equation from can be derived from the analogous results for equation (5).
In this Letter the fast diffusion equation (3) is investigated from the symmetry point of view. The nonclassical symmetries of the corresponding potential equation (5) are completely classified with respect to its Lie symmetry group. As a result, new wide classes of potential nonclassical symmetries of equation (3) are found. Some classes of potential nonclassical symmetries prove to be connected with usual nonclassical ones on the solution set of potential system (4). The set of exact non-Lie solutions constructed in is supplemented with the similar ones. It is shown that all known non-Lie solutions of the fast diffusion equation are exhausted by ones which can be constructed with the above potential nonclassical symmetries.
Our Letter is organized as follows. First of all (Section 2) we adduce results on Lie and potential symmetries of (3), including discrete ones. It is important since classical symmetries really are partial cases of nonclassical symmetries and below we solve the problem on finding only pure nonclassical symmetries which are not equivalent to classical ones. Moreover, our approach is based on application of the notion of equivalence of nonclassical symmetries with respect to a transformation group, which is developed and investigated in Section 3. Usage of equivalence with respect to the complete Lie invariance group including the discrete symmetries plays a significant role in simplification of proof, testing and improving presentation of the main result (Theorem 1, Section 4). In spite of the techniques applied in and similarly to , we use the single potential equation (5) instead of potential system (4), to produce potential nonclassical symmetries of (3). After the โno-goโ case of the zero coefficient of $`_t`$ is discussed, all the reduction operators having the nonvanishing coefficient of $`_t`$ are classified. Connection between partial classes of usual and potential reduction operators of (1) is studied in Section 5. In Section 6 the known Lie solutions of (3) and (5) are collected. A list of non-Lie solutions is supplemented with the similar ones. Connections between exact solutions and different ways of their construction are discussed shortly. In conclusion some recent results on nonclassical symmetries of equations (6) are announced.
## 2 Lie and potential symmetries of fast diffusion equation
The Lie invariance algebra
$$A_1=_t,_x,t_t+u_u,x_x2u_u$$
of equation (3) was found in . The complete Lie invariance group $`G_1`$ of (3) is generated by both continuous one-parameter transformation groups with infinitesimal operators from $`A_1`$ and two involution transformations of alternating sign in the sets $`\{t,u\}`$ and $`\{x\}`$. Action of any element from $`G_1`$ on the function $`u`$ is given by the formula
$$\stackrel{~}{u}(t,x)=\epsilon _3^1\epsilon _4^2u(\epsilon _3t+\epsilon _1,\epsilon _4x+\epsilon _2),$$
where $`\epsilon _1`$, โฆ, $`\epsilon _4`$ are arbitrary constants, $`\epsilon _3\epsilon _40`$ .
The Lie symmetry properties of (3) are common for diffusion equations. Uncommonness of equation (3) from the symmetry point of view becomes apparent after introducing the potential $`v`$ and considering potential system (4) or potential equation (5). Point and nonclassical symmetries of (4) or (5) are called *potential* and *potential nonclassical* symmetries of (3) correspondingly.
The Lie invariance algebra
$$A_2=_t,_x,_v,t_t+v_v,x_xv_v$$
of equation (5) and the corresponding connected Lie symmetry group are quite ordinary for nonlinear filtration equations. However, equation (5) is distinguished for its discrete symmetries since it possesses, besides two usual sign changes in the variable sets $`\{t,v\}`$ and $`\{x,v\}`$, the hodograph transformation $`\stackrel{~}{t}=t`$, $`\stackrel{~}{x}=v`$, $`\stackrel{~}{v}=x`$. These three involutive transformations together with the continuous one-parameter transformation groups having infinitesimal operators from $`A_2`$ generate the complete Lie invariance group $`G_2`$ of (5). Therefore, $`G_2`$ consists of the transformations
$$\begin{array}{c}\stackrel{~}{t}=\epsilon _3t+\epsilon _1,\stackrel{~}{x}=\epsilon _4x+\epsilon _2,\stackrel{~}{v}=\epsilon _3\epsilon _{4}^{}{}_{}{}^{1}v\text{and}\hfill \\ \stackrel{~}{t}=\epsilon _3t+\epsilon _1,\stackrel{~}{x}=\epsilon _3\epsilon _{4}^{}{}_{}{}^{1}v,\stackrel{~}{v}=\epsilon _4x+\epsilon _2,\hfill \end{array}$$
where $`\epsilon _1`$, โฆ, $`\epsilon _4`$ are arbitrary constants, $`\epsilon _3\epsilon _40`$.
A similar result is true for system (4). Namely, it is invariant with respect to the following transformation
$$\stackrel{~}{t}=t,\stackrel{~}{x}=v,\stackrel{~}{u}=u^1,\stackrel{~}{v}=x$$
(7)
which is additional to the usual Lie symmetry group $`G_1`$ of equation (3) and is called the potential hodograph transformation of this equation.
It can be proved that the set of Lie invariant solutions of equation (3) is closed under transformation (7).
## 3 Equivalence of reduction operators <br>with respect to transformation groups
The notion of nonclassical symmetry was introduced in 1969 . A precise and rigorous definition was suggested later (see e.g. ).
Consider an $`r`$th order differential equation $``$ of the form $`L(t,x,u_{(r)})=0`$ for the unknown function $`u`$ of two independent variables $`t`$ and $`x`$. Here $`u_{(r)}`$ denotes the set of all the derivatives of the function $`u`$ with respect to $`t`$ and $`x`$ of order not greater than $`r`$, including $`u`$ as the derivative of order zero. Within the local approach the equation $``$ is treated as an algebraic equation in the jet space $`J^{(r)}`$ of order $`r`$ and is identified with the manifold of its solutions in $`J^{(r)}`$:
$$=\{(t,x,u_{(r)})J^{(r)}|L(t,x,u_{(r)})=0\}.$$
The set of (first-order) differential operators of the general form
$$Q=\tau (t,x,u)_t+\xi (t,x,u)_x+\eta (t,x,u)_u,(\tau ,\xi )(0,0),$$
will be denoted by $`๐ฌ`$. Here and below $`_t=/t`$, $`_x=/x`$ and $`_u=/u`$. Subscripts of functions denote differentiation with respect to the corresponding variables.
Two differential operators $`\stackrel{~}{Q}`$ and $`Q`$ are called *equivalent* if they differ by a multiplier being a non-vanishing function of $`t`$, $`x`$ and $`u`$: $`\stackrel{~}{Q}=\lambda Q`$, where $`\lambda =\lambda (t,x,u)`$, $`\lambda 0`$. The equivalence of operators will be denoted by $`\stackrel{~}{Q}Q`$. Denote also the result of factorization of $`๐ฌ`$ with respect to this equivalence relation by $`๐ฌ_\mathrm{f}`$. Elements of $`๐ฌ_\mathrm{f}`$ will be identified with their representatives in $`๐ฌ`$.
The first-order differential function $`Q[u]:=\eta (t,x,u)\tau (t,x,u)u_t\xi (t,x,u)u_x`$ is called the characteristic of the operator $`Q`$. The characteristic PDE $`Q[u]=0`$ (called also the *invariant surface condition*) has two functionally independent integrals $`\zeta (t,x,u)`$ and $`\omega (t,x,u)`$. Therefore, the general solution of this equation can be implicitly presented in the form $`F(\zeta ,\omega )=0`$, where $`F`$ is an arbitrary function of its arguments.
The characteristic equations of equivalent operators have the same set of solutions. And vice versa, any family of two functionally independent functions of $`t`$, $`x`$ and $`u`$ is a complete set of integrals of the characteristic equation of a differential operator. Therefore, there exists a one-to-one correspondence between $`๐ฌ_\mathrm{f}`$ and the set of families of two functionally independent functions of $`t`$, $`x`$ and $`u`$, which is factorized with respect to the corresponding equivalence. (Two families of the same number of functionally independent functions of the same arguments are considered equivalent if any function from one of the families is functionally dependent on functions from the other family.)
Since $`(\tau ,\xi )(0,0)`$ we can assume without loss of generality that $`\zeta _u0`$ and $`F_\zeta 0`$ and resolve the equation $`F=0`$ with respect to $`\zeta `$: $`\zeta =\phi (\omega )`$. This implicit representation of the function $`u`$ is called an *ansatz* corresponding to the operator $`Q`$.
Denote the manifold defined by the set of all the differential consequences of the characteristic equation $`Q[u]=0`$ in $`J^{(r)}`$ by $`๐ฌ^{(r)}`$, i.e.
$$๐ฌ^{(r)}=\{(t,x,u_{(r)})J^{(r)}|D_t^\alpha D_x^\beta Q[u]=0,\alpha ,\beta \{0\},\alpha +\beta <r\},$$
where $`D_t=_t+u_{\alpha +1,\beta }_{u_{\alpha \beta }}`$ and $`D_x=_x+u_{\alpha ,\beta +1}_{u_{\alpha \beta }}`$ are the operators of total differentiation with respect to the variables $`t`$ and $`x`$, the variable $`u_{\alpha \beta }`$ of the jet space $`J^{(r)}`$ corresponds to the derivative $`^{\alpha +\beta }u/t^\alpha x^\beta `$.
###### Definition 1.
The differential equation $``$ is called *conditionally invariant* with respect to the operator $`Q`$ if the relation $`Q_{(r)}L(t,x,u_{(r)})|_{๐ฌ^{(r)}}=0`$ holds, which is called the *conditional invariance criterion*. Then $`Q`$ is called an operator of *conditional symmetry* (or $`Q`$-conditional symmetry, nonclassical symmetry, etc.) of the equation $``$.
In Definition 1 the symbol $`Q_{(r)}`$ stands for the standard $`r`$th prolongation of the operator $`Q`$ : $`Q_{(r)}=Q+_{0<\alpha +\beta r}\eta ^{\alpha \beta }_{u_{\alpha \beta }}`$, where $`\eta ^{\alpha \beta }=D_t^\alpha D_x^\beta Q[u]+\tau u_{\alpha +1,\beta }+\xi u_{\alpha ,\beta +1}`$.
The equation $``$ is conditionally invariant with respect to the operator $`Q`$ iff the ansatz constructed with this operator reduces $``$ to an ordinary differential equation . So, we will also call operators of conditional symmetry by reduction operators of $``$.
###### Lemma 1 ().
If the equation $``$ is conditionally invariant with respect to the operator $`Q`$, then it is conditionally invariant with respect to any operator which is equivalent to $`Q`$.
The set of reduction operators of the equation $``$ is a subset of $`๐ฌ`$ and so will be denoted by $`๐ฌ()`$. In view of Lemma 1, $`Q๐ฌ()`$ and $`\stackrel{~}{Q}Q`$ imply $`\stackrel{~}{Q}๐ฌ()`$, i.e. $`๐ฌ()`$ is closed under the equivalence relation on $`๐ฌ`$. Therefore, factorization of $`๐ฌ`$ with respect to this equivalence relation can be naturally restricted on $`๐ฌ()`$ that results in the subset $`๐ฌ_\mathrm{f}()`$ of $`๐ฌ_\mathrm{f}`$. As in the whole set $`๐ฌ_\mathrm{f}`$, we identify elements of $`๐ฌ_\mathrm{f}()`$ with their representatives in $`๐ฌ()`$. In this approach the problem of complete description of reduction operators for the equation $``$ is nothing but the problem of finding $`๐ฌ_\mathrm{f}()`$.
We can essentially simplify and order classification of reduction operators, additionally taking into account Lie symmetry transformations of an equation or equivalence transformations of a whole class of equations.
###### Lemma 2.
Any point transformation of $`t`$, $`x`$ and $`u`$ induces a one-to-one mapping of $`๐ฌ`$ into itself. Namely, the transformation $`g`$: $`\stackrel{~}{t}=T(t,x,u)`$, $`\stackrel{~}{x}=X(t,x,u)`$, $`\stackrel{~}{u}=U(t,x,u)`$ generates the mapping $`g_{}:๐ฌ๐ฌ`$ such that the operator $`Q`$ is mapped to the operator $`g_{}Q=\stackrel{~}{\tau }_{\stackrel{~}{t}}+\stackrel{~}{\xi }_{\stackrel{~}{x}}+\stackrel{~}{\eta }_{\stackrel{~}{u}}`$, where $`\stackrel{~}{\tau }(\stackrel{~}{t},\stackrel{~}{x},\stackrel{~}{u})=QT(t,x,u)`$, $`\stackrel{~}{\xi }(\stackrel{~}{t},\stackrel{~}{x},\stackrel{~}{u})=QX(t,x,u)`$, $`\stackrel{~}{\eta }(\stackrel{~}{t},\stackrel{~}{x},\stackrel{~}{u})=QU(t,x,u)`$. If $`Q^{}Q`$ then $`g_{}Q^{}g_{}Q`$. Therefore, the corresponding factorized mapping $`g_\mathrm{f}:๐ฌ_\mathrm{f}๐ฌ_\mathrm{f}`$ also is well-defined and one-to-one.
###### Definition 2 ().
The differential operators $`Q`$ and $`\stackrel{~}{Q}`$ are called equivalent with respect to a group $`G`$ of point transformations if there exists $`gG`$ for which the operators $`Q`$ and $`g_{}\stackrel{~}{Q}`$ are equivalent. Notation: $`Q\stackrel{~}{Q}modG.`$
###### Lemma 3.
Given any point transformation $`g`$ of the equation $``$ to an equation $`\stackrel{~}{}`$, $`g_{}`$ maps $`๐ฌ()`$ to $`๐ฌ(\stackrel{~}{})`$ in a one-to-one manner. The same statement is true for the factorized mapping $`g_\mathrm{f}`$ from $`๐ฌ_\mathrm{f}()`$ to $`๐ฌ_\mathrm{f}(\stackrel{~}{})`$.
###### Corollary 1.
Let $`G`$ be a Lie symmetry group of the equation $`.`$ Then the equivalence of operators with respect to the group $`G`$ generates equivalence relations in $`๐ฌ()`$ and in $`๐ฌ_\mathrm{f}()`$.
Consider the class $`|_๐ฎ`$ of equations $`_\theta `$: $`L(t,x,u_{(r)},\theta (t,x,u_{(r)}))=0`$ parameterized with the parameter-functions $`\theta =\theta (t,x,u_{(r)}).`$ Here $`L`$ is a fixed function of $`t`$, $`x`$, $`u_{(r)}`$ and $`\theta .`$ $`\theta `$ denotes the tuple of arbitrary (parametric) functions $`\theta (t,x,u_{(r)})=(\theta ^1(t,x,u_{(r)}),\mathrm{},\theta ^k(t,x,u_{(r)}))`$ running the set $`๐ฎ`$ of solutions of the system $`S(t,x,u_{(r)},\theta _{(q)}(t,x,u_{(r)}))=0`$. This system consists of differential equations on $`\theta `$, where $`t`$, $`x`$ and $`u_{(r)}`$ play the role of independent variables and $`\theta _{(q)}`$ stands for the set of all the partial derivatives of $`\theta `$ of order not greater than $`q`$. In what follows we call the functions $`\theta `$ arbitrary elements. Denote the point transformations group preserving the form of the equations from $`|_๐ฎ`$ by $`G^{}`$.
Let $`P`$ denote the set of the pairs each of which consists of an equation $`_\theta `$ from $`|_๐ฎ`$ and an operator $`Q`$ from $`๐ฌ(_\theta )`$. In view of Lemma 3, action of transformations from the equivalence group $`G^{}`$ on $`|_๐ฎ`$ and $`\{๐ฌ(_\theta )|\theta ๐ฎ\}`$ together with the pure equivalence relation of differential operators naturally generates an equivalence relation on $`P`$.
###### Definition 3.
Let $`\theta ,\theta ^{}๐ฎ`$, $`Q๐ฌ(_\theta )`$, $`Q^{}๐ฌ(_\theta ^{})`$. The pairs $`(_\theta ,Q)`$ and $`(_\theta ^{},Q^{})`$ are called $`G^{}`$-equivalent if there exists $`gG^{}`$ which transforms the equation $`_\theta `$ to the equation $`_\theta ^{}`$, and $`Q^{}g_{}Q`$.
Classification of reduction operators with respect to $`G^{}`$ will be understood as classification in $`P`$ with respect to the above equivalence relation. This problem can be investigated in the way that is similar to usual group classification in classes of differential equations. Namely, we construct firstly the reduction operators that are defined for all values of the arbitrary elements. Then we classify, with respect to the equivalence group, the values of arbitrary elements for each of that the equation $`_\theta `$ admits additional reduction operators.
In an analogues way we also can introduce equivalence relations on $`P`$, which are generated by either generalizations of usual equivalence groups or all admissible point transformations (called also form-preserving ones ) in pairs of equations from $`|_๐ฎ`$.
## 4 Reduction operators <br>of nonlinear filtration equation
In this section we describe $`G_2`$-inequivalent reduction operators of the potential fast diffusion equation (5). Here reduction operators have the general form $`Q=\tau _t+\xi _x+\theta _v`$, where $`\tau `$, $`\xi `$ and $`\theta `$ are functions of $`t`$, $`x`$ and $`v`$, and $`(\tau ,\xi )(0,0)`$. Since (5) is an evolution equation, there are two principally different cases of finding $`Q`$: $`\tau 0`$ and $`\tau =0`$.
In the case $`\tau =0`$ we have $`\xi 0`$, and up to the usual equivalence of reduction operators we can assume that $`\xi =1`$, i.e. $`Q=_x+\theta _v`$. The conditional invariance criterion implies only one determining equation on the coefficient $`\theta `$
$$\theta \theta _t=\theta _{xx}+2\theta \theta _{xv}+\theta ^2\theta _{vv}\theta ^1(\theta _x)^22\theta _x\theta _v\theta (\theta _v)^2$$
which is reduced with a non-point transformation to equation (5), where $`\theta `$ becomes a parameter. That is why the case $`\tau =0`$ is called the โno-goโ one. It is characteristic for evolution equations in general. First the โno-goโ case was completely investigated for the one-dimensional linear heat equation in . It was proved that the problem of finding the conditional symmetry operators with the vanishing coefficient of $`_t`$ is reduced to solving the initial equation. In the proof was extended to the class of $`(1+1)`$-dimensional evolution equations and in this result was generalized for evolution equations with $`n`$ space variables.
Let us note that โno-goโ has to be treated as impossibility of exhaustive solving of the problem. At the same time, imposing additional constraints on the coefficient $`\theta `$, one can construct a number of particular examples of operators with $`\tau =0`$ and then apply them to finding exact solutions of the initial equation. It is the approach that was used in for fast diffusion equation (3). Since the determining equation has more independent variables and, therefore, more freedom degrees, it is more convenient often to guess a simple solution or a simple ansatz for the determining equation, which can give a parametric set of complicated solutions of the initial equation. (Similar situation is for Lie symmetries of first-order ordinary differential equations.)
Consider the case $`\tau 0`$ which admits complete solving unlike the previous case. We can assume $`\tau =1`$ up to the usual equivalence of reduction operators. Then the determining equations for the coefficients $`\xi `$ and $`\theta `$ have the form
$$\begin{array}{c}\xi _{vv}=\xi \xi _v,\xi _t=2\xi _{xv}\theta _{vv}\theta _v\xi +\theta \xi _v\xi \xi _x,\hfill \\ \theta _{xx}=\theta \theta _x,\theta _t=2\theta _{xv}\xi _{xx}\xi _x\theta +\xi \theta _x\theta \theta _v.\hfill \end{array}$$
(10)
###### Theorem 1.
A complete list of $`G_2`$-inequivalent non-Lie reduction operators of the potential fast diffusion equation (5) is exhausted by the following ones:
$`1._t+\epsilon _x+f(\omega )_v,\text{where}\omega =x+\epsilon t;`$
$`2._t+f(\omega )(_x+_v),\text{where}\omega =x+v;`$
$`3._t+\xi _x+(\phi _t+\phi _x\xi )_v,\text{where}\xi ={\displaystyle \frac{2}{v+\phi }},\phi \{t+e^x,tf(x)\};`$
$`4._t+\xi _x{\displaystyle \frac{\chi _t+\chi _x\xi }{1+\chi ^2}}_v,\text{where}\xi =2{\displaystyle \frac{1+\chi \mathrm{tan}v}{\mathrm{tan}v\chi }},`$
$`\chi \{\mathrm{tan}(2t)\mathrm{tanh}x,\mathrm{coth}(2t)\mathrm{cot}x\};`$
$`5._t+\xi _x{\displaystyle \frac{\chi _t+\chi _x\xi }{1\chi ^2}}_v,\text{where}\xi =2{\displaystyle \frac{1\chi \mathrm{tanh}v}{\mathrm{tanh}v\chi }},`$
$`\chi \{\mathrm{tanh}(2t)\mathrm{tanh}x,\mathrm{tanh}(2t)\mathrm{coth}x,\mathrm{coth}(2t)\mathrm{coth}x,{\displaystyle \frac{e^{2x}\mathrm{tanh}2t+1}{e^{2x}\mathrm{tanh}2t}},{\displaystyle \frac{2e^{2x}e^{4t}}{2+e^{2x}+e^{4t}}}\}.`$
Here $`\epsilon \{0,\mathrm{\hspace{0.17em}1}\}`$, $`f`$ is an arbitrary nonconstant solution of the ordinary differential equation $`f_{\omega \omega }=ff_\omega `$, i.e. $`f\{2/\omega ,2\mathrm{cot}\omega ,2\mathrm{tanh}\omega ,2\mathrm{coth}\omega \}modG_2.`$
###### Proof.
Here we only outline a sketch of proof. Any solution of the equation $`\xi _{vv}=\xi \xi _v`$ belongs to the set $`\{\phi ,2/(v+\phi ),2\mu \mathrm{cot}\omega ,2\mu \mathrm{tanh}\omega ,2\mu \mathrm{coth}\omega \},`$ where $`\omega =\mu (v+\phi )`$, $`\mu `$ and $`\phi `$ are arbitrary functions of $`t`$ and $`x`$, $`\mu 0`$. The second equation of (10) is a linear inhomogeneous second-order ordinary differential equation with respect to $`\theta `$, where $`v`$ is the independent variable and $`t`$ and $`x`$ are assumed parameters. It is possible to construct its partial exact solution without irrational singularities for any above value of $`\xi `$. The solutions of the corresponding homogeneous equation have irrational singularities if $`\xi _v0`$. In view of the other equations of (10), the part of $`\theta `$ containing such singularities has to vanish identically. Moreover, $`\mu =const`$, i.e. $`\mu =1modG_2`$, and $`\phi `$ satisfies an overdetermined system of differential equation in $`t`$ and $`x`$. Integration of it for all above values of $`\xi `$ and classification of obtained solutions up to equivalence with respect to $`G_2`$ with excluding Lie cases result in the statement of the theorem. โ
All operators from Theorem 1 are potential nonclassical symmetries of equation (3).
## 5 Connection between classes of <br>nonclassical and potential nonclassical symmetries
Let us investigate connection between reduction operators of equations (3) and (5).
As mentioned in the introduction, one of the problems studied in was construction of partial classes of reduction operators for diffusion equations (2) in the โno-goโ case when operators can be reduced to the form $`_x+\eta (t,x,u)_u`$. Namely, Gandarias proposed to look for the coefficient $`\eta `$ with the ansatz
$$\eta =\frac{\eta ^1(t,x)u+\eta ^2(t,x)}{f(u)},$$
(11)
where $`f(u)u^\alpha `$. After substituting (11) in the determining equation for $`\eta `$ and splitting with respect to $`u`$, one obtains an overdetermined system for the functions $`\eta ^1`$ and $`\eta ^2`$. For the fast diffusion equation (3) this system has the form
$$\eta _{xx}^2=\eta ^2\eta _x^2,\eta _t^2=\eta ^2\eta _x^1\eta ^1\eta _x^2+\eta _{xx}^1,\eta _t^1=\eta ^1\eta _x^1.$$
(12)
System (12) can be derived from (10) with reduction by the group of translations with respect to $`v`$, i.e. with assuming that $`\xi `$ and $`\theta `$ do not depend on $`v`$ and re-denoting $`\xi =\eta ^1`$, $`\theta =\eta ^2`$.
This observation can be easily explained in a rigorous way for any pair of equations (1) and (6) with the same function $`f`$.
Consider reduction operators $`Q=_t+\xi _x+\theta _v`$ and $`Q^{}=_x+\eta _u`$ of equations (6) and (1) correspondingly, where the coefficients $`\xi `$ and $`\theta `$ depend only on $`t`$ and $`x`$, the coefficient $`\eta `$ is defined by (11). The conditional invariance criterion applied to equation (6) (or (1) ) and the operator $`Q`$ ($`Q^{}`$) implies the following determining equation on $`\xi `$ and $`\theta `$ ($`\eta ^1`$ and $`\eta ^2`$):
$$(\xi \xi _xv_x{}_{}{}^{2}(\xi _x\theta +\xi \theta _x)v_x+\theta \theta _x)f^{}(v_x)$$
$$+((\xi _t+2\xi \xi _x)v_x\theta _t2\theta \xi _x)f(v_x)+(\xi _{xx}v_x+\theta _{xx})(f(v_x))^2=0$$
$$(\text{or}(\eta ^1u+\eta ^2)(\eta _x^1u+\eta _x^2)f^{}(u)$$
$$((\eta _t^12\eta ^1\eta _x^1)u+\eta _t^02\eta ^0\eta _x^1)f(u)+(\eta _{xx}^1u+\eta _{xx}^2)(f(u))^2=0)$$
which has to be additionally split with respect to $`v_x`$ ($`u`$). It is obvious that the systems obtained in the both cases after splitting coincide under the supposition $`\eta ^1=\xi `$, $`\eta ^2=\theta `$. The characteristic equation $`Q[v]=\theta v_t\xi v_x=0`$ can be rewritten on the manifold of solutions of the potential system
$$v_x=u,v_t=f(u)u_x$$
(13)
in the form
$$u_x\frac{\xi u+\theta }{f(u)}=0$$
and coincides in this way with the characteristic equation $`Q^{}[u]=0`$.
Therefore, the following proposition is true.
###### Proposition 1.
$`Q=_t+\xi _x+\theta _v`$, where $`\xi =\xi (t,x)`$ and $`\theta =\theta (t,x)`$, is a reduction operator of equation (6) iff
$$Q^{}=_x+\frac{\xi u+\theta }{f(u)}_u$$
is a reduction operator of equation (1).
System (13) establishes connection between the corresponding sets of invariant solutions.
## 6 Exact solutions of fast diffusion <br>and nonlinear filtration equations
All invariant solutions of (3) and (5), which were earlier constructed in closed forms with the classical Lie method, were collected e.g. in . A complete list of $`G_1`$-inequivalent solutions of such type is exhausted by the following ones:
$$\begin{array}{cc}& 1)u=\frac{1}{1+\epsilon e^{x+t}},v=\mathrm{ln}|e^x+\epsilon e^t|;\hfill \\ & 2)u=e^x,v=e^x+t;\hfill \\ & 3)u=\frac{1}{xt+\mu te^{x/t}},v=\mathrm{ln}|t|+\frac{d\vartheta }{\vartheta 1+\mu e^\vartheta }|_{\vartheta =x/t};\hfill \\ & 4)u=\frac{2t}{x^2+\epsilon t^2},v|_{\epsilon =0}=\frac{2t}{x},v|_{\epsilon =1}=2\mathrm{arctan}\frac{x}{t},v|_{\epsilon =1}=\mathrm{ln}|\frac{xt}{x+t}|;\hfill \\ & 5)u=\frac{2t}{\mathrm{cos}^2x},v=2t\mathrm{tan}x;\hfill \\ & 6)u=\frac{2t}{\mathrm{cosh}^2x},v=2t\mathrm{tanh}x;\hfill \\ & 7)u=\frac{2t}{\mathrm{sinh}^2x},v=2t\mathrm{coth}x.\hfill \end{array}$$
(14)
Here $`\epsilon `$ and $`\mu `$ are arbitrary constants, $`\epsilon \{1,0,1\}modG_1`$. The below arrows denote the possible transformations of solutions (14) to each other by means of the potential hodograph transformation (7) up to translations with respect to $`x`$ :
$$\begin{array}{cc}& \mathrm{\hspace{0.33em}1})_{\epsilon =0};1)_{\epsilon =1}1)_{\epsilon =1,x+t<0};\mathrm{\hspace{0.33em}1})_{\epsilon =1,x+t>0};2)3)_{\mu =0,x>t};\hfill \\ & \mathrm{\hspace{0.33em}4})_{\epsilon =0};5)4)_{\epsilon =4};6)4)_{\epsilon =4,|x|<2|t|};7)4)_{\epsilon =4,|x|>2|t|}.\hfill \end{array}$$
The sixth connection was known earlier . If $`\mu 0`$ solution 3) from list (14) is mapped by (7) to the solution
$$8)u=t\vartheta (\omega )t+\mu te^{\vartheta (\omega )},\omega =x\mathrm{ln}|t|,$$
which is invariant with respect to the algebra $`t_t+_x+u_u`$. Here $`\vartheta `$ is the function determined implicitly by the formula $`(\vartheta 1+\mu e^\vartheta )^1๐\vartheta =\omega .`$
Some classes of non-Lie exact solutions of (3) were obtained in . These solutions and the ones similar to them can be represented uniformly over the complex field as compositions of two simple waves moving with the same โvelocitiesโ in opposite directions:
$$\begin{array}{cc}& u=\frac{\alpha ^2}{\beta }(\mathrm{cot}(\alpha x+\beta t+\gamma )+\mathrm{cot}(\alpha x\beta t+\delta ))\hfill \\ & =\frac{\alpha ^2}{\beta }\frac{2\mathrm{sin}(2\beta t+\gamma \delta )}{\mathrm{cos}(2\beta t+\gamma \delta )\mathrm{cos}(2\alpha x+\gamma +\delta )},\hfill \end{array}$$
(15)
where $`\alpha `$, $`\beta `$, $`\gamma `$ and $`\delta `$ are complex constants, $`\alpha \beta 0`$. It can be proved that function (15) takes real values (for real $`x`$ and $`t`$) iff up to transformations from $`G_1`$
$$(\alpha ,\beta ,\gamma ,\delta )\{(1,1,0,0),(i,i,0,0),(i,i,\pi /2,0),(i,i,\pi /2,\pi /2),(i,1,0,0),(1,i,0,0)\}.$$
Using representation (15) and the above values of tuples $`(\alpha ,\beta ,\gamma ,\delta )`$, we obtain the following solutions of fast diffusion equation (3) and nonlinear filtration equation (5):
$$\begin{array}{cc}& 1^{})u=\mathrm{cot}(xt)\mathrm{cot}(x+t)=\frac{2\mathrm{sin}2t}{\mathrm{cos}2t\mathrm{cos}2x},v=\mathrm{ln}|\frac{\mathrm{sin}(xt)}{\mathrm{sin}(x+t)}|;\hfill \\ & 2^{})u=\mathrm{coth}(xt)\mathrm{coth}(x+t)=\frac{2\mathrm{sinh}2t}{\mathrm{cosh}2x\mathrm{cosh}2t},v=\mathrm{ln}|\frac{\mathrm{sinh}(xt)}{\mathrm{sinh}(x+t)}|;\hfill \\ & 3^{})u=\mathrm{coth}(xt)\mathrm{tanh}(x+t)=\frac{2\mathrm{cosh}2t}{\mathrm{sinh}2x\mathrm{sinh}2t},v=\mathrm{ln}|\frac{\mathrm{sinh}(xt)}{\mathrm{cosh}(x+t)}|;\hfill \\ & 4^{})u=\mathrm{tanh}(xt)\mathrm{tanh}(x+t)=\frac{2\mathrm{sinh}2t}{\mathrm{cosh}2x+\mathrm{cosh}2t},v=\mathrm{ln}|\frac{\mathrm{cosh}(xt)}{\mathrm{cosh}(x+t)}|;\hfill \\ & 5^{})u=\mathrm{cot}(ix+t)\mathrm{cot}(ixt)=\frac{2\mathrm{sin}2t}{\mathrm{cosh}2x\mathrm{cos}2t},v=2\mathrm{arctan}(\mathrm{cot}t\mathrm{tanh}x);\hfill \\ & 6^{})u=i\mathrm{cot}(x+it)i\mathrm{cot}(xit)=\frac{2\mathrm{sinh}2t}{\mathrm{cosh}2t\mathrm{cos}2x},v=2\mathrm{arctan}(\mathrm{coth}t\mathrm{tan}x).\hfill \end{array}$$
(All twoโs in the latter expressions for $`u`$ can be moved over with scale transformations from $`G_1`$.)
Transformation (7) acts on the set of solutions $`1^{})`$$`6^{})`$ in the following way:
$$\begin{array}{cc}& 1^{})_{\mathrm{cos}2t<\mathrm{cos}2x}5^{})|_{tt+\pi /2,xx/2,v2v};1^{})_{\mathrm{cos}2t>\mathrm{cos}2x}5^{})|_{xx/2,v2v\pi };\hfill \\ & 2^{})_{|x|<|t|}4^{})|_{xx/2,v2v};\mathrm{\hspace{0.33em}2}^{})_{|x|>|t|}|_{xx/2,v2v};\hfill \\ & \mathrm{\hspace{0.33em}3}^{})_{x<t}|_{xx/2,v2v};3^{})_{x>t}3^{})_{x>t}|_{xx/2,v2v};\mathrm{\hspace{0.33em}6}^{})|_{xx/2,v2v}.\hfill \end{array}$$
These actions can be interpreted in terms of actions of transformation (7) on the nonclassical symmetry operators which correspond to solutions $`1^{})`$$`6^{})`$.
In Rosenau took advantage of additive separation of variables for the potential fast diffusion equation (5) and constructed solution $`4^{})`$. Using the generalized conditional symmetry method, Qu found solutions which can be written in forms $`1^{})`$ and $`6^{})`$. After rectifying computations in two cases from , one can find also solutions $`2^{})`$ and $`5^{})`$. Solutions $`1^{})`$, $`3^{})`$ and $`4^{})`$ were obtained in , at least, in one from the above forms, but equivalence of these forms was not shown there.
One of techniques which can be applied for finding the above solutions is reduction by conditional symmetry operators of the form $`Q=_x+(\eta ^1(t,x)u+\eta ^2(t,x))u_u`$ (see for details). Namely, solutions $`1^{})`$, $`3^{})`$ and $`4^{})`$ are obtained with the following operators:
$$_x+(u^22\mathrm{cot}(xt)u)_u,_x+(u^22\mathrm{coth}(xt)u)_u\text{and}_x+(u^22\mathrm{tanh}(xt)u)_u.$$
We supplement the list of solutions adduced in with similar ones, namely, with $`2^{})`$ and $`5^{})`$. Solution $`2^{})`$ can be also constructed by reduction with the second above operator. Real solutions $`5^{})`$ and $`6^{})`$ correspond to the similar operators
$$_x+(iu^22\mathrm{coth}(xit)u)_u\text{and}_x(iu^22i\mathrm{coth}(tix)u)_u$$
with complex-valued coefficients.
All reductions performed with reduction operators from Theorem 1 or with equivalent ones result in solutions which are equivalent to the listed Lie solutions 1)โ7) or solutions $`1^{})`$$`6^{})`$. For example, the operators
$$_t_x2\mathrm{cot}(xt)_v,_t_x2\mathrm{coth}(xt)_v\text{and}_t_x2\mathrm{tanh}(xt)_v$$
lead to solutions $`1^{})`$, $`3^{})`$ and $`4^{})`$ correspondingly (see also Section 5).
## 7 Conclusion
In this Letter we present classification of reduction operators for nonlinear filtration equation (5) with summary of necessary notions and statements, a basic sketch of the proof and a list of constructed exact solutions including both Lie and non-Lie ones. Since (5) is the potential equation of fast diffusion equation (3), all the obtained operators are potential nonclassical symmetries of (3). Moreover, most of them are nonprojectible on the space of the independent variables $`t`$ and $`x`$ that leads to technically cumbersome implicit reductions of (5) to ordinary differential equations. Now we optimize the proof and hope to realize it in a quite compact form. Presentation of the proof and reduction technics will be subjects of a future paper.
We continue our investigation on potential reduction operators of the nonlinear diffusion equations from class (1). In some sense, equation (3) is singular in this class with the potential nonclassical symmetry point of view. More precisely, as a result of joint work with Prof. Sophocleous the following theorem have been proved recently.
###### Theorem 2.
Nonlinear filtration equations (6) admit non-Lie reduction operators with non-vanishing coefficients of $`_t`$ only in the case of the Fujitaโs nonlinearities
$$f(v_x)=\frac{1}{av_x{}_{}{}^{2}+bv_x+c},$$
where $`a`$, $`b`$ and $`c`$ are constants.
Let us note that there are exactly three $`G^{}`$-inequivalent cases of the Fujitaโs nonlinearities:
$$f(v_x)=1,f(v_x)=\frac{1}{v_x},f(v_x)=\frac{1}{v_x^2+1}.$$
The equivalence group $`G^{}`$ of class (6) is formed by the transformations
$$\stackrel{~}{t}=\epsilon _1t+\epsilon _2,\stackrel{~}{x}=\epsilon _1^{}x+\epsilon _2^{}v+\epsilon _3^{},\stackrel{~}{v}=\epsilon _1^{\prime \prime }x+\epsilon _2^{\prime \prime }v+\epsilon _3^{\prime \prime },\stackrel{~}{f}=\epsilon _1^1(\epsilon _1^{}+\epsilon _2^{}v_x)^2f,$$
where $`\epsilon _1,`$ $`\epsilon _2,`$ $`\epsilon _i^{},`$ $`\epsilon _i^{\prime \prime }`$ $`(i=1,2,3)`$ are arbitrary constants, $`\epsilon _1(\epsilon _1^{}\epsilon _2^{\prime \prime }\epsilon _2^{}\epsilon _1^{\prime \prime })0.`$ The nonclassical (conditional) symmetries of the $`(1+1)`$-dimensional linear heat equation ($`f=1`$) were completely studied in . Analogous investigation of the second case ($`f=v_x^1`$) is carried out in this Letter. Therefore, to complete classification of reduction operators in the class of nonlinear filtration equations (6) with respect to $`G^{}`$ (see Definition 3), it is enough to describe reduction operators of the equation with the latter nonlinearity, and we achieved significant progress in solving this problem.
### Acknowledgements
The authors are grateful to Prof. C. Sophocleous for useful discussions and interesting comments. ROP and OOV thank University of Cyprus for hospitality and support during writing this paper. The research of ROP was supported by Austrian Science Fund (FWF), Lise Meitner project M923-N13. The research of OOV and NMI was partially supported by the grant of the President of Ukraine for young scientists GF/F11/0061. NMI acknowledges financial support from National Sciences and Engineering Council of Canada and Department of Mathematics of the University of British Columbia. The authors also wish to thank the referees for careful reading and suggestions for improvement of this Letter.
|
warning/0506/quant-ph0506233.html
|
ar5iv
|
text
|
# Stopped light with storage times greater than one second using EIT in a solid
## Abstract
We report on the demonstration of light storage for times greater than a second in praseodymium doped Y<sub>2</sub>SiO<sub>5</sub> using electromagnetically induced transparency. The long storage times were enabled by the long coherence times possible for the hyperfine transitions in this material. The use of a solid state system also enabled operation with the probe and coupling beam counter propagating, allowing easy separation of the two beams. The efficiency of the storage was low because of the low optical thickness of the sample, as is discussed this deficiency should be easy to rectify.
Quantum memory, Electromagnetically induced transparency,slow light, Coherent Spectroscopy, Rare-earth
Some of the most significant advances in quantum information processing have been made using quantum optics-based techniques. For example, working practical quantum cryptosystems already exist and there have been demonstrations of linear optics quantum computing OโBrien et al. (2003), quantum teleportation, quantum non-demolition measurements Pryde et al. (2004), quantum feedback and control Stockton et al. (2004). To proceed further it is necessary to have devices such as single photon sources, quantum memories and quantum repeaters, where quantum information is exchanged in a controlled fashion between light fields and material systems. It has been proposed that both the required control and strong coupling can be readily achieved using an ensemble approach, where the light field interacts with a large number of identical atoms. Such a ensemble based approaches now exist for single photon sources Duan et al. (2001), โcatโ state sources Paternostro et al. (2003), quantum memories Fleischhauer and Lukin (2002); Moiseev and Kroll (2001); Eisaman et al. (2004) and quantum repeaters Duan et al. (2001). Experiments have demonstrated heralded single photon sources Kuzmich et al. (2003); Jiang et al. (2004) and the mapping of quantum information on a light field onto spin states of an atomic ensemble Julsgaard et al. (2004). Experiments using electromagnetic induced transparency have demonstrated the storage and recall of optical pulses Liu et al. (2001); Phillips et al. (2001).
The quantum systems used for these ensemble based demonstrations have almost exclusively been atomic vapors. An issue with these demonstrations is that even for a laser cooled ensembles, the atomic motion impacts on the devicesโ performance. Ensembles of solid-state optical centers provide an alternative to atomic systems where the relative motion is zero. In this paper we investigate the use of solid-state system for ensemble based quantum optics and highlight its usefulness by stopping a light pulse using electromagnetically induced transparency (EIT). Unlike an earlier experiment Turukhin et al. (2001), the current demonstration highlights for the first time two advantages of using optically active solid state centers: a one thousand fold increase in storage time and the ability to operate with a less restrictive beam geometry.
When storing light using EIT characteristics of the field are recorded as a spin wave in the ensemble. The storage time is determined by the coherence times of the hyperfine transitions. In principle coherence times for hyperfine transitions in atomic systems can be very long and many minutes have been measured in ion traps Fisk et al. (1995). However, these long coherence times in large ensembles suitable for EIT have not been achieved. Transit time broadening in vapor cells and magnetic inhomogeneity in trapped systems mean that the longest that light has been stored atomic systems is a few milliseconds. In contrast, in earlier work we have demonstrated techniques to obtain hyperfine coherence times of tens of seconds in Pr:Y<sub>2</sub>SiO<sub>5</sub> Fraval et al. (2004a, b). Here we show it is possible to utilize these long coherence times to stop light for similar lengths of time.
EIT is sensitive to atomic movement, with the spin wave being scrambled once the atoms have moved significantly compared to the wavevector mismatch between the probe and the coupling beams. To minimize this wavevector mismatch experiments in atomic systems typically operate with the beams co-propagating. Because the probe and coupling beams are close in frequency, in this configuration, the wavevector mismatch is typically less than 1 cm<sup>-1</sup>. A consequence of this co-propagating operation is that it is difficult to separate the probe and the coupling beam. In a solid-state system, where the optical centers are locked in a crystal lattice, co-propagating operation is not required, in the present work the probe and the coupling beams are counter-propagating. With counter-propagating beams it is easier to separate the probe and coupling beam whilst maintaining optimum overlap.
The experimental setup is shown in FIG. 1. Because of the narrow 2500 Hz optical homogeneous linewidth of the <sup>3</sup>H<sub>4</sub>$``$<sup>1</sup>D<sub>2</sub> transition in Pr$`{}_{}{}^{3}+`$:Y<sub>2</sub>SiO<sub>5</sub> a highly frequency stabilized dye laser was required for the experiment not to be limited by laser jitter. The laser used was a modified Coherent 699 dye laser with a linewidth 200 Hz over 1 second time scales. The laser output was split into two beams, one of which was frequency shifted and gated by two AOMs and used as the probe beam. The other beam was frequency shifted and gated using a double pass AOM setup. This beam was used for the coupling and repumping fields. This coupling/repumup beams was aligned on a beam-splitter to go through the sample counter-propagating with the probe. The spare port of this right-most beam splitter was used to combine a local oscillator beam with the transmitted probe beam, enabling the heterodyne detection of the signal.
The sample used was the same as that used in reference Fraval et al. (2004b) and consisted of 0.05% Praseodymium doped in Y<sub>2</sub>SiO<sub>5</sub>. It was 4 mm thick along the direction of light propagation. The sample was mounted in a bath liquid helium cryostat. Three orthogonal super-conducting magnets were used to apply a DC magnetic field to the sample and a six turn rf coil was used to apply a rf field.
The dominant dephasing mechanism for the hyperfine states of the Pr<sup>3+</sup> ions is random Zeeman shifting due to fluctuating magnetic fields from the yttrium nuclei. Dramatic increases in coherence times can be achieved by operating at a magnetic field where the transition frequency is insensitive to magnetic field changes to first order Fraval et al. (2004a). The magnetic field required is 78 mT in an orientation described in Ref. Fraval et al. (2004a). Once the magnetic field is obtained the remaining fluctuations have reasonably long correlation times. This situation enables the effective use of dynamic decoherence control (DDC) techniques Viola (2004) and coherence times in excess of half a minute have been demonstrated Fraval et al. (2004b).
An energy level diagram showing the transitions driven during the experiments is shown in Fig. 2. While the optical inhomogeneous line widths is a few GHz. The narrow homogeneous linewidth (of order 1 kHz) and long hyperfine population lifetimes (of order 1 minute) enabled the experiment to be carried out on an ensemble with a much smaller range of optical frequencies. At the beginning of each shot a sequence of the five optical frequencies (labelled โRโ in Fig. 2) was applied repeatedly. The repump frequencies were applied sequentially rather that all at once to avoid the possibility of darks states and nonlinear mixing of the different frequencies in the AOM. The gap in time between the repumping and each experimental shot was long enough to ensure that ions had no remaining optical coherence. This repumping procedure prepared an ensemble of ions in the desired hyperfine state and gives a narrow adsorption with an inhomogeneous width of 100 kHz when measured by sweeping a week probe in frequency (line given by dots in Fig. 3). When the coupling beam was applied a narrow transparency was obtained in the absorption of the weak probe (solid trace in Fig. 3).
The repumping beams were applied after each shot and the 300 kHz span shown was swept in 4 ms. The transmitted probe beam was detected as a heterodyne beat signal and the bandwidth of the RF detector was comparable to 300 kHz, the extra noise at each end of the spectrum came from dividing out this frequency response. For coupling intensities above 1 mW the EIT was observed to depend linearly on the amplitude of the coupling beam. The limiting EIT width at low intensity was 10 kHz, corresponding to the hyperfine inhomogeneous linewidth. Below 1 mW the EIT transmission decreased with decreasing coupling intensity.
It can bee seen from FIG. 3 that the peak absorption of our ensemble is only about 15% and, as is discussed below, this limits the efficiency of the storing process.
The time sequence for the light storage demonstration is shown on the left of FIG. 4. A 20 $`\mu `$s long probe pulse was applied and then the 10 mW coupling beam was turned off to transfer the optical coherence onto the spin transition. As in the previous solid state stopped light Turukhin et al. (2001) experiment RF rephasing pulses were used to rephase the inhomogeneous broadening in the spin transition. Although one RF rephasing pulse is enough to rephase the spin-wave it also flips the spin-waveโs direction. Therefore when not using co-propagating beams, as is the case here, it is necessary to use an even number of rephasing pulses.
The size of the pulse of light recalled as a function of delay can be shown with and without dynamic decoherence control (DDC) and the results are shown in FIG. 4. The decay constants for the stored signal output were 0.35 seconds without DDC and 2.3 seconds with DDC. These decay rates were comparable to measurements of $`T_2`$ made using the same method as Fraval et al. Fraval et al. (2004a). The difference between the present measurements of $`T_2`$ and those obtained by Fraval et al. Fraval et al. (2004a) is attributed to not having tuned the magnetic field as carefully as was achieved by Fraval et al.
Shown in the inset of FIG. 4 is the intensity of the output pulse as the intensity of the input probe pulse is varied. From the graph it can be seen that the storage process is linear at low powers and starts to saturate at higher powers once the input pulse becomes a significant fraction of a $`\pi /2`$ pulse. This demonstration of linearity is important. Previous solid state experiments Turukhin et al. (2001) have been restricted by laser frequency jitter to using probe pulses with areas greater than $`\pi `$. At such high powers effects such as self induced transparency (SIT) McCall and Hahn (1969) cannot be ignored.
While the effect was linear and scaled to low powers, the efficiency was low, of the order of 1%. This in part can be improved with better timing and shaping of the probe and coupling waveforms. However the main reason for the low efficiency is the low optical absorption at the probe frequency and the accompanying modest group delay.
The sample used for this experiment was only 4 mm thick, longer samples as well as multi-pass cells and cavities are simple means to increase the optical absorption. Preliminary measurements on a samples with a range of praseodymium concentrations ala suggest that at least two or three fold increases in the optical thickness can be achieved by increasing the concentration without significantly increasing the inhomogeneous broadening of the hyperfine transition.
As it is a goal of this line of research to store and retrieve quantum mechanical states it worthwhile to consider the effect that rephasing pulses would have on few photon states stored in the hyperfine coherences. It has been asserted in a theoretical investigation of quantum information storage in the solid state Johnsson and Molmer (2004) that one would not be able to apply the RF $`\pi `$ pulses with sufficient accuracy. This is not the view of the authors of this paper. In Ref. Johnsson and Molmer (2004) it was assumed that the $`\pi `$ pulse would have to be applied with an accuracy close to 1 part in $`N`$ (where $`N`$ is the number of atoms) in order that the few photon pulse not be swamped by light caused by inaccuracies of the $`\pi `$ pulse. However this light will be emitted randomly rather than in the very precise spatio-temporal mode of the output pulse. This should enable the output pulse to be easily separated from the background with very high efficiency.
In conclusion, we have demonstrate stopped light in Pr:Y<sub>2</sub>SiO<sub>5</sub> for time scales of several seconds which is three orders of magnitude longer than any obtained previously. Based on previous measurements of $`T_2`$ it should be possible to extend this storage time by at least one more order of magnitude.
For the first time stopped light has been demonstrated in a solid with the coupling and probe beams counter propagating. This configuration is desirable as it allows easy separation of the two beams. However, it is only practical if the atoms movement during the storage time is small compared to the optical wavelength. Even for ultra-cold systems this places significant limits storage time. In a solid where the atoms are locked into position this isnโt a problem.
The efficiency of the storage process required for a quantum memory should be obtainable by increasing the density of the dopant ions and by increasing the interaction length.
The authors would like to thank Philip Hemmer for helpful discussions. We would like to acknowledge the support of the Australian Research Council and the Australian Department of Defense.
|
warning/0506/astro-ph0506346.html
|
ar5iv
|
text
|
# CONSTRAINTS ON HOT METALS IN THE VICINITY OF THE GALAXY
## 1 Introduction
Present-day structures such as galaxies and clusters of galaxies are believed to have condensed from within partially collapsing filaments of primordial matter. The legacy of the structure formation epoch includes shock-heated filaments as well as structures and these filaments should thread the modern universe. Simulations of structure formation indicate that around half of the baryonic matter at low-redshift lives in such filaments, shock-heated to $`10^{57}`$K (see e.g. Cen & Ostriker (1999); (2001) and references therein). Although more recent work by Kang et al. (2005) suggests that a a lower temperature component (T$`<10^5`$ K) of the shock-heated filaments may be more significant than previously thought. The โwarmโ ($`10^{56}`$K) component of this warm/hot intergalactic medium (WHIGM) has been observed in absorption in the UV band (see e.g. Tripp et al. (2000) and references therein). However, much of the WHIGM is expected to be hotter than this, so the high spectral resolution X-ray detectors such as those aboard Chandra and XMM-Newton are best placed for investigating the โhotโ component of the missing baryons.
The clearest X-ray spectral signature of hot gas in the vicinity of our Galaxy consists of absorption features imprinted in spectra of X-ray bright active galactic nuclei (AGN) at $`z=0`$ in the observed frame (Hellsten et al., 1998; Perna & Loeb, 1998). One of the most important new results from the Chandra and XMM-Newton X-ray telescopes, has been the discovery of hot, low density, highly ionized gas in the vicinity of our Galaxy (at $`z=0`$) and possibly beyond (at $`z>0`$). X-ray absorption due to local ($`z=0`$) hot gas was discovered recently (, 2002, 2002, 2002; Rasmussen et al., 2002; , 2003; McKernan et al., 2003a; , 2003b, 2004). Of course, such absorption need not be due to local WHIGM. Absorption by hot gas at $`z=0`$ could be due to local Galactic gaseous structures or infalling High Velocity Clouds (see e.g. Sembach et al. (2003); Collins, Shull & Giroux (2004) and references therein). Even if there are no known local structures along the sightline to a particular AGN, the location of the absorbing gas is still ambiguous. Around half of all type I AGN exhibit strong absorption in the soft X-ray and UV bands due to partially ionized, optically thin, outflowing, circum-nuclear material known as the โwarm absorberโ (see e.g. (2005) and references therein). Since the warm absorbers are outflowing, absorption which is actually local to our Galaxy at $`z0`$ could be misinterpreted as a warm absorber outflow coinciding with the cosmological recession velocity (cz) of the AGN (see e.g. Fig.3 of (2004)). Confusion with AGN outflow is a worse problem for WHIGM at intermediate redshifts ($`z>0`$). The few apparently robust detections of absorbers at redshifts intermediate between our Galaxy and the AGN rest-frame, e.g. (2005) could be variable (Ravasio et al. (2005)) suggestive of an origin in an ionized outflow from the host AGN, rather than WHIGM.
In this paper, we investigate a sample of AGN observed with the high energy transmission gratings (Markert et al., 1995) on board Chandra . The uniform analysis of the data from these AGN and the results of the analysis, in particular the characterization of the AGN continua and the warm absorption in AGN, have been discussed in detail by (, 2005). In (2004) we investigated this sample of AGN spectra for absorption due to local, highly ionized Oxygen. Here we extend that study by searching for absorption in the vicinity of the Galaxy by metals other than Oxygen. Our aim is to further the systematic study of the hot local gas and to begin constraining the temperatures and the relative metal abundances in the hot, local gas.
## 2 The Sample and Data Analysis
Table 1 lists the AGN sample assembled by (2005). Also listed in Table 1 are the AGN redshifts (from NED<sup>1</sup><sup>1</sup>1http://nedwww.ipac.caltech.edu/forms/byname.html using 21cm H i radiation measurements where possible), the AGN Galactic latitude and longitude (also from NED), the Galactic column density and the total exposure times of the spectra. The sample, including selection criteria are discussed in detail in (2005). The Chandra data were reprocessed and analyzed according to the methods outlined in (, 2005).
Here we extend the study of (, 2004) by searching for evidence of absorption by metals less abundant than Oxygen. Table 2 lists the relative solar abundances of the metals relevant for this study. Iron is the next most abundant metal after those listed in Table 2, but there is a forest of Fe L- and M-shell transitions in the soft X-ray band (see e.g. Behar et al. (2001)). There is therefore considerable ambiguity in Fe absorption line identification due to blending. Furthermore, higher order transitions of more abundant elements can also be mis-identified as Fe transitions. Therefore we limited our search to the strongest absorption transitions in the soft X-ray band in the most abundant highly stripped ions (not O vii and O viii since we have studied these elsewhere) namely: N vii , Ne ix , Ne x , Mg xi , Mg xii , Si xiii and Si xiv respectively. Table 3 lists the transitions that we investigated in this study, including their rest-frame wavelengths and oscillator strengths.
Once we measured the discrete absorption profiles, we used the extrapolated linear approximation to the curves-of-growth<sup>2</sup><sup>2</sup>2 The linear part of the curves-of-growth implies that $`N_{ion}=1.13\times 10^{17}EW/f\lambda ^2`$ where $`N_{ion}`$ is the ionic column density ($`\mathrm{cm}^2`$), EW is the equivalent width of the absorption feature (in mร
), f is the oscillator strength of the transition and $`\lambda `$ is in ร
. to obtain a lower limit on the ionic column density ($`N_{ion}`$), if a *lower limit* on the EW of the absorption feature is available. Such a lower limit on $`N_{ion}`$ is valid for any value of the velocity width (b) of the absorber. Where no lower limit on the EW exists, absorption is not significant (at 90$`\%`$ confidence). However, in this case, for an assumed b value, we can use the *upper limit* on the EW to get an upper limit on $`N_{ion}`$. In such cases, we assumed a velocity width of $`b100`$ km $`\mathrm{s}^1`$, since this is roughly the smallest width of a feature that the MEG can resolve although it is considerably larger than the average value of $`<b>=40\pm 13`$ km $`\mathrm{s}^1`$ found by Sembach et al. (2003) in signatures of local O vi absorption.
## 3 Spectral Fitting
We used XSPEC v.11.2.0 for spectral fitting to the MEG spectra. All spectral fitting was carried out based on the best-fitting continuum models from (, 2005). Spectral fitting was carried out in the 0.5-5 keV energy band, excluding the 2.0-2.5 keV region, which suffers from systematics as large as $`20\%`$ in the effective area due to limitations in the calibration of the X-ray telescope <sup>3</sup><sup>3</sup>3http://asc.harvard.edu/udocs/docs/POG/MPOG/node13.html. We analyzed data binned at $`0.02`$ร
, which is approximately the MEG FWHM spectral resolution ($`0.023`$ร
). This MEG spectral resolution corresponds to FWHM velocities of $`280`$ and $`560\mathrm{k}\mathrm{m}\mathrm{s}^1`$ at observed energies of 0.5 and 1.0 keV respectively. We used the C-statistic (Cash, 1976) for finding best-fit model parameters and quote 90$`\%`$ confidence, one-parameter errors.
We proceeded to fit the MEG spectra for the discrete absorption transitions in Table 3 by adding an inverted Gaussian model component to the best-fitting continuum models detailed in (2005). The width of the inverted Gaussian was chosen to be $`>100`$ km $`\mathrm{s}^1`$, which is approximately the lower limit of the instrumental velocity resolution. We fixed the redshift of the Gaussian components at $`z=0`$ and allowed the rest-energy of the component to vary by $`\pm 1200`$ km $`\mathrm{s}^1`$ from the rest-frame energies of the transitions listed in Table 3. The allowed velocity range is identical to that used by Sembach et al. (2003) and (2004) in searches for highly ionized Oxygen absorption in the vicinity of the Milky Way.
## 4 Results
Of the fifteen AGN sightlines in our sample, only the sightlines to NGC 4051 and MCG-6-30-15 exhibit absorption features within $`\pm 1200`$ km $`\mathrm{s}^1`$ of their rest-frame energy at $`z=0`$ at $`99\%`$ confidence ($`\mathrm{\Delta }C11.3`$ for three additional parameters). The sightlines to F9, NGC 4593, NGC 3227 and MCG-6-30-15 exhibit absorption features within $`\pm 1200`$ km $`\mathrm{s}^1`$ of their rest-frame energy at $`z=0`$ at $`90\%`$ but $`99\%`$ confidence. Of the metals we searched for, only Ne, Mg and Si absorption signatures were detected at $`>90\%`$ confidence.
Tables 4 shows the best-fitting model parameters for the strongest detections of Ne ix (r) and Ne x Ly$`\alpha `$ absorption features. The strongest absorption signatures lie along the sightlines to NGC 4051 and MCG-6-30-15 respectively, where at least one of the Ne absorption features has been detected at $`99\%`$ confidence. Table 5 similarly details the best-fit model parameters from a detection of Si xiv Ly$`\alpha `$ at $`>3\sigma `$ significance and a detection of Mg xii Ly$`\alpha `$ at $`>90\%`$ confidence along the sightline to MCG-6-30-15. Listed in Tables 4 and 5 are the equivalent widths (EW) of the absorption features and (in brackets) the improvement in the C-statistic upon addition of the inverted Gaussian model component to the continuum. Also listed in Tables 4 and 5 are the velocity offsets from $`z=0`$ of the respective Gaussian centroids and limits on the respective ionic column densities as estimated from a curve-of-growth analysis as outlined above in ยง2.
Although the features listed in Tables 4 and 5 are statistically significant, the kinematics of most of these features could be consistent with origin in an AGN outflow. The absorption features detected towards NGC 4051 for example, are barely kinematically consistent with an origin in gas at $`z=0`$, or indeed with an origin in the same gas (within $`90\%`$ errors, see Table 4). Nevertheless, at $`cz=725`$ km $`\mathrm{s}^1`$, NGC 4051 is very near and with a Chandra gratings energy resolution of FWHM $`500`$ km $`\mathrm{s}^1`$ at the energy of the Ne ix (r) transition (0.922 keV), it is kinematically difficult to distinguish X-ray absorption intrinsic to the NGC 4051 outflow from X-ray absorption due to hot local gas. Some or most of this hot gas may be associated with the warm absorbing outflow in this AGN.
Along the MCG-6-30-15 sightline, only the Ne ix (r) absorption feature is kinematically consistent with $`z=0`$ absorption. The Ne x Ly$`\alpha `$ absorption feature in Table 4 (at $`+615_{235}^{+295}`$ km $`\mathrm{s}^1`$), coincides kinematically with the Si xiv Ly$`\alpha `$ and Mg xii Ly$`\alpha `$ absorption features in Table 5 (at $`+615`$ and $`+530`$ km $`\mathrm{s}^1`$), but these features are not kinematically consistent with an origin in hot local gas at $`z=0`$. Thus, we conclude that the Ne x Ly$`\alpha `$ , Si xiv Ly$`\alpha `$ and Mg xii Ly$`\alpha `$ features along the sightline to MCG-6-30-15 originate in a warm absorber outflowing from the AGN at $`1710`$ km $`\mathrm{s}^1`$ rather than hot local gas (see also (2005)). Therefore, in order to study the hot *local* Ne x Ly$`\alpha `$ along the sightline to MCG-6-30-15, we searched for upper limits on absorption due to Ne x Ly$`\alpha `$ by adding a narrow inverted Gaussian to the spectrum at $`65`$ km $`\mathrm{s}^1`$ from $`z=0`$ rest-frame energy (to compare with the corresponding Ne ix (r) absorption). We obtained upper limits on absorption due to Mg xii Ly$`\alpha `$ and Si xiv Ly$`\alpha `$ similarly by adding a narrow inverted Gaussian at the $`z=0`$ rest-frame energy of the transition. Likewise, for NGC 4051, we obtained upper limits on absorption due to Ne ix and Ne x by adding narrow inverted Gaussians at the rest-frame energy of the respective absorption transitions at $`z=0`$.
Fig. 1 is a multipanel plot showing velocity profiles of the Ne ix (r) and Ne x Ly$`\alpha `$ absorption features along the sightlines to NGC 4051 and MCG-6-30-15. The profiles are centered on the Ne ix (r) transition energy (0.9220 keV) and the Ne x Ly$`\alpha `$ transition energy (1.0218 keV) in the LSR respectively (both energies are denoted by vertical dashed lines at 0 km $`\mathrm{s}^1`$). The vertical dotted line in Fig. 1(a,b) at $`+110`$ km $`\mathrm{s}^1`$ denotes the weighted mean offset velocity of the O vii and O viii absorption along this sightline detected by (2004). The vertical dash-dot line in Fig. 1(a,b) at $`+725`$ km $`\mathrm{s}^1`$ denotes the recessional velocity (cz) of NGC 4051. Superposed on the data is the best-fit inverted Gaussian absorption line model (from Table 4) and continuum (horizontal solid line).
Of the less significant absorption features detected (at $`90\%`$ confidence but $`<99\%`$ confidence), those along the sightlines towards F9 and NGC 4593 are kinematically coincident with their rest-energies at $`z=0`$ and are therefore likely to correspond to hot local gas. A Si xiv Ly$`\alpha `$ feature detected at an offset velocity of $`525_{600}^{+450}`$ km $`\mathrm{s}^1`$ along the sightline towards NGC 3227 is more likely to kinematically correspond to an outflow at $`1700`$ km $`\mathrm{s}^1`$ from the AGN (cz$`=1160`$ km $`\mathrm{s}^1`$) than absorption due to hot, local gas. We obtained upper limits on absorption due to Si xiv along this sightline by fitting a narrow inverted Gaussian to the continuum at the rest-energy of Si xiv Ly$`\alpha `$ at $`z=0`$.
### 4.1 Local, hot gas versus AGN outflow?
NGC 4051 and MCG-6-30-15 are two of the closest AGN in our sample, at recessional velocities of $`cz=726`$ km $`\mathrm{s}^1`$ and $`cz=2325`$ km $`\mathrm{s}^1`$ respectively. Since the spectra of Type I AGN typically exhibit hot gas outflowing at several hundred km $`\mathrm{s}^1`$, the proximity of these two AGN, raises the important issue of confusion between absorption due to hot gas at $`z=0`$ and that due to a warm absorber (see also the discussion in (2004)). The proximity of NGC 4051 and MCG-6-30-15 (and NGC 3227) and the limited gratings spectral resolution makes it difficult to distinguish X-ray absorption features along these sightlines due to hot local gas from those due to warm absorbing outflows. On the other hand, the more distant the AGN, the more likely it is that absorption signatures at offset velocities very close to their $`z=0`$ rest-frame energy *are* due to hot, local gas. The relation in Figure 3 of (2004) shows this effect quite dramatically. Thus, outflows previously thought to be associated with more distant AGN, e.g. PG 1211+143 (Pounds et al., 2003) and PDS 456 (Reeves et al., 2003) are kinematically much more likely to correspond to hot, local gas. We note that the surprisingly high Fe column densities towards both of these AGN might be accounted for by the intersection of both of these sightlines with the limb of the local Northern Polar Spur (NPS) structure. A third sightline through the NPS, towards MCG-6-30-15 also exhibits strong absorption due to a large column of highly ionized Fe that is kinematically consistent with a local origin (Young et al., 2005). The NPS is a local feature believed to correspond to the superposition of several supernovae and is clearly seen in X-ray emission maps (see e.g. Snowden et al. (1997)) and in maps of polarization towards nearby stars (Mathewson & Ford, 1970; Heiles & Jenkins, 1976; Axon & Ellis, 1976). An alternative hypothesis is that the NPS may actually be a much larger feature subtended at the Galactic center Sofue (2000); Bland-Hawthorn & Cohen (2003), although we shall not consider this hypothesis further here. If the progenitor supernovae of the NPS were rich in pure Fe, this could account for the anomalous column of Fe along these sightlines. We intend to investigate this possibility in future work.
### 4.2 Column densities of hot local metals
Table 6 lists the limits on the ionic column densities for all the ions from spectral fitting. Most of the results in Table 6 are upper limits, indicating that discrete absorption features due to local gas are not detected at $`>90\%`$ confidence in most cases. None of the sightlines exhibited absorption due to local N at $`>90\%`$ confidence. The spectra of three AGN, namely NGC 3227, NGC 3516 and NGC 7314, were too heavily absorbed at $`0.8`$ keV even to obtain meaningful limits on N vii absorption. We found that assuming different values of the velocity width (e.g. $`b50`$ or $`200`$ km $`\mathrm{s}^1`$), led to small changes in estimates of $`N_{ion}`$ in Table 6, $`\mathrm{log}(\mathrm{\Delta }N_{ion})<0.2`$ for $`b=50`$ km $`\mathrm{s}^1`$ and $`\mathrm{log}(\mathrm{\Delta }N_{ion})<0.1`$ for $`b=200`$ km $`\mathrm{s}^1`$ respectively using the linear part of the curves-of-growth.
Sembach et al. (2003) and (2004) associate local, highly ionized O absorption with known local structures such as the Magellanic Stream (MS), Complex C and Extreme Positive North (EPn), as well as a diffuse Local Group (LG) and (potentially) the warm/hot IGM. Several of the sightlines in Table 6 can be identified with these structures: one (F9) with the MS, three (NGC 4593, NGC 4051, NGC 3227) are identified with EPn and one (MCG-6-30-15) which we associate with the NPS. In the southern Galactic hemisphere, the sightline to Akn 564 passes through the Magellanic Stream extension (MSe), but this may also be LG (Sembach et al., 2003).
Sembach et al. (2003) conclude that most of the high velocity (v $``$ 100-400 km $`\mathrm{s}^1`$) O vi gas in the vicinity of the galaxy is created by collisional ionization. Furthermore, gas in the low redshift IGM is far more likely to be collisionally ionized than photoionized (Heckman et al., 2002). Therefore, if we assume that the hot gas in the vicinity of the Galaxy is in collisional ionization equilibrium (CIE), and that the absorption signatures discussed here and in (2004) are due to the same gas, it is possible to establish temperature constraints on the gas, whether it is local to our Galaxy or low redshift WHIGM. Sutherland & Dopita (1993) calculate $`\mathrm{N}_{\mathrm{Ne}\mathrm{IX}}`$/$`\mathrm{N}_{\mathrm{Ne}\mathrm{X}}`$ and $`\mathrm{N}_{\mathrm{Si}\mathrm{XIII}}`$/$`\mathrm{N}_{\mathrm{Si}\mathrm{XIV}}`$ for gas in CIE, so where there are statistically significant lower limits on an ionic column along a given sightline, we can constrain the temperature of local gas along that sightline with some confidence. However, from searching for seven absorption transitions along each of fifteen different sightlines through the hot, local gas, there are only three *lower* limits on ionic column densities. The corresponding temperature limits (assuming CIE) using the $`\mathrm{N}_{\mathrm{Ne}\mathrm{IX}}`$/$`\mathrm{N}_{\mathrm{Ne}\mathrm{X}}`$ ratio are: $`T<10^{6.75}`$ K towards MCG-6-30-15, $`T<10^{6.70}`$ K towards NGC 4593 and $`T>10^{6.35}`$ K towards F9. The sightline to F9 provides a *lower* limit on the temperature because we detect Ne x Ly$`\alpha `$ and not Ne ix (r) .
## 5 Comparison with signatures of Oxygen absorption
(2004) showed that seven of the fifteen sightlines in our sample exhibit discrete absorption features due to local O vii and O viii . Sembach et al. (2003) found local O vi ($`\lambda 1031.926`$ ร
) absorption along 59 of 102 sightlines towards UV bright AGN/QSOs at high Galactic latitudes ($`|b|30^{}`$). Several of the 59 sightlines coincide with sightlines in our sample (F9, NGC 5448, Mkn 509, Akn 564), of which three (F9, Mkn 509, Akn 564) show significant O vi absorption. In Figure 2, we show the sightlines to the fifteen AGN in our sample in Hammer-Aitoff projection. Crosses indicate non-detection of *any* highly ionized local gas (Ne, Mg, Si and O) along the sightline. Triangles indicate detection (at $`>90\%`$ confidence) of highly ionized local gas (Ne or Si) with no corresponding highly ionized O. Diamonds indicate the presence of local, highly ionized O along the sightline, but non-detection of local, highly ionized Ne, Mg and Si. Filled-in circles indicate sightlines along which highly ionized Ne or Si has been detected (at $`>90\%`$ confidence) *and* which show absorption due to local, highly ionized O.
Of the three sightlines which exhibit absorption by local, highly ionized Ne at $`>90\%`$ confidence (listed in Table 6), two sightlines (F9 and NGC 4593) also exhibit absorption by O vii and/or O viii at $`>90\%`$ confidence (, 2004). The sightline towards F9 also exhibits O vi absorption (Sembach et al., 2003). The sightlines to MCG-6-30-15 does not exhibit absorption due to highly ionized Oxygen. In (2004) we constrained the temperature in local gas from limits on $`N_{OVIII,VII,VI}`$ and an assumption of CIE. Towards F9, we found $`10^{5.75}<T<10^{6.35}`$K. This is marginally inconsistent with our estimate of $`T>10^{6.35}`$K using N(Ne ix , Ne x ) in ยง4 above. However, the condition that $`T<10^{6.35}`$K along this sightline is derived assuming a b-parameter of precisely $`100`$ km $`\mathrm{s}^1`$ using $`N_{OVIII}`$ in a curve-of-growth analysis. A choice of a slightly larger b-parameter (which would be allowed by the data) would yield a temperature upper limits compatible with that derived using N (Ne ix , Ne x ). The combined results using N(O vi , O vii , O viii , Ne ix & Ne x ) along the sightline to F9, suggests that the temperature in the local hot gas in the MS is close to $`T10^{6.35}`$ K and that there is some velocity broadening of the hot gas along this sightline. The temperature of the gas along the sightlines to NGC 5448 and Akn 564 was also constrained by (2004) to be $`T>10^{6.2}`$ K and $`T<10^{6.1}`$K respectively. However, the absence of significant absorption along the sightlines to NGC 5548 and Akn 564 in this study (see Table 6), means that we cannot provide additional constraints on the temperature of gas along these sightlines.
## 6 Possible constraints on metallicities in hot local gas
The temperature ranges derived from the ratios of column densities in ยง4 above and by (Sembach et al., 2003) and (, 2004) suggest that we can assume that most of the O, Ne, Mg and Si in the hot gas in the vicinity of the Galaxy is in H-like and He-like ions. If we assume that the gas is in CIE, then for a range of temperatures, such as those discussed here, the fraction in H-like or He-like ions is much larger than in other ionic states (Sutherland & Dopita, 1993) (e.g. in the case of Oxygen ions, O viii / O ix $``$ 1, O vii / O vi $``$ 1). Thus, in a simple approximation, the largest upper (or lower) limit on the column density of the He-like or H-like ions of a particular element is approximately the upper (lower) bound on the amount of the element present.
Therefore, given a lower limit to the column density N(X) of an element X, using this simple approximation, we can use upper limits on the column density N(Y) of element Y to obtain a lower limit on N(X)/N(Y) along a sightline. Limits on relative metal abundances in the hot local gas, allow us to compare the results with solar relative abundances (see Table 2) and thereby establish limits on the composition of the hot gas. The relative depletion or enrichment of certain metals can provide clues to the environment, formation and evolution of the hot gas in local structures and the hot IGM. Note that it is not possible to obtain reliable upper limits on the ratio N(X)/N(Y), since an upper limit on N(X) is potentially much more inaccurate than a lower limit (a firm lower bound on the column), due to the possibility of additional columns of more or less ionized gas than those considered here.
Tables 7 and 8 list constraints on the metal abundances relative to O and Ne respectively, in the hot gas in the vicinity of our Galaxy. Clearly the lower limits on the relative abundances in the local, hot gas are in general too small to be interesting, with a few exceptions. In Table 7, the lower limit on \[O/N\] along the sightline to F9 is close to the solar value, while the other ratios are one or two orders of magnitude lower, which may suggest a relative underabundance of N in the MS. Also in Table 7, the lower limit on \[O/Si\] along the sightline to NGC 3783 is close to the solar ratio, while the other ratios are significantly lower, which may be hinting at a relative underabundance of Si along this sightline. In Table 8, \[Ne/N\] along the sightline to F9 *exceeds* the solar value by $`3.9`$, and \[Ne/Si\] along the sightline to MCG $``$6$``$30$``$15 is roughly solar. The other Ne ratios are one or two orders of magnitude below their solar values. This again suggests an underabundance of N in the hot gas in the MS.
## 7 Conclusions
Half of the baryonic matter in the local universe seems to be missing. The search for the hottest component of the missing matter has only been possible with the latest generation of high spectral resolution X-ray telescopes. Hot gas in the vicinity of the Galaxy may be due to local WHIGM or it may reside either in a hot Galactic halo or locally in a thick disk and has only recently begun to be studied in the X-ray band. We assembled a small sample of type I AGN observed with the high resolution X-ray gratings on board Chandra and we have applied a uniform analysis to detect soft X-ray absorption by hot gas in the vicinity of our Galaxy. This study is an extension of our previous study of O vii and O viii absorption by hot local gas (, 2004).
Three of the fifteen sightlines in our sample (towards F9, NGC 4593 and MCG-6-30-15 respectively) exhibit Ne ix (r) or Ne x Ly$`\alpha `$ absorption due to hot, local gas at $`90\%`$ confidence. We identify these absorption features with the hot phase of local structures. Such local structures are either in the disk of the Galaxy (e.g. the Local Bubble or superbubbles such as the NPS) or lie above the disk in HVCs or other structures, such as a Galactic halo. Hot gas in a thick disk is expected to have low offset velocity from $`z=0`$ (typically $`100`$ km $`\mathrm{s}^1`$) and hot gas above the disk in HVCs or in a Galactic halo is expected to have a higher offset velocity from $`z=0`$ (typically $`100400`$ km $`\mathrm{s}^1`$). The sightlines to the AGN in our sample lie well away from the Galactic plane ($`b30^{}`$) and Chandra does not possess the velocity resolution to distinguish between kinematic signatures of a thick disk or Galactic halo. Therefore we cannot tell whether absorption is local to the disk, or further out in a halo, where there are no known local structures along a sightline. Absorption studies in the UV band with FUSE indicate that there is a thick Galactic disk of O vi , with a scale height of $`2.3`$ kpc, as well as a patchy overdensity of O vi in the northern Galactic hemisphere from $`b45^{}`$ to $`90^{}`$ due to the Local Bubble and superbubbles (including the NPS) (Savage et al., 2003). X-ray spectral studies of Galactic X-ray binaries confirm that there is a hot, thick disk with scale height of $`12`$ kpc associated with our Galaxy (see e.g. (2004); Yao & Wang (2005)). The column densities inferred from (2004); Yao & Wang (2005) are consistent with the limits inferred here, so it may be that most or all of the hot gas locally is associated with a Galactic โthick diskโ of hot gas, with enrichment along particular sightlines in the northern Galactic hemisphere due to local superbubbles such as the NPS.
There can be considerable ambiguity in distinguishing local hot absorbing gas and hot absorbing outflows from AGN, especially with the limited spectral resolution of the Chandra gratings. This is particularly true of some of the AGN in the present study. Two sightlines in our study (towards MCG-6-30-15 and NGC 4051 respectively) exhibit absorption features within $`\pm 1200`$ km $`\mathrm{s}^1`$ of their rest-energies at $`z=0`$ at confidence levels $`>99\%`$. However, the kinematics of the absorption signatures suggest that these features are most likely due to absorption in hot gas outflowing from the respective background AGN. Nevertheless, the more distant the AGN, the more likely that absorption signatures in the AGN spectrum which kinematically coincide with the AGN recession velocity are due to hot, local gas. This is demonstrated by the relation in Fig. 3 of (, 2004). Two of the AGN in Fig. 3 of (, 2004), namely PG 1211+143 and PDS 456, also exhibit absorption due to a very large local column of Fe, which is surprising. We point out that absorption by the local NPS bubble, if Fe-enriched, could account for the very large column density of Fe along the sightline to these AGN, and towards MCG-6-30-15 as well. Thus, hot gas in the disk of the Galaxy may account for much of the $`z=0`$ absorption in the soft X-ray spectra of several AGN.
We assumed collisionally ionized equilibrium (CIE) to use limits on ionic column densities to derive limits on the temperatures in the hot, local gas. We find that in the southern Galactic hemisphere, the sightline to F9 through the MS yields $`T>10^{6.35}`$K, which is marginally consistent with our temperature limits from N(O vii )/N(O viii ) (, 2004). Likewise, in the northern Galactic hemisphere we find temperature limits of $`T<10^{6.75}`$ K towards MCG-6-30-15 and $`T<10^{6.70}`$ K towards NGC 4593. For a range of temperatures, such as those discussed here, derived limits on the H- and He-like ion column densities allow us to establish constraints on the relative abundances of N, Ne, Mg and Si (as well as O from (2004)) along the sightlines in our sample. In general, we found lower limits on the relative abundances of the metals that are one or two orders of magnitude below the corresponding solar values (see Tables 7 and 8). However, we do find evidence for a relative *underabundance* of N in the hot gas in the MS along the sightline towards F9.
The question of whether the hot gas detected in the vicinity of the Galaxy is in part (or at all) associated with the missing matter remains as yet unanswered. The environment of the Galaxy is complex, consisting of satellite galaxies (Magellanic Clouds), tidal trails (the Magellanic Stream), infalling HVCs and outflows from supernovae in the Galaxy. Standard theories of structure formation in the universe suggest that our Local Group (our Galaxy and M31) should have formed from a low density filament of gas, the heated remnants of which should persist today. The local filament should be metal-enriched by outflows from structures that formed within it. However, the upper limits on column densities that we derive here and in (2004) seem to be consistent with estimates of column densities from studies of Galactic halos in edge-on normal spiral galaxies (Strickland et al., 2004). Thus, our first systematic (but hardly comprehensive) look at the hot, X-ray absorbing component of gas in the vicinity of our Galaxy suggests that our Galaxy is embedded in a hot halo or hot thick disk.
## Acknowledgments
We gratefully acknowledge support from NSF grant AST0205990 (BM) and NASA grant AR4-5009X issued by CXC operated by SAO under NASA contract NAS8-39073 (TY). We made use of the HEASARC on-line data archive services, supported by NASA/GSFC and also of the NASA/IPAC Extragalactic Database (NED), operated by the Jet Propulsion Laboratory, CalTech, under contract with NASA. Thanks to the Chandra instrument and operations teams for making the observations possible. Thanks to M. Coleman Miller, David Strickland and Andy Fabian for useful discussions and thanks to Ken Sembach for bringing to our attention the uncertainty in the solar Ne abundance. Thanks to the anonymous referee for their comments which helped improve this paper.
|
warning/0506/math0506091.html
|
ar5iv
|
text
|
# Asymptotic estimates of the norms of positive definite Tรถplitz matrices and detection of quasi-periodic components of stationary random signals
## 1. Introduction
This paper is motivated by the studies \[VGM01\] devoted to the problem of detection of hidden unstable components in random neutron signals measured in boiling water nuclear reactors. We assume that a signal from a monitored system forms, during a sufficiently long time interval, a real-valued stationary random process $`\xi \left(t\right),`$ $`t`$, with discrete time, such that the means
$$\xi \left(t\right)=0,\xi ^2\left(t\right)=1\text{ }.$$
The correlation function of such a process
$$b\left(t\right):=\xi \left(t\right)\xi \left(0\right)=\xi \left(t+t^{}\right)\xi \left(t^{}\right),t,t^{}\text{ ,}$$
is a real-valued sequence, which admits the representation
(1)
$$b\left(t\right)=_\pi ^\pi \mathrm{exp}\left(it\theta \right)๐\sigma \left(\theta \right)\text{ },$$
where $`\sigma \left(\theta \right)`$ is a non-decreasing bounded function on $`[\pi ,\pi ]`$ \[R67\]. By virtue of our assumptions
$$b\left(0\right)=\sigma \left(\pi \right)\sigma \left(\pi \right)=1\text{ },$$
and for any $`\theta _1,`$ $`\theta _2`$ such that $`0\theta _1\theta _2\pi `$, we have
(2)
$$\sigma \left(\theta _2\right)\sigma \left(\theta _1\right)=\sigma \left(\theta _1\right)\sigma \left(\theta _2\right)\text{ }.$$
In general, the spectral distribution function $`\sigma \left(\theta \right)`$ which determines the process correlation function by (1), can be split into a sum
(3)
$$\sigma \left(\theta \right)=\sigma _c\left(\theta \right)+\sigma _d\left(\theta \right)$$
of a continuous non-decreasing function $`\sigma _c\left(\theta \right)`$ and a non-decreasing step function $`\sigma _d\left(\theta \right)`$, and such a representation is unique up to constant contributions in $`\sigma _c\left(\theta \right)`$ and $`\sigma _d\left(\theta \right)`$ \[R67\]. Notice that both functions, $`\sigma _c\left(\theta \right)`$ and $`\sigma _d\left(\theta \right)`$, satisfy the condition (2). Actually, the problem, formulated in \[VGM01\], was to find, in a real-time operation mode, whether the spectral distribution function of the random signal $`\sigma \left(\theta \right)`$ contains or does not contain a non-trivial component $`\sigma _d\left(\theta \right)`$.We will call here for brevity the random process (signal) $`\xi \left(t\right)`$ stable if its spectral distribution function $`\sigma \left(\theta \right)`$ is continuous and unstable otherwise. In other words, the process is unstable if and only if
$$\sigma _d\left(\pi \right)\sigma _d\left(\pi \right)>0\text{ }.$$
The main task of the present work is to formulate criteria of stability of the process in terms of its correlation function $`b\left(t\right)`$. As a main tool we use the sequence of positive definite Tรถplitz matrices $`Q_N=\left(b\left(jk\right)\right)_{j,k=0}^{N1}`$. This paper is organized in the following way.
In Section 2 we find the principal asymptotic contribution of the Hilbert-Schmidt norm $`Q_N_2`$ with $`N\mathrm{}`$, and show that the process is stable if and only if this contribution is $`o\left(N\right)`$.
Section 3 contains a similar criterion, but with the Hilbert norm $`Q_N`$ instead of $`Q_N_2`$. We prove here that if $`N\mathrm{}`$, then $`Q_N=mN+o(N)`$, where $`m`$ is the maximal jump of $`\sigma \left(\theta \right)`$.
In Section 4 both criteria are generalized for continuous time processes or for positive integral operators with difference kernels.
In Section 5 the efficiency of the above stability criteria for signal processing is discussed and illustrated by application to simulated signals and real neutron signals emitted by the Forsmark 1&2 boiling water reactor.
## 2. Asymptotic form of the Hilbert-Schmidt norm of truncated correlation matrices and the stability criterion
Let us denote by $`\left\{Q_N\right\},N=1,2,\mathrm{}`$, the sequence of Tรถplitz matrices $`\left(b\left(jk\right)\right)_{j,k=0}^{N1}`$ and let $`Q_N_2`$ be the Hilbert-Schmidt norm of $`Q_N`$:
(4)
$$Q_N_2=\left(\underset{j,k=0}{\overset{N1}{}}b^2\left(jk\right)\right)^{\frac{1}{2}}=\left(Nb^2\left(0\right)+2\underset{k=1}{\overset{N1}{}}\left(Nk\right)b^2\left(k\right)\right)^{\frac{1}{2}}\text{ }.$$
Our assumptions imply that
$$b^2\left(k\right)=\left|_\pi ^\pi \mathrm{exp}\left(ik\theta \right)๐\sigma \left(\theta \right)\right|^2b^2\left(0\right)=1\text{ }.$$
Therefore $`Q_N_2N`$.
###### Theorem 2.1.
The process $`\xi \left(t\right)`$ is stable if and only if
$$\underset{N\mathrm{}}{lim}\frac{1}{N}Q_N_2=0\text{ }.$$
###### Proof.
Introduce
$$\mathrm{\Phi }_N\left(\theta \right)=\underset{\pi }{\overset{\pi }{}}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\sigma _c\left(\theta ^{}\right)$$
and
$`\mathrm{\Psi }_N\left(\theta \right)`$ $`=`$ $`{\displaystyle \underset{\pi }{\overset{\pi }{}}}{\displaystyle \frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}}๐\sigma _d\left(\theta ^{}\right)=`$
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha }{}}\left({\displaystyle \frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta _\alpha \right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta _\alpha \right)}}+{\displaystyle \frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta +\theta _\alpha \right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta +\theta _\alpha \right)}}\right)m_\alpha \text{ },`$
where $`\left\{\pm \theta _\alpha \right\}`$ is the set of jump points of $`\sigma \left(\theta \right)`$ or, what is the same, of the points of growth of $`\sigma _d\left(\theta \right),`$
$$\frac{1}{2}m_\alpha =\sigma \left(\theta _\alpha +0\right)\sigma \left(\theta _\alpha 0\right)=\sigma _d\left(\theta _\alpha +0\right)\sigma _d\left(\theta _\alpha 0\right)\text{ }.$$
Due to (1) and (4), we have
$$\frac{1}{N^2}Q_N_2^2=\frac{1}{N^2}\underset{j,k=0}{\overset{N1}{}}b^2\left(jk\right)=$$
(5)
$$\frac{1}{N^2}\underset{\pi }{\overset{\pi }{}}\underset{\pi }{\overset{\pi }{}}\left(\underset{j,k=0}{\overset{N1}{}}\mathrm{exp}\left(i\left(jk\right)\left(\theta \theta ^{}\right)\right)\right)๐\sigma \left(\theta ^{}\right)๐\sigma \left(\theta \right)=$$
$$\underset{\pi }{\overset{\pi }{}}\mathrm{\Phi }_N\left(\theta \right)๐\sigma _c\left(\theta \right)+2\underset{\pi }{\overset{\pi }{}}\mathrm{\Phi }_N\left(\theta \right)๐\sigma _d\left(\theta \right)+\underset{\pi }{\overset{\pi }{}}\mathrm{\Psi }_N\left(\theta \right)๐\sigma _d\left(\theta \right)\text{ }.$$
Let
$$\omega \left(\delta \right)=\underset{\left|\theta \theta ^{}\right|\delta }{\mathrm{max}}\left|\sigma _c\left(\theta \right)\sigma _c\left(\theta ^{}\right)\right|\text{ }.$$
The continuity of $`\sigma _c\left(\theta \right)`$ implies $`\omega \left(\delta \right)\underset{\delta 0}{}0`$.
Since
(6)
$$\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}1$$
and
$$\left|\mathrm{sin}x\right|\frac{2}{\pi }\left|x\right|\text{ },\left|x\right|\frac{\pi }{2}\text{ }\left(\mathrm{mod2}\pi \right)\text{ },$$
then, for any $`0<\delta <\frac{2}{\pi }`$, we have
$$\mathrm{\Phi }_N\left(\theta \right)=\underset{\left|\theta \theta ^{}\right|<\delta \text{ }\left(\mathrm{mod2}\pi \right)}{}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\sigma _c\left(\theta ^{}\right)+$$
$$\underset{\left|\theta \theta ^{}\right|>\delta \text{ }\left(\mathrm{mod2}\pi \right)}{}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\sigma _c\left(\theta ^{}\right)$$
(7)
$$2\omega \left(\delta \right)+\frac{\pi ^2}{N^2}\left[\sigma _c\left(\pi \right)\sigma _c\left(\pi \right)\right]2\omega \left(\delta \right)+\frac{\pi ^2}{N^2\delta ^2}\text{ }.$$
Therefore, by an appropriate choice of $`\delta `$ and $`N`$, the first two integrals on the right-hand part of (5) can be made arbitrarily small. Hence, these integrals tend to zero as $`N\mathrm{}`$.
Rewrite now the third integral on the right-hand side of (5) in the form
$$\underset{\pi }{\overset{\pi }{}}\mathrm{\Psi }_N\left(\theta \right)๐\sigma _d\left(\theta \right)=\frac{1}{2}\underset{\alpha }{}m_\alpha ^2+$$
$$\frac{1}{2}\underset{\alpha ^{}\alpha }{}m_\alpha m_\alpha ^{}\left(\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta _\alpha \theta _\alpha ^{}\right)}{\mathrm{sin}^2\frac{1}{2}\left(\theta _\alpha \theta _\alpha ^{}\right)}+\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta _\alpha +\theta _\alpha ^{}\right)}{\mathrm{sin}^2\frac{1}{2}\left(\theta _\alpha +\theta _\alpha ^{}\right)}\right)\text{ },$$
and assume that $`\sigma _d\left(\theta \right)`$ has a finite number of jumps, then we can conclude, taking into account (6), (7) and the inequality
$$\underset{\alpha }{}m_\alpha =\sigma _d\left(\pi \right)\sigma _d\left(\pi \right)\sigma \left(\pi \right)\sigma \left(\pi \right)=1$$
that
(8)
$$\underset{N\mathrm{}}{lim}\frac{1}{N^2}Q_N_2^2=\frac{1}{2}\underset{\alpha }{}m_\alpha ^2\text{ }.$$
Thus,
(9)
$$\begin{array}{c}\underset{๐ผ}{\mathrm{max}}m_\alpha \sqrt{\frac{1}{2}\underset{\alpha }{}m_\alpha ^2}=\underset{N\mathrm{}}{lim}\frac{1}{N}Q_N_2\hfill \\ \sqrt{\underset{๐ผ}{\mathrm{max}}m_\alpha }\sqrt{\frac{1}{2}\underset{\alpha }{}m_\alpha }\sqrt{\underset{๐ผ}{\mathrm{max}}m_\alpha }\text{ }.\hfill \end{array}$$
The generalization of (8) and, hence, of (9) for the case of $`\sigma _d\left(\theta \right)`$ having infinitely many jumps can be obtained by continuity using the standard method employed in the next section.โ
The condition of Theorem 2.1 in special cases can be specified.
###### Proposition 2.2.
If the process is stable and its spectral distribution function $`\sigma \left(\theta \right)`$ satisfies the Hรถlder condition:
(10)
$$\left|\sigma \left(\theta _1\right)\sigma \left(\theta _2\right)\right|A\left|\theta _1\theta _2\right|^\nu \text{ ,}\mathrm{\hspace{0.33em}0}<\nu 1\text{,}\pi \theta _1,\theta _2\pi ,$$
then
$$Q_N_2\underset{N\mathrm{}}{=}O\left(N^{\frac{2}{2+\nu }}\right)\text{ }.$$
###### Proof.
Using the estimate (7) and the assumptions of the proposition we deduce that
$$Q_N_2\sqrt{2A\delta ^\nu N^2+\frac{\pi ^2}{\delta ^2}},\mathrm{\hspace{0.33em}0}<\delta \frac{\pi }{2}\text{ }.$$
By the minimization of the latter estimate in $`\delta `$ we obtain that
$$Q_N_2\sqrt{3}\pi ^{\frac{\nu }{2+\nu }}A^{\frac{1}{2+\nu }}N^{\frac{2}{2+\nu }}\text{ }.$$
## 3. Asymptotic form of the maximal eigenvalue of a truncated correlation matrix
Let us denote by $`\lambda _m\left(N\right)`$ the maximal eigenvalue of a positive (i.e., positive definite) matrix $`Q_N`$, $`\lambda _m\left(N\right)=Q_N`$. The condition of Theorem 2.1 admits the following weakening.
###### Theorem 3.1.
The process $`\xi \left(t\right)`$ is stable if and only if
$$\underset{N\mathrm{}}{lim}\frac{\lambda _m\left(N\right)}{N}\left(=\underset{N\mathrm{}}{lim}\frac{1}{N}Q_N\right)=0\text{ }.$$
###### Proof.
The Hilbert norm $`A`$ of any square matrix (or any nuclear operator $`A)`$ satisfies the inequality
(11)
$$A\frac{A_2}{A_1}\text{ },$$
where $`A_1`$ and $`A_2`$ are the nuclear and Hilbert-Schmidt norms, respectively. Recall that $`A`$ for positive definite $`A`$ coincides with the maximal eigenvalue $`\lambda _m\left(A\right)`$ of $`A`$ and for $`A0`$ (11) takes the form
(12)
$$\lambda _m\left(A\right)\frac{\mathrm{Tr}A^2}{\mathrm{Tr}A}\text{ }.$$
The application of (12) and (8) to $`Q_N`$ by virtue of the equality $`\mathrm{Tr}Q_N=Nb\left(0\right)=N`$ gives
(13)
$$\underset{N\mathrm{}}{\underset{ยฏ}{lim}}\frac{\lambda _m\left(N\right)}{N}\frac{1}{2}\underset{\alpha }{}m_\alpha ^2\text{ }.$$
On the other hand,
$$\lambda _m\left(N\right)\sqrt{\mathrm{Tr}Q_N^2}\text{ }.$$
Hence,
(14)
$$\overline{\underset{N\mathrm{}}{lim}}\frac{\lambda _m\left(N\right)}{N}\sqrt{\frac{1}{2}\underset{\alpha }{}m_\alpha ^2}\text{ }.$$
The inequalities (13), (14) can be specified.
###### Theorem 3.2.
Given the sequence of maximal eigenvalues (norms) $`\left\{\lambda _m\left(N\right)\right\}`$ of positive definite Tรถplitz matrices $`\left\{Q_N\right\}`$, generated by a non-negative measure $`d\sigma \left(\theta \right)`$ (3), it holds that
$$\underset{N\mathrm{}}{lim}\frac{\lambda _m\left(N\right)}{N}=\underset{๐ผ}{\mathrm{max}}m_\alpha \text{ .}$$
###### Proof.
The Tรถplitz matrix $`Q_N`$ generated by the non-decreasing function (3) is the sum of non-negative Tรถplitz matrices $`Q_N^{\left(c\right)}`$ and $`Q_N^{\left(d\right)}`$, generated by non-decreasing functions $`\sigma _c`$ and $`\sigma _d`$, respectively. Let us denote by $`\lambda _m^{\left(c\right)}\left(N\right)`$ and $`\lambda _m^{\left(d\right)}\left(N\right)`$ the maximal eigenvalues (norms) of the matrices $`Q_N^{\left(c\right)}`$ and $`Q_N^{\left(d\right)}`$, respectively. Since $`Q_N`$ $`Q_N^{\left(d\right)}`$, then
$$\lambda _m^{\left(d\right)}\left(N\right)\lambda _m\left(N\right)=Q_N^{\left(d\right)}+Q_N^{\left(c\right)}Q_N^{\left(d\right)}+Q_N^{\left(c\right)}=\lambda _m^{\left(d\right)}\left(N\right)+\lambda _m^{\left(c\right)}\left(N\right)\text{ }.$$
By virtue of Theorem 3.1, $`\lambda _m^{\left(c\right)}\left(N\right)=o\left(N\right)`$. Hence it remains to prove that
(15)
$$\underset{N\mathrm{}}{lim}\frac{\lambda _m^{\left(d\right)}\left(N\right)}{N}=\underset{๐ผ}{\mathrm{max}}m_\alpha \text{ }.$$
To this end, let us consider first the case of $`\sigma _d\left(\theta \right)`$ having only a finite number $`2s`$ of points of growth. We will not use in this proof the fact that the jump points of $`\sigma _d\left(\theta \right)`$ are located symmetrically with respect to the point $`\theta =0`$. The Tรถplitz matrix $`Q_N^{\left(d\right)}=\left(b_d\left(jk\right)\right)_{j,k=0}^{N1},`$ $`2sN`$, generated by $`\sigma _d`$, can be represented in this case in the form
(16)
$$Q_N^{\left(d\right)}=\underset{\alpha =1}{\overset{2s}{}}m_\alpha (,๐_\alpha )๐_\alpha \text{ },$$
where
$$(,๐_\alpha )๐_\alpha =\left(\mathrm{exp}\left(i\left(jk\right)\theta _\alpha \right)\right)_{j,k=0}^{N1}$$
are $`N\times N`$ matrices of unit rank, so that $`Q_N`$ transforms a $`N\times 1`$ column vector $`๐ฑ=\left(x_j\right)_{j=0}^{N1}`$ into
(17)
$$Q_N^{\left(d\right)}๐ฑ=\underset{\alpha =1}{\overset{2s}{}}m_\alpha (๐ฑ,๐_\alpha )๐_\alpha \text{ },$$
where $`(,)`$ is the scalar product in the linear space of $`N\times 1`$ column vectors $`_{}`$ defined in a standard way:
$$(๐ฑ,๐ฒ)=\underset{j=0}{\overset{N1}{}}x_j\overline{y}_j,๐ฑ=\left(x_j\right)_{j=0}^{N1},๐ฒ=\left(y_j\right)_{j=0}^{N1}\text{ }.$$
Notice that the vectors $`\left\{๐_\alpha \right\}`$ are linearly independent. Indeed, suppose that there is a set of complex numbers $`\left\{z_\alpha \right\}`$ such that
(18)
$$\underset{\alpha =1}{\overset{s}{}}z_\alpha ๐_\alpha =0\text{ }.$$
Due to (18), the numbers $`z_\alpha `$ satisfy the homogeneous system
$$\underset{\alpha =1}{\overset{s}{}}\mathrm{exp}\left(ik\theta _\alpha \right)z_\alpha =0,k=0,1,\mathrm{},2s1.$$
But the determinant of this system is the Van der Monde determinant, which vanishes if and only if among the numbers $`\left\{\mathrm{exp}\left(ik\theta _\alpha \right)\right\}`$ there are equals. The latter is impossible by our assumption. Hence all $`z_\alpha =0`$.
Let $`\lambda `$ be a non-zero eigenvalue of $`Q_N^{\left(d\right)}`$ and $`๐ก_\lambda `$ be a corresponding non-zero eigenvector:
(19)
$$\underset{\alpha =1}{\overset{s}{}}m_\alpha (๐ก_\lambda ,๐_\alpha )๐_\alpha =\lambda ๐ก_\lambda \text{ }.$$
By (19) $`๐ก_\lambda `$ admits the representation:
$$๐ก_\lambda =\underset{\alpha =1}{\overset{s}{}}z_\alpha ๐_\alpha \text{ },$$
where $`z_\alpha `$ are some complex numbers, not all of which are equal to zero. Put
$$\eta _\alpha =\sqrt{m_\alpha }(๐ก_\lambda ,๐_\alpha )\text{ }.$$
By virtue of (19), not all numbers $`\eta _\alpha =0`$. Taking the scalar products of both sides of (19) with all vectors $`\sqrt{m_\alpha }๐_\alpha `$, we obtain the following homogeneous system for $`\eta _\alpha `$:
(20)
$$\underset{\alpha ^{}=1}{\overset{s}{}}\sqrt{m_\alpha m_\alpha ^{}}(๐_\alpha ^{},๐_\alpha )\eta _\alpha ^{}=\lambda \eta _\alpha \text{ }.$$
Thus, the non-zero eigenvalues of $`Q_N^{\left(d\right)}`$ coincide, with account of their multiplicities, with the eigenvalues of the $`2s\times 2s`$ Hermitian positive definite matrix
(21)
$$A_N=\left(\sqrt{m_\alpha m_\alpha ^{}}(๐_\alpha ^{},๐_\alpha )\right)_{\alpha ,\alpha ^{}=1}^{2s}\text{ }.$$
Notice that by definition of the vectors $`๐_\alpha `$, we have
(22)
$$(๐_\alpha ^{},๐_\alpha )=\frac{\mathrm{exp}\left(iN\left(\theta _\alpha ^{}\theta _\alpha \right)\right)1}{\mathrm{exp}\left(i\left(\theta _\alpha ^{}\theta _\alpha \right)\right)1},\alpha ^{}\alpha ;(๐_\alpha ,๐_\alpha )=N\text{ }.$$
Hence, the matrix $`A_N`$ is the sum of the diagonal matrix
$$A_{1,N}:=\left(Nm_\alpha \delta _{\alpha \alpha ^{}}\right)_{\alpha ,\alpha ^{}=1}^{2s}$$
and the Hermitian matrix $`A_{2,N}`$ with zero diagonal elements and non-diagonal elements $`\sqrt{m_\alpha m_\alpha ^{}}(๐_\alpha ^{},๐_{,\alpha }),`$ $`\alpha \alpha ^{}`$. By (22) the non-diagonal elements of $`A_{2,N}`$ are uniformly bounded:
$$\left|\sqrt{m_\alpha m_\alpha ^{}}(๐_\alpha ^{},๐_\alpha )\right|2\left(\underset{\alpha ^{}\alpha }{\mathrm{max}}\left|\theta _\alpha ^{}\theta _\alpha \right|^1\right)\left(\underset{๐ผ}{\mathrm{max}}m_\alpha \right)\text{ },$$
and, hence,
$$A_{2,N}\left(4s2\right)\left(\underset{\alpha ^{}\alpha }{\mathrm{max}}\left|\theta _\alpha ^{}\theta _\alpha \right|^1\right)\left(\underset{๐ผ}{\mathrm{max}}m_\alpha \right)\text{ }.$$
Therefore,
$$\left[N\left(4s2\right)\left(\underset{\alpha ^{}\alpha }{\mathrm{max}}\left|\theta _\alpha ^{}\theta _\alpha \right|^1\right)\right]\left(\underset{๐ผ}{\mathrm{max}}m_\alpha \right)A_{1,N}A_{2,N}$$
$$A_NA_{1,N}+A_{2,N}\left[N+\left(4s2\right)\left(\underset{\alpha ^{}\alpha }{\mathrm{max}}\left|\theta _\alpha ^{}\theta _\alpha \right|^1\right)\right]\left(\underset{๐ผ}{\mathrm{max}}m_\alpha \right)\text{ }.$$
We see that
(23)
$$\lambda _m^{\left(d\right)}\left(N\right)=A_N\underset{N\mathrm{}}{=}N\underset{๐ผ}{\mathrm{max}}m_\alpha +O\left(1\right)\text{ }.$$
To prove the relation (15) for a non-decreasing step function $`\sigma _d\left(\theta \right),\sigma _d\left(\pi \right)\sigma _d\left(\pi \right)1`$, having infinitely many points of jump, we take a small $`\epsilon >0`$ and split $`\sigma _d\left(\theta \right)`$ into a sum $`\sigma _{1d}\left(\theta \right)+\sigma _{2d}\left(\theta \right)`$ of non-decreasing step functions $`\sigma _{1,d}\left(\theta \right)`$ and $`\sigma _{2,d}\left(\theta \right)`$, where, as before, $`\sigma _{1d}\left(\theta \right)`$ has a finite number of jump points and $`\sigma _{2d}\left(\theta \right)`$ is such that
$$\underset{\pi }{\overset{\pi }{}}๐\sigma _{2,d}\left(\theta \right)<\epsilon <\underset{๐ผ}{\mathrm{max}}m_\alpha \text{ }.$$
With respect to this split, we represent the Tรถplitz matrix $`Q_N^{\left(d\right)}`$ as the sum $`Q_N^{(1,d)}+Q_N^{(2,d)}`$ of non-negative Tรถplitz matrices generated by $`\sigma _{1,d}\left(\theta \right)`$ and $`\sigma _{2,d}\left(\theta \right)`$, respectively. Notice that by construction
(24)
$$Q_N^{(2,d)}\mathrm{Tr}Q_N^{(2,d)}<N\epsilon \text{ }.$$
Besides,
$$\lambda _m^{(1,d)}\left(N\right)=Q_N^{(1,d)}\lambda _m^{\left(d\right)}\left(N\right)\lambda _m^{(1,d)}\left(N\right)+Q_N^{(2,d)}\text{ }.$$
Applying the estimate (23) to $`Q_N^{(1,d)}`$ and taking into account the inequality (24) for $`N\mathrm{}`$ yields
$$N\underset{๐ผ}{\mathrm{max}}m_\alpha +O\left(1\right)=\lambda _m^{(1,d)}\left(N\right)\lambda _m^{\left(d\right)}\left(N\right)N\left(\underset{๐ผ}{\mathrm{max}}m_\alpha +\epsilon \right)+O\left(1\right)\text{ }.$$
Finally,
$$\underset{๐ผ}{\mathrm{max}}m_\alpha \underset{N\mathrm{}}{\underset{ยฏ}{lim}}\frac{\lambda _m^{\left(d\right)}\left(N\right)}{N}\overline{\underset{N\mathrm{}}{lim}}\frac{\lambda _m^{\left(d\right)}\left(N\right)}{N}\underset{๐ผ}{\mathrm{max}}m_\alpha +\epsilon \text{ },$$
where $`\epsilon >0`$ can be taken arbitrarily small.โ
###### Remark 3.3.
For the Hilbert norm $`C`$ of any $`N\times N`$ matrix $`C=\left(c_{jk}\right)_{j,k=0}^{N1}`$ the following estimate holds:
(25)
$$C\mathrm{max}\{\underset{๐}{\mathrm{max}}\underset{p=0}{\overset{N1}{}}\left|c_{jp}\right|,\underset{๐}{\mathrm{max}}\underset{p=0}{\overset{N1}{}}\left|c_{pk}\right|\}.$$
If $`C=\left(c_{jk}\right)_{j,k=0}^{N1}`$ is a Tรถplitz matrix, then by (25)
(26)
$$C\left|c_0\right|+\underset{p=1}{\overset{N1}{}}\left(\left|c_p\right|+\left|c_p\right|\right)\text{ }.$$
Thus for the Tรถplitz matrices $`Q_N`$, which are under consideration here, Theorem 3.2 by virtue of (26), gives
(27)
$$\frac{1}{2}\underset{๐ผ}{\mathrm{max}}m_\alpha =\underset{N\mathrm{}}{lim}\frac{1}{2N}Q_N\underset{N\mathrm{}}{\underset{ยฏ}{lim}}\frac{1}{N}\underset{p=0}{\overset{N1}{}}\left|b\left(p\right)\right|\text{ }.$$
Therefore the condition
(28)
$$\underset{N\mathrm{}}{lim}\frac{1}{N}\underset{p=0}{\overset{N1}{}}\left|b\left(p\right)\right|=0$$
is sufficient for the process stability.
## 4. Extension to continuous time processes
Real signals are, certainly, continuous time processes, $`\xi \left(t\right)`$. The correlation function $`b\left(t\right)`$ of a process having a finite second moment $`\xi ^2\left(t\right)`$ is a Hermitian positive function. As such, it admits the representation
(29)
$$b\left(t\right)=_{\mathrm{}}^{\mathrm{}}\mathrm{exp}\left(i\lambda t\right)๐\vartheta \left(\lambda \right)\text{ },$$
where $`\vartheta \left(\lambda \right)`$ is a bounded non-decreasing function on the real axis. Like for the discrete time processes, $`\vartheta \left(\lambda \right)`$ can be represented, in general, as the sum
$$\vartheta \left(\lambda \right)=\vartheta _c\left(\lambda \right)+\vartheta _d\left(\lambda \right)$$
of a non-decreasing continuous function $`\vartheta _c\left(\lambda \right)`$ and a non-decreasing step function $`\vartheta _d\left(\lambda \right)`$, and we call the process stable if $`\vartheta \left(\lambda \right)`$ is continuous and unstable otherwise. To investigate the instability characteristics of a continuous time process, we consider instead of the Tรถplitz matrices $`Q_N`$, the set of non-negative integral operators
(30)
$$\left(B_Tf\right)\left(t\right)=_0^Tb\left(ts\right)f\left(s\right)๐s,\mathrm{\hspace{0.33em}0}<T<\mathrm{}\text{ },$$
in the Hilbert spaces $`L_2(0,T)`$. Since $`b\left(t\right)`$ is a continuous function, all these operators are nuclear and their nuclear and Hilbert-Schmidt norms $`B_T_1`$ and $`B_T_2`$ are given by the expressions
(31)
$$B_T_1=Tb\left(0\right)=_{\mathrm{}}^{\mathrm{}}๐\vartheta \left(\lambda \right)\text{ },$$
$$B_T_2=\sqrt{2T\underset{0}{\overset{T}{}}\left(1\frac{t}{T}\right)\left|b\left(t\right)\right|^2๐t}=$$
(32)
$$\sqrt{\underset{\mathrm{}}{\overset{\mathrm{}}{}}\underset{\mathrm{}}{\overset{\mathrm{}}{}}\frac{4}{\left(\lambda \lambda ^{}\right)^2}\mathrm{sin}^2\frac{\left(\lambda \lambda ^{}\right)T}{2}d\vartheta \left(\lambda ^{}\right)๐\vartheta \left(\lambda \right)}\text{ }.$$
Let us denote, as before, by $`\left\{m_\alpha \right\}`$ the set of jumps of $`\vartheta \left(\lambda \right)`$. Using (31), (32) and the arguments similar to those employed in the proofs of Theorems 2.1 and 3.2, we obtain the following criteria of stability of a continuous time process $`\xi \left(t\right)`$.
###### Theorem 4.1.
A stationary continuous time process $`\xi \left(t\right)`$ is stable if and only if its correlation function $`b\left(t\right)`$ possesses the property:
$$\underset{T\mathrm{}}{lim}\frac{2}{T}\underset{0}{\overset{T}{}}\left(1\frac{t}{T}\right)\left|b\left(t\right)\right|^2๐t=0\text{ }.$$
Otherwise,
$$\underset{T\mathrm{}}{lim}\frac{2}{T}\underset{0}{\overset{T}{}}\left(1\frac{t}{T}\right)\left|b\left(t\right)\right|^2๐t=\underset{\alpha }{}m_\alpha ^2.$$
###### Theorem 4.2.
A stationary continuous time process $`\xi \left(t\right)`$ is stable if and only if the Hilbert norms $`B_T`$ of integral operators (30), where $`b\left(t\right)`$ is the correlation function of the process, are such that
$$\underset{T\mathrm{}}{lim}\frac{1}{T}B_T=0\text{ }.$$
Otherwise,
(33)
$$\underset{T\mathrm{}}{lim}\frac{1}{T}B_T=\underset{๐ผ}{\mathrm{max}}m_\alpha \text{ }.$$
###### Remark 4.3.
For the norm of the integral operator $`B_T`$ the following estimate:
$$B_T2\underset{0}{\overset{T}{}}\left|b\left(t\right)\right|๐t$$
holds. As it stems from (33) the relation
$$\underset{T\mathrm{}}{lim}\frac{1}{T}\underset{0}{\overset{T}{}}\left|b\left(t\right)\right|๐t=0\text{ },$$
guarantees the stability of the process $`\xi \left(t\right)`$ .
## 5. Application to processing of random signals
### 5.1. Detection of quasi-periodic components in a random signal
The jumps of the spectral distribution function $`\vartheta \left(\lambda \right)`$ of a stationary process is, in general, a sign of appearance of undamped oscillation components in a signal and the points of discontinuity of $`\vartheta \left(\lambda \right)`$ are either the frequencies of such components themselves or directly related to them. Notice that, due to physical reasons, the measurement of $`\xi \left(t\right)`$ is possible only at discrete moments of time with a step $`\mathrm{\Delta }`$. If $`t`$ in (29) is an integer multiple of $`\mathrm{\Delta }`$, then it is evident that
(34) $`b\left(t\right)`$ $`=`$ $`{\displaystyle _\mathrm{\Omega }^\mathrm{\Omega }}\mathrm{exp}\left(it\theta \right)๐\sigma \left(\theta \right),\mathrm{\Omega }={\displaystyle \frac{\pi }{\mathrm{\Delta }}}\text{ },`$
$`\sigma \left(\theta \right)`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\left[\vartheta \left(\theta +2n\mathrm{\Omega }\right)\vartheta \left(2n\mathrm{\Omega }\mathrm{\Omega }\right)\right],\mathrm{\Omega }\theta <\mathrm{\Omega }\text{ }.`$
The function $`\sigma \left(\theta \right)`$ is bounded and non-decreasing in the interval $`[\mathrm{\Omega },\mathrm{\Omega }]`$. If $`\vartheta \left(\lambda \right)`$ loses its continuity at the points $`\pm \lambda _1,\pm \lambda _2,`$ . . . , then $`\sigma \left(\theta \right)`$ has a non-void set of discontinuity points
(35)
$$\left\{\pm \theta _j^{}=\left(๐\left(\frac{\pm \lambda _j}{2\mathrm{\Omega }}+\frac{1}{2}\right)\frac{1}{2}\right)2\mathrm{\Omega }\right\}[\mathrm{\Omega },\mathrm{\Omega }]\text{ },$$
where $`๐\left(x\right)`$ is the fractional part of the number $`x`$. (In general, $`\pm \theta _{j_1}^{}`$ coincides with every $`\pm \theta _{j_2}^{}`$ such that $`\lambda _{j_1}\lambda _{j_2}`$ is a multiple of $`2\mathrm{\Omega }`$.) Therefore, in general, the jump of $`\sigma \left(\theta \right)`$ at a point $`\theta _j^{}`$ is the sum of the jumps of $`\vartheta \left(\lambda \right)`$ at all co-images of $`\theta _j^{}`$ under the mapping (35).) Taking $`\mathrm{\Delta }`$ as the time measurement unit, we return to the representation (1) of $`b\left(t\right)`$ for integer $`t`$. Thus, the spectral distribution function $`\sigma \left(\theta \right)`$ inherits all discontinuities of $`\vartheta \left(\lambda \right)`$ from the interval $`[\mathrm{\Omega },\mathrm{\Omega }]`$ and also may get new ones at the points (calculated according to (35)) related to the discontinuity points of $`\vartheta \left(\lambda \right)`$ outside this interval. We see that the spectral distribution function for the discrete time process obtained in such a way from a continuous time process, has a non-trivial component $`\sigma _d`$ if and only if the corresponding spectral distribution function of the initial discrete time process has non-zero jumps on some set of points. In other words, the values of a random continuous time process measured at discrete moments of time, form a stable discrete time process if and only if the initial process is stable.
The correlation function of the discrete time process delivers not only the described gauge of instability of the process, but also the following tool for the detection of the points $`\left\{\pm \theta _\alpha \right\}`$, which are the discontinuity points of $`\sigma \left(\theta \right)`$. Put
$$\mathrm{\Theta }_N\left(\theta \right)=b\left(0\right)+2\underset{k=0}{\overset{N1}{}}\left(1\frac{k}{N}\right)b\left(k\right)\mathrm{cos}k\theta =$$
(36)
$$=\underset{\pi }{\overset{\pi }{}}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\sigma \left(\theta ^{}\right)\text{ }.$$
It is not difficult to see that
(37)
$$\underset{N\mathrm{}}{lim}\frac{1}{N}\mathrm{\Theta }_N\left(\theta \right)=\sigma \left(\theta +0\right)\sigma \left(\theta 0\right)\text{ }.$$
Further, take a sufficiently large $`N1`$ and split the interval $`[\pi ,\pi ]`$ into equal segments of longitude $`\delta `$ such that $`N\delta 1`$. Let $`\sigma \left(\theta \right)`$ have a jump $`m_\alpha `$ within the interval $`(l\delta ,\left(l+1\right)\delta )`$. Since
$$\frac{5}{6}\left|x\right|\left|x\right|\left(1\frac{x^2}{6}\right)\left|\mathrm{sin}x\right|\left|x\right|\text{ },$$
then, for $`\left|\theta l\delta \right|\delta `$, we have
$$\mathrm{\Theta }_N\left(\theta \right)\underset{l\delta }{\overset{\left(l+1\right)\delta }{}}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{N^2\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\sigma \left(\theta ^{}\right)\frac{5}{6}N\underset{l\delta }{\overset{\left(l+1\right)\delta }{}}๐\sigma \left(\theta ^{}\right)\frac{5}{6}Nm_\alpha \text{ }.$$
On the other hand, if the continuous part $`\sigma _c\left(\theta \right)`$ of $`\sigma \left(\theta \right)`$ satisfies the condition of Proposition 2.2, then one can use the arguments similar to those employed in the proof of this proposition to show that
$$\mathrm{\Theta }_N\left(\theta \right)=O\left(N^{\frac{2}{2+\nu }}\right)$$
at the points remote from the jumps of $`\sigma \left(\theta \right)`$. Hence, the values of $`\mathrm{\Theta }_N\left(l\delta \right)`$, where $`l`$ is an integer which satisfies the condition $`\pi l\delta \pi `$, at the distances $`\delta `$ from the jump points of $`\sigma \left(\theta \right)`$, must be visible as larger than those at the distances$``$ $`l_0\delta `$, where $`Nl_0\delta 1`$.
The assertion of Theorems 2.1 and 3.2 can be used for the detection of symptoms of emerging instabilities of a random process, which can be considered as stationary for long time intervals. The method consists in the construction of the correlation function of the process from a piece of its time series from the beginning of observation to a rather far off moment of time $`\mathrm{{\rm Y}}`$ in the future. Set, as usually,
$$b\left(k\right)=\frac{1}{\mathrm{{\rm Y}}k}\underset{p=0}{\overset{\mathrm{{\rm Y}}k}{}}\xi \left(p+k\right)\xi \left(p\right)m^2,m=\frac{1}{\mathrm{{\rm Y}}}\underset{p=0}{\overset{\mathrm{{\rm Y}}}{}}\xi \left(p\right)\text{ },$$
and compute, for a sufficiently large $`N\mathrm{{\rm Y}}`$, the numbers
(38)
$$\frac{1}{N^2}Q_N_2^2=\frac{1}{N}+\frac{2}{N}\underset{k=1}{\overset{N1}{}}\left(1\frac{k}{N}\right)\frac{b^2\left(k\right)}{b^2\left(0\right)},\frac{1}{N}\underset{p=0}{\overset{N1}{}}\left|\frac{b\left(p\right)}{b\left(0\right)}\right|\text{ },$$
or the numbers
$$\frac{1}{T}\underset{0}{\overset{T}{}}\left(1\frac{t}{T}\right)\left|b\left(t\right)\right|^2๐t,\frac{1}{T}\underset{0}{\overset{T}{}}\left|b\left(t\right)\right|๐t$$
for a continuous time process. An explicit tendency of any of these numbers to be bounded, for increasing $`N`$, from below by certain positive numbers, is a serious evidence of the process instability.
The following example demonstrates that the manifestation of such a tendency begins the sooner in $`N`$ the larger is the contribution of the oscillating components generated by $`d\sigma _d\left(\theta \right)`$ into $`b\left(0\right)`$.
Let the correlation function of a stationary random process be given by the expressions:
(39) $`b\left(0\right)`$ $`=`$ $`1;b\left(k\right)=b\left(k\right)={\displaystyle \underset{\alpha =1}{\overset{s}{}}}m_\alpha \mathrm{cos}k\theta _\alpha ,k=1,2,\mathrm{}\text{ },`$
$`1`$ $``$ $`s<\mathrm{},\mathrm{\hspace{0.33em}0}<{\displaystyle \underset{\alpha =1}{\overset{s}{}}}m_\alpha <1,\mathrm{\hspace{0.33em}0}<\theta _1,\mathrm{},\theta _s<\pi \text{ }.`$
The spectral distribution function $`\sigma \left(\theta \right)`$ of such a process is the sum of the spectral distribution function $`\sigma _c\left(\theta \right)`$ of the โwhite noiseโ,
$$d\sigma _c\left(\theta \right)=\frac{p}{2\pi }d\theta ,p=1\underset{\alpha =1}{\overset{s}{}}m_\alpha \text{ },$$
and the step function $`\sigma _d\left(\theta \right)`$, the jump points of which are $`\left\{\pm \theta _\alpha \right\}`$, and
$$\sigma _d\left(\theta _\alpha +0\right)\sigma _d\left(\theta _\alpha 0\right)=\sigma _d\left(\theta _\alpha +0\right)\sigma _d\left(\theta _\alpha 0\right)=\frac{1}{2}m_\alpha \text{ }.$$
To calculate the right-hand side of (5) in this special case, notice that
$$\frac{1}{2\pi }\underset{\pi }{\overset{\pi }{}}\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta \theta ^{}\right)}{\mathrm{sin}^2\frac{1}{2}\left(\theta \theta ^{}\right)}๐\theta ^{}N\text{ }.$$
Therefore we see that now $`\mathrm{\Phi }_N\left(\theta \right)=pN^1`$ and thus
(40)
$$\frac{1}{N^2}Q_N_2^2=\frac{1}{2}\underset{\alpha =1}{\overset{s}{}}m_\alpha ^2+\frac{p\left(2p\right)}{N}+$$
$$\frac{1}{2N^2}\underset{\alpha ^{}\alpha }{}m_\alpha m_\alpha ^{}\left(\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta _\alpha \theta _\alpha ^{}\right)}{\mathrm{sin}^2\frac{1}{2}\left(\theta _\alpha \theta _\alpha ^{}\right)}+\frac{\mathrm{sin}^2\frac{1}{2}N\left(\theta _\alpha +\theta _\alpha ^{}\right)}{\mathrm{sin}^2\frac{1}{2}\left(\theta _\alpha +\theta _\alpha ^{}\right)}\right)\text{ }.$$
Hence, the first term on the right hand side of (40) becomes dominant for
$$N2p\left(2p\right)\left(\underset{\alpha =1}{\overset{s}{}}m_\alpha ^2\right)^1\text{ }.$$
### 5.2. Numerical results
Let us apply the latter results to the investigation of real signals obtained from the Forsmark 1&2 boiling water reactor (BWR) \[VGM01\], and also to simulated random signals.
In Fig.1 and Fig2 we display the sequences of the values
(41)
$$\frac{1}{N^2}Q_N_2^2=\frac{1}{N^2}\underset{j,k=0}{\overset{N1}{}}b^2\left(jk\right)$$
for the Tรถplitz matrices constructed for two different shots of real signals obtained in the Forsmark BWR, named in \[VGM01\] as c4\_lprm3 and c4\_lprm22. The sampling interval of these signals equal $`0`$ $`.08`$ $`s`$, and both of them consist of 4209 points. In \[VGM01\] several methods were used to obtain the corresponding decay ratio ($`DR`$) value. The decay ratio is a parameter related to the signal stability: the signal is more unstable if its $`DR`$ is higher, see Appendix for details. The mean values of the $`DR`$ given by these methods are $`0`$ $`.90`$ for c4\_lprm3 and $`0`$ $`.51`$ for c4\_lprm22.
We observe the difference between a more unstable shot, c4\_lprm3 and a more stable one, c4\_lprm22. As expected, for higher $`DR`$, the sequence tends to zero slower.
To determine the points of discontinuity of $`\sigma \left(\theta \right)`$ we have also calculated the function
$$\frac{1}{N}\mathrm{\Theta }_N\left(\theta \right)=\frac{b\left(0\right)}{N}+\frac{2}{N}\underset{k=0}{\overset{N1}{}}\left(1\frac{k}{N}\right)b\left(k\right)\mathrm{cos}k\theta $$
for different values of $`N`$. The corresponding results are provided at Fig.3 and Fig.4, they demonstrate different behavior of $`\frac{1}{N}\mathrm{\Theta }_N\left(\theta \right)`$ with $`N`$ growing, and for different degrees of stability measured by $`DR`$. In both cases the number of segments into which the interval \[-$`\pi `$ , $`\pi `$\] was split, was equal to 3000. It can be observed that for c4\_lprm22, a more stable signal, the function $`\frac{1}{N}\mathrm{\Theta }_N\left(\theta \right)`$ tends to zero with increasing $`N`$ much more rapidly.
The peaks of the signal correspond to the instability frequencies. The main peak for c4\_lprm3 is obtained for $`\theta =0.24`$ $`rad`$, which corresponds to the frequency of $`f=0.48`$ $`Hz`$. On the other hand, for c4\_lprm22 the main peak is located at $`\theta =0.27`$ $`rad`$ ($`f=0.53`$ $`Hz`$). We can compare these results with those obtained in \[VGM01\]. The range of main instability frequencies obtained there is $`[0.480,0.495]`$ $`Hz`$ for c4\_lprm3 and $`[0.450,0.557]`$ $`Hz`$ for c4\_lprm22. The method allows to locate secondary instability frequencies situated quite close. In Fig. 3 there is a secondary peak located at $`\theta =0.27`$ $`rad`$, it coincides with the main peak of the shot c4\_lprm22. Notice also that the present approach to the detection of instabilities is model-free.
In addition, we have also simulated signals with different degrees of stability using the model suggested in \[SVM88\], i.e., from the following continuous Langevin model:
(42)
$$\ddot{\xi }\left(t\right)+c\dot{\xi }\left(t\right)+U\left(\xi \right)=F\left(t\right),$$
with $`F(t)`$ being a Gaussian colored external force such that
(43)
$$\dot{F}\left(t\right)+\tau ^1F\left(t\right)=\tau ^1W\left(t\right),$$
and
$$U\left(\xi \right)=a_1\xi +a_2\xi ^2+a_3\xi ^3.$$
Here $`c`$ is a damping constant, $`a_j`$, $`j=1,2,3,`$ are some model constant parameters, and $`\tau `$ is the correlation time, while $`W\left(t\right)`$ is a white Gaussian noise with the correlation function
$$W(t)W(t^{})=D\delta (tt^{}).$$
The parameters of the model, $`c,`$ $`a_1,`$ $`a_2,`$ $`a_3,`$ $`D`$ and $`\tau `$, are directly related to the stability of the signal: $`a_1`$ coincides with $`w^2`$, $`w=2\pi f`$ being the fundamental frequency, and the parameters $`c`$ and $`a_1`$ determine $`DR`$, see Appendix.
In order to compare the results obtained for the simulated signals with those obtained previously for the real ones, we constructed the simulated signals with the values of the parameters chosen to produce the values of $`f`$ and $`DR`$ similar to those of the signals c4\_lprm3 and c4\_lprm22, and with the same sampling interval, $`0.08`$ $`s`$, and the same number of points, 4209. In Fig.5 and Fig.6 the sequences of the values of the Tรถplitz matrices for the models with $`c=0.689`$ ($`DR=0.5`$) and $`c=0.105`$ ($`DR=0.9`$) are gathered (the values of other parameters are: $`a_1=9`$ $`.87,`$which corresponds to $`f=0`$ $`.5`$ $`Hz`$, $`a_2=a_3=0,`$ $`D=500,`$ $`\tau =0`$ $`.6`$).
As expected, the behavior of these signals is close to that obtained for the real signals, the sequences tend slower to zero for higher $`DR`$.
For the same simulated signals, the discontinuity points are shown in Fig.7 and Fig.8. As before, the number of segments in which the interval \[-$`\pi `$ , $`\pi `$\] was split was equal to 3000.
Again, for lower $`DR`$ the function $`\frac{1}{N}\mathrm{\Theta }_N\left(\theta \right)`$ tends to zero with $`N`$ much more rapidly. The main peak is now obtained at $`\theta =0`$ $`.24`$ ($`f=0`$ $`.48`$ $`Hz)`$ for the model with $`DR=0`$ $`.5`$, and at $`\theta =0`$ $`.25`$ rad $`(f=0`$ $`.50`$ $`Hz)`$ for the model with $`DR=0`$ $`.9`$.
### 5.3. Appendix
The notion of decay ratio ($`DR`$), a basic parameter in the analysis of reactor stability, is deduced from the oscillatory model (42), but with \[STV97\]
(44)
$$U\left(\xi \right)=w^2\xi \text{ .}$$
The $`DR`$ is a measure of the system damping defined as the ratio between two consecutive maxima of the signal, for the model (42), (44) it is a constant parameter,
$$DR=\mathrm{exp}\left\{\frac{2\pi c}{\sqrt{4\omega ^2c^2}}\right\}\text{ }.$$
Neutronic signals are very noisy and, in general, their behavior cannot be fitted to that of a continuous second-order system, hence the $`DR`$ in reality is not a constant, and its value depends on the model used to evaluate it \[VGM01\]. Nevertheless, it gives a hint on the system stability.
|
warning/0506/cond-mat0506121.html
|
ar5iv
|
text
|
# Quantum Phase Transitions in Spin-1/2 Ising Chain in Regularly Alternating Transverse Field: Spin Correlation Functions
## Abstract
We consider the spin-$`\frac{1}{2}`$ Ising chain in a regularly alternating transverse field to examine the effects of regular alternation on the quantum phase transition inherent in the quantum Ising chain. The number of quantum phase transition points strongly depends on the specific set of the Hamiltonian parameters but never exceeds $`2p`$ where $`p`$ is the period of alternation. Calculating the spin correlation functions numerically (for long chains of up to 5400 sites) and determining the critical exponents we have demonstrated that two types of critical behavior are possible. In most cases the square-lattice Ising model universality class occurs, however, a weaker singularity may also take place.
journal: SCES โ04
,
We consider the spin-$`\frac{1}{2}`$ Ising chain in a transverse field assuming that the Hamiltonian parameters (i.e. the nearest neighbor interactions and the on-site fields) vary regularly along the chain with a finite period of alternation $`p`$. The Hamiltonian of the model reads
$`H={\displaystyle \underset{n}{}}2I_ns_n^xs_{n+1}^x+{\displaystyle \underset{n}{}}\mathrm{\Omega }_ns_n^z`$ (1)
and the sequence of parameters in (1) is
$`I_1\mathrm{\Omega }_1I_2\mathrm{\Omega }_2\mathrm{}I_p\mathrm{\Omega }_pI_1\mathrm{\Omega }_1I_2\mathrm{\Omega }_2\mathrm{}I_p\mathrm{\Omega }_p\mathrm{}.`$
Recently, it has been shown that a regularly alternating transverse field may strongly influence the quantum phase transition inherent in the uniform chain . In what follows we consider the chain of period 2 with $`\mathrm{\Omega }_{1,2}=\mathrm{\Omega }\pm \mathrm{\Delta }\mathrm{\Omega }`$ and $`I_n=1`$. Then the quantum phase transition points follow from the condition
$`\left(\mathrm{\Omega }^{}+\mathrm{\Delta }\mathrm{\Omega }\right)\left(\mathrm{\Omega }^{}\mathrm{\Delta }\mathrm{\Omega }\right)=\pm 1.`$ (2)
Eq. (2) gives two critical fields $`\mathrm{\Omega }^{}=\pm \sqrt{\mathrm{\Delta }\mathrm{\Omega }^2+1}`$ if $`\mathrm{\Delta }\mathrm{\Omega }<1`$ or four critical fields $`\mathrm{\Omega }^{}=\pm \sqrt{\mathrm{\Delta }\mathrm{\Omega }^2\pm 1}`$ if $`\mathrm{\Delta }\mathrm{\Omega }>1`$. Moreover, if $`\mathrm{\Delta }\mathrm{\Omega }=1`$ in addition to $`\mathrm{\Omega }^{}=\pm \sqrt{2}`$ we have one more critical field $`\mathrm{\Omega }^{}=0`$. Further we restrict ourselves to the case $`\mathrm{\Delta }\mathrm{\Omega }=1`$ and examine the critical behavior in the vicinity of the two representative critical points $`\mathrm{\Omega }^{}=\sqrt{2}`$ and $`\mathrm{\Omega }^{}=0`$. This is by no means a restrictive example since only two types of critical behavior are expected for the model (1) . The critical behavior is characterized by a set of critical exponents. Namely, the order parameter (Ising magnetization) vanishes as $`m^x\left(\mathrm{\Omega }^{}\mathrm{\Omega }\right)^\beta `$. The two-point spin correlation function $`s_n^xs_{n+2r}^xs_n^xs_{n+2r}^x`$ in the limit $`r\mathrm{}`$ decays as $`\left(2r\right)^\eta \mathrm{exp}\left(\frac{2r}{\xi }\right)`$, where $`\xi |\mathrm{\Omega }\mathrm{\Omega }^{}|^\nu `$. The transverse static susceptibility diverges as $`\chi ^z|\mathrm{\Omega }\mathrm{\Omega }^{}|^\alpha `$ and the energy gap disappears as $`\mathrm{\Delta }|\mathrm{\Omega }\mathrm{\Omega }^{}|^{\nu z}`$. From the analytical calculations we know that in the vicinity of $`\mathrm{\Omega }^{}=\sqrt{2}`$ we have $`\nu z=1`$ and $`\alpha =0`$ ($`\chi ^z`$ exhibits a logarithmic singularity) whereas in the vicinity of $`\mathrm{\Omega }^{}=0`$ we have $`\nu z=2`$ and $`\alpha =2`$ ($`\frac{^2\chi ^z}{\mathrm{\Omega }^2}`$ exhibits a logarithmic singularity). In the present paper we complete these analytical findings by precise numeric results for other critical exponents $`\eta `$, $`\beta `$ and $`\nu `$ computing two-point spin correlation functions for chains of a few thousand sites .
We start our analysis considering $`s_n^xs_{n+2r}^x`$ at the critical fields $`\mathrm{\Omega }^{}=0`$ and $`\mathrm{\Omega }^{}=\sqrt{2}`$. The results reported in the main plot in Fig. 1
clearly indicate the power-law decay of spin correlations with $`\eta =\frac{1}{4}`$ at both critical points $`\mathrm{\Omega }^{}=0`$ (squares) and $`\mathrm{\Omega }^{}=\sqrt{2}`$ (diamonds). We calculate the on-site $`x`$-magnetization $`m_j^x`$, $`m_{j}^{x}{}_{}{}^{2}=lim_r\mathrm{}s_j^xs_{j+2r}^x`$, for finite chains of $`N`$ sites putting $`j=\frac{1}{4}N`$, $`2r=\frac{1}{2}N`$. $`m_j^x`$ vanishes at both critical fields as $`N^{\frac{1}{8}}`$ as can be seen in the inset in Fig. 1 (squares refer to $`\mathrm{\Omega }^{}=0`$, diamonds refer to $`\mathrm{\Omega }^{}=\sqrt{2}`$).
We turn to the critical region. The results for $`m^x=\frac{1}{2}\left(m_1^x+m_2^x\right)`$ are shown in Fig. 2 (see also ).
In the vicinity of $`\mathrm{\Omega }^{}=0`$ (Fig. 2a) we observe a power-law vanishing of the order parameter with $`\beta =\frac{1}{4}`$. In the vicinity of $`\mathrm{\Omega }^{}=\sqrt{2}`$ (Fig. 2b) we observe a power-law vanishing of the order parameter with $`\beta =\frac{1}{8}`$. Finally, in Fig. 3
we present the results for the correlation length. While in the vicinity of $`\mathrm{\Omega }^{}=0`$ we find $`\nu =2`$ (Fig. 3a) in the vicinity of $`\mathrm{\Omega }^{}=\sqrt{2}`$ we clearly observe $`\nu =1`$ (Fig. 3b).
To summarize, combining the analytical and numerical results we have found that the spin-$`\frac{1}{2}`$ Ising chain in a regularly alternating transverse field may exhibit the critical behavior which belongs to the square-lattice Ising model universality class with $`\beta =\frac{1}{8}`$, $`\nu =1`$, $`\eta =\frac{1}{4}`$, $`\alpha =0`$, $`z=1`$ (as the considered chain at $`\mathrm{\Omega }^{}=\sqrt{2}`$) and a weaker singularity with $`\beta =\frac{1}{4}`$, $`\nu =2`$, $`\eta =\frac{1}{4}`$, $`\alpha =2`$, $`z=1`$ (as the considered chain at $`\mathrm{\Omega }^{}=0`$). Let us finally note, that the spin correlation functions required for the estimation of exponents $`\beta `$, $`\nu `$, $`\eta `$ should be possible to calculate analytically as in the uniform case. We left this problem for future studies.
The authors thank J. Richter and O. Zaburannyi for collaboration in the earlier stage of this study and Ja. Ilnytskyi for discussions. Preliminary results of this study were presented at the 5th Small Triangle Meeting on theoretical physics (Medzev, Slovak Republic, 2003). O. D. thanks the organizers for support and M. Jaลกฤur and J. Streฤka for discussions.
|
warning/0506/astro-ph0506667.html
|
ar5iv
|
text
|
# A search for HI in some peculiar faint dwarf galaxies
## 1 Introduction
Here we present a search for HI in three peculiar dwarf galaxies, POX 186, SC 24 and KKR 25. Based, in part, on previous, single dish HI observations, these galaxies have been classified as a blue compact dwarf (BCD), a dwarf irregular and a transition galaxy respectively.
POX 186 is an unusually compact BCD galaxy with a linear size of only $`300`$ pc<sup>1</sup><sup>1</sup>1 Following Guseva et al. (2004) we assume a distance of 18.5 Mpc for this galaxy.. HST observations (Corbin & Vacca 2002) found an unusual asymmetry in the galaxy which they interpreted as a tidal feature. They hence argued for POX 186 being the result of a recent ($`<10^8`$ yr) collision between two sub-galactic sized clumps, and thus representative of a dwarf galaxy in formation. However this conclusion remains controversial as, from a study of ionized gas emission from the galaxy, Guseva et al.(2004) argued that the asymmetry in the morphology of POX 186 comes from a starburst-driven gaseous shell and not from tidal arms. As signatures of tidal interactions are often most prominent in HI, a deep HI image of POX 186 would be crucial in resolving this issue. While single dish observations did not detect any HI from the galaxy (Kunth et al. 1988), the limit on the HI mass of the galaxy of $`2.0\times 10^7`$ M is somewhat higher than the expected HI mass for the galaxy ($`6\times 10^6`$ M, assuming M$`{}_{\mathrm{HI}}{}^{}/\mathrm{L}_\mathrm{B}`$ $`2`$, which is a typical value for BCDs of luminosity comparable to POX 186 $``$ e.g. Pustilnik et al. 2002).
KKR 25 (M$`{}_{\mathrm{B}}{}^{}9.96`$) has been classified as a โtransitionโ galaxy because of its significant HI content ($`1.2\times 10^6\mathrm{M}_{}`$ $``$ Huchtmeier et al. 2003) and the presence of a population of faint blue stars (Karachentsev et al. 2001). However, Grebel et al.(2003) point out that properties of KKR 25 are consistent with those of dwarf spheroidal (dSph) galaxies, as no H$`\alpha `$ was detected from the galaxy and it also lies in the same region of the metallicity-luminosity plane as occupied by dwarf spheroidal galaxies. Based on the I magnitude of the tip of the red giant branch, Karachentsev et al. (2001) estimated the distance to KKR 25 to be 1.86 Mpc.
The dwarf irregular galaxy SC 24 is the lowest luminous member of the Sculptor group and is also one of the faintest (M$`{}_{\mathrm{B}}{}^{}`$8.39) irregular galaxies known. This galaxy was discovered by Cรดte et al. (1997), as a part of a survey of dwarf galaxies of the Sculptor group. Based on the distance-velocity relationship for the Sculptor group galaxies, Skillman et al.(2003) derived a distance of 1.66 Mpc to SC 24.
Our new Giant Metrewave Radio Telescope (GMRT) observations failed to detect HI in any of these galaxies. In what follows, we discuss our new GMRT observations, and the impact of our non detection of HI on the classification of these galaxies.
## 2 Observations, analysis and results
HI 21cm observations of KKR 25, SC 24 and POX 186 were conducted with the GMRT (Swarup et al. 1991). The setup for the observations is given in Table 1.
The data were reduced in the usual way using standard tasks in classic AIPS. The GMRT has a hybrid configuration which simultaneously provides both high angular resolution ($`3^{^{\prime \prime }}`$ if one uses baselines between the arm antennas) as well as sensitivity to extended emission (from baselines between the antennas in the central array). Data cubes were therefore made using various (u,v) cutoffs, allowing a search for HI and continuum emission at various spatial resolutions (given in Table 2).
For all the galaxies, the data cubes were examined for line emission at a variety of spectral resolutions $``$ in all cases no significant emission was found. Besides visual inspection, the AIPS task SERCH was also used to search for line emission in all data cubes. No statistically significant feature was detected in the cubes. To derive the final limit on the HI mass for those galaxies with previous single dish HI detections, the data were smoothed to the velocity width of the single dish spectrum. In particular, for KKR 25 a velocity width of 15 km s<sup>-1</sup>was used (Huchtmeier et al. 2003), while velocity widths of 55 km s<sup>-1</sup>and 21 km s<sup>-1</sup>were used for SC 24 (see Sect. 3.2). For POX 186, in the absence of a previous single dish detection a velocity width of $``$20 km s<sup>-1</sup>(a typical velocity width for such faint galaxies e.g. Begum & Chengalur 2004, Begum et al. 2003) was used. Finally, continuum images at $`4^{^{\prime \prime }}\times 3^{^{\prime \prime }}`$ resolution were also made for all the galaxies by averaging central 80 channels. No continuum was detected in any galaxy. The derived 3$`\sigma `$ limits from the continuum images are given in Table 2.
Table 2 summarizes our results for a representative selection of spatial resolutions for three galaxies. Col. (1) represents the galaxy name, (2) the spatial resolution defined by the half-power beam width (HPBW) of the synthesised beam, (3) RMS noise corresponding to this spatial resolution, (4) the 5$`\sigma `$ upper limit on the HI mass (5) velocity resolution used for deriving the limits ($`\mathrm{\Delta }`$v) on the HI mass and (6) derived $`3\sigma `$ limit from the continuum image (RMS<sub>cont</sub>) .
## 3 Discussion
The non detection of HI with the GMRT in KKR 25 and SC 24, in conflict with the previous single dish detections of these galaxies, could be because the HI emission has resolved out by the GMRT. However, our past experience with the GMRT in successfully imaging dwarf galaxies with HI flux and optical sizes similar to SC 24 and KKR 25 (Begum et al. (2005) in preparation), makes such a possibility unlikely. Further, as discussed below, for both galaxies, a case can be made for confusion of emission from the galaxy with local HI.
### 3.1 KKR 25
The Leiden/Dwingeloo survey of galactic neutral hydrogen (Hartman 1994) detects HI in the direction of KKR 25 in the same velocity range as detected by the single dish observations for this galaxy, it is hence likely that the HI emission that has been associated with this galaxy actually arises from local gas. The upper limit on the HI mass from the current GMRT observations is comparable to the limits obtained for typical dwarf spheroidal galaxies (Mateo 1998 and references therein).
### 3.2 SC 24
There have been two separate claims of detection of HI in SC 24, viz. Cรดte et al. (1997) and Skillman et al. (2003). The single dish flux measured by Cรดte et al. (1997) is 11.8 Jy km s<sup>-1</sup>, with a width at the 50% level $`\mathrm{\Delta }\mathrm{V}_{50}`$$``$ 55.0 km s<sup>-1</sup>. On the other hand, the HI flux estimated from the HIPASS database at the position of SC 24, is only 3.2 Jy km s<sup>-1</sup>with $`\mathrm{\Delta }\mathrm{V}_{50}21`$km s<sup>-1</sup>(Skillman et al. 2003).
We note that in the case of the Cรดte et al. (1997) observations, the signal to noise of the HI spectrum is poor and that there is a strong possibility of confusion with local HI. Indeed, the HIPASS survey detects HI from the Magellanic stream in the direction of SC 24 at the radial velocity of the galaxy (Putman 2003).
If one goes by the current GMRT observations, the upper limit on the HI mass of SC 24 is similar to the typical HI mass limits for dSph galaxies (see Mateo 1998 and references therein). However the optical appearance of the galaxy suggest that it is a dwarf irregular galaxy (Karachentsev et al. 2004). Further, unlike other member galaxies of the Sculptor group, the HST observations failed to resolve this galaxy into stars (Karachentsev, I. D., private communication). Hence, SC 24 is more likely to be a distant galaxy.
### 3.3 POX 186
Our GMRT observations for POX 186 confirms the previous single dish non detection, albeit with a much better limit on the HI mass. The upper limit on the HI mass of the galaxy computed<sup>2</sup><sup>2</sup>2The velocity width and confidence limit used to calculate the upper limit on the HI mass for POX 186 is not specified in the paper from the single dish observations is $`2\times 10^7\mathrm{M}_{}`$ (Kunth et al. 1988), whereas the 5$`\sigma `$ upper limit on the HI mass derived from our interferometric observations is $`2.4\times 10^6\mathrm{M}_{}`$. Since HI in dwarf galaxies typically extends to $``$ 2 times the Holmberg diameter (e.g. Hunter 1997), and the angular size of the optical emission is $`3^{\prime \prime }`$, (Corbin & Vacca 2002), the upper limit we quote is derived from the $`7^{\prime \prime }\times 7^{\prime \prime }`$ resolution data cube, and for a velocity width of $`20`$ km s<sup>-1</sup>. HST observations of POX 186 found a young OB stellar cluster with an estimated age of $`10^610^7`$ yr (Corbin & Vacca, 2002). We searched for HI emission near the location of this cluster at the highest angular resolution of our data (4$`{}_{}{}^{\prime \prime }\times 3^{\prime \prime }`$) but did not find any statistically significant emission. Our 5$`\sigma `$ upper limit (again for a velocity width of $`20`$ km/s) is $`1.6\times 10^6\mathrm{M}_{}`$. The absence of HI in POX 186 is some-what puzzling, as these young stars must be associated with some neutral HI. From a statistical study of a sample of BCDs in various environments, Pustilnik et al. (2002) found that BCDs in voids have a higher M<sub>HI</sub>/L<sub>B</sub> than those in higher density regions and that there is also a trend for increasing M<sub>HI</sub>/L<sub>B</sub> with decreasing L<sub>B</sub>. From these correlations, the expected M<sub>HI</sub>/L<sub>B</sub> for POX 186 is $`2`$. Hence, our derived upper limit on the HI mass of POX 186, (which corresponds to M<sub>HI</sub>/L$`{}_{\mathrm{B}}{}^{}0.8`$) means that the galaxy has a somewhat smaller M<sub>HI</sub>/L<sub>B</sub> than is typical for BCDs. However it should be noted that the scatter in M<sub>HI</sub>/L<sub>B</sub> for BCDs at a given luminosity is large, and further that the correlations found by Pustilnik et al. (2002) were computed for galaxies brighter than M$`{}_{\mathrm{B}}{}^{}16.0`$ mag, and it is not clear whether the fainter BCDs follow a same trend as the brighter ones. Given the small size of the galaxy and the fact that a significant amount of ionized gas has been detected in POX 186 (Guseva et al. 2004), it is possible that a sizeable fraction of HI in the galaxy has been ionized by the recent burst of star formation.
## 4 Conclusions
To conclude, despite a deep search, we find no HI emission associated with POX 186, SC 24 and KKR 25. The non-detection of HI in SC 24 and KKR 25 suggests that previous single dish measurements were affected by confusion with the galactic emission. Our stringent limits on the HI mass of KKR 25 indicate that it is a normal dwarf spheroidal galaxy, whereas SC 24 is more likely to be a distant galaxy. In the case of POX 186, the derived HI mass limit is somewhat smaller than is typical for BCDs.
## Acknowledgments
We are grateful to I. D. Karachentsev for providing important information about SC 24 which helped us in interpreting our results. The GMRT is operated by the National Center for Radio Astrophysics of the Tata Institute of Fundamental Research.
|
warning/0506/hep-ph0506328.html
|
ar5iv
|
text
|
# Probing Color Response - Wakes in a Color Plasma
## 1 Introduction
QCD Jet quenching in relativistic heavy ion collisions has been proposed as an important possible signal of the creation of a quark-gluon plasma . It is extensively studied at RHIC.
The radiative energy loss of the leading parton due to the emission of a secondary partonic shower has been the main emphasis in the theoretical studies so far. For an overview of the quantum field theoretical calculations of the radiative energy loss and the experimental implications, see e. g. .
The study focussed on a different aspect of the QCD jet physics in medium, namely the properties of the wake of current density, charge density and (chromo-)electirc and magnetic field configurations that is induced by a charged particle traveling through a QCD medium with high momentum. The present contribution relies on that work .
The coherent behavior of the plasma as a reaction of a charged particle traveling through it is investigated by calculating the plasmaโs response to that external current. The calculation is restricted to only one weak, external current which points in a fixed direction in color group space. Methods of linear response theory are applied. The medium is assumed to be isotropic and homogenous. In such a framework quantum and non-ablian effects are included indirectly via the dielectric functions, $`ฯต_L`$ and $`ฯต_T`$. Linear response theory implies that the dielectric functions are not in turn modified by the effects of the external current.
Two qualitatively different scenerarios are investigated: In the first one, one assume that the plasmon is in a high temperature regime. There the gluon self-energies can be described using a high temperature expansion $`T\omega ,k`$. In the second scenario we study what are the significant physical effects if the plasma is strongly coupled (sQGP). In that scenario the plasma might be described best in terms of a quantum liquid.
It will be shown that the wake of a jet in such a quantum liquid can exhibit a characteristic cone-structure in the charge- and current-densities under certain conditions for the plasmonโs dispersion relation and assuming a โsupersonicalโ velocity of the jet . Since it can be expected that the phenomenon of these Mach-cones leads to correlations in the directed emission of secondary partons from the plasmon. Observable consequences for particle correlations of this effect will be discussed in detail later in the present contribution.
The idea of determining the sound velocity of plasmon excitations of the expanding plasma from the emission pattern of the plasma particles traveling at an angle with respect to the jet axis has been discussed in . Our consideration of Mach-cones have also been motivated in parts by earlier studies of wakes induced by fast electrically charged projectiles in the electron plasma of metallic targets , as well as by recent work exploring the induction of a conical flow by a jet in a sQGP . This latter study invokes a hydrodynamical description of the energy deposited by a quenched jet in the medium and emphasizes the emergence of a Mach cone.
This proceeding contribution is organized as follows: first, in section 2, I give a short reminder of some general properties of plasma physics and derive the equation for the induced charge- and current-density in the considered situation. In section 3 I discuss the wake as obtained in a pQGP. In section 4 the wake obtained in a strongly coupled QGP is discussed. Possible observable consequences for particle correclations are studied in section 5. In section 6 I end with conclusions and a short acknowledgment.
The reader solely interested in the results related to a Mach cone structure might actually skip the next two sections and continue at section 4.
## 2 Plasma physics
The formalism for linear response theory of a plasma can straightforwardly be generalized from electromagnetic plasma physics (see e. g. ).
A dielectric medium can be characterized by the components the dielectric tensor $`ฯต_{ij}`$, which can be decomposed into $`ฯต_L`$ and $`ฯต_T`$ for an isotropic and homogenous medium via the longitudinal and transverse orthonormal projectors $`๐ซ_{L,ij}=k_ik_j/k^2`$ and $`๐ซ_T=1๐ซ_L`$ with respect to the momentum vector $`\stackrel{}{k}`$. Invoking a quantum field theoretical description one can relate these dielectic functions to the self-energies $`\mathrm{\Pi }_L`$ and $`\mathrm{\Pi }_T`$ of the in-medium gluon
$`ฯต_L(\omega ,k)=1{\displaystyle \frac{\mathrm{\Pi }_L(\omega ,k)}{K^2}},`$ (1)
$`ฯต_T(\omega ,k)=1{\displaystyle \frac{\mathrm{\Pi }_T(\omega ,k)}{\omega ^2}},`$ (2)
where $`K^2=\omega ^2k^2`$.
The knowledge of these dielectric tensors allows to relate the external current $`\stackrel{}{j}_{\mathrm{ext}}`$ to the total chromoelectric field $`\stackrel{}{E}_{\mathrm{tot}}^a`$ in the QCD plasma via:
$`\left[ฯต_L๐ซ_L+\left(ฯต_T{\displaystyle \frac{k^2}{\omega ^2}}\right)๐ซ_T\right]\stackrel{}{E}_{\mathrm{tot}}^a(\omega ,\stackrel{}{k})={\displaystyle \frac{4\pi }{i\omega }}\stackrel{}{j}_{\mathrm{ext}}^a(\omega ,\stackrel{}{k}).`$ (3)
The color charge induced by the external charge distribution is
$`\rho _{\mathrm{ind}}=\left({\displaystyle \frac{1}{ฯต_L}}1\right)\rho _{\mathrm{ext}}.`$ (4)
The induced color charge density can also be calculated from the induced scalar potential via a Poisson equation:
$`\mathrm{\Phi }_{\mathrm{ind}}={\displaystyle \frac{4\pi }{k^2}}\rho _{\mathrm{ind}},`$ (5)
if one works in a gauge where the vector potential is transverse to the momentum .
Since one can relate the total chromo-electric field to the induced charge in linear response theory by
$`\stackrel{}{j}_{\mathrm{ind}}^a=i\omega (1ฯต)\stackrel{}{E}_{\mathrm{tot}}^a/(4\pi ),`$ (6)
a direct relation between the external and the induced current can be derived using Eqn. (3):
$`\stackrel{}{j}_{\mathrm{ind}}^a=\left[\left({\displaystyle \frac{1}{ฯต_L}}1\right)๐ซ_L+{\displaystyle \frac{1ฯต_T}{ฯต_T\frac{k^2}{\omega ^2}}}๐ซ_T\right]\stackrel{}{j}_{\mathrm{ext}}^a.`$ (7)
The induced charge and the induced current obey a continuity equation:
$`i\stackrel{}{k}\stackrel{}{j}_{\mathrm{ind}}i\omega \rho _{\mathrm{ind}}=0.`$ (8)
For the following it is helpful to specify the current and charge densities associated with a color charge as Fourier transform of a point charge moving along a straight-line trajectory with constant velocity $`\stackrel{}{v}`$:
$`\stackrel{}{j}_{\mathrm{ext}}^a`$ $`=`$ $`2\pi q^a\stackrel{}{v}\delta (\omega \stackrel{}{v}\stackrel{}{k}),`$ (9)
$`\stackrel{}{\rho }_{\mathrm{ext}}^a`$ $`=`$ $`2\pi q^a\delta (\omega \stackrel{}{v}\stackrel{}{k})`$ (10)
where $`q^a`$ is its color charge defined by $`q^aq^a=C\alpha _s`$, with the strong coupling constant $`\alpha _s=g^2/4\pi `$ and the quadratic Casimir invariant $`C`$ (which is either $`C_F=4/3`$ for quarks or $`C_A=3`$ for a gluon). In this model description changes of the color charge due to interactins while propagating through the medium are disregarded by fixing the chargeโs orientation in color space The induced charge- $`\rho _{\mathrm{ind}}`$ for this situation reads in cylindrical coordinates:
$`\rho _{\mathrm{v},\mathrm{ind}}(\rho ,z,t)={\displaystyle \frac{m_g^3}{(2\pi )^2v}}q^a{\displaystyle _0^{\mathrm{}}}๐\kappa ^{}\kappa ^{}J_0(\kappa ^{}\rho m_g){\displaystyle _{\mathrm{}}^{\mathrm{}}}๐\omega ^{}\mathrm{exp}\left[i\omega ^{}\left({\displaystyle \frac{z}{v}}t\right)m_g\right]\left({\displaystyle \frac{1}{ฯต_L}}1\right),`$ (11)
where $`k=\sqrt{\kappa ^2+\omega ^2/v^2}`$, $`\kappa =\kappa ^{}m_g`$ and $`\omega =\omega ^{}m_g`$, showing that the induced charge density $`\rho _{\mathrm{v},\mathrm{ind}}`$ is proportional to $`m_g^3`$. The cylindrical symmetry around the jet axis restricts the form of the current density vector $`\stackrel{}{j}_{\mathrm{ind}}`$. It has only non-vanishing components parallel to the beam axis, $`\stackrel{}{j}_{\mathrm{v},\mathrm{ind}}`$, and radially perpendicular to it, $`\stackrel{}{j}_{\rho ,\mathrm{ind}}`$:
$`\stackrel{}{j}_{\mathrm{v},\mathrm{ind}}(\rho ,z,t)`$ $`=`$ $`{\displaystyle \frac{m_g^3}{(2\pi )^2v^2}}q^a{\displaystyle _0^{\mathrm{}}}d\kappa ^{}\kappa ^{}J_0(\kappa ^{}\rho m_g)\times `$
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐\omega ^{}\mathrm{exp}\left[i\omega ^{}\left({\displaystyle \frac{z}{v}}t\right)m_g\right]\left[\left({\displaystyle \frac{1}{ฯต_L}}1\right){\displaystyle \frac{\omega ^2}{k^2}}+{\displaystyle \frac{1ฯต_T}{ฯต_T\frac{k^2}{\omega ^2}}}\left(v^2{\displaystyle \frac{\omega ^2}{k^2}}\right)\right],`$
$`\stackrel{}{j}_{\rho ,ind}(\rho ,z,t)`$ $`=`$ $`{\displaystyle \frac{im_g^3}{(2\pi )^2v}}q^a{\displaystyle _0^{\mathrm{}}}d\kappa ^{}\kappa ^{}J_1(\kappa ^{}\rho m_g)\times `$ (12)
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}๐\omega ^{}\mathrm{exp}\left[i\omega ^{}\left({\displaystyle \frac{z}{v}}t\right)m_g\right]{\displaystyle \frac{\omega \kappa }{k^2}}\left[\left({\displaystyle \frac{1}{ฯต_L}}1\right)\left({\displaystyle \frac{1ฯต_T}{ฯต_T\frac{k^2}{\omega ^2}}}\right)\right].`$
Again, the components of the current density are proportional to $`m_g^3`$.
In order to deepen the understanding of the results obtained in the different scenarios it is helpful to discuss the corresponding plasmon dispersion relations for a given dielectric function. In order to determine non-trivial solutions of Equation (3), the following determinant has to vanish:
$`\mathrm{det}\left|ฯต_L๐ซ_L+(ฯต_T{\displaystyle \frac{k^2}{\omega ^2}})๐ซ_T\right|=0.`$ (13)
This equation governs the dispersion relation for the waves in the medium. Since it can be diagonalized into purely longitudinal and transverse parts, dispersion relations for the longitudinal and transverse dieelectric functions can be derived :
$`ฯต_L`$ $`=`$ $`0,`$ (14)
$`ฯต_T`$ $`=`$ $`(k/\omega )^2.`$ (15)
These equations determine the longitudinal and transverse plasma modes. The longitudinal equation is also the dispersion relation for density-fluctuations in the plasma, namely space-charge fields which could be spontaneously excited in the plasma without an application of external disturbances .
## 3 Charge wake in a QGP in the high temperature approximation
In this section the first scenario is studied in detail. Here the medium is assumed to be in the high temperature regime where the gluon self-energies are described by the leading order of the high temperature expansion $`T\omega ,k`$ . The self-energies derived within the HTL approximation have been shown to be gauge invariant and the dielectric functions are therefore also gauge invariant. The dielectric functions read explicitly :
$`ฯต_L`$ $`=`$ $`1+{\displaystyle \frac{2m_g^2}{k^2}}\left[1{\displaystyle \frac{1}{2}}x\left(\mathrm{ln}\left|{\displaystyle \frac{x+1}{x1}}\right|i\pi \mathrm{\Theta }\left(1x^2\right)\right)\right],`$
$`ฯต_T`$ $`=`$ $`1{\displaystyle \frac{m_g^2}{\omega ^2}}\left[x^2+{\displaystyle \frac{x(1x^2)}{2}}\left(\mathrm{ln}\left|{\displaystyle \frac{x+1}{x1}}\right|i\pi \mathrm{\Theta }\left(1x^2\right)\right)\right],`$ (16)
where $`x=\omega /k`$.
These dielectric function leads to longitudinal and transverse plasma modes, that are determined by the dispersion relations (16). They can only appear in the time-like sector of the $`\omega ,k`$ plane , see Fig. 1. Collective excitations do not contribute to the charge and current density profile of the wake. Mach cones do not appear, but the charge carries a screening color cloud along with it. Fig. 2 illustrates this physically intuitive result numerically. It shows the charge density of a colored parton traveling with $`v=0.99c`$ in cylindrical coordinates. The screening cloud is concentrated in the vincinty of the particle. In the static case where the current density vector vanishes, the induced charge density can be calculated analytically, it has a Yukawa-like shape $`\mathrm{exp}(\sqrt{2m_gr})`$.
Despite the fact that Mach cones do not appear in the charge density wake, the particle still suffers energy loss due to elastic collisions in the medium. This effect of energy dissipation has been studied in in detail. The integrand determining the energy loss contributes only where $`|x|<1`$, and therefore does not get contributions from frequencies where collective plasma modes exist. This is consistent with the fact that such modes are not excited in the induced charge and current densities.
## 4 Charge wake induced in a strongly coupled QGP
Here we assume that the plasma is in a strongly coupled regime having properties of a quantum liquid. Since there is a lack of theoretical methods for first principle calculations for the dielectric functions in this regime, we use a simple model calculation in this scenario. Nonetheless, the simplified model is constructed in such a way that it allows for general conclusion quite independent from the exact form of the dielectric functions.
The notion of a sQGP suggest low dissipation at small momenta (โhydro modesโ) and large dissipation at high $`k_c`$. The existence of a critical momentum $`k_c`$ is assumed which separates the regimes of collective and single particle modes in the quantum liquid. Below $`k_c`$ are plasmon excitations and dissipation is assumed to be neglible for simplicity. Since we are interested in collective effects, the region above $`k_c`$ is neglected.
To be specific, it is assumed that the longitudinal dielectric function of the strongly coupled plasma in the $`k<k_c`$ regime is
$`\omega _\mathrm{L}=\sqrt{u^2k^2+\omega _p^2},`$ (17)
where $`\omega _p`$ denotes the plasma frequency and $`u`$ the speed of plasmon propagation which is assumed to be constant. The most important property of this dispersion relation for our purposes is, that it extends into the space-like region of the $`\omega ,k`$ plane above some $`k`$. For the dispersion relation (17) this is the case for $`k>\omega _p/\sqrt{(}1u^2)`$. This is different for the high-temperature plasma, where longitudinal and transverse plasma modes only appear in the time-like region, $`|x|=|\omega /k|>1`$. In the quantum liquid scenario one can expect that the modes with low phase velocity $`|x|<u`$ suffer severe Landau damping because they accelerate the slower moving charges and decelerate those moving faster than the wave . A charge moving with a velocity that is lower than the speed of plasmon propagation can only excite these modes and not the modes with intermediate phase velocities $`u<|x|<1`$, which are undamped . The qualitative properties of the color wake can in this case expected to be analogous to those of the high temperature plasma, namely that the charge carries only a screening color cloud with it and Mach cones do not appear.
If the colored parton travels with a velocity $`v>u`$ that is higher than the speed of sound in the medium, modes with an intermediate phase velocity $`u<|x|<1`$ can be excited. The emission of these plasma oscillations induced by supersonically traveling particles is analogous to Cherenkov radiation. This can be expected to lead to the emergence of Mach cones in the induced charge density cloud with the opening angle
$`\mathrm{\Delta }\mathrm{\Phi }=\mathrm{arccos}\left(u/v\right).`$ (18)
The effect is well known in solid state physics and is analogous to the Mach cones induced by fast heavy ions in electron plasmas . The physics of shock waves in relativistic heavy-ion collisions has also been discussed in .
To be specific the following dielectric function is assumed in accordance with (17):
$`ฯต_L=1+{\displaystyle \frac{\omega _p^2/2}{u^2k^2\omega ^2+\omega _p^2/2}}(kk_c).`$ (19)
Note that this differs from the classical, hydrodynamical dielectric function of Bloch , since the latter one is singular at small $`k`$ and $`\omega `$ due to phonon contributions which cannot mix with colored plasmons.
This dielectric function (19) has the plasmon mode (see Eqn. 17) which extends into the space-like region of the $`\omega ,k`$ plane, see Fig. 1. The qualitative induced wake structure in such a quantum liquid scenario is general, namely a Mach shock wave structure for a โsupersonicallyโ traveling color source as discussed above. In that sense the principal findings can be expected to hold generally for a quantum liquid with a plasmon branch similiar to (19) independent of the exact form of the dielectric function.
To be specific a speed of plasmon propagation of $`u/c=1/\sqrt{3}`$ is assumed (compared that to $`u/c=\sqrt{3/5}`$ in the HTL approximation and to $`u/c\sqrt{0.2}`$ for a hadron resonance gas).
Figures 4a,b show the induced charge density for a colored particle traveling โsupersonicallyโ at $`v=0.99c>u`$ which is calculated using eq. (11). The integration area has been restricted to the region $`k<k_c=2\omega _p`$<sup>1</sup><sup>1</sup>1Note that the expressions in do not correspond to $`k<k_c`$, but $`\kappa <k_c`$. . Numerical consistency has been checked also via calculation of the induced current densities and the continuity equation. On the other hand a โsubsonicallyโ traveling particle with $`v=0.55c<v`$ induces a charge density profile which is qualitativly analogous to the high-temperature plasma, namely a comoving screening cloud, see Fig. 3.
## 5 Observable consequences
It can be expected that the phenomenon of Mach cones in the charge density translates into Mach cones in the particle density and should eventually lead to a directed emission of secondary partons from the plasma. This effect has an analogoy in solid state physics where Mach cones induced by fast heavy ions in an electron plasma lead to an emission of electrons that have been carried within the wake . This effect has been studied experimentally .
If that scenario of a strongly coupled QCD plasma is realized in the matter created in an ultrarelativistic heavy ion collision, one could expect to observe these cones in the actual distribution of secondary particles associated with jets at RHIC, a signature also proposed in .
Indeed preliminary data from the PHENIX and STAR experiments (see e.g. Fig. 1 in and preliminary data from PHENIX) show such effects in the background distribution of secondary particles in the azimuthal angle $`\mathrm{\Delta }\varphi `$. The peak near zero degree corresponds to secondaries from the outmoving jet, particles from the companion jet result in a distribution with a clear maximum near $`\mathrm{\Delta }\varphi =\pi `$ for pp collisions, where no medium effects are present. In Au-Au collisions there is a distribution with two maxima around the $`\mathrm{\Delta }\varphi =\pi `$ position. These are located at $`\mathrm{\Delta }\varphi \pi \pm 1.1`$. One can argue that such an effect could possibly be traced back to a Mach shock front (predicted in a hydrodynamical framework) traveling with the side-away jet leading to maxima in the distribution at about $`\mathrm{\Delta }\mathrm{\Phi }=\pi \pm \mathrm{arccos}(u/v)`$.
Given the confirmation of this effect in the data mentioned, this would clearly indicate that the first scenario, viz. a pQCD plasma, is not realized in RHIC experiments. A further decision on the possible occurence of Mach cone structures could be deduced by studying correlations in the secondaries as proposed by the PHENIX collaboration. Indication of such a Mach cone structure could reveal properties of the plasma and itโs plasmons in general, which is a fascinating perspective. The speed of plasma propagation could be also determined. In fact if the maxima at $`\mathrm{\Delta }\varphi \pi \pm 1.1`$ are experimentally confirmed, it would correspond to $`u/c\sqrt{0.2}`$ . It is interesting to note, that a study of the angular structure of the collisional energy losses of a hard jet in the medium would also support such an observation: the incident hard jetโs scattering angle vanishes in the relativistic limit - leading to the jetโs propagation along a straight line - whereas the expectation value of the scattering angle $`\mathrm{\Theta }`$ of a struck โthermalโ particle is $`\mathrm{\Theta }1.04`$ which is close to $`1.1`$.
## 6 Conclusions
The properties of the charge density wake of a colored hard partonic jet traveling through a QGP plasma in the framework of linear response theory has been discussed. Two different scenarios have been studied, namely a high temperature QGP at $`TT_c`$ described in the HTL approximation, and a description of a strongly coupled QGP (sQGP) behaving as a quantum liquid. It was found that the structure of the wake corresponds to a screening color cloud traveling with the particle in the case of the high temperature plasma and in the case of a quantum fluid, if the velocity of sound in the plasma is not exceeded by the jet in the latter case. The structure of the wake is changed considerably in comparison to the former cases, if the jetโs velocity exceeds the plasmaโs speed of sound and the collective modes have a dispersion relation extending in the space-like region. Then the induced parton density exhibits the characteristics of Mach waves trailing the jet at the Mach angle.
It is argued that this effect could be used to constrain theoretically possible scenarios by experimental analysis via the measurments of angular distribution of the secondary particle cones of jet events in RHIC. First indications of the observation of a Mach shock phenomenon in RHIC in the quenched jetโs secondary particle distribution from PHENIX and STAR data were discussed.
In general secondary particle distributions can be used to provide methods of probing the QCD plasmaโs collective excitations experimentally.
## Acknowledgements
The results presented here rely on work done in a collaboration with Berndt Mรผller whom I want to thank in the first place. I thank Purnendu Chakraborty and Munshi G. Mustafa for drawing my attention to a numerical inconsistency regarding an earlier version of Fig. 2. I also want to thank Steffen Bass, Jorge Casalderry, Rainer Fries, Roy Lacey, Abhijit Majumder, and Wolf Holzmann for interesting discussions. This work was supported in part by U. S. Department of Energy under grants DE-FG02-96ER40945 and DE-FG02-05ER41367. I thank the Alexander von Humboldt Foundation for support as a Feodor Lynen Fellow.
|
warning/0506/gr-qc0506004.html
|
ar5iv
|
text
|
# Conformally Flat Noncircular Spacetimes
## I Introduction
One of the challenges in general relativity is the search for interior configurations describing isolated rotating bodies supporting their corresponding exterior gravitational fields. Usually, the description of these rotating configurations is done by means of circular spacetimes, i.e., stationary axisymmetric spacetimes where the metric, in addition to be time and rotation-angle independent, possesses also as isometries the inversions of time and of the rotation angle. It is worthwhile to mention that almost all the stationary axisymmetric configurations reported in the literature belong to this class.
Nevertheless, there is no room for the circular idealization when considering rotating neutron stars Gourgoulhon:1993 surrounded by strong toroidal magnetic fields ranging $`10^{16}`$ to $`10^{17}`$ G, see also Ioka:2003dd ; Ioka:2003nh and references therein. The circularity condition is a very severe restriction, which fails to hold in spacetimes allowing the existence of toroidal magnetic fields and meridional flows Ioka:2003nh . Thus, to deal with such astrophysical configurations one has to abandon the fulfillment of the circularity condition and consider in consequence the wider noncircular class of spacetimes.
Besides the above astrophysically motivated reasons to study noncircular configurations, there are also purely theoretical ones. As soon as Schwarzschild published his exterior spherically symmetric static solution, he was able to determine its interior solution modeled trough a perfect fluid with homogeneous density. Later, on the light of the Petrov classification, it was established that the Schwarzschild solution belongs to Petrov typeโD, while the interior Schwarzschild solution falls in the conformally flat family. In 1973 Kerr reported his famous stationary axisymmetric gravitational field corresponding to a field created by a rotating body; this solution also belongs to Petrov typeโD. The search for the interior solution to the exterior Kerr metric began since that time. Collinson established that a conformally flat stationary axisymmetric spacetime is necessarily static Collinson:1976 . Some stationary axisymmetric PetrovโD metrics coupled to perfect fluid distributions have been reported in the literature, but non of them allows for the matching with the Kerr metric. Recently, the results of Ref. Vera:2003cn indicate that the matching between an interior noncircular spacetime with an exterior circular one is at least technically possible. This fact opens the possibility of searching for interior solutions within the noncircular class.
Moreover, the general metric for conformally flat stationary cyclic symmetric circular metric has been reported recently Garcia:2002gj , see on this respect also Barnes:2003gz ; Garcia:2003is . For this metric, being cyclic but not axisymmetric, the circularity theorem Kundt:1966 does not hold because of the lack of a rotation axis. The next step in complexity, which is the main goal of the present work, is to determine the metric for conformally flat stationary cyclic symmetric noncircular spacetimes. In particular, from this more general result one is able to derive the particular circular branch, and if one requires the existence of an axis of symmetry one recovers the staticity property of the considered class of metrics, thus the Collinson theorem is just included within a more general result.
In the next section the mathematical preliminaries needed in order to study the spacetimes under consideration are introduced. Specifically, the physical and geometrical details behind the concepts of stationarity, cyclic symmetry, axisymmetry, circularity, and staticity are clearly stated in order to make the work self-contained. In Sec. III the conformal flatness constraints, consisting in the vanishing of the complex Weyl components, are fully integrated for any stationary cyclic symmetric spacetime including in general noncircular contributions. Sec. IV is devoted to revise the conditions guarantying staticity on the obtained spacetimes, and in Sec. V the locus of the axis of symmetry, for the configurations allowing one, is found. It is concluded that both physical situations occur for the same configuration. Some conclusions are given in Sec. VI. In appendix A the explicit form of the complex components of the Weyl tensor for a general, not necessarily circular, stationary cyclic symmetric spacetime are given. In appendix B, the problem is addressed for a singular case ($`a=b`$) not included in the generic treatment.
## II Stationary Cyclic Symmetric Spacetimes
In this section we characterize stationary cyclic symmetric spacetimes, see for example Ref. Heusler:1996 for the original definitions. A spacetime is *stationary* if it admits an asymptotically timelike Killing field $`๐`$. A spacetime is called *cyclic symmetric* if it is invariant under the action of the oneโparameter group $`SO(2)`$, it is assumed that the corresponding Killing field $`๐`$ with closed integral curves is spacelike. A cyclic symmetric spacetime is named *axisymmetric* if the fixed point set of the $`SO(2)`$ action, i.e., the *rotation axis*, is nonempty. A spacetime is called *stationary cyclic symmetric (axisymmetric)* if it is both stationary and cyclic symmetric (axisymmetric) and if the Killing fields $`๐`$ and $`๐`$ commute.
A stationary cyclic symmetric (axisymmetric) spacetime is said to be *circular* if the $`2`$-surfaces orthogonal to the Killing fields $`๐`$ and $`๐`$ are integrable. This is equivalent to satisfy the Frobenius integrability conditions
$`๐๐๐
๐`$ $`=`$ $`0,`$
$`๐๐๐
๐`$ $`=`$ $`0.`$ (1)
The circularity property means that locally the gravitational field is not only independent of time and the rotation angle, but, it is also invariant under the simultaneous inversion of time and the angle. Almost all the literature related to stationary cyclic symmetric (axisymmetric) spacetimes concerns only with the circular case. This is due in part for simplicity, since in this case it is possible to use the Lewis-Papapetrou ansatz for the metric.
As a last definition, a stationary spacetime is said to be *static* if the Killing field $`๐`$ is hypersurface orthogonal. This occurs if and only if it satisfies
$$๐๐
๐=0,$$
(2)
and it is equivalent to demand that locally the gravitational field is not only time-independent but it is also invariant under time-reversal.
In this work we are interested in noncircular spacetimes, i.e, general stationary cyclic symmetric spacetimes not necessarily restricted to satisfy the Frobenius integrability conditions (1). The metric of such spacetimes can be written as
$`๐`$ $`=`$ $`\mathrm{e}^{2Q}({\displaystyle \frac{1}{a+b}}[๐
๐+a๐
๐+\mathrm{Im}(M๐
๐)]`$ (3)
$`\times [๐
๐b๐
๐\mathrm{Im}(N๐
๐)]`$
$`+\mathrm{e}^{2P}๐
๐๐
\overline{๐}),`$
where $`a`$, $`b`$, $`P`$, and $`Q`$ are real functions and $`M`$ and $`N`$ are complex ones. Here the bar means complex conjugation, and $`\mathrm{Im}`$ (respectively $`\mathrm{Re}`$) denotes the imaginary (respectively real) part of a complex quantity. All the functions depend on the coordinates $`z`$ and $`\overline{z}`$ only, since the Killing fields realizing the stationary and cyclic isometries are $`๐=\mathbf{}_๐`$ and $`๐=\mathbf{}_๐`$. The above metric has eight independent real functions, hence, diffeomorphism invariance allows to make two other gauge elections. This metric (3) is invariant under the rescaling $`zg(z)^1dz`$ together with the redefinitions $`PP\mathrm{ln}\sqrt{g(z)\overline{g}(\overline{z})}`$, $`MMg(z)`$, and $`NNg(z)`$. We shall fix the gauge just after exploiting this special symmetry in our calculations.
The noncircularity of the above metric can be realized from the fact that the following quantities are not necessarily zero
$`(๐๐๐
๐)`$ $`=`$ $`{\displaystyle \frac{\mathrm{e}^{2(PQ)}}{2(a+b)}}\mathrm{Re}({\displaystyle \frac{(MN)}{\overline{z}}}`$ (4)
$`{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{(ab)}{\overline{z}}}),`$
$`(๐๐๐
๐)`$ $`=`$ $`{\displaystyle \frac{\mathrm{e}^{2(PQ)}}{2(a+b)}}\mathrm{Re}(a{\displaystyle \frac{M}{\overline{z}}}+b{\displaystyle \frac{N}{\overline{z}}}`$ (5)
$`{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{(ab)}{\overline{z}}}),`$
where the star stands for the Hodge dual.
In order to evaluate the Weyl tensor it is more convenient to use the Newman-Penrose formalism. One starts writing the metric as
$$๐=2๐^1๐^22๐^3๐^4,$$
(6)
using a complex null tetrad, which in the present case is chosen as
$`๐^1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\mathrm{e}^{QP}๐
๐,`$ (7a)
$`๐^2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\mathrm{e}^{QP}๐
\overline{๐},`$ (7b)
$`๐^3`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{\mathrm{e}^Q}{\sqrt{a+b}}}\left(๐
๐b๐
๐{\displaystyle \frac{N๐
๐\overline{N}๐
\overline{๐})}{2i}}\right),`$ (7c)
$`๐^4`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{\mathrm{e}^Q}{\sqrt{a+b}}}\left(๐
๐+a๐
๐+{\displaystyle \frac{M๐
๐\overline{M}๐
\overline{๐}}{2i}}\right).`$ (7d)
The related Weyl complex components are given in the Appendix A.
## III Solving the Conformal Flatness Constraints
In order to find the general class of conformally flat stationary cyclic symmetric spacetimes we demand the vanishing of all the complex components of the Weyl tensor, i.e., $`\mathrm{\Psi }_0=\mathrm{\Psi }_4=\mathrm{\Psi }_2=\mathrm{\Psi }_1=\mathrm{\Psi }_3=0`$, see Appendix A.
The complex components $`\mathrm{\Psi }_0`$ and $`\mathrm{\Psi }_4`$ are the same than in the circular case ($`M=0=N`$) studied in Refs. Collinson:1976 ; Garcia:2002gj . Hence, initially, we shall apply the same strategy of these references. First, the vanishing of the following combinations
$`\mathrm{\Psi }_0\overline{\mathrm{\Psi }}_4`$ $`=`$ $`2(a+b)\mathrm{e}^{2Q}`$ (8)
$`\times {\displaystyle \frac{}{z}}\left({\displaystyle \frac{\mathrm{e}^{2P}}{(a+b)^2}}{\displaystyle \frac{(ab)}{z}}\right)=0,`$
$`\mathrm{\Psi }_0{\displaystyle \frac{b}{z}}+\overline{\mathrm{\Psi }}_4{\displaystyle \frac{a}{z}}`$ $`=`$ $`2(a+b)\mathrm{e}^{2(QP)}`$ (9)
$`\times {\displaystyle \frac{}{z}}\left({\displaystyle \frac{\mathrm{e}^{4P}}{(a+b)^2}}{\displaystyle \frac{a}{z}}{\displaystyle \frac{b}{z}}\right)=0,`$
give rise to the following first order conditions
$`{\displaystyle \frac{a}{z}}{\displaystyle \frac{b}{z}}`$ $`=`$ $`\overline{g}(\overline{z})(a+b)^2e^{2P},`$ (10)
$`{\displaystyle \frac{a}{z}}{\displaystyle \frac{b}{z}}`$ $`=`$ $`\overline{h}(\overline{z})(a+b)^2e^{4P},`$ (11)
where $`g`$ and $`h`$ are integration functions.
Secondly, since the functions $`a`$, $`b`$, and $`P`$ are real Eq. (10) implies
$$g(z)\frac{}{z}(ab)=\overline{g}(\overline{z})\frac{}{\overline{z}}(ab).$$
(12)
Rescaling the complex coordinate by,
$$(\tau ,\sigma ,z,\overline{z})(\tau ,\sigma ,g(z)^1dz,\overline{g}(\overline{z})^1d\overline{z}),$$
(13)
the above equation allows to conclude that in terms of the new complex coordinate: $`ab=F(z+\overline{z})`$.
Using the rescaling (13), together with the redefinitions $`PP\mathrm{ln}\sqrt{g(z)\overline{g}(\overline{z})}`$ and $`\overline{h}(\overline{z})\overline{h}(\overline{z})/\overline{g}(\overline{z})^2`$, in Eqs. (10) and (11) is equivalent to put $`g(z)=1`$ without loosing generality. This is due to the fact that metric (3), as it was previously anticipated, is invariant under such changes if we consider also the redefinitions $`MMg(z)`$ and $`NNg(z)`$.
Combining the fact that $`ab=F(z+\overline{z})`$ with the imaginary part of the component $`\mathrm{\Psi }_2`$,
$$\mathrm{Im}(\mathrm{\Psi }_2)=\frac{\mathrm{e}^{2(Q+P)}}{i(a+b)^2}\left(\frac{a}{z}\frac{b}{\overline{z}}\frac{a}{\overline{z}}\frac{b}{z}\right)=0,$$
(14)
we conclude that if $`F\mathrm{const}.`$ then $`a=a(z+\overline{z})`$, $`b=b(z+\overline{z})`$, and consequently $`P=P(z+\overline{z})`$ by virtue of Eq. (10).
In what follows we analyze the generic case $`\mathrm{d}F/\mathrm{d}x0`$, leaving the study of the special case $`\mathrm{d}F/\mathrm{d}x=0`$ for the Appendix B.
With regard to the other integration function appearing in Eq. (11), since the left hand side of this equation is real then $`\overline{h}(\overline{z})=h(z)=\mathrm{const}.ฯตk^2`$, where $`ฯต\pm 1`$ just encodes the sign of the constant (see Ref. Garcia:2002gj for the transcendence of this sign).
Using the real and imaginary parts of $`z`$ as coordinates, $`z=x+iy`$, Eqs. (10) and (11) are now expressed by
$`{\displaystyle \frac{\mathrm{d}a}{\mathrm{d}x}}{\displaystyle \frac{\mathrm{d}b}{\mathrm{d}x}}`$ $`=`$ $`(a+b)^2\mathrm{e}^{2P},`$ (15)
$`{\displaystyle \frac{\mathrm{d}a}{\mathrm{d}x}}{\displaystyle \frac{\mathrm{d}b}{\mathrm{d}x}}`$ $`=`$ $`ฯตk^2(a+b)^2\mathrm{e}^{4P}.`$ (16)
We are now ready to fix the gauge. We choose that the redefined complex functions $`M`$ and $`N`$, after the rescaling (13), are real functions. That is, our gauge elections are $`\overline{M}=M`$ and $`\overline{N}=N`$. It is important to emphasize that until is correct to make such election from the beginning, it is useless since we loose the scaling freedom in the metric which allows to fix one of the above integration functions. In terms of the real coordinates $`x`$ and $`y`$ the metric is written in this gauge as
$`๐`$ $`=`$ $`\mathrm{e}^{2Q}({\displaystyle \frac{1}{a+b}}(๐
๐+a๐
๐+M๐
๐)`$ (17)
$`\times (๐
๐b๐
๐N๐
๐)`$
$`+\mathrm{e}^{2P}(๐
๐^2+๐
๐^2)),`$
where now we have six real functions, $`M`$, $`N`$, and $`Q`$ depending on $`x`$ and $`y`$, and $`a`$, $`b`$, and $`P`$ depending just on $`x`$. In the above coordinates the four gauge elections are $`g_{\tau x}=g_{\sigma x}=g_{yx}=0`$ and $`g_{xx}g_{yy}`$ is proportional to the remaining noncircular components.
Let us infer some consequences concerning the noncircular components of the metric. Using that $`a=a(x)`$, $`b=b(x)`$, and $`P=P(x)`$ the real part of the following combination is written as
$$\mathrm{Re}(\mathrm{\Psi }_3\mathrm{\Psi }_1)=\frac{\sqrt{a+b}\mathrm{e}^{2Q+3P}}{8}\frac{^2}{xy}\left(\frac{M+N}{a+b}\right)=0,$$
(18)
which implies that the following function is separable in $`x`$ and $`y`$, i.e.,
$$\frac{M+N}{a+b}=F_1(x)+F_2(y),$$
(19)
where $`F_1`$ and $`F_2`$ are undetermined functions. Isolating $`N`$ from the above expression and inserting it in the real part of $`\mathrm{\Psi }_1`$ we obtain
$$\mathrm{Re}(\mathrm{\Psi }_1)=\frac{\mathrm{e}^{2Q+3P}}{8\sqrt{a+b}}\frac{^2}{xy}\left[M(F_1+F_2)a\right]=0,$$
(20)
and hence
$$M(x,y)=[F_1(x)+F_2(y)]a(x)+F_3(x)+F_4(y),$$
(21)
where $`F_3`$ and $`F_4`$ is another pair of undetermined functions. Using Eq. (19), $`N`$ is given by
$$N(x,y)=[F_1(x)+F_2(y)]b(x)F_3(x)F_4(y).$$
(22)
The dependence on the coordinate $`y`$ of the functions $`M`$ and $`N`$ (and of the whole problem) is encoded in the functions $`F_2`$ and $`F_4`$. However, such functions can be eliminated by shifting appropriately the Killing coordinates, i.e., the coordinate change
$$(\tau ,\sigma ,x,y)(\tau +F_4(y)dy,\sigma +F_2(y)dy,x,y),$$
(23)
is equivalent to put $`F_2=0=F_4`$. We would like to emphasize that due to the same argument the functions $`F_1`$ and $`F_3`$ must be determined up to the addition of constants factors; such constant factors can be included within the definitions of $`F_2`$ and $`F_4`$ and eliminated by the previous transformation. In summary, the only dependence on the coordinate $`y`$ of metric (3) appears in the conformal factor $`Q`$.
Now, we apply the same strategy of Ref. Garcia:2002gj , namely we first redefine the functions $`a`$ and $`b`$ by
$`a+b`$ $`=2kY,`$ $`a`$ $`=k(Y+X),`$
$`ab`$ $`=2kX,`$ $`b`$ $`=k(YX).`$ (24)
Using the new functions $`X`$ and $`Y`$, Eqs. (15) and (16) are rewritten as
$`{\displaystyle \frac{\mathrm{d}X}{\mathrm{d}x}}`$ $`=`$ $`2kY^2\mathrm{e}^{2P},`$ (25)
$`\left({\displaystyle \frac{\mathrm{d}Y}{\mathrm{d}x}}\right)^2\left({\displaystyle \frac{\mathrm{d}X}{\mathrm{d}x}}\right)^2`$ $`=`$ $`4ฯตk^2Y^2\mathrm{e}^{4P}.`$ (26)
Equation (25) suggests to choose a new coordinate $`x`$ defined by
$$(\tau ,\sigma ,x,y)(\tau ,\sigma ,2kY^2\mathrm{e}^{2P}dx,y).$$
(27)
The general solutions of Eqs. (25) and (26) in terms of the new coordinate $`x`$ are
$`X(x)`$ $`=`$ $`x,`$ (28)
$`Y^2(x)`$ $`=`$ $`(xx_0)^2ฯต.`$ (29)
Up to now we have integrated seven equations; the four related with $`\mathrm{\Psi }_0`$ and $`\mathrm{\Psi }_4`$, the imaginary part of $`\mathrm{\Psi }_2`$, and the real parts of $`\mathrm{\Psi }_3\mathrm{\Psi }_1`$ and $`\mathrm{\Psi }_1`$. Hence, it remains to integrate the other three equations which allows to specify the functions $`P`$, $`F_1`$, and $`F_3`$. These equations are the following
$$\mathrm{\Psi }_3\mathrm{\Psi }_1=\frac{ik^{3/2}Y^{7/2}\mathrm{e}^{2QP}}{\sqrt{2}}\frac{\mathrm{d}^2}{\mathrm{d}x^2}(F_3+kxF_1)=0,$$
(30)
which integrates just as $`F_3=kxF_1`$, since the two related integration constants can be absorbed within the definitions of the functions $`F_1`$ and $`F_3`$ and eliminated by a shifting of the Killing coordinates as in the transformation (23). Also we use
$$\mathrm{\Psi }_2=\frac{k^2Y^3\mathrm{e}^{2Q}}{3}\frac{\mathrm{d}^2}{\mathrm{d}x^2}\left(Y\mathrm{e}^{2P}+\frac{k}{2}Y^2F_{1}^{}{}_{}{}^{2}\right)=0,$$
(31)
determining the function $`P`$ as
$$\mathrm{e}^{2P}=\frac{C_0+C_1x}{Y}\frac{k}{2}YF_{1}^{}{}_{}{}^{2}.$$
(32)
The last equation is
$$\mathrm{\Psi }_3+\mathrm{\Psi }_1=\frac{ik^{5/2}Y^{5/2}\mathrm{e}^{2QP}}{\sqrt{2}}\frac{\mathrm{d}^2}{\mathrm{d}x^2}(F_1Y^2)=0,$$
(33)
giving
$$F_1(x)=\frac{K_0+K_1x}{(xx_0)^2ฯต}.$$
(34)
As last step, in order to write the obtained metric in simple form we make the following coordinate transformation, see Ref. Garcia:2002gj ,
$$(\tau ,\sigma ,x,y)(\sqrt{2}(\tau +kx_0\sigma ),\sqrt{2}k\sigma ,xx_0,2ky),$$
(35)
together with the next redefinition of the conformal factor $`QQ+\frac{1}{4}\mathrm{ln}(16k^4Y^2)`$ and also simple redefinitions of the involved constants. The final form of the most general conformally flat stationary cyclic symmetric metric is
$`๐`$ $`=`$ $`e^{2Q(x,y)}(k(๐
๐^2+2x๐
๐๐
๐+ฯต๐
๐^2)`$ (36)
$`+{\displaystyle \frac{๐
๐^2}{(C_0+C_1x)(x^2ฯต)k(K_0+K_1x)^2}}`$
$`+2k(K_0+K_1x)๐
๐๐
๐+(C_0+C_1x)๐
๐^2).`$
It is easy to note that for $`K_0=0=K_1`$ we recover the circular metrics of Ref. Garcia:2002gj . Instead, for $`K_00`$ and $`K_10`$ the above metric is noncircular as follows from the following quantities
$`(๐๐๐
๐)`$ $`=`$ $`k^2\mathrm{e}^{2Q}(K_0+K_1x),`$
$`(๐๐๐
๐)`$ $`=`$ $`k^2\mathrm{e}^{2Q}(ฯตK_1+K_0x).`$ (37)
## IV Staticity
As it was defined in Sec. II, the spacetimes derived in the previous section would be static if there exist a timelike linear combination of the Killing fields,
$$๐_\text{s}=A\frac{\mathbf{}}{\mathbf{}๐}+B\frac{\mathbf{}}{\mathbf{}๐},$$
(38)
satisfying the staticity condition (2). For metric (36) such condition becomes
$`0=(๐_\text{s}๐
๐_\text{s})`$ $`=`$ $`k[B(K_0AฯตK_1B)๐
๐`$ (39)
$`+B(K_1AK_0B)๐
๐`$
$`+(A^2ฯตB^2)๐
๐].`$
It is straightforward to realize that we are in the presence of static configurations only if
$$ฯต=1\mathrm{and}K_1=\pm K_0,$$
(40)
in which case the hypersurface orthogonal Killing fields are proportional to $`๐_\text{s}=\mathbf{}\mathbf{/}\mathbf{}๐\pm \mathbf{}\mathbf{/}\mathbf{}๐`$.
## V Axisymmetry
Now we turn our attention to the existence of a rotation axis, i.e., there are conformally flat stationary axisymmetric configurations within the class (36). It follows from the definition of Sec. II that the rotation axis is the spacetime region where the cyclic Killing field $`๐`$ vanishes. For metric (36) its general stationary and cyclic Killing fields are a linear combination of the vectors $`\mathbf{}_๐`$ and $`\mathbf{}_๐`$. Hence, performing the transformation $`(\tau =\alpha t+\beta \varphi ,\sigma =\gamma t+\delta \varphi )`$, where $`\alpha \delta \beta \gamma 0`$, the Killing fields are written as $`๐=\mathbf{}_๐`$ and $`๐=\mathbf{}_\mathit{\varphi }`$, respectively. In terms of the new coordinates, all the metric components $`g_{\varphi \mu }=๐(๐,\mathbf{}_๐)`$ must vanish on the axis, which implies the following set of algebraic equations
$`g_{\varphi t}`$ $`=`$ $`ke^{2Q}[(\alpha \delta +\beta \gamma )x+\alpha \beta +ฯต\gamma \delta ]=0,`$ (41a)
$`g_{\varphi \varphi }`$ $`=`$ $`ke^{2Q}(\beta ^2+2\beta \delta x+ฯต\delta ^2)=0,`$ (41b)
$`g_{\varphi y}`$ $`=`$ $`ke^{2Q}\delta (K_0+K_1x)=0.`$ (41c)
Isolating $`x`$ from Eq. (41b) and inserting the result in Eq. (41a) we obtain
$$\frac{(\alpha \delta \beta \gamma )(\beta ^2ฯต\delta ^2)}{\beta \delta }=0,$$
(42)
since $`\alpha \delta \beta \gamma 0`$ the above equation has nontrivial solutions only if $`ฯต=1`$. Using the above conditions in the remaining Eq. (41c) we obtain
$$\delta (K_1K_0)=0,$$
(43)
which implies that the related rotation axis must be located at
$$x=1.$$
(44)
Summarizing, metric (36) describes a stationary axisymmetric spacetime only for $`ฯต=1`$ and $`K_1=\pm K_0`$, i.e., the conditions for the existence of the rotation axis are the same than guaranty than the spacetime is static, see conditions (40).
As a consequence, the known Collinson theorem Collinson:1976 ; Garcia:2002gj not only is generalized to include configurations which are not necessarily circular, but, it is part of a more general statement: any conformally flat stationary cyclic symmetric spacetime, even a noncircular one, is additionally axisymmetric if and only if it is also static.
## VI Conclusions
In this paper we study all the stationary cyclic symmetric spacetimes which are at the same time conformally flat. In contrast to previous work on the subject we consider also noncircular configurations. The conformal flatness is imposed by demanding the vanishing of the Weyl tensor. The resulting constraints are extremely involved by the inclusion of noncircular contributions, and leave us with a system of ten nonlinear pdes, see Appendix A. However, it is still possible to achieve its full integration as we show in detail at Sec. III. The class of obtained spacetimes is fully determined up to a conformal factor which respects the spacetime symmetries, and several integration constants. In particular, two integration constants characterize the noncircular behavior of these spacetimes, when they vanish we recover the circular configurations obtained in Ref. Garcia:2002gj by two of the authors.
We investigate the conditions allowing the existence of a rotation axis in the resulting configurations. The static spacetimes within the class are also considered. It results that the parameters election for these two physical situations is the same: $`ฯต=1`$ and $`K_1=\pm K_0`$, i.e., the involved spacetimes are axisymmetric if and only if they are also static. Hence, one of the main result of the paper can be summarized in the following
> Theorem: *Any conformally flat stationary cyclic symmetric spacetime, even a noncircular one, is additionally axisymmetric if and only if it is also static.*
With regard to the properly cyclic symmetric class (with no rotation axis, and by the above theorem containing necessarily nonstatic spacetimes) it will be interesting to investigate what kind of sources can solve Einstein equations with gravitational fields within this class. In the case that be possible to retain the noncircular contributions in this process, the derived configurations would corresponds to the first exact noncircular gravitational fields found in the literature.
###### Acknowledgements.
We thank M. Hassaรฏne for discussions. This work has been partially supported by FONDECYT Grants 1040921, 7040190, and 1051064, CONACyT Grants 38495E and 34222E, CONICYT/CONACyT Grant 2001-5-02-159, and MECESUP Grant FSM 0204. The generous support of Empresas CMPC to the Centro de Estudios Cientรญficos (CECS) is also acknowledged. CECS is a Millennium Science Institute and is funded in part by grants from Fundaciรณn Andes and the Tinker Foundation.
## Appendix A Weyl Complex Components
For the null tetrad (7) the complex components of the Weyl tensor are given by
$`\mathrm{\Psi }_0`$ $`=`$ $`{\displaystyle \frac{2\mathrm{e}^{2(Q+P)}}{a+b}}\left[{\displaystyle \frac{^2a}{z^2}}+2{\displaystyle \frac{P}{z}}{\displaystyle \frac{a}{z}}{\displaystyle \frac{2}{a+b}}\left({\displaystyle \frac{a}{z}}\right)^2\right],`$ (45)
$`\overline{\mathrm{\Psi }}_4`$ $`=`$ $`{\displaystyle \frac{2\mathrm{e}^{2(Q+P)}}{a+b}}\left[{\displaystyle \frac{^2b}{z^2}}+2{\displaystyle \frac{P}{z}}{\displaystyle \frac{b}{z}}{\displaystyle \frac{2}{a+b}}\left({\displaystyle \frac{b}{z}}\right)^2\right],`$ (46)
$`6\mathrm{\Psi }_2`$ $`=`$ $`{\displaystyle \frac{2\mathrm{e}^{2(Q+P)}}{(a+b)^2}}\left(2(a+b)^2{\displaystyle \frac{^2P}{z\overline{z}}}+5{\displaystyle \frac{a}{z}}{\displaystyle \frac{b}{\overline{z}}}{\displaystyle \frac{a}{\overline{z}}}{\displaystyle \frac{b}{z}}\right)`$ (47)
$`{\displaystyle \frac{4\mathrm{e}^{2Q+4P}}{a+b}}\mathrm{Re}\left({\displaystyle \frac{M}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{a}{\overline{z}}}\right)`$
$`\times \mathrm{Re}\left({\displaystyle \frac{N}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{b}{\overline{z}}}\right),`$
$`2\mathrm{\Psi }_1`$ $`=`$ $`{\displaystyle \frac{\mathrm{e}^{2Q+P}}{i\sqrt{a+b}}}{\displaystyle \frac{}{z}}\left[\mathrm{e}^{2P}\mathrm{Re}\left({\displaystyle \frac{M}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{a}{\overline{z}}}\right)\right]`$ (48)
$`{\displaystyle \frac{\mathrm{e}^{2Q+3P}}{i(a+b)^{3/2}}}{\displaystyle \frac{a}{z}}`$
$`\times \mathrm{Re}\left[{\displaystyle \frac{M}{\overline{z}}}3{\displaystyle \frac{N}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}\left({\displaystyle \frac{a}{\overline{z}}}3{\displaystyle \frac{b}{\overline{z}}}\right)\right],`$
$`2\overline{\mathrm{\Psi }}_3`$ $`=`$ $`{\displaystyle \frac{i\mathrm{e}^{2Q+P}}{\sqrt{a+b}}}{\displaystyle \frac{}{z}}\left[\mathrm{e}^{2P}\mathrm{Re}\left({\displaystyle \frac{N}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}{\displaystyle \frac{b}{\overline{z}}}\right)\right]`$ (49)
$`{\displaystyle \frac{i\mathrm{e}^{2Q+3P}}{(a+b)^{3/2}}}{\displaystyle \frac{b}{z}}`$
$`\times \mathrm{Re}\left[{\displaystyle \frac{N}{\overline{z}}}3{\displaystyle \frac{M}{\overline{z}}}{\displaystyle \frac{M+N}{a+b}}\left({\displaystyle \frac{b}{\overline{z}}}3{\displaystyle \frac{a}{\overline{z}}}\right)\right].`$
## Appendix B The case $`a=b`$
In section III it was shown that $`ab=F(x)`$ and the complete study of the case $`\mathrm{d}F/\mathrm{d}x0`$ was performed. Here, we concentrate in the special case $`F=\mathrm{const}.`$, such constant can be putted to zero by shifting the timelike coordinate and redefining function $`a`$ or function $`b`$. Hence, this case is equivalent to have $`a=b`$. Since now $`\mathrm{\Psi }_0=\overline{\mathrm{\Psi }}_4`$, equation (8) \[or its consequence (25)\] is pointless, which invalidates the coordinate transformation (27). This is the reason why this case must be studied separately. The vanishing of the $`\mathrm{\Psi }_0`$ component for $`a=b`$ implies
$$\mathrm{\Psi }_0=\mathrm{e}^{2Q}\frac{}{z}\left(\mathrm{e}^{2P}\frac{}{z}\mathrm{ln}a\right)=0\frac{a}{z}=\overline{g}(\overline{z})a\mathrm{e}^{2P}.$$
(50)
We can take $`g(z)=1`$ again, just applying the coordinate transformation (13) together with the relevant redefinitions of the functions $`P`$, $`M`$, and $`N`$. Applying the same arguments than in section III we conclude that $`a`$ and $`P`$ are functions of coordinate $`x`$ only. Choosing now the gauge elections $`\overline{M}=M`$ and $`\overline{N}=N`$ we end with metric (17) evaluated in $`b=a`$. Equations (18) and (20) and their solutions (21) and (22) are the same just considering that $`b=a`$.
Since $`a`$ and $`P`$ are functions of $`x`$ only, the first order equation (50) suggests to take $`a`$ as a new spatial coordinate by
$$(\tau ,\sigma ,x,y)(\tau ,\sigma ,a=\mathrm{exp}(\mathrm{e}^{2P}\mathrm{d}x),y).$$
(51)
In terms of this coordinate Eqs. (30) and (33) are written in this case as
$`\mathrm{\Psi }_3\mathrm{\Psi }_1`$ $`=`$ $`{\displaystyle \frac{ia^{5/2}\mathrm{e}^{2QP}}{4\sqrt{2}}}{\displaystyle \frac{\mathrm{d}}{\mathrm{d}a}}\left({\displaystyle \frac{1}{a}}{\displaystyle \frac{\mathrm{d}F_3}{\mathrm{d}a}}\right)=0,`$ (52)
$`\mathrm{\Psi }_3+\mathrm{\Psi }_1`$ $`=`$ $`{\displaystyle \frac{i\mathrm{e}^{2QP}}{4\sqrt{2}\sqrt{a}}}{\displaystyle \frac{\mathrm{d}}{\mathrm{d}a}}\left(a^3{\displaystyle \frac{\mathrm{d}F_1}{\mathrm{d}a}}\right)=0,`$ (53)
giving
$$F_1(a)=\frac{K_0}{a^2},F_3(a)=K_1a^2.$$
(54)
Using these expressions, Eq. (31) becomes
$$\mathrm{\Psi }_2=\frac{a^2\mathrm{e}^{2Q}}{12}\frac{\mathrm{d}}{\mathrm{d}a}\left[\frac{1}{a}\frac{\mathrm{d}}{\mathrm{d}a}\left(a\mathrm{e}^{2P}+\frac{K_{0}^{}{}_{}{}^{2}}{2a^2}\frac{K_{1}^{}{}_{}{}^{2}a^4}{2}\right)\right]=0,$$
(55)
and integrates as
$$\mathrm{e}^{2P(a)}=\frac{\alpha +\beta a^2}{a}\frac{K_{0}^{}{}_{}{}^{2}}{2a^3}+\frac{K_{1}^{}{}_{}{}^{2}a^3}{2}.$$
(56)
Finally, for a general stationary cyclic symmetric spacetime with $`a=b`$, the conformally flat metrics are given by
$`๐`$ $`=`$ $`e^{2Q(a,y)}[{\displaystyle \frac{๐
๐^2}{a}}+a๐
๐^2`$ (57)
$`+{\displaystyle \frac{๐
๐^2}{a(\alpha +\beta a^2)\frac{K_{0}^{}{}_{}{}^{2}}{a}+K_{1}^{}{}_{}{}^{2}a^5}}`$
$`+2({\displaystyle \frac{K_0}{a}}๐
๐K_1a๐
๐)๐
๐+{\displaystyle \frac{\alpha +\beta a^2}{a}}๐
๐^2].`$
where we have rescaled the Killing coordinates and some of the constants. For $`K_0=0=K_1`$ we recover the static metric of Ref. Garcia:2002gj . In general, the above metric describes a static spacetime for $`K_0=0`$. Additionally, it is incompatible with the existence of a rotation axis, hence, this class is not in contradiction with the Collinson theorem.
|
warning/0506/math0506352.html
|
ar5iv
|
text
|
# A model category for local po-spaces
## 1. Introduction
The motivation for this paper stems from the study of concurrent processes accessing shared resources. Such systems were originally described by discrete models based on graphs, possibly equipped with some additional information \[Mil80\]. The precision of these models suffers, however, from an inaccuracy in distinguishing between concurrent and non-deterministic executions. It turned out that a satisfactory way to organize this information can be based on cubical sets, giving rise to the notion of *Higher-Dimensional Automata* or HDAโs \[Gou96, Gou02\]. HDAโs live in slice categories of $`\mathrm{๐๐๐๐ญ}`$, the category of cubical sets and their morphisms.
A different view, which has its origins in Dijkstraโs notion of *progress graphs* \[Dij68\], takes the flow of time into account. The difficulty here is to adequately model the fact that time is irreversible as far as computation is concerned. On the other hand, one would like to identify execution paths corresponding to (at least) the same sequence of acquisitions of shared resources. However, in order not to lose precision, this notion of homotopy is also subject to the constraint above of the irreversibility of time. There are two distinct approaches, both based on topological spaces.
One approach, advocated by P. Gaucher, is to topologize the sets of paths between the states of an automaton, which technically amounts to an enrichment with no units \[Gau03\]. The intuition behind the setup is to distinguish between *spatial* and *temporal* deformations of computational paths. The related framework of *Flows* has clear technical advantages from a (model-)categorical point of view.
The other approach, advocated by Fajstrup, Goubault, Raussen and others, is to topologize partially ordered states of automata. Such objects are called partially-ordered spaces or *po-spaces* (also *pospaces*)<sup>1</sup><sup>1</sup>1M. Grandis uses a related approach \[Gra03\] in which the underlying topological space comes with a class of directed paths. However these spaces are not partially-ordered, even locally.. The advantage of using po-spaces is that there is a very simple and intuitive way to express directed homotopy or *dihomotopy* \[Gou03, FGR99\].
However, the price paid is that po-spaces cannot model executions of (concurrent) programs with loops. The solution is to order the underlying topological space only *locally*. Such objects are called *local po-spaces* and the notion of dihomotopy becomes more intricate in this context. Nevertheless, practical reasons like tractability call for a good notion of equivalence in the category of local po-spaces. Put differently, it would be useful to be able to replace a given local po-space model with a simpler local po-space which nevertheless preserves the relevant computer-scientific properties.
In this paper, we study these questions in the framework of Quillenโs (closed) model categories \[Qui67, Hov99, Hir03\]. Briefly, a model category is a category with all small limits and colimits and three distinguished classes of morphisms called *weak equivalences*, *cofibrations* and *fibrations*. Weak equivalences that are also cofibrations or fibrations are called *trivial cofibrations* and *trivial fibrations*, respectively. These morphisms satisfy four axioms that allow one to apply the machinery of homotopy theory to the category. This machinery allows a rigorous study of equivalences. We remark that there are other frameworks for studying equivalence. However model categories have the most developed theory, and have succeeded in illuminating many diverse subjects.
Our aim is to construct a model category of locally partial-ordered spaces as a foundation for the study of concurrent systems. This is technically difficult because locally partial-ordered spaces are not closed under taking colimits. We will define a category $`\mathrm{๐๐๐}`$ of local po-spaces, which embeds into the category $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ of simplicial presheaves on local po-spaces. The objects of $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ are contravariant functors from $`\mathrm{๐๐๐}`$ to the category of simplicial sets and the morphisms are the natural transformations. This embedding is given by a Yoneda embedding (see Definition 2.17),
$$\overline{y}:\mathrm{๐๐๐}\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐}).$$
We now briefly describe some technical conditions on model categories which strengthen our theorems. For more details see Definitions 8.2 and 8.4 and \[Hov99, Hir03\]. A model category is *proper* if the weak equivalences are closed under both pushouts with cofibrations and pullbacks with fibrations. It is *left proper* if the first condition holds. A model category is *cofibrantly generated* if the model category structure is induced by a set of generating cofibrations and a set of generating trivial cofibrations, both of which permit the small object argument. A *cellular* model category is a cofibrantly generated model category in which the cell complexes are well behaved. A *simplicial* model category $``$ is a model category enriched over simplicial sets, which for any $`X`$ and any simplicial set $`K`$ has objects $`XK`$ and $`X^K`$ which satisfy various compatibility conditions.
###### Theorem 1.1.
The category $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ has a proper, cellular, simplicial model structure in which
* the cofibrations are the monomorphisms,
* the weak equivalences are the *stalkwise equivalences*, and
* the fibrations are the morphisms which have the right lifting property with respect to all trivial cofibrations.
Furthermore among morphisms coming from $`\mathrm{๐๐๐}`$ (using the Yoneda embedding $`\mathrm{๐๐๐}\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$), the weak equivalences are precisely the isomorphisms.
The model structure on $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ is Jardineโs model structure \[Jar87, Jar96\] on the category of simplicial presheaves on a small *Grothendieck site*. We show that $`\mathrm{๐๐ก๐ฏ}(\mathrm{๐๐๐})`$ is a *Grothendieck topos* which has *enough points*. Under this condition, Jardine showed that the weak equivalences are the *stalkwise equivalences*.
This model category can be thought of as a localization of the universal injective model category of local po-spaces \[Joy84, Dug01, DHI04\]. While in general the weak equivalences are interesting and nontrivial \[Jar87\], this is not true for those coming from $`\mathrm{๐๐๐}`$. To obtain a more interesting category from the point of view of concurrency we would like to localize with respect to directed homotopy equivalences. In \[Bub04\] it is argued that the relevant equivalences are the directed homotopy equivalences relative to some *context*. The context is a local po-space $`A`$ and the directed homotopy equivalences rel $`A`$ are a set of morphisms in $`๐\mathrm{๐๐๐}`$.
We combine this approach with Theorem 1.1 as follows. First we remark that $`A`$ embeds in $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ as $`\overline{y}(A)`$. Next the model structure on $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ induces a model structure on the coslice category $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$. Finally one can take the left Bousfield localization of this model category with respect to the directed homotopy equivalences rel $`A`$.
###### Theorem 1.2.
Let $`=\{\overline{y}(f)|f\text{ is a directed homotopy equivalence rel }A\}`$. Then the category $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ has a left proper, cellular model structure in which
* the cofibrations are the monomorphisms,
* the weak equivalences are the $``$-local equivalences, and
* the fibrations are those morphisms which have the right lifting property with respect to monomorphisms which are $``$-local equivalences.
Recall that, given a topological space $`Z`$, *รฉtale bundles over $`Z`$* are maps $`WZ`$ which are local homeomorphisms. Let $`๐ช(Z)`$ be $`Z`$โs locale of open subsets and recall that sheaves over $`Z`$ are functors $`๐ช(Z)^{op}\mathrm{๐๐๐ญ}`$ that enjoy a good gluing property. There is a well-known correspondence between รฉtale bundles and sheaves. We establish a directed version of this correspondence, which may be of independent interest.
###### Theorem 1.3.
Let $`Z\mathrm{๐๐๐}`$. Let $`\mathrm{๐๐ญ๐๐ฅ๐}(Z)`$ be the category of di-รฉtale bundles over $`Z`$, i.e. the category of bundles which are local dihomeomorphisms. Let $`๐ช(Z)`$ be the category of open subobjects of $`Z`$. There is an equivalence of categories:
$$\mathrm{\Gamma }:\mathrm{๐๐ญ๐๐ฅ๐}(Z)\mathrm{๐๐ก๐ฏ}(๐ช(Z)):\mathrm{\Lambda }.$$
Acknowledgments. The authors would like to thank Eric Goubault, Emmanuel Haucourt, Kathryn Hess and Phil Hirschhorn for helpful discussions and suggestions.
## 2. Background
This section contains some known definitions and facts we build on. We start by stating the definition of a model category in subsection 2.1. Next we review the basics on presheaves in subsection 2.2 and on sheaves in subsection 2.3. We then recall the notions of topoi and geometric morphisms in subsection 2.4 and of stalks in subsection 2.5. Our main reference for this material is \[MLM92\]. Subsection 2.6 is devoted to some important model structures on $`\mathrm{๐ฌ๐๐๐ญ}^{^{op}}`$, the category of simplicial presheaves over a category $``$. The material is drawn from \[Jar87, Jar96, DHI04\].
### 2.1. Model categories
Recall that a morphism $`i:AB`$ has the left lifting property with respect to a morphism $`p:XY`$ if in every commutative diagram
there is a morphism $`h:BX`$ making the diagram commute. Also $`f`$ is a retract of $`g`$ if there is a commutative diagram:
###### Definition 2.1.
A *model category* is a category with all small limits and colimits that has three distinguished classes of morphisms: $`๐ฒ`$, called the *weak equivalences*; $`๐`$, called the *cofibrations*; and $``$, called the *fibrations*, which together satisfy the axioms below. We remark that morphisms in $`๐ฒ`$ $``$ $`๐`$, and $`๐ฒ`$ $``$ $``$, are called trivial cofibrations and trivial fibrations, respectively.
1. Given composable morphisms $`f`$ and $`g`$ if any of the two morphisms $`f`$, $`g`$, and $`gf`$ are in $`๐ฒ`$, then so is the third.
2. If $`f`$ is a retract of $`g`$ and $`g`$ is in $`๐ฒ`$, $`๐`$ or $``$, then so is $`f`$.
3. Cofibrations have the left-lifting property with respect to trivial fibrations, and trivial cofibrations have the left-lifting property with respect to fibrations.
4. Every morphism can be factored as a cofibration followed by a trivial fibration, and as a trivial cofibration followed by a fibration. These factorizations are functorial.
### 2.2. Presheaves
Recall that a presheaf $`P`$ on $``$ is just a functor $`P\mathrm{๐๐๐ญ}^{^{op}}`$. In particular, โhom-ingโ
$$\begin{array}{cccc}(\mathrm{\_},C):\hfill & ^{op}\hfill & & \mathrm{๐๐๐ญ}\hfill \\ & X\hfill & & (X,C)\hfill \end{array}$$
gives rise to a presheaf and further to the Yoneda embedding
$$\begin{array}{cccc}y:\hfill & \hfill & & \mathrm{๐๐๐ญ}^{^{op}}\hfill \\ & C\hfill & & (\mathrm{\_},C).\hfill \end{array}$$
This embedding is *dense*, i.e.
$$Pcolim(y\pi )$$
canonically for any presheaf $`P`$, where $`\pi :(yP)`$ is the projection from the comma-category $`yP`$. Recall that a presheaf in the image of the Yoneda-embedding (up to equivalence) is called *representable*.
### 2.3. Sheaves
###### Definition 2.2.
A *sieve* on $`M`$ is a subfunctor $`S(\mathrm{\_},M)`$. A *Grothendieck topology* $`J`$ on $``$ assigns to each $`M`$ a collection $`J(M)`$ of sieves on $`M`$ such that
1. (maximal sieve) $`(\mathrm{\_},M)J(M)`$ for all $`M`$;
2. (stability under pullback) if $`g:MN`$ and $`SJ(N)`$, then $`(g\mathrm{\_})^{}(S)J(M)`$ as given by
3. (transitivity) if $`SJ(M)`$ and $`R`$ is a sieve on $`M`$ such that $`(f\mathrm{\_})^{}(R)J(U)`$ for all $`f:UM`$ in the image of $`S`$, then $`RJ(M)`$;
We say that a sieve $`S`$ on $`M`$ is a *covering sieve* or a *cover of* $`M`$ whenever $`SJ(M)`$.
###### Remark 2.3.
Unwinding definition 2.2 pinpoints a sieve as a right ideal, i.e. a set of arrows $`S`$ with codomain $`M`$ such that $`fSfhS`$ whenever the codomain of $`h`$, $`\mathrm{cod}(h)=\mathrm{dom}(f)`$, the domain of $`f`$. From this point of view, pulling back a sieve $`S`$ on $`M`$ by an arrow $`N\stackrel{๐}{}M`$ amounts to building the set
$$f^{}(S)\stackrel{def}{=}\{h|\mathrm{cod}(h)=N,fhS\}.$$
It is then immediate how to rephrase a Grothendieck topology in terms of right ideals.
###### Definition 2.4.
Let $`J`$ be a Grothendieck topology on $``$. A presheaf $`P\mathrm{๐๐๐ญ}^{^{op}}`$ is a *sheaf* with respect to $`J`$ provided any natural transformation $`\theta :SP`$ uniquely extends through $`y(M)`$ as in
for all $`SJ(M)`$ and all $`M`$. $`J`$ is *subcanonical* if the representable presheaves are sheaves.
###### Remark 2.5.
Let $`\theta :SP`$ be a natural transformation from a sieve $`S`$ to a presheaf $`P`$. If one sees $`S`$ as a right ideal $`S=\{u_j:M_jM\}`$, then $`\theta `$ amounts to a function that assigns to every $`u_j:M_jMS`$ an element $`a_jP(M_j)`$ such that
$$P(v)(a_j)=a_k$$
for all $`v:M_kM_j`$ and for all $`u_k=u_jvS`$. Such a function is called a *matching family* for $`S`$ of elements of $`P`$. A matching family $`a_jP(M_j)`$ admits an *amalgamation* $`aP(M)`$ if
$$P(u_j)(a)=a_j$$
for all $`u_j:M_jMS`$. From this point of view, the Yoneda lemma characterizes a *sheaf* as a presheaf such that every matching family has a unique amalgamation for all $`SJ(M)`$ and all $`M`$.
A Grothendieck topology is a huge object. In practice, a generating device is used.
###### Definition 2.6.
A *basis* $`K`$ for a Grothendieck topology assigns to each object $`M`$ a collection $`K(M)`$ of families of morphisms with codomain $`M`$ such that
1. all isomorphisms $`f:UM`$ are contained in $`K(M)`$,
2. given a morphism $`g:NM`$ and $`\{f_i:U_iM\}K(M)`$, then the family of pullbacks $`\{\pi _2:U_i\times _MNN\}K(N)`$, and
3. given $`\{f_i:U_iM\}K(M)`$ and for each $`i`$, $`\{h_{ij}:A_{ij}U_i\}K(U_j)`$, then the family of composites $`\{f_ih_{ij}:A_{ij}M\}K(M)`$.
###### Remark 2.7.
Given a basis $`K`$ for a Grothendieck topology one generates the corresponding Grothendieck topology $`J`$ by defining
$$VJ(M)\text{there is }UK(M)\text{ such that }UV.$$
As expected, the sheaf condition can be rephrased in terms of a basis.
As an example, consider the case $`=๐ช(X)`$ with $`X`$ a topological space and $`๐ช(X)`$ its locale of opens. The basis of the *open-cover* (Grothendieck) topology is, as expected, given by open coverings of the opens.
###### Theorem 2.8.
Let $`\mathrm{๐๐ก๐ฏ}(,J)`$ be the full subcategory of $`\mathrm{๐๐๐ญ}^{^{op}}`$ whose objects are sheaves for $`J`$. The inclusion functor $`i:\mathrm{๐๐ก๐ฏ}(,J)\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ has a left adjoint $`a`$ called the *associated sheaf functor* or *sheafification*. This left adjoint preserves finite limits.
Theorem 2.8 is listed as Theorem III.5.1 in \[MLM92\]. There are several equivalent ways to construct the associated sheaf functor, the most classical one being the โplus-constructionโ applied twice.
###### Remark 2.9.
A cover on $`M`$ amounts to a cocone in $``$ with vertex $`M`$. The associated sheaf functor maps these cocones onto colimiting ones. Moreover, it is universal with respect to this property.
### 2.4. Topoi
###### Definition 2.10.
A category $``$ *has exponentials* provided that for all $`X`$, the functor $`\mathrm{\_}\times X:`$ has a right adjoint denoted $`(\mathrm{\_})^X`$, so that
$$(Y\times X,Z)(Y,Z^X).$$
Suppose now $``$ has a terminal object $`1`$, and has finite limits. A *subobject classifier* is a monomorphism $`\text{true}:1\mathrm{\Omega }`$ such that for every monomorphism $`s:SX`$, there is a unique morphism $`\varphi _S`$ such that pullback of true along $`\varphi _S`$ yields $`s`$:
The category $``$ is a *topos* if it has exponentials and a subobject classifier.
A subobject classifier is obviously unique (up to isomorphism). Furthermore, a topos has all finite colimits, though this is not easy to prove. It would take pages to enumerate all the remarkable features of a topos, see \[Joh77\] for an introduction to the lore of the material. Let us just say that topoi as introduced by Grothendieck and his collaborators had a very strong algebro-geometrical flavor \[AGV72\], yet the rich structure is relevant not only for for algebraic geometers but for logicians as well \[Law63, Law64, Law73\].
###### Definition 2.11.
A site $`(,J)`$ is a small category $``$ equipped with a Grothendieck topology $`J`$. A *Grothendieck topos* is a category equivalent to the category $`\mathrm{๐๐ก๐ฏ}(,J)`$ of sheaves on $`(,J)`$.
The following are well known.
###### Proposition 2.12.
1. A Grothendieck topos is a topos;
2. $`\mathrm{๐๐๐ญ}`$ is a topos;
3. $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ is a topos for any $``$.
###### Definition 2.13.
Let $``$ and $``$ be topoi. A *geometric morphism* $`g:`$ is a pair of adjoint functors
such that the left adjoint $`g^{}`$ is left-exact (that is, it preserves finite limits). The right adjoint is called *direct image* and the left one *inverse image*.
As an example, $`i:\mathrm{๐๐ก๐ฏ}(,J)\mathrm{๐๐๐ญ}^{^{op}}`$ is the direct image part of a geometric morphism. Notice that the convention for a geometric morphism is to have the direction of its direct image part.
###### Definition 2.14.
A (geometric) *point* in a topos $``$ is a geometric morphism
$$p:\mathrm{๐๐๐ญ}$$
(we write $`p`$ by abuse of notation). A topos $``$ *has enough points* if given $`fg:PQ`$ there is a point $`p`$ such that $`p^{}fp^{}g\mathrm{๐๐๐ญ}`$.
### 2.5. Stalks and germs
###### Definition 2.15.
Let $`(,J)`$ be a site, $`a:\mathrm{๐๐๐ญ}^{^{op}}\mathrm{๐๐ก๐ฏ}(,J)`$ the associated sheaf functor and $`x\mathrm{๐๐ก๐ฏ}(,J)`$ a point. The *stalk functor* at $`x`$ is given by
$$stalk_x\stackrel{def}{=}x^{}a:\mathrm{๐๐๐ญ}^{^{op}}\mathrm{๐๐๐ญ}.$$
Given a presheaf $`F`$, we say that $`stalk_x(F)`$ is the stalk of $`F`$ at $`x`$. As an example, consider again the case $`=๐ช(X)`$ with $`X`$ a (this time) Hausdorff topological space and $`๐ช(X)`$ its locale of opens equipped with the open-cover topology. Let $`\mathrm{๐๐ก๐ฏ}(X)`$ be the corresponding topos of sheaves. It can be shown that any geometric point $`x:\mathrm{๐๐๐ญ}\mathrm{๐๐ก๐ฏ}(X)`$ corresponds to a โphysicalโ point $`x^{}X`$. The stalk of $`F\mathrm{๐๐๐ญ}^{๐ช(X)^{\text{op}}}`$ at $`x`$ is then given by
$$stalk_x(F):=\underset{U๐ช(X),x^{}U}{colim}F(U).$$
Write $`\mathrm{germ}_{x,U}:F(U)stalk_x(F)`$ for the canonical map at $`U`$ ($`\mathrm{germ}_x`$ when $`U`$ is clear from the context). We call the equivalence class $`\mathrm{germ}_{x,U}(s)`$ of $`s`$ in $`stalk_x(F)`$ the *germ of $`s`$ at $`x`$*. Obviously,
$$stalk_x(F)=\{\mathrm{germ}_{x,U}(s)|U๐ช(X),x^{}U,sF(U)\}.$$
### 2.6. Simplicial Presheaves
For the rest of this section, let $``$ be a small category with a Grothendieck topology $`J`$ such that $`\mathrm{๐๐ก๐ฏ}(,J)`$ has *enough points*.
Let $`\mathrm{\Delta }`$ be the simplicial category which has objects $`[n]=\{0,1,\mathrm{},n\}`$ for $`n0`$, and whose morphisms are the maps such that $`xy`$ implies that $`f(x)f(y)`$. Then $`\mathrm{๐ฌ๐๐๐ญ}`$ is the category $`\mathrm{๐๐๐ญ}^{\mathrm{\Delta }^{\mathrm{op}}}`$. This category has a well-known model structure (see \[Hov99\] for example) where $`๐ฒ_{\mathrm{๐ฌ๐๐๐ญ}}`$ are the morphisms whose geometric realization is a weak homotopy equivalence and $`๐_{\mathrm{๐ฌ๐๐๐ญ}}`$ are the monomorphisms.
Objects of $`\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$ are called *simplicial presheaves* on $``$ since
$$\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}=\left(\mathrm{๐๐๐ญ}^{\mathrm{\Delta }^{op}}\right)^{^{op}}\mathrm{๐๐๐ญ}^{\mathrm{\Delta }^{op}\times ^{op}}\left(\mathrm{๐๐๐ญ}^{^{op}}\right)^{\mathrm{\Delta }^{op}}.$$
There is an embedding
$$\begin{array}{cccc}\hfill \kappa :& \mathrm{๐๐๐ญ}^{^{op}}\hfill & & \mathrm{๐ฌ๐๐๐ญ}^{^{op}}\hfill \\ & F\hfill & & \kappa _F\hfill \end{array}$$
where $`\kappa _F`$ is constant *levelwise* i.e. $`\left(\kappa _F\right)\left(C\right)_n\stackrel{def}{=}F\left(C\right)`$ for all $`n`$, and all the face and degeneracy maps are the identity. There is a further embedding
$$\begin{array}{cccc}\hfill \gamma :& \mathrm{๐ฌ๐๐๐ญ}\hfill & & \mathrm{๐ฌ๐๐๐ญ}^{^{op}}\hfill \\ & K\hfill & & \gamma _K\hfill \end{array}$$
where $`\gamma _K`$ is constant *objectwise* i.e. $`\gamma _K\left(C\right)\stackrel{def}{=}K`$ for all $`C`$.
Recall that for $`C`$ and $`F\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$, the Yoneda lemma gives the isomorphism $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}(y(C),F)F(C)`$, where $`y`$ is the Yoneda embedding (see Section 2.2). In the simplicial case we have the following variation, which can be proved using the same idea used in the proof of the Yoneda lemma.
###### Proposition 2.16.
(Bi-Yoneda) Let $`C`$ and $`F\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$. There is an isomorphism
$$\mathrm{๐ฌ๐๐๐ญ}^{^{op}}(\kappa _{y\left(C\right)}\times \gamma _{\mathrm{\Delta }\left[n\right]},F)F\left(C\right)_n$$
natural in all variables.
###### Definition 2.17.
Using the Yoneda embedding $`y:\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ for presheaves one can define an embedding
$$\overline{y}:\stackrel{๐ฆ}{}\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}\stackrel{๐
}{}\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}$$
for simplicial presheaves. The functor $`\overline{y}`$ is also called a Yoneda embedding.
There are two Quillen equivalent model structures on $`\mathrm{๐ฌ๐๐๐ญ}^{^{op}}`$ which are in a certain sense *objectwise:*
* the *projective* model structure $`\mathrm{๐ฌ๐๐๐ญ}_{prj}^{^{op}}`$ where $`๐ฒ_{prj}`$ and $`_{prj}`$ are objectwise (that is, $`f:PQ๐ฒ_{prj}(_{prj})`$ if and only if for all $`C`$, $`f(C):P(C)Q(C)๐ฒ_{\mathrm{๐ฌ๐๐๐ญ}}(_{\mathrm{๐ฌ๐๐๐ญ}})`$ ), and
* the *injective* model structure $`\mathrm{๐ฌ๐๐๐ญ}_{inj}^{^{op}}`$ where $`๐ฒ_{inj}`$ and $`๐_{inj}`$ are objectwise.
These were studied by Bousfield and Kan \[BK72\] and Joyal \[Joy84\], respectively.
###### Proposition 2.18.
Both $`\mathrm{๐ฌ๐๐๐ญ}_{prj}^{^{op}}`$ and $`\mathrm{๐ฌ๐๐๐ญ}_{inj}^{^{op}}`$ are proper, simplicial, cellular model categories. All objects are cofibrant in the latter. The identity functor is a left Quillen equivalence from the projective model structure to the injective model structure.
The injective one is more handy when it comes down to calculating homotopical localizations, yet the fibrant objects are easier to grasp in the projective one<sup>2</sup><sup>2</sup>2They are objectwise Kan..
Using the stalk functor for presheaves, one can define a simplicial stalk functor for simplicial presheaves.
###### Definition 2.19.
The *simplicial stalk functor* at a point $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$ is given by
$$\begin{array}{cccc}\hfill (\mathrm{\_})_p:& \mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}\hfill & & \mathrm{๐ฌ๐๐๐ญ}\hfill \\ & P\hfill & & \{stalk_p(P_n)\}_{n0}.\hfill \end{array}$$
A morphism $`f:PQ\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$ is a *stalkwise equivalence* if $`f_p:P_pQ_p\mathrm{๐ฌ๐๐๐ญ}`$ is a weak equivalence for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$.
Jardine \[Jar87\] proved the existence of a local version of Joyalโs injective model structure. Since we will only be interested in the special case where $`\mathrm{๐๐ก๐ฏ}()`$ has enough points, we will not recall the definition of local weak equivalences.
###### Theorem 2.20 (\[Jar87, Jar96\]).
Let $``$ be a small category with a Grothendieck topology. Then $`\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$the category of simplicial presheaves on $``$ has a proper, simplicial, cellular model structure in which
* the cofibrations are the monomorphisms, i.e. the levelwise monomorphisms of presheaves,
* the weak equivalences are the *local weak equivalences*, and
* the fibrations are the morphisms which have the right lifting property with respect to all trivial cofibrations.
Furthermore, if the Grothendieck topos $`\mathrm{๐๐ก๐ฏ}()`$ has enough points, then the local weak equivalences are the stalkwise equivalences.
Jardineโs model structure can be seen to be cellular since it can also be constructed as a left Bousfield localization of the injective model structure \[DHI04\].
## 3. Local po-spaces
The focus of this section is to provide the reader with the main definitions and constructions. We define a small category of local po-spaces $`\mathrm{๐๐๐}`$ and state some of the properties, most of which are proved in the later sections. We show that Theorem 1.1 follows from these properties and a theorem of Jardine.
To simplify the analysis, we will only work with topological spaces which are subspaces of $`^n`$ for some $`n`$, since this provides more than enough generality for studying concurrent systems. The main technical advantage of this setting is that we obtain small categories.
###### Definition 3.1.
1. Let $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ be the category whose objects are subspaces of $`^n`$ for some $`n`$, and whose morphisms are continuous maps.
2. Let $`\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$ be the category whose objects are *po-spaces*: that is $`U\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ together with a *partial order* (a reflexive, transitive, anti-symmetric relation) $``$ such that $``$ is a closed subset of $`U\times U`$ in the product topology.
3. For any $`M\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ define an *order-atlas* on $`M`$ to be an open cover<sup>3</sup><sup>3</sup>3That is, for all $`i`$, $`U_i`$ is open as a subspace of $`M`$ and $`M=_iU_i`$. $`U=\{U_i\}`$ of $`M`$ indexed by a set $`I`$, where $`U_i\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$. These partial orders are compatible: $`_i`$ agrees with $`_j`$ on $`U_iU_j`$ for all $`i,jI`$. We will usually omit the index set from the notation.
4. Let $`U`$ and $`U^{}`$ be two order atlases on $`M`$. Say that $`U^{}`$ is a *refinement* of $`U`$ if for all $`U_iU`$, and for all $`xU_i`$, there exists a $`U_j^{}U^{}`$ such that $`xU_j^{}U_i`$ and for all $`a,bU_j^{}`$, $`a_j^{}b`$ if and only if $`a_ib`$.
5. Say that two order atlases are *equivalent* if they have a common refinement. This is an equivalence relation: reflexivity and symmetry follow from the definition. For transitivity, if $`U`$ and $`U^{}`$ have a refinement $`V=\{V_i\}`$ and $`U^{}`$ and $`U^{\prime \prime }`$ have a refinement $`W=\{W_j\}`$, let $`T=\{V_iW_j\}`$. One can check that $`T`$ is an order atlas of $`M`$ and that is a refinement of $`U^{}`$ and $`U^{\prime \prime }`$.
Any po-space $`(U,)`$ is a local po-space with the equivalence class of order atlases generated by the order atlas $`\{U\}`$. As a further example, we remark that any discrete space has a unique equivalence class of order-atlases.
###### Definition 3.2.
Let $`\mathrm{๐๐๐}`$ be the category of local po-spaces described as follows. The objects, called *local po-spaces*, are all pairs $`(M,๐ฐ)`$ where $`M`$ is an object in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and $`๐ฐ`$ is an equivalence class of *order-atlases* of $`M`$. The morphisms, called *dimaps* are described as follows. $`f\mathrm{๐๐๐}((M,๐ฐ),(N,๐ฑ))`$ if and only if $`f\mathrm{๐๐ฉ๐๐๐๐ฌ}(M,N)`$ and for all $`V=\{V_j\}_{jJ}๐ฑ`$ there is a $`U=\{U_i\}_{iI}๐ฐ`$ such that for all $`iI`$, $`jJ`$, for all $`x,yU_if^1(V_j)`$,
$$x_{U_i}yf(x)_{V_j}f(y).$$
###### Remark 3.3.
This condition is not necessarily true for arbitrary $`U๐ฐ`$. For example, take $`M=\{1,1\}`$ with $`๐ฐ`$ the unique equivalence class of order atlases generated by the order atlas $`U=\{\{1\},\{1\}\}`$. Let $`f=\mathrm{Id}_M:(M,๐ฐ)(M,๐ฐ)`$. Now let $`M_+`$ be the po-space on $`M`$ with the ordering $`11`$ and let $`M_{}`$ be the po-space on $`M`$ with the ordering $`11`$. Then $`\{M_+\}๐ฐ`$ and $`\{M_{}\}๐ฐ`$ (both have $`U`$ as a common refinement). However, even though $`1,1M_+f^1(M_{})`$,
$$1_{M_+}1\text{ but }f(1)_M_{}f(1).$$
###### Remark 3.4.
It is easy to check that a dimap of po-spaces is also a dimap of local po-spaces. Thus $`\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$ the category of po-spaces is a subcategory of $`\mathrm{๐๐๐}`$.
###### Remark 3.5.
Subobjects in $`\mathrm{๐๐๐}`$ .
If $`(M,๐ฐ)\mathrm{๐๐๐}`$, then a subspace $`LM\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ inherits local po-space structure as follows. Let $`U=\{U_i\}๐ฐ`$ and let $`W=\{W_i\}`$ where $`W_i=LU_i`$ and $`W_i`$ has the partial order inherited from $`U_i`$. Then $`W`$ is an open cover of $`L`$ and the partial orders are compatible. That is $`W`$ is an order atlas. Let $`๐ฒ`$ be the equivalence class of $`W`$.
We claim that $`๐ฒ`$ does not depend on the choice of $`U`$. Let $`\stackrel{~}{U}=\{\stackrel{~}{U}_i\}๐ฐ`$, let $`\stackrel{~}{W}_i=L\stackrel{~}{U}_i`$, and let $`\stackrel{~}{W}=\{\stackrel{~}{W}_i\}`$. $`U`$ and $`\stackrel{~}{U}`$ have a common refinement $`\widehat{U}=\{\widehat{U}_i\}`$. Let $`\widehat{W}_i=L\widehat{U}_i`$ and let $`\widehat{W}=\{\widehat{W}_i\}`$. Then one can check that $`\widehat{W}`$ is a common refinement of $`W`$ and $`\stackrel{~}{W}`$. So the equivalence class of $`\stackrel{~}{W}`$ is also $`๐ฒ`$.
Next we claim that there is a dimap $`\iota :(L,๐ฒ)(M,๐ฐ)`$ given by the inclusion $`\iota :LM`$. Let $`U=\{U_k\}๐ฐ`$, let $`W_k=LU_k`$, and let $`W=\{W_k\}`$. Then $`W๐ฒ`$. Let $`x,yW_j\iota ^1(U_k)=W_jLU_k=W_jW_k.`$ Note that $`\iota (x)=x`$ and $`\iota (y)=y`$. Then
$$x_{W_j}yx_{W_k}yx_{U_k}y.$$
Therefore when $`LM\mathrm{๐๐ฉ๐๐๐๐ฌ}`$, then there is an induced inclusion $`(L,๐ฒ)(M,๐ฐ)\mathrm{๐๐๐}`$.
The remark above will be used implicitly and without reference in Section 6.
###### Definition 3.6.
A collection of dimaps $`\{\varphi _j:(M_j,๐ฐ^j)(M,๐ฐ)\}`$ $`\mathrm{๐๐๐}`$ is an *open dicover* if
1. $`\{\varphi _j:M_jM\}`$ is an open cover, and
2. for each $`j`$, $`๐ฐ^j`$ is the local po-space structure inherited from $`(M,๐ฐ)`$.
###### Remark 3.7.
The local po-space structures inherited by the subspaces of $`(M,๐ฐ)`$ are compatible. So if $`\{\varphi _j:(M_j,๐ฐ^j)(M,๐ฐ)\}`$ is a open cover, then for each $`j`$, there is a $`U^j=\{U_k^j\}๐ฐ^j`$ such that $`U^{}=\{U_k^j\}_{j,k}`$ is an order atlas for $`M`$ and $`U^{}๐ฐ`$.
The following is easy to check.
###### Lemma 3.8.
$`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and $`\mathrm{๐๐๐}`$ are small categories.
Define $`U:\mathrm{๐๐๐}\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ to be the forgetful functor defined on objects and morphisms as follows $`(M,U)M`$ and $`\phi \phi `$.
Define $`F:\mathrm{๐๐ฉ๐๐๐๐ฌ}\mathrm{๐๐๐}`$ as follows. If $`M`$ is an object in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ , then let $`F(M)=(M,\overline{M}_\varphi )`$, where $`\overline{M}_\varphi `$ is the equivalence class of $`M_\varphi =\{M\}`$ with $`x_Myx=y`$. If $`f:MN\mathrm{๐๐ฉ๐๐๐๐ฌ}`$, then $`F(f)=f:(M,\overline{M}_\varphi )(N,\overline{N}_\varphi )`$. This is a dimap since for any $`V=\{V_j\}\overline{N}_\varphi `$ with $`x,yf^1V_j`$, $`x_Myx=yf(x)=f(y)f(x)_{V_j}f(y)`$.
###### Remark 3.9.
Note that $`U`$ is faithful and $`F`$ includes $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ as a full subcategory of $`\mathrm{๐๐๐}`$ .
###### Proposition 3.10.
$`F:\mathrm{๐๐ฉ๐๐๐๐ฌ}\mathrm{๐๐๐}:U`$ is an adjunction.
###### Proof.
Let $`M`$ be an object in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and $`(N,\overline{V})\mathrm{๐๐๐}`$. We claim that there is a natural bijection
$$\mathrm{๐๐๐}(F(M),(N,\overline{V}))\mathrm{๐๐ฉ๐๐๐๐ฌ}(M,U(N,\overline{V})).$$
We need to show that there is a natural bijection
$$\theta :\mathrm{๐๐ฉ๐๐๐๐ฌ}(M,N)\stackrel{}{}\mathrm{๐๐๐}((M,\overline{M}_\varphi ),(N,\overline{V})).$$
If $`f\mathrm{๐๐๐}((M,\overline{M}_\varphi ),(N,\overline{V}))`$, then $`f\mathrm{๐๐ฉ๐๐๐๐ฌ}(M,N)`$ such that for any $`V=\{V_j\}\overline{V}`$, for all $`j`$, $`f|_{f^1(V_j)}`$ satisfies $`x_Myf(x)_{V_j}f(y)`$. Since $`x_My`$ if and only if $`x=y`$ this last condition is vacuous. Thus the bijection is simply $`\theta :ff`$.
To show naturality let $`\alpha :(N,\overline{V})(N^{},\overline{V^{}})\mathrm{๐๐๐}`$ and $`\xi :M^{}M\mathrm{๐๐ฉ๐๐๐๐ฌ}`$. Then
$$\theta (U(\alpha )f\xi )=\alpha f\xi =\alpha \theta (f)\xi .$$
###### Remark 3.11.
$`\mathrm{๐๐๐}`$ does not have colimits.
Consider the product of the directed circle and an interval. Now collapse the top circle of this cylinder. The vertex of the resulting cone does not have a local partial order.
## 4. The open-dicover topology
We define the open cover Grothendieck topology for $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and the open dicover Grothendieck topology for $`\mathrm{๐๐๐}`$ in the following lemma. The proof of the lemma follows directly from the definition of a basis for a Grothendieck topology.
###### Lemma 4.1.
1. $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ has a *Grothendieck topology* whose basis is given by the open covers. For $`M\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ let $`K(M)=\{\text{open covers of }M\}`$. Let $`J`$ be the Grothendieck topology generated by $`K`$. Call $`J`$ the *open cover topology*.
2. Analogously, $`\mathrm{๐๐๐}`$ has a Grothendieck topology whose basis is given by the open dicovers in $`\mathrm{๐๐๐}`$. Let $`K((M,๐ฐ))=\{\text{open dicovers of }(M,๐ฐ)\}`$. Call the Grothendieck topology generated by $`K`$ the *open-dicover topology*.
In Section 3, we defined a Grothendieck topology to be subcanonical if every representable presheaf a sheaf. In this section, we will prove that the open-dicover topology is subcanonical.
The following proposition shows that if a Grothendieck topology is generated by a basis $`K`$, then to see if a presheaf is a sheaf it suffices to check the basis. For the definition of matching families and amalgamations see Remark 2.5.
###### Proposition 4.2 (\[MLM92, Proposition III.4.1\]).
Let $``$ be a small category with a Grothendieck topology $`J`$ generated by a basis $`K`$. Then a presheaf $`P\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ is a sheaf for $`J`$ if and only if for every $`M`$ and every cover $`\{\varphi _j:M_jM\}K(M)`$, every matching family for $`\{\varphi _j\}`$ of elements of $`P`$ has a unique amalgamation.
###### Example 4.3.
Let $`N\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and $`y(N)=\mathrm{๐๐ฉ๐๐๐๐ฌ}(,N)\mathrm{๐๐๐ญ}^{\mathrm{๐๐ฉ๐๐๐๐ฌ}^{\mathrm{op}}}`$. Let $`\varphi _j:M_jM`$ be an open cover, and let $`\alpha _j:M_jN`$ be a matching family. Then $`\varphi _j`$ has a unique amalgamation $`\varphi :MN`$. Therefore $`y(N)`$ is a sheaf for the open cover topology, and hence the open cover topology is subcanonical.
###### Proposition 4.4.
In the open-dicover topology $`J`$ for local po-spaces every representable presheaf is a sheaf. That is $`J`$ is subcanonical.
###### Proof.
Consider the representable presheaf
$$y((N,\overline{V}))=\mathrm{๐๐๐}(,(N,\overline{V}))\mathrm{๐๐๐ญ}^{\mathrm{๐๐๐}^{\mathrm{op}}}.$$
By Proposition 4.2, $`y((N,\overline{V}))`$ is a sheaf if and only if for all open dicovers $`\{\varphi _j\}K((M,\overline{U}))`$, any matching family
$$\{\alpha _j:(M_j,\overline{U}_j)(N,\overline{V})\}$$
has a unique amalgamation $`\alpha :(M,\overline{U})(N,\overline{V})`$. That is, there is a map $`\alpha `$ such that the diagrams
commute in $`\mathrm{๐๐๐}`$ for all $`j`$.
Let $`\{\alpha _j\}`$ be such a matching family for an open dicover $`\{\varphi _j\}`$. Since $`\{\varphi _j\}`$ is an open dicover, then by Remark 3.7 for each $`j`$ there is a $`U^j=\{U_k^j\}\overline{U}_j`$ such that $`U^{}=\{U_k^j\}_{j,k}`$ is an order atlas and $`U^{}\overline{U}`$.
By definition $`\{\varphi _j:M_jM\}`$ is a cover in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ and $`\{\alpha _j:M_jN\}`$ is a matching family. Therefore there is a unique amalgamation $`\alpha :MN\mathrm{๐๐ฉ๐๐๐๐ฌ}`$. That is, there is a map $`\alpha `$ such that
commutes in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ for all $`j`$. It remains to show that $`\alpha `$ is a dimap. Let $`V=\{V_l\}\overline{V}`$. Since $`\alpha _j:(M_j,\overline{U}_j)(N,\overline{V})\mathrm{๐๐๐}`$, there is a $`\stackrel{~}{U}^j=\{\stackrel{~}{U}_k^j\}_k\overline{U}^j`$ such that for all $`k,l`$,
$$\text{for all }x,y\stackrel{~}{U}_k^j\alpha _j^1(V_l),x_{\stackrel{~}{U}_k^j}y\alpha _j(x)_{V_l}\alpha _j(y).$$
Now for each $`j`$, let $`\widehat{U}^j=\{\widehat{U}_k^j\}_k\overline{U}_j`$ be a common refinement of $`\stackrel{~}{U}^j`$ and $`U^j`$. Then since $`\widehat{U}^j`$ is a refinement of $`\stackrel{~}{U}^j`$,
(1)
$$\text{for all }x,y\widehat{U}_k^j\alpha _j^1(V_l),x_{\widehat{U}_k^j}y\alpha _j(x)_{V_l}\alpha _j(y),$$
and since $`\widehat{U}^j`$ is a refinement of $`U^j`$, if we define $`U=\{\widehat{U}_k^j\}_{j,k}`$, then $`U\overline{U}`$.
Since $`\alpha `$ is an amalgamation of $`\{\alpha _j\}`$ in $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ if $`x\widehat{U}_k^jM`$, then $`\alpha (x)=\alpha _j(x)`$ and for all $`l`$, $`\widehat{U}_k^j\alpha _j^1(V_l)=\widehat{U}_k^j\alpha ^1(V_l)`$. Therefore using (1) for all $`k,l`$,
$$\text{for all }x,y\widehat{U}_k^j\alpha ^1(V_l),x_{\widehat{U}_k^j}y\alpha (x)_{V_l}\alpha (y).$$
That is $`\alpha `$ is a dimap. Therefore $`\alpha :(M,\overline{U})(N,\overline{V})`$ is a unique amalgamation of $`\{\alpha _j\}`$. โ
## 5. Equivalence of sheaves and di-รฉtale bundles
In this section $``$ is either $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ or $`\mathrm{๐๐๐}`$ with the Grothendieck topology generated by open (di)covers.
###### Notation 5.1.
We will use $`A\stackrel{}{\mathrm{open}}B`$ to denote that $`A`$ is an open subset of $`B`$.
###### Notation 5.2.
Let $`Z`$ and let $`F\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$. Choose $`xU\stackrel{}{\mathrm{open}}Z`$ and $`sF(U)`$. Then for open subobjects of $`U`$, $`L\stackrel{i}{}U`$, we have $`F(i):F(U)F(L)`$ and we will use the notation
$$s|_L:=F(i)(s).$$
Recall that $`stalk_x(F)=colim_{xL\stackrel{}{\mathrm{open}}U}F(L)`$ and $`\mathrm{germ}_x(s)`$ is the equivalence class represented by $`s`$ in $`stalk_x(F)`$.
###### Definition 5.3.
Given $`Z`$, a *bundle* over $`Z`$ is just a morphism $`p:WZ`$. An *(di)รฉtale bundle* is a bundle which is a *local (di)homeomorphism*. That is, given $`yW`$ there is some open set $`VW`$ such that $`p(V)`$ is open in $`Z`$ and $`p|_V`$ is an isomorphism in $``$.
A morphism of (รฉtale) bundles $`p:WZ`$ and $`p:W^{}Z`$ is a morphism $`\theta :WW^{}`$ such that the following diagram commutes:
Let $`\mathrm{๐๐ญ๐๐ฅ๐}(Z)`$ denote the category of (di)รฉtale bundles over $`Z`$. In addition let $`๐ช(Z)`$ denote the category of open subobjects of $`Z`$, where the objects are open subobjects of $`Z`$ and the morphisms are the inclusions.
###### Theorem 5.4 (Theorem 1.3).
Let $`Z`$. Then there is an equivalence of categories
$$\mathrm{\Gamma }:\mathrm{๐๐ญ๐๐ฅ๐}(Z)\mathrm{๐๐ก๐ฏ}(๐ช(Z)):\mathrm{\Lambda }.$$
###### Proof.
It is well known that the statement of Theorem 1.3 is true when $`=\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ (see for example \[MLM92, Corollary II.6.3\]). We will show that this equivalence between รฉtale bundles on topological spaces and sheaves on topological spaces extends to local po-spaces.
First we describe the functors $`\mathrm{\Gamma }`$ and $`\mathrm{\Lambda }`$ in the case where $`=\mathrm{๐๐ฉ๐๐๐๐ฌ}`$. The functor $`\mathrm{\Gamma }`$ assigns to each bundle $`W\stackrel{๐}{}Z`$ the presheaf of cross-sections:
$`P:๐ช(Z)^{\mathrm{op}}`$ $``$ $`\mathrm{๐๐๐ญ}`$
$`U`$ $``$ $`\{s:UW|ps=\mathrm{Id}_U\}`$
$`U\stackrel{\theta }{}V`$ $``$ $`\theta ^{}(\theta ^{}(t)=t\theta ).`$
One can check that if $`p`$ is รฉtale, then $`P`$ is in fact a sheaf \[MLM92, p.79\]. Thus $`\mathrm{\Gamma }`$ restricts to a functor $`\mathrm{\Gamma }:\mathrm{๐๐ญ๐๐ฅ๐}(Z)\mathrm{๐๐ก๐ฏ}(๐ช(Z))`$.
Given a presheaf $`P:๐ช(Z)^{\mathrm{op}}\mathrm{๐๐๐ญ}`$, $`\mathrm{\Lambda }(P)`$ is the bundle $`W\stackrel{๐}{}Z`$ where
$$W=\{\mathrm{germ}_xs|xU\stackrel{}{\mathrm{open}}Z,sP(U)\}\text{ and }p:\mathrm{germ}_xsx.$$
A basis for the topology on $`W`$ is given by the sets $`\dot{s}(U)`$, where $`U`$ is an open set in $`Z`$, $`sP(U)`$ and
$`\dot{s}:U`$ $``$ $`\mathrm{\Lambda }(P)`$
$`x`$ $``$ $`\mathrm{germ}_xs.`$
Using this topology, $`p:WZ`$ is a continuous map. Again, one can check that if $`P`$ is a sheaf, then $`W\stackrel{๐}{}Z`$ is in fact an รฉtale bundle \[MLM92, p.85\]. So $`\mathrm{\Lambda }`$ restricts to a functor $`\mathrm{\Lambda }:\mathrm{๐๐ก๐ฏ}(๐ช(Z))\mathrm{๐๐ญ๐๐ฅ๐}(Z)`$.
Now we will show that $`\mathrm{\Gamma }`$ and $`\mathrm{\Lambda }`$ can be similarly defined in the case where $`=\mathrm{๐๐๐}`$. Let $`p:(W,\overline{T})(Z,\overline{U})`$ be an รฉtale bundle of local po-spaces. The definition of $`\mathrm{\Gamma }`$ is exactly the same: $`\mathrm{\Gamma }((W,\overline{T})\stackrel{๐}{}(Z,\overline{U}))`$ is the sheaf of cross-sections.
Given a sheaf $`P`$ on a local po-space $`(Z,\overline{U})`$, $`\mathrm{\Lambda }(P)=(W\stackrel{๐}{}Z)`$ is an รฉtale bundle of topological spaces. To extend $`\mathrm{\Lambda }`$ to local po-spaces it remains to define a local order on $`W`$ and show that this makes $`p`$ a dimap.
###### Lemma 5.5.
$`W`$ has a canonical local po-space structure such that $`p`$ is a dimap.
###### Proof.
Recall that the sets $`\dot{s}(U)`$ defined above form a basis for the topology of $`W`$. Choose an order atlas $`\{(U_i,_i)\}\overline{U}`$ for $`Z`$. For each open sub-po-space $`VU_i`$ and each $`sP(V)`$, $`\dot{s}(V)W`$ is a po-space under the relation
$$\mathrm{germ}_xs_{\dot{s}(V)}germ_ys\text{ if and only }x_iy.$$
This is well-defined since $`\{U_i\}`$ is an order-atlas, and it makes $`\dot{s}(V)`$ a po-space since $`\dot{s}:U_i\dot{s}(U_i)`$ is a homeomorphism.
We claim that
$$T:=\{\dot{s}(V)|V\stackrel{}{\mathrm{open}}U_i,sP(V)\}$$
is an order atlas on $`W`$. First we need to show that it is an open cover. Each of the sets is open by construction. If $`U๐ช(Z)`$ and $`sP(U)`$, consider $`\mathrm{germ}_xs`$. Since $`\{U_i\}`$ is an open cover of $`Z`$, for some $`i`$, $`xU_i`$. Let $`V=UU_i`$. Then $`\mathrm{germ}_xs=\mathrm{germ}_xs|_V\dot{(s|_V)}(V)`$. Therefore $`T`$ is an open cover of $`W`$.
Finally we need to show that the orders are compatible. For $`k=1,2`$ let $`V_k\stackrel{}{\mathrm{open}}U_{i_k}\stackrel{}{\mathrm{open}}Z`$, and $`s_kP(V_k)`$. Assume $`g_1,g_2\dot{s_1}(V_1)\dot{s}(V_2)`$. That is, $`g_1=\mathrm{germ}_{x_1}s_1`$ $`=\mathrm{germ}_{x_1}s_2`$ and $`g_2=\mathrm{germ}_{x_2}s_1=\mathrm{germ}_{x_2}s_2`$. For $`k=1,2`$,
$$g_1_{\dot{s_k}(V_k)}g_2x_1_{i_k}x_2.$$
Since $`\{U_i\}`$ is an order-atlas, the order $`_{i_1}`$ and $`_{i_2}`$ are compatible. Therefore the orders $`_{\dot{s_1}(V_1)}`$ and $`_{\dot{s_1}(V_1)}`$ are compatible, and $`T`$ is an order-atlas on $`W`$.
Let $`\overline{T}`$ be the equivalence class of order atlases of $`T`$. We claim that $`\overline{T}`$ does not depend on the choice of $`U\overline{U}`$.
Let $`U,U^{}\overline{U}`$, then $`U`$ and $`U^{}`$ have a common refinement $`U^{\prime \prime }`$. Let $`T,T^{},T^{\prime \prime }`$ be the corresponding order-atlases for $`W`$ constructed as above. We will show that $`T^{\prime \prime }`$ is a refinement of $`T`$.
Let $`A\stackrel{}{\mathrm{open}}U_jU`$, $`sP(A)`$ and $`\mathrm{germ}_xs\dot{s}(A)`$. Then there is some $`U_k^{\prime \prime }U^{\prime \prime }`$ such that $`xU_k^{\prime \prime }`$ and $`U_k^{\prime \prime }`$ is a sub-po-space of $`U_j`$. Let $`A^{\prime \prime }=AU_k^{\prime \prime }`$. It follows that $`\dot{(s|_{A^{\prime \prime }})}(A^{\prime \prime })\dot{s}(A)`$, and $`\mathrm{germ}_xs=\mathrm{germ}_x(s|_{A^{\prime \prime }})\dot{(s|_{A^{\prime \prime }})}(A^{\prime \prime })T^{\prime \prime }`$. Since $`U_k^{\prime \prime }`$ is a sub-po-space of $`U_j`$ it follows that $`\dot{(s|_{A^{\prime \prime }})}(A^{\prime \prime })`$ is a sub-po-space of $`\dot{s}(A)`$. Thus $`T^{\prime \prime }`$ is a refinement of $`T`$.
Similarly $`T^{\prime \prime }`$ is a refinement of $`T^{}`$ and is hence a common refinement of $`T`$ and $`T^{}`$. Therefore $`\overline{T}`$ does not depend on the choice of $`U\overline{U}`$.
Finally we will show that the projection $`p:WZ`$ given by $`\mathrm{germ}_xsx`$ is a dimap. Let $`U\overline{U}`$ be an order-atlas on $`Z`$. Let $`T`$ be the order-atlas on $`W`$ constructed above from $`U`$. Observe that $`T\overline{T}`$, since $`\overline{T}`$ does not depend on the choice of $`U\overline{U}`$. Let $`U_jU`$, let $`A\stackrel{}{\mathrm{open}}U_iU`$, and let $`sP(A)`$. Assume that
$$\mathrm{germ}_{x_1}s,\mathrm{germ}_{x_2}s\dot{s}(A)p^1(U_j).$$
Then $`x_1,x_2U_iU_j`$. By the construction of $`T`$ and since $`U`$ is an order atlas,
$$\mathrm{germ}_{x_1}s_{\dot{s}(A)}\mathrm{germ}_{x_2}sx_1_{U_i}x_2x_1_{U_j}x_2.$$
Therefore $`\mathrm{\Lambda }`$ can be extended to local po-spaces. โ
Thus we have maps
$$\mathrm{\Gamma }:\mathrm{๐๐ญ๐๐ฅ๐}(Z)\mathrm{๐๐ก๐ฏ}(๐ช(Z)):\mathrm{\Lambda }.$$
To show that they give an equivalence of categories we will show that for a sheaf $`P`$ and an รฉtale space $`W\stackrel{๐}{}Z`$ there are natural isomorphisms
$$ฯต_W:\mathrm{\Lambda }\mathrm{\Gamma }WW\text{ and }\eta _P:P\mathrm{\Gamma }\mathrm{\Lambda }P.$$
Recall that elements of $`\mathrm{\Lambda }\mathrm{\Gamma }W`$ are of the form $`\dot{s}(x)=\mathrm{germ}_xs`$, where $`s:UW`$ satisfies $`ps=\mathrm{Id}_U`$ and $`xU`$. Define $`ฯต_W`$ to be the map $`\dot{s}xsx`$. We will show this is an isomorphism by constructing an inverse $`\theta _W`$. Let $`yW`$ and let $`x=py`$. Since $`W`$ is รฉtale there exists $`yV\stackrel{}{\mathrm{open}}W`$ such that $`p|_V:V\stackrel{}{}p(V)`$. Let $`q=(p|_V)^1`$. Then define $`\theta _W(y)=\mathrm{germ}_xq=\dot{q}x`$. Then we claim $`\theta _W`$ is an inverse for $`ฯต_W`$. Indeed
$$ฯต_W\theta _Wy=ฯต_W\dot{q}x=qx=y.$$
Also for all $`\dot{s}x\mathrm{\Lambda }\mathrm{\Gamma }W`$, $`\theta _Wฯต_W\dot{s}x=\theta _Wsx=\mathrm{germ}_xt`$, where t is a restriction of $`s`$. So $`\mathrm{germ}_xt=\mathrm{germ}_xs=\dot{s}x`$.
Finally we claim that $`ฯต_W`$ and $`\theta _W`$ are dimaps. First choose $`T=\{T_k\}\overline{T}`$ and $`U=\{U_i\}\overline{U}`$ such that $`p`$ satisfies the dimap condition. ยฟFrom $`T`$ construct the canonical order atlas of the form $`\{\dot{s}V\}`$ for $`\mathrm{\Lambda }\mathrm{\Gamma }W`$ as in the proof of Lemma 5.5. Now let $`\dot{s}x_1,\dot{s}x_2\dot{s}Vฯต_W^1(T_k)`$. Then by construction,
$$\dot{s}x_1_{\dot{s}V}\dot{s}x_2x_1_{U_i}x_2.$$
Since $`s`$ satisfies the dimap condition this implies that $`sx_1_{T_k}sx_2`$ which is the same as $`ฯต_W\dot{s}x_1_{T_k}ฯต_W\dot{s}x_2`$. Thus $`ฯต_W`$ is a dimap. Next let $`y_1,y_2T_k\theta _W^1(\dot{s}V)=T_kฯต_W(\dot{s}V)=T_ksV`$. Then there are $`x_1,x_2V`$ such that $`y_1=sx_1`$ and $`y_2=sx_2`$. Since $`p`$ satisfies the dimap condition
$$y_1_{T_k}y_2py_1_{U_i}py_2.$$
But this is the same as $`x_1_{U_i}x_2`$ which implies that $`\dot{s}x_1_{\dot{s}V}\dot{s}x_2`$. Therefore $`\theta _W`$ is a dimap.
The proof that the morphism $`\eta _P`$ is a bijection is the same as the proof in the case of topological spaces \[MLM92, Theorem II.5.1\]. โ
## 6. Points
In this section $``$ is either $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ or $`\mathrm{๐๐๐}`$ with the Grothendieck topology generated by open (di)covers.
Let $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ and $`\mathrm{๐๐ก๐ฏ}()`$ be the topoi of presheaves and sheaves on $``$. Recall that the inclusion functor $`i:\mathrm{๐๐ก๐ฏ}()\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ has a right adjoint $`a`$ called the associated sheaf functor. Recall from Definition 2.15 that if $`p`$ is a point in $`\mathrm{๐๐ก๐ฏ}()`$ and $`\alpha \mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$, then $`stalk_p(F)=p^{}a(\alpha )`$.
Let $`Z`$. Then $`Z`$ is a topological space or a local po-space and we can choose any point (in the usual sense) $`xZ`$. Define
$`p_x^{}:\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ $`\mathrm{๐๐๐ญ}`$
$`F`$ $`\underset{xL\stackrel{}{\mathrm{open}}Z}{colim}F(L)`$
where the colimit is taken over all open subsets of $`Z`$ containing $`x`$. See Remark 3.5 for a discussion of subobjects in $`\mathrm{๐๐๐}`$.
Given a functor $`p^{}:\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}\mathrm{๐๐๐ญ}`$ there is an induced functor
$$A:\stackrel{๐ฆ}{}\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}\stackrel{p^{}}{}\mathrm{๐๐๐ญ},$$
where $`y`$ is the Yoneda embedding defined on objects and morphisms by $`Z(,Z)`$ and $`\phi (,\phi )`$.
Given a functor $`A:\mathrm{๐๐๐ญ}`$ one can define induced adjoint functors $`p^{}:\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}\mathrm{๐๐๐ญ}`$ and $`p_{}:\mathrm{๐๐๐ญ}\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ ($`p^{}=_{}A`$ and $`p_{}=(A,)`$, see \[MLM92, Section VII.2\] ).
###### Definition 6.1.
1. The functor $`A:\mathrm{๐๐๐ญ}`$ is *flat* if the corresponding $`p^{}`$ is left exact.
2. $`A`$ is *continuous* if $`A`$ sends each covering sieve to an epimorphic family of functions. That is, if $`S`$ is a covering sieve, then the family of functions $`\{A(\phi )|\phi S\}`$ is jointly surjective.
###### Proposition 6.2 (\[MLM92, Corollary VII.5.4\]).
Using the correspondence above, $`p`$ is a point in $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ if and only if $`A`$ is flat. Furthermore $`p`$ descends to a point in $`\mathrm{๐๐ก๐ฏ}()`$ if and only if $`A`$ is flat and continuous.
###### Proposition 6.3.
$`p_x`$ defined above descends to a point in $`\mathrm{๐๐ก๐ฏ}()`$ .
(2)
###### Proof.
Let $`A_x=p_x^{}y`$, where $`y`$ is the Yoneda embedding.
First we show that $`p_x^{}`$ is left exact, that is it preserves finite limits. Let $`F\times _GH`$ be a pullback in $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ .
$`p_x^{}(F\times _GH)`$ $`=`$ $`\underset{xLZ}{colim}(F\times _GH)(L)`$
$`=`$ $`\underset{xLZ}{colim}F(L)\times _{G(L)}H(L)`$
$`=`$ $`colimF(L)\times _{colimG(L)}colimH(L)`$
$`=`$ $`p_x^{}F\times _{p_x^{}G}p_x^{}H`$
The third equality holds because $`colim`$ commutes with pullbacks in $`\mathrm{๐๐๐ญ}`$, and the others are by definition. Thus $`A`$ is flat and $`p_x^{}`$ is a point in $`\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ .
Next we show that $`A_x`$ is continuous. Let $`\{Y_i\stackrel{\phi _i}{}N\}`$ be a covering sieve for $`N`$ in $``$ . Recall that $`A_x=p_x^{}y`$. Let $`(\phi _i)_{}`$ denote composition with $`\phi _i`$. For each arrow in the covering sieve,
$`p_x^{}y(Y_i\stackrel{\phi _i}{}N)`$ $`=`$ $`p_x^{}((,Y_i)\stackrel{(\phi _i)_{}}{}(,N))`$
$`=`$ $`\underset{xLZ}{colim}((L,Y_i)\stackrel{(\phi _i)_{}}{}(L,N))`$
$`=`$ $`y(Y_i)_x\stackrel{(\phi _i)_{}}{}y(N)_x.`$
We claim that this is an epimorphic family of functions in $`\mathrm{๐๐๐ญ}`$ . Let $`fy(N)_x`$. Then there is an open subspace $`L`$ such that $`xLZ`$ and $`f`$ is represented by a morphism $`f^{}(L,N)`$. Since $`\{Y_i\}`$ covers $`N`$, $`f^{}(x)Y_k`$ for some $`k`$. Let $`K=(f^{})^1(Y_k)`$. Then $`K`$ is open and $`xKLZ`$. Furthermore $`f^{}|_K(K,Y_k)`$ which represents an element $`f^{\prime \prime }y(Y_k)_x`$, and $`(\phi _k)_{}f^{\prime \prime }=f`$. Hence we have an epimorphic family as claimed. Thus $`A`$ is continuous and $`p_x`$ descends to a point in $`\mathrm{๐๐ก๐ฏ}()`$ . โ
Abusing notation we will also denote the induced functor in diagram (2) by $`p_x^{}`$. With this abuse of notation, the stalk of $`F\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}`$ at $`x`$ is given by $`stalk_x(F)=p_x^{}a(F)=p_x^{}(F)`$. Note that $`stalk_x(F)=\{\mathrm{germ}_x(s)|xU\stackrel{}{\mathrm{open}}Z,sF(U)\}`$.
###### Theorem 6.4.
The points $`p_x`$ defined above provide enough points for $`\mathrm{๐๐ก๐ฏ}()`$. That is, given $`fg:PQ\mathrm{๐๐ก๐ฏ}()`$, there is an $`Z`$ and a $`xZ`$ such that $`p_x^{}fp_x^{}g:p_x^{}Pp_x^{}Q\mathrm{๐๐๐ญ}`$.
###### Proof.
Given $`Z`$ and either $`P\mathrm{๐๐ก๐ฏ}()`$ or $`f\mathrm{Mor}\mathrm{๐๐ก๐ฏ}()`$, let $`P_Z`$ or $`f_Z`$ denote the restriction to $`\mathrm{๐๐ก๐ฏ}(๐ช(Z))`$.
Assume that $`fg:PQ\mathrm{๐๐ก๐ฏ}()`$. Thus there is some $`Z`$ such that $`f_Zg_Z:P_ZQ_Z\mathrm{๐๐ก๐ฏ}(๐ช(Z))`$.
By Theorem 1.3, this is equivalent to saying that the corresponding maps between รฉtale spaces are not equal. That is,
$$\mathrm{\Lambda }f_Z\mathrm{\Lambda }g_Z:\mathrm{\Lambda }P_Z\mathrm{\Lambda }Q_Z\mathrm{๐๐ญ๐๐ฅ๐}(Z).$$
Thus there is some point $`y\mathrm{\Lambda }P_Z`$ such that $`\mathrm{\Lambda }f_Z(y)\mathrm{\Lambda }g_Z(y)`$.
By the definition of $`\mathrm{\Lambda }`$, $`y=\mathrm{germ}_xs`$ for some $`xU\stackrel{}{\mathrm{open}}Z`$ and $`sP_Z(U)`$. That is $`ystalk_x(P)=p_x^{}P`$. Therefore $`p_x^{}fg_x^{}g:p_x^{}Pp_x^{}Q`$. โ
## 7. Stalkwise equivalences
Let $`(,\tau )`$ be a site with a subcanonical Grothendieck topology such that $`\mathrm{๐๐ก๐ฏ}()`$ has enough points and let $`\overline{y}:\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$ be the Yoneda embedding. Recall the definition of stalkwise equivalence in Definition 2.19 which uses the simplicial stalk functor $`()_p`$. Also recall the Yoneda embedding $`\overline{y}:\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$ given in Definition 2.17. Let $`\phi :XY`$.
###### Lemma 7.1.
$`\overline{y}(\phi )`$ is a stalkwise equivalence if and only if for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$, $`p^{}ay(\phi )\mathrm{๐๐๐ญ}`$ is an isomorphism.
###### Proof.
Let $`p`$ be a point in $`\mathrm{๐๐ก๐ฏ}()`$ . Recall that the simplicial stalk of $`\overline{y}(\phi )`$ at $`p`$ is given by
$$(\overline{y}(\phi ))_p=\{stalk_p(\overline{y}(\phi )_n)\}_{n0}=\{p^{}ay(\phi )\}_{n0},$$
which is simplicially constant. Thus $`\overline{y}(\phi )_p\mathrm{๐ฌ๐๐๐ญ}`$ is an isomorphism if and only if $`p^{}ay(\phi )\mathrm{๐๐๐ญ}`$ is an isomorphism. โ
###### Lemma 7.2.
If the Grothendieck topology $`\tau `$ is subcanonical, then the composite functor $`\stackrel{๐ฆ}{}\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}\stackrel{๐}{}\mathrm{๐๐ก๐ฏ}()`$ is faithful.
###### Proof.
By the Yoneda lemma, $`y`$ is full and faithful. Since $`\tau `$ is subcanonical $`\mathrm{im}(y)\mathrm{๐๐ก๐ฏ}()`$. Furthermore $`ai:\mathrm{๐๐ก๐ฏ}()\mathrm{๐๐ก๐ฏ}()`$ is naturally isomorphic to the identity functor \[MLM92, Corollary III.5.6\]. Thus $`ay`$ is naturally isomorphic to $`y`$ which is faithful. โ
###### Theorem 7.3.
Let $`\phi :XY`$ and assume that $`\overline{y}(\phi )`$ is a stalkwise equivalence. Then $`\phi `$ is bijective.
The proof of this theorem is split into the following two propositions.
###### Proposition 7.4.
Let $`\phi :XY`$ and assume that $`\overline{y}(\phi )`$ is a stalkwise equivalence. Then $`\phi `$ is epi.
###### Proof.
For $`i=1,2`$, let $`\psi _i:YZC`$ be a morphism such that $`\psi _1\phi =\psi _2\phi :XZ`$. Then for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$ , $`p^{}ay(\psi _1\phi )=p^{}ay(\psi _2\phi )`$. ยฟFrom this it follows that
$$p^{}ay(\psi _1)p^{}ay(\phi )=p^{}ay(\psi _2)p^{}ay(\phi ).$$
But by Lemma 7.1 $`p^{}ay(\phi )`$ is a set isomorphism, so in particular it is epi. Therefore $`p^{}ay\psi _1=p^{}ay\psi _2`$ for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}().`$ Since $``$ has enough points, $`ay\psi _1=ay\psi _2`$. By Lemma 7.2 $`ay`$ is faithful, thus $`\psi _1=\psi _2`$. Therefore $`\phi `$ is epi. โ
###### Proposition 7.5.
Let $`\phi :XY`$ and assume that $`\overline{y}(\phi )`$ is a stalkwise equivalence. Then $`\phi `$ is mono.
###### Proof.
For $`i=1,2`$, let $`\psi _i:WXC`$ be a morphism such that $`\phi \psi _1=\phi \psi _2:WY`$. As in the proof of the previous proposition, for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$ ,
$$p^{}ay(\phi )p^{}ay(\psi _1)=p^{}ay(\phi )p^{}ay(\psi _2).$$
Again by Lemma 7.1, $`p^{}ay(\phi )`$ is mono. Therefore $`p^{}ay\psi _1=p^{}ay\psi _2`$ for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$ . Since $``$ has enough points, $`ay\psi _1=ay\psi _2`$. By Lemma 7.2 $`ay`$ is faithful, thus $`\psi _1=\psi _2`$. Therefore $`\phi `$ is mono. โ
Let $``$ = $`\mathrm{๐๐ฉ๐๐๐๐ฌ}`$ or $`\mathrm{๐๐๐}`$ with the open cover topology. By Example 4.3 and Proposition 4.4 this topology is subcanonical.
Recall from Section 6 that if $`Z`$ and $`xZ`$, then
(3)
$$\begin{array}{cc}\hfill p_x^{}:\mathrm{๐๐๐ญ}^{^{\mathrm{op}}}& \mathrm{๐๐๐ญ}\hfill \\ \hfill F& \underset{xL\stackrel{}{\mathrm{open}}Z}{colim}F(L)\hfill \end{array}$$
descends to a point in $`\mathrm{๐๐ก๐ฏ}()`$ (where the colimit is taken over open subspaces of $`Z`$ which contain $`x`$).
###### Theorem 7.6.
Let $`\phi :XY`$. Then $`\overline{y}(\phi )`$ is a stalkwise equivalence if and only if $`\phi `$ is an isomorphism in $``$.
###### Proof.
($``$) If $`\phi `$ is an isomorphism, then for all points $`p`$ in $`\mathrm{๐๐ก๐ฏ}()`$ $`p^{}ay(\phi )`$ is an isomorphism. Hence by Lemma 7.1 $`\overline{y}(\phi )`$ is a stalkwise equivalence.
($``$) Assume that $`\overline{y}(\phi )`$ is a stalkwise equivalence. Then by Theorem 7.3, $`\phi `$ is a bijection.
Let $`xY`$. Let $`p_x`$ be the corresponding point defined in (3). Then
$$p_x^{}ay(\phi ):\underset{xLY}{colim}(L,X)\stackrel{\phi _{}}{}\underset{xLY}{colim}(L,Y)\mathrm{๐๐๐ญ}$$
is a bijection. Let $`f:YY`$ be given by $`f=\mathrm{Id}_Y`$. Let $`\overline{f}=[f]colim_{xLY}(L,Y)`$. Let $`\overline{g}=(\phi _{})^1(\overline{f})`$. Then there is some $`xWY`$ such that $`\overline{g}`$ has a representative $`g(W,X)`$.
Let $`f^{}=\phi _{}g=\phi g`$. Then $`[f^{}]=\phi _{}[g]=[f]`$. Therefore there exists $`xSY`$ such that $`SYW`$ and $`f^{}|_S=f|_S=\mathrm{Id}_Y|_S`$.
Let $`\psi =g|_S`$. Therefore $`\phi \psi =\mathrm{Id}_S`$. Let $`T=\mathrm{im}(\psi )`$. Then $`\phi |_T\psi =\mathrm{Id}_S`$ and $`\phi |_T`$ is a bijection. Hence $`\phi |_T:TS`$ is an isomorphism, where $`xS`$.
Finally this construction can be repeated for all $`xY`$. For each $`xY`$ there is a $`xS_xY`$ and there is a map
$$\psi _x:S_xX\text{ such that }\psi _x=(\phi |_{\mathrm{im}(\psi _x)})^1.$$
Since $`\phi `$ is a bijection, all local inverses must agree. That is, $`\{\psi _x:S_xX\}`$ is a matching family on the open cover $`\{S_x\}`$ of $`Y`$. Since the topology is subcanonical, there is a unique amalgamation $`\psi :YX`$. It remains to show that $`\psi `$ is an inverse for $`\phi `$.
$$\text{For all }S_x,\phi \psi |_{S_x}=\phi \psi _x=\mathrm{Id}_{S_x}.$$
Therefore $`\phi `$ is an isomorphism in $``$ . โ
## 8. Model categories for local po-spaces
### 8.1. A model category for local po-spaces
Using our results on $`\mathrm{๐๐๐}`$, Theorem 1.1 will now follow directly from Jardineโs model structure (Theorem 2.20).
###### Proof of Theorem 1.1.
The open dicovers induce a Grothendieck topology on the small category $`\mathrm{๐๐๐}`$. Applying Theorem 6.4, the Grothendieck topos $`\mathrm{๐๐ก๐ฏ}(\mathrm{๐๐๐})`$ has enough points. So by Jardineโs Theorem (Theorem 2.20), $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ has a proper, simplicial, cellular model structure in which
* the cofibrations are the monomorphisms, i.e. the levelwise monomorphisms of presheaves,
* the weak equivalences are the stalkwise equivalences, and
* the fibrations are the morphisms which have the right lifting property with respect to all trivial cofibrations.
Finally by Theorem 7.6 the weak equivalences coming from $`\mathrm{๐๐๐}`$ (via the Yoneda embedding) are precisely the isomorphisms. โ
### 8.2. Localization
Our main motivation for constructing a model category for local po-spaces was to model concurrent systems. In particular we would like to be able to define and understand equivalences of concurrent systems using such a model category. However our model structure on $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ does not have any non-trivial equivalences among the morphisms coming from $`\mathrm{๐๐๐}`$. To obtain a model category more directly useful for studying concurrency, we need to localize with respect to a set of morphisms. In particular we want morphisms which preserve certain computer-scientific information.
How to best choose such morphisms is an important question and has been studied in \[Bub04\]. For the sake of simplicity that paper studied only the category $`\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$ of po-spaces (a subcategory of $`\mathrm{๐๐๐}`$). There it was shown that the set of morphisms which should be equivalences depends on the *context*. That is, instead of choosing equivalences for $`\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$ one should be choosing equivalences for the coslice category or undercategory $`๐\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$ of po-spaces under a po-space $`A`$, where $`A`$ is called the context.
This result can be easily extended to our setting. First we remark that if we choose a local po-space $`A`$ then the undercategory $`๐\mathrm{๐๐๐}`$ is the category whose objects are dimaps $`\iota _M:A(M,\overline{U})`$ and whose morphisms are dimaps $`f:(M,\overline{U})(N,\overline{V})`$ such that the following diagram commutes:
Next, $`\overline{y}(A)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ and the undercategory $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ is the category whose objects are morphisms of simplicial presheaves $`\iota _\alpha :\overline{y}(A)\alpha `$ and whose morphisms are morphisms of simplicial presheaves $`f:\alpha \beta `$ such that the following diagram commutes:
Since $`\overline{y}:\mathrm{๐๐๐}\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ is a functor
$$\overline{y}(\iota _M):\overline{y}(A)\overline{y}(M,\overline{U})\text{ and }\overline{y}(\iota _N)=\overline{y}(f\iota _M)=\overline{y}(f)\overline{y}(\iota _M).$$
Hence $`๐\mathrm{๐๐๐}`$ embeds as a subcategory of $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$.
Define morphisms in $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ to be weak equivalences, cofibrations and fibrations if and only if they are weak equivalence, cofibrations and fibrations in $`\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$. Then this makes $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ into a model category (see \[Hir03, Theorem 7.6.5\]).
We will show that this model category is again proper and cellular. We will need the following definitions and a theorem of Kan.
###### Definition 8.1.
* Let $``$ be a category and $`I`$ be a set of maps in $`.`$ A *relative $`I`$-cell complex* is a map that can be constructed by a transfinite composition of pushouts of elements of $`I`$.
* An object $`A`$ is *small relative to a collection of morphisms $`๐`$* in $``$ if there exists a cardinal $`\kappa `$ such that for all regular cardinals $`\lambda \kappa `$ and for all $`\lambda `$-sequences
$$X_0X_1X_2\mathrm{}X_\beta \mathrm{}$$
with $`X_\beta X_{\beta +1}`$ in $`๐`$ for $`\beta +1<\lambda `$, the set map
$$\underset{\beta <\lambda }{colim}(A,X_\beta )(A,\underset{\beta <\lambda }{colim}X_\beta )$$
is an isomorphism.
###### Definition 8.2.
A model category $``$ is cofibrantly generated if there are sets $`I`$ and $`J`$ such that
* the domains of $`I`$ are small relative to the relative $`I`$-cell complexes,
* the domains of $`J`$ are small relative to the relative $`J`$-cell complexes,
* the fibrations have the right lifting property with respect to $`J`$, and
* the trivial fibrations have the right lifting property with respect to $`I`$.
We say that $``$ is cofibrantly generated by $`I`$ and $`J`$.
###### Definition 8.3.
* Let $``$ be a model category cofibrantly generated by $`I`$ and $`J`$. An object $`A`$ is *compact* if there is a cardinal $`\gamma `$ such that for all relative $`I`$-cell complexes $`f:XY`$ with a particular presentation, every map $`AY`$ factors through a subcomplex of size at most $`\gamma `$.
* $`f:AB`$ is an *effective monomorphism* if $`f`$ is the equalizer of the inclusions $`BB_AB`$.
###### Definition 8.4.
A *cellular* model category is a model category cofibrantly generated by $`I`$ and $`J`$ such that
* the domains and codomains of elements of $`I`$ and $`J`$ are compact,
* the domains of elements of $`J`$ are small relative to relative $`I`$-cell complexes, and
* the cofibrations are effective monomorphisms.
###### Theorem 8.5 (\[Hir03, Theorem 11.3.2\]).
Let $``$ be a model category cofibrantly generated by the sets $`I`$ and $`J`$, and let $`๐ฉ`$ be a bicomplete category such that there exists a pair of adjoint functors $`F:๐ฉ:U`$. Define $`FI=\{Fu|uI\}`$ and $`FJ=\{Fv|vJ\}`$. If
1. the domains of $`FI`$ and $`FJ`$ are small relative to $`FI`$-cell and $`FJ`$-cell, respectively, and
2. $`U`$ maps relative $`FJ`$-cell complexes to weak equivalences,
then $`๐ฉ`$ has a model category structure cofibrantly generated by $`FI`$ and $`FJ`$ such that $`f`$ is a weak equivalence in $`๐ฉ`$ if and only if $`Uf`$ is a weak equivalence in $``$, and $`(F,U)`$ is a Quillen pair.
###### Theorem 8.6.
Let $``$ be a model category and let $`A`$. Then $`๐`$ has a model structure where a morphism is a weak equivalence, cofibration or fibration in $`๐`$ if and only if $`f`$ is a weak equivalence, cofibration or fibration, respectively, in $``$. If $``$ is proper, cofibrantly generated or cellular, then so is $`๐`$.
###### Remark 8.7.
For a more detailed proof we invite the reader to regard Hirschhornโs note \[Hir05\].
###### Proof.
That $`๐`$ has the stated model structure follows from the definitions (see \[Hir03, Theorem 7.6.5\]).
Pushouts and pullbacks in $`๐`$ can be formed by taking pushouts and pullbacks of the underlying morphisms in $``$, and then taking the induced maps from $`A`$. It thus follows that if $``$ is proper so is $`๐`$.
Assume $``$ is cofibrantly generated by $`I`$ and $`J`$. The method for showing that $`๐`$ is cofibrantly generated will be to apply Theorem 8.5 to the following adjoint functors:
$$F:(๐):U$$
where for $`B`$ and $`f:BC`$,
$$F(B)=\begin{array}{c}\hfill \text{}\end{array},F(f)=\begin{array}{c}\hfill \text{}\end{array}$$
and $`U`$ is the forgetful functor
$$U\left(\begin{array}{c}\hfill \text{}\end{array}\right)=B,U\left(\begin{array}{c}\hfill \text{}\end{array}\right)=B\stackrel{๐}{}C.$$
Define $`FI=\{Fu|uI\}`$ and $`FJ=\{Fv|vJ\}`$.
The main observation for the proof is that for a morphism $`u`$ in $``$, the pushout of $`Fu`$ is obtained from the pushout of $`u`$ in $``$. That is,
$$\begin{array}{c}\hfill \text{}\end{array}\text{ where }P\text{ is defined by }\begin{array}{c}\hfill \text{}\end{array}$$
From this it follows that for a set of morphisms $`S`$ in $``$, the underlying morphisms of a relative $`FS`$-complex are a relative $`S`$-complex.
Hence the conditions on $`๐`$ in Theorem 8.5 and the definition of a cellular model category (Definition 8.4) are all inherited from the corresponding conditions in $``$.
Finally one can check that the model category structure given by Theorem 8.5 coincides with the one in the statement of the theorem. โ
Let $``$ denote the model structure above on $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$. Since $``$ is cellular we can apply left Bousfield localization \[Hir03\] to this model structure $``$ with respect to a set of morphisms which will preserve the computer-scientific properties we are interested in. In \[Bub04\], one inverted the set of *dihomotopy equivalences* in $`๐\mathrm{๐๐จ๐๐ฉ๐๐๐๐ฌ}`$. So in our setting we will let $`I`$ be the set of *dihomotopy equivalences* in $`๐\mathrm{๐๐๐}`$ defined below. We will invert the set $`=\{\overline{y}(f)|fI\}\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$.
###### Definition 8.8.
* Let $`\stackrel{}{I}`$ be the po-space $`([0,1],)`$ where $``$ is the usual total order on $`[0,1]`$. Given dimaps $`f,g:(M,\overline{U})(N,\overline{V})๐\mathrm{๐๐๐}`$, $`\varphi `$ is a *dihomotopy* from $`f`$ to $`g`$ if $`\varphi :(M,\overline{U})\times \stackrel{}{I}(N,\overline{V})`$, $`\varphi |_{(M,\overline{U})\times \{0\}}=f`$, $`\varphi |_{(M,\overline{U})\times \{1\}}=g`$, and for all $`aA`$, $`\varphi (\iota _M(a),t)=\iota _N(a)`$. In this case write $`\varphi :fg`$.
* The symmetric, transitive closure of dihomotopy is an equivalence relation. Write $`fg`$ if there is a chain of dihomotopies $`ff_1f_2\mathrm{}f_ng`$.
* A dimap $`f:(M,\overline{U})(N,\overline{V})`$ is a *dihomotopy equivalence* if there is a dimap $`g:(N,\overline{V})(M,\overline{U})`$ such that $`gf\mathrm{Id}_M`$ and $`fg\mathrm{Id}_N`$.
The left Bousfield localization of $``$ with respect to $``$ provides a model structure on $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ in which the weak equivalences are the $``$-local equivalences (see \[Hir03\]), the cofibrations are the cofibrations in $``$ and the fibrations are morphisms which have the right lifting property with respect to morphisms which are both cofibrations and $``$-local equivalences.
###### Theorem 8.9 (Theorem 1.2).
Let $`=\{\overline{y}(f)|f`$ is a directed homotopy equivalence rel $`A\}`$. The category $`\overline{๐ฒ}(๐)\mathrm{๐ฌ๐๐ซ๐}(\mathrm{๐๐๐})`$ has a left proper, cellular model structure in which
* the cofibrations are the monomorphisms,
* the weak equivalences are the $``$-local equivalences, and
* the fibrations are those morphisms which have the right lifting property with respect to monomorphisms which are $``$-local equivalences.
We claim that this model category provides a good model for studying concurrency. An analysis of this model category will be the subject of future research.
## Appendix A Hypercovers
Suppose now $``$ is small and equipped with a Grothendieck topology, i.e. we have a site $`(,\tau )`$. The $`\stackrel{ห}{\text{C}}\text{ech}`$ structure $`\mathrm{๐ฌ๐๐๐ญ}_{\stackrel{ห}{c}\left(\tau \right)}^{^{op}}`$ is obtained from the projective structure by homotopically localizing the comparison morphisms given by the $`\stackrel{ห}{\text{C}}\text{ech}`$ covers with respect to $`\tau `$ or, up-to homotopy, from the injective structure by localizing at the same set of morphisms.
###### Definition A.1.
Let $`U=\left\{U_i\stackrel{u_i}{}X\right\}_{iI}J\left(X\right)`$ be a cover. Let $`i_pI`$ for each $`0pn`$ and $`U_{i_0\mathrm{}i_n}`$ be the *wide pullback* of the $`u_{i_p}`$โs, i.e. the limiting object of the diagram
The $`\stackrel{ห}{C}ech`$ nerve $`\stackrel{ห}{U}`$ of $`U`$ is the simplicial presheaf given by
$$\stackrel{ห}{U}_n\stackrel{def}{=}\underset{i_0,\mathrm{},i_nI}{}y\left(U_{i_0\mathrm{}i_n}\right)$$
###### Remark A.2.
For any $`n`$, $`X`$ and $`UJ\left(X\right)`$ there is a morphism
$$u_{i_0\mathrm{}i_n}:U_{i_0\mathrm{}i_n}X$$
and a diagram of presheaves
where $`E_{U,X,n}`$ is given by universal property. The $`E_{U,X,n}`$ assemble to a morphism of simplicial presheaves
$$E_{U,X}:\stackrel{ห}{U}\kappa _{y\left(X\right)}$$
###### Remark A.3.
Given $`UJ\left(X\right)`$ seen as a subcategory of the slice $`/X`$, there is the evident functor
$$\begin{array}{cccc}\hfill \delta _U:& U\hfill & & \mathrm{๐ฌ๐๐๐ญ}^{^{op}}\hfill \\ & u_i\hfill & & \kappa _{y\left(U_i\right)}\hfill \end{array}$$
###### Proposition A.4.
Localizing $`\mathrm{๐ฌ๐๐๐ญ}_{inj}^{^{op}}`$ at the sets
1. $`\{E_{U,X}X,UJ\left(X\right)\}`$ ;
2. $`\{hocolim\left(\delta _U\right)\kappa _{y\left(X\right)}X,UJ\left(X\right)\}`$;
3. $`\{\kappa \left(\iota _U\right)X,UJ\left(X\right)\}`$where, given $`X`$ and $`R`$ a sieve on $`X`$, $`\iota _R:Ry\left(X\right)`$ is the corresponding inclusion of presheaves;
4. $`\{\eta _F:Fj\left(F\right)F\mathrm{๐ฌ๐๐๐ญ}^{^{op}}\}`$where $`j:\mathrm{๐ฌ๐๐๐ญ}^{^{op}}\mathrm{๐ฌ๐๐๐ญ}^{^{op}}`$ is the objectwise sheafification functor;
yields the same model structure $`\mathrm{๐ฌ๐๐๐ญ}_{\stackrel{ห}{c}\left(\tau \right)}^{^{op}}`$. The same holds for the projective version.
Finally, there is a model structure $`\mathrm{๐ฌ๐๐๐ญ}_{hyp\left(\tau \right)}^{^{op}}`$ obtained from the projective structure by homotopically localizing at the set of the comparison morphisms given by hypercovers with respect to $`\tau `$. This model structure is Quillen equivalent to Jardineโs model structure (Theorem 2.20) on $`\mathrm{๐ฌ๐๐๐ญ}^{^{\mathrm{op}}}`$ \[DHI04, Theorem 1.2\]. As with the $`\stackrel{ห}{\text{C}}\text{ech}`$ structure, there is also an injective version. Since $`\stackrel{ห}{\text{C}}\text{ech}`$ covers are particular hypercovers, there is the series of inclusions
$$๐ฒ_{prj}๐ฒ_{\stackrel{ห}{c}\left(\tau \right)}๐ฒ_{hyp\left(\tau \right)}$$
and a similar series for the injective version. It is in general the case that $`๐ฒ_{\stackrel{ห}{c}\left(\tau \right)}๐ฒ_{hyp\left(\tau \right)}`$, yet equality holds in some important particular cases like the smooth Nisnevitch site (c.f. \[DHI04, Example A10\]). It is an interesting question whether or not $`๐ฒ_{\stackrel{ห}{c}(\tau )}=๐ฒ_{hyp(\tau )}`$ for local po-spaces.
|
warning/0506/hep-th0506014.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
One of the simplest and most accessible forms of flux compactification is given by M-theory on $`\mathrm{K3}\times \mathrm{K3}`$. This was first a analyzed in . The fluxes may preserve the full $`N=4`$ supersymmetry, or break some or all of the supersymmetry. We will be concerned with the case where this flux breaks the supersymmetry to $`N=2`$.
The F-theory limit of this theory yields an $`N=1`$ theory in four dimensions and is dual, via the construction of , to the type IIB string on $`\mathrm{K3}\times (T^2/_2)`$, where the $`_2`$ action includes an orientifolding reflection on the world-sheet.
This theory, mainly in the orientifold language, was analyzed in . The fluxes themselves obstruct many of the moduli of $`\mathrm{K3}\times \mathrm{K3}`$ but, at least if one uses the rules of the supergravity limit described in , one cannot fix all of the moduli. It is believed that there are possibilities of flux obstruction beyond those found in supergravity , but the rules for this are not yet understood properly so we will not consider this possibility in this paper.
The fluxes select a preferred complex structure on $`\mathrm{K3}\times \mathrm{K3}`$ and a given choice of flux determines this complex structure uniquely. There remain up to 20 undetermined complexified Kรคhler moduli for each K3 surface. We will show that, in certain cases, all of these remaining moduli are generically fixed by M5-brane instanton corrections to the superpotential.
It has been realized recently that fluxes may modify Wittenโs analysis of which divisors M5-instantons may wrap to give non-trivial effects. This allows for interesting instanton effects even in simple geometries, such as a tori and K3 surfaces, where naรฏvely one might not expect such things.
In particular, in an explicit counting of fermionic zero modes on M5 with the background (2,2) primitive flux $`G`$ was performed. The generalized condition for the non-vanishing instanton corrections to the superpotential in this case requires that the new, flux dependent index of the Dirac operator equals to one, $`\chi __D(G)=1`$. Here $`\chi __D(G)=\chi __D(h^{(0,2)}n)`$ where $`\chi __D`$ is the arithmetic genus of the divisor and $`0nh^{(0,2)}`$ is a number of solutions of a certain constraint equation which the fermionic zero modes have to satisfy in presence of fluxes. In absence of fluxes this condition is reduced back to Wittenโs condition that $`\chi __D=1`$.
In particular, for the case of $`\mathrm{K3}\times \mathrm{K3}`$ 4-fold and divisors of the form $`\mathrm{K3}\times ^1`$ without fluxes $`\chi __D=2`$ and no instanton corrections to the superpotential are possible. In presence of the background (2,2) primitive flux $`G`$ it was established in that $`n=h^{(0,2)}`$, and therefore $`\chi __D(G)=1`$ and instantons corrections to the superpotential are possible. The same result for $`\mathrm{K3}\times \mathrm{K3}`$ was obtained in .
Oddly enough, we will see that if one is overzealous and tries to leave fewer than 20 Kรคhler moduli unfixed by the flux, the possibility arises that the instanton effects might be unable to fix some of the remaining moduli.
Our interest in the model of M-theory compactified on $`\mathrm{K3}\times \mathrm{K3}`$ is two-fold. First of all, this is a relatively simple model, well-understood in the framework of IIB string theory and 4d gauged supergravity . The geometry of K3 surfaces is far-better understood than generic CalabiโYau threefolds and fourfolds and so this model can be analyzed more thoroughly than the many previous examples with all moduli fixed. Secondly, this model has practical applications to cosmology of D3/D7 brane inflation in type IIB string on $`\mathrm{K3}\times (T^2/_2)`$ .
The geometry of fluxes on $`\mathrm{K3}\times \mathrm{K3}`$ is a very beautiful subject and has connections with number theory as analyzed in . Here we will show that this allows for a complete analysis of all possibilities. In the case that the flux is purely of the type that breaks half the supersymmetries, we list all 13 possibilities that arise. Of these, only 8 correspond to orientifolds.
In section 2 we will analyze the conditions imposed on the K3 surfaces by a flux which breaks half the supersymmetry. This contains some very pretty mathematics associated to โattractive K3 surfacesโ. In section 3 we discuss the role of M5-brane instantons and argue that all the moduli will be generically fixed, except possibly in some cases where a particular choice of flux is made. We conclude in section 4.
## 2 Moduli Spaces and Fluxes
In this section we review the analysis of fluxes for M-theory on $`\mathrm{K3}\times \mathrm{K3}`$ and its F-theory limit. The latter is equivalent to an orientifold of the type IIB string on $`\mathrm{K3}\times (T^2/_2)`$. While this has been analyzed quite extensively in , we present a slightly different approach which more closely follows which we believe is a little more efficient.
### 2.1 M Theory
Let us begin with M-theory on $`S_1\times S_2`$, where each $`S_j`$ is a K3 surface. For compactification on an 8-manifold $`X`$, an element of $`G`$-flux may be present. This $`G`$-flux is subject to a quantization condition , which asserts that, in our case<sup>1</sup><sup>1</sup>1We have absorbed a factor of $`2\pi `$ into $`G`$ compared to much of the rest of the literature.
$$GH^4(S_1\times S_2,).$$
(1)
A consistent theory must contain M2-branes and/or nonzero $`G`$-flux in this background satisfying
$$n_{\text{M2}}+\frac{1}{2}G^2=24.$$
(2)
The M2-branes will not break any supersymmetry, but the $`G`$-flux may. The supergravity analysis of showed that $`G`$ must be primitive and of type (2,2) in order that any supersymmetry be preserved. Any such integral 4-form may be decomposed
$$G=G_0+G_1,$$
(3)
where
$$\begin{array}{cc}\hfill G_0& =\underset{\alpha =1}{\overset{M}{}}\omega _1^{(\alpha )}\omega _2^{(\alpha )}\hfill \\ \hfill G_1& =\mathrm{Re}(\gamma \mathrm{\Omega }_1\overline{\mathrm{\Omega }}_2),\hfill \end{array}$$
(4)
and $`\omega _j^{(\alpha )}`$ are (cohomology classes<sup>2</sup><sup>2</sup>2We only discuss cohomology classes of forms in this paper but we will usually not state this explicitly to avoid cluttering notation. of) integral primitive (1,1)-forms on $`S_j`$, $`\mathrm{\Omega }_j`$ is the holomorphic 2-form on $`S_j`$ and $`\gamma `$ is a complex number which must be chosen to make the last term integral.
There are essentially three kinds of moduli which arise in such a compactification:
1. Deformations of the K3 surfaces $`S_1`$ and $`S_2`$.
2. Motion of the M2-branes.
3. Deformations of vector bundles with nonabelian structure group associated to enhanced gauge symmetries arising from singular points in $`S_1`$ and $`S_2`$.
By assuming, from now on, that $`n_{M2}=0`$ and that our K3 surfaces are smooth, we will restrict attention to only the first kind of modulus in this paper.
The moduli space of M-theory on $`\mathrm{K3}\times \mathrm{K3}`$ is of the form $`\mathrm{M}_1\times \mathrm{M}_2`$, where each factor is associated to one of the K3 surfaces. If no supersymmetry is broken by fluxes, each of the $`\mathrm{M}_j`$ factors is a quaternionic Kรคhler manifold. Ignoring instanton corrections, each $`\mathrm{M}_j`$ is of the form
$$\mathrm{O}(\mathrm{\Gamma }_{4,n_j})\backslash \mathrm{O}(4,n_j)/(\mathrm{O}(4)\times \mathrm{O}(n_j)),j=1,2$$
(5)
where $`\mathrm{\Gamma }_{4,n_j}`$ is a lattice of signature $`(4,n_j)`$. The values of $`n_j20`$ will be determined by the choice of flux $`G`$. The space (5) should be viewed as the Grassmannian of space-like 4-planes in $`\mathrm{\Pi }_j\mathrm{\Gamma }_{4,n_j}`$ divided out by the discrete group of automorphisms of the lattice $`\mathrm{\Gamma }_{4,n_j}`$.
The Grassmannian (5) is familiar (for $`n_j=20`$) from the moduli space of $`N=(4,4)`$ superconformal field theories associated to the sigma model with a K3 target space $`S`$. In this case, the degrees of freedom parametrized by the conformal field theory are given by a Ricci flat metric on $`S`$ together with a choice of $`BH^2(S,\mathrm{U}(1))`$. The choice of metric on a K3 surface of volume one is given by a space-like 3-plane $`\mathrm{\Sigma }H^2(S,)=^{3,19}`$. The 3-plane $`\mathrm{\Sigma }`$ is spanned by the real and imaginary parts of $`\mathrm{\Omega }`$, and the Kรคhler form $`J`$. The extra data of the $`B`$-field and volume extend this to a choice of space-like 4-plane $`\mathrm{\Pi }H^{}(S,)=^{4,20}`$. We refer to and references therein for a full account of this.
Even though M-theory itself has no $`B`$-field, the M5-brane wrapped on one $`\mathrm{K3}`$ gives us an effective $`B`$-field for compactification on the other K3. Hence the form (5). We refer to for examples.
A K3 surface is a hyperkรคhler manifold and thus has a choice of complex structures for a fixed metric. This choice corresponds to specifying the direction of $`J`$ in $`\mathrm{\Sigma }`$. Since supersymmetries are constructed for complex structures, this multiplicity of complex structures implies the existence of a specific extended supersymmetry.
If $`G_1=0`$ in (3) then the condition that $`G`$ be primitive and of type (2,2) preserves the freedom to rotate $`\mathrm{\Omega }`$ and $`J`$ within $`\mathrm{\Sigma }`$. Thus, values of $`G`$ purely of the form $`G_0`$ preserve the full $`N=4`$ supersymmetry in three dimensions .
If the term $`G_1`$ in (3) is non-trivial then we destroy the symmetry of rotations within $`\mathrm{\Sigma }`$ and the supersymmetry is broken to $`N=2`$. This is the case of interest and we therefore assume, from now on, that $`G_1`$ is nonzero.
### 2.2 Attractive K3 surfaces
For now, let us assume that $`G`$ is purely of type $`G_1`$, i.e., $`G_0=0`$. Let
$$\mathrm{\Omega }_j=\alpha _j+i\beta _j.$$
(6)
for $`\alpha _j,\beta _jH^2(S_j,)`$. From $`_{S_j}\mathrm{\Omega }_j\overline{\mathrm{\Omega }}_j>0`$ and $`_{S_j}\mathrm{\Omega }_j\mathrm{\Omega }_j=0`$, it follows that<sup>3</sup><sup>3</sup>3We use the implicit inner product $`a.b=_Sab`$.
$$\begin{array}{cc}\hfill \alpha _j^2& =\beta _j^2>0\hfill \\ \hfill \alpha _j.\beta _j& =0\hfill \\ \hfill \alpha _j& \beta _j\hfill \end{array}$$
(7)
We also have
$$G=\alpha _1\alpha _2+\beta _1\beta _2,$$
(8)
where we set $`\gamma =1`$ in (4) by rescaling $`\mathrm{\Omega }_1`$. Let $`๐_j`$ be a 2-plane in $`H^2(S_j,)`$ spanned by $`\alpha _j`$ and $`\beta _j`$. We claim
###### Theorem 1
$`๐_1`$ and $`๐_2`$ are uniquely determined by $`G`$.
To prove this we first use the Kรผnneth formula which tells us that
$$H^4(S_1\times S_2,)H^0(S_1,)H^4(S_2,)H^2(S_1,)H^2(S_2,)H^4(S_1,)H^0(S_2,).$$
(9)
We know from (4) that $`G`$ lies entirely in the second term on the right-hand side of (9). Let us assume we are given $`G,\alpha _j,\beta _j`$ solving
$$G=\alpha _1\alpha _2+\beta _1\beta _2.$$
(10)
Now try to find other solutions of the form
$$G=(\alpha _1+\alpha _1^{})(\alpha _2+\alpha _2^{})+(\beta _1+\beta _1^{})(\beta _2+\beta _2^{}).$$
(11)
It follows that
$$\alpha _1\alpha _2^{}+\alpha _1^{}\alpha _2+\alpha _1^{}\alpha _2^{}+\beta _1\beta _2^{}+\beta _1^{}\beta _2+\beta _1^{}\beta _2^{}=0.$$
(12)
Let $`\pi _1`$ be the projection
$$\pi _1:H^2(S_1,)H^2(S_1,)/\mathrm{Span}(\alpha _1,\beta _1).$$
(13)
Thus
$$\pi _1(\alpha _1^{})(\alpha _2+\alpha _2^{})+\pi _1(\beta _1^{})(\beta _2+\beta _2^{})=0.$$
(14)
The only solution is to put $`\pi _1(\alpha _1^{})=\pi _1(\beta _1^{})=0`$, which corresponds to not rotating $`๐_1`$ at all; or putting $`\alpha _2+\alpha _2^{}`$ or $`\beta _2+\beta _2^{}`$ equal to zero, or making $`\alpha _2+\alpha _2^{}`$ and $`\beta _2+\beta _2^{}`$ collinear. The latter conditions would make the new $`๐_2`$, spanned by $`\alpha _1+\alpha _1^{}`$ and $`\beta _1+\beta _1^{}`$ violate (7). We may also reverse the rรดles of $`๐_1`$ and $`๐_2`$ in the argument. This completes the proof of theorem 1.
The statement that $`๐_1`$ and $`๐_2`$ are fixed by $`G`$ means that the complex structures of $`S_1`$ and $`S_2`$ and uniquely determined by a choice of flux.
The next thing we prove is
###### Theorem 2
The K3 surfaces $`S_1`$ and $`S_2`$ whose complex structures are fixed by $`G`$ are forced to both be attractive.
Before we prove this, we first review the definition of an attractive<sup>4</sup><sup>4</sup>4The standard mathematical term is โsingularโ but as this is such a singularly misleading term, we prefer to follow Mooreโs choice of language from . K3 surface. The Picard lattice of a K3 surface is given by the lattice $`H^{1,1}(S_j)H^2(S_j,)`$. The Picard number $`\rho (S_j)`$ is defined as the rank of this lattice. The surface $`S_j`$ is said to be attractive if $`\rho (S_j)=20`$, the maximal value.
Let us define
$$\mathrm{{\rm Y}}_j=\left(H^{2,0}(S_j)H^{0,2}(S_j)\right)H^2(S_j,),$$
(15)
which is the intersection of the 2-plane $`๐_j`$ with the lattice $`H^2(S_j,)`$ in the space $`H^2(S_j,)`$. For a generic K3 surface $`\mathrm{{\rm Y}}_j`$ will be completely trivial, but the maximal rank of $`\mathrm{{\rm Y}}_j`$ is 2. The โtranscendental latticeโ is defined as the orthogonal complement of the Picard lattice in $`H^2(S_j,)`$. If, and only if, the rank of $`\mathrm{{\rm Y}}_j`$ is 2, the transcendental lattice will coincide with $`\mathrm{{\rm Y}}_j`$ and the K3 surface $`S_j`$ will be attractive. We therefore need to prove that $`\mathrm{{\rm Y}}_j`$ is rank 2.
Let $`e_k^j`$, $`k=1,\mathrm{},22`$ be an integral basis for $`H^2(S_j,)`$. Expanding
$$\begin{array}{cc}\hfill \alpha _j& =\underset{k}{}a_{jk}e_k^j\hfill \\ \hfill \beta _j& =\underset{k}{}b_{jk}e_k^j\hfill \\ \hfill G& =\underset{kl}{}N_{kl}e_k^1e_l^2,\hfill \end{array}$$
(16)
where $`a_{jk}`$ and $`b_{jk}`$ are real numbers and $`N_{kl}`$ are integers (since $`G`$ is an integral 4-form). Then (8) becomes
$$a_{1k}a_{2l}+b_{1k}b_{2l}=N_{kl},\text{for all }k,l.$$
(17)
Fixing $`l`$, the above equation may be read as saying that a real combination of $`\alpha _1`$ and $`\beta _1`$ lies on a lattice point of $`H^2(S_1,)`$. By varying $`l`$ we get 22 different such combinations. The fact that $`\alpha _2`$ and $`\beta _2`$ are linearly independent means that all these lattice points cannot be collinear. Thus $`๐_1`$ contains a 2-dimensional lattice. Similarly $`๐_2`$ contains a 2-dimensional lattice and we complete the proof of theorem 2.
Attractive K3 surfaces were completely classified in . They were shown to be in one-to-one correspondence with $`\mathrm{SL}(2,)`$-equivalence classes of positive-definite even integral binary quadratic forms. Such a quadratic form can be written in terms of a matrix
$$Q=\left(\begin{array}{cc}2a& b\\ b& 2c\end{array}\right),$$
(18)
where $`a,b,c`$, $`a>0`$, $`c>0`$, and $`detQ=4acb^2>0`$. Two forms $`Q`$ and $`Q^{}`$ define an equivalent K3 surface if, and only if, $`Q^{}=M^TQM`$, for some $`M\mathrm{SL}(2,)`$.
Let $`\mathrm{{\rm Y}}_j`$ be spanned (over the integers) by integral vectors $`p_j`$ and $`q_j`$. The above lattice is then
$$Q_j=\left(\begin{array}{cc}p_j^2& p_j.q_j\\ p_j.q_j& q_j^2\end{array}\right),$$
(19)
We are free to rescale $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$ (since they are only defined up to complex multiplication) so that
$$\mathrm{\Omega }_j=p_j+\tau _jq_j,$$
(20)
for a complex number $`\tau _j`$, which is fixed by the condition $`\mathrm{\Omega }_j^2=0`$ to be
$$\tau _j=\frac{p_j.q_j+i\sqrt{detQ_j}}{q_j^2}.$$
(21)
Note that this choice of rescaling means we cannot now assume $`\gamma =1`$ in (4). We then obtain
$$G=\left(\mathrm{Re}(\gamma )p_1p_2+\mathrm{Re}(\gamma \tau _1)q_1p_2+\mathrm{Re}(\gamma \overline{\tau }_2)q_2p_1+\mathrm{Re}(\gamma \tau _1\overline{\tau }_2)q_1q_2\right).$$
(22)
Consider the condition imposed by the integrality of $`G`$. Since $`p_1p_2`$ is integral and primitive<sup>5</sup><sup>5</sup>5Primitive in the sense that it is not an integral multiple of a lattice element. we must have $`\mathrm{Re}(\gamma )`$. The other terms on (22) put further conditions of $`\gamma `$. It is easy to show that a consistent choice of $`\gamma `$ making each term in (22) integral is possible if and only if $`\sqrt{det(Q_1Q_2)}`$ is an integer. That is,
###### Theorem 3
A pair of attractive K3 surfaces $`S_1`$ and $`S_2`$ will correspond to a choice of integral $`G`$-flux if and only if $`det(Q_1Q_2)`$ is a perfect square.
Finally we need to impose the tadpole condition $`\frac{1}{2}G^2=24`$. We compute
$$G^2=\frac{1}{4}(\gamma \mathrm{\Omega }_1\overline{\mathrm{\Omega }}_2+\overline{\gamma }\mathrm{\Omega }_2\overline{\mathrm{\Omega }}_1)^2,$$
(23)
and use the fact that $`\mathrm{\Omega }_j^2=0`$ so only the cross term in the square is not vanishing. Therefore
$$G^2=\frac{1}{2}|\gamma |^2\mathrm{\Omega }_1\overline{\mathrm{\Omega }}_1\mathrm{\Omega }_2\overline{\mathrm{\Omega }}_2.$$
(24)
Using $`\mathrm{\Omega }_j=p_j+\tau _jq_j`$ we find
$$\frac{1}{2}G^2=\frac{|\gamma |^2det(Q_1Q_2)}{q_1^2q_2^2}=24.$$
(25)
Solving (25) together with the integrality of (22) provides all possibilities of flux compactifications with $`G=G_1`$. We may prove that there is a finite number of attractive K3 surfaces that yield solutions to this equation as follows. We use the following theorem from :
###### Theorem 4
In the equivalence class of the matrix (18) under the action of $`\mathrm{SL}(2,)`$, assuming that $`det(Q)`$ is not a perfect square, one can always find a representative matrix satisfying
$$|b||c||a|.$$
(26)
In our case $`det(Q)`$ is negative and so, clearly, not a perfect square. Thus we may restrict attention to matrices satisfying the above bounds. Putting
$$Q_1=\left(\begin{array}{cc}2a& b\\ b& 2c\end{array}\right),Q_2=\left(\begin{array}{cc}2d& e\\ e& 2f\end{array}\right),$$
(27)
yields
$$det(Q_1)=4acb^24acac=3ac.$$
(28)
Similarly, $`det(Q_2)3df`$. Thus (25) yields
$$\frac{1}{2}G^2\frac{9}{4}|\gamma |^2ad.$$
(29)
Suppose $`\mathrm{Re}(\gamma )0`$. Then $`|\gamma |^21`$, since $`\mathrm{Re}(\gamma )`$. In this case we are done since $`a`$ and $`d`$ are positive integers and $`b,c,e,f`$ are constrained by theorem 26. On the other hand, if $`\gamma `$ is purely imaginary, then the integrality of the second term in (22) forces
$$\frac{\mathrm{Im}(\gamma )\sqrt{det(Q_1)}}{2c}.$$
(30)
Obviously $`\mathrm{Im}(\gamma )`$ cannot be zero since then $`|\gamma |^2`$ would be zero. We then have
$$\frac{1}{2}G^2\frac{cdet(Q_2)}{f}3dc.$$
(31)
This bounds $`c`$ and $`d`$. Similarly we may use the third term in (22) to bound $`a`$ and $`f`$. Thus we complete the proof that there are only a finite number of possibilities for $`a,b,c,d,e,f`$, and thus only a finite number of attractive K3 surfaces whenever $`G^2`$ is bounded.
In fact, it is not hard to perform a computer search to yield the full list of possibilities. For $`\frac{1}{2}G^2=24`$, i.e., $`G_0=0`$, there are 13 possibilities up to $`\mathrm{SL}(2,)`$ equivalence which we list in table 1. The column labeled โO?โ will be explained in section 2.3. In principle, a given pair of attractive K3 surfaces might admit many, but finitely many, inequivalent choices of $`G`$. In our case, where the numbers are quite small, this never happens.
One may, of course, obtain other possibilities by considering a nonzero $`G_0`$. In this case, we solve the same problem for $`\frac{1}{2}G^2<24`$. As one might expect, the number of possibilities for a given $`\frac{1}{2}G^2<24`$ are somewhat fewer than above.
As stated above, the complex structures of $`S_1`$ and $`S_2`$ are fixed. What remains unfixed is the Kรคhler form and $`B`$-field degree of freedom. Using the assumption $`G_0=0`$ in (3), all 20 such complex degrees of freedom remain undetermined by the fluxes. A non-trivial choice of $`G_0`$ will fix some of these 20 remaining moduli.
The choice of $`G`$ fixes a 2-plane within $`\mathrm{\Pi }_j`$ spanned by the real and imaginary parts of $`\mathrm{\Omega }_j`$. This means that the moduli space (5) is reduced to
$$\mathrm{M}_j\mathrm{O}(\mathrm{\Gamma }_{2,n_j})\backslash \mathrm{O}(2,n_j)/(\mathrm{O}(2)\times \mathrm{O}(n_j)).$$
(32)
If $`G_0=0`$ then $`n_1=n_2=20`$. If $`G_0`$ is nonzero, these numbers will decrease.
### 2.3 The Orientifold
Now let us turn our attention to the related question of orientifolds on $`\mathrm{K3}\times (T^2/_2)`$. One obtains this orientifold via F-theory.
Begin with M-theory on $`S_1\times S_2`$ (ignoring flux for now) to obtain an $`N=4`$ theory in three dimensions as above. Assume that $`S_2`$ is an elliptic K3 surface with a section. Let $`\pi :S_2B`$ denote this elliptic fibration of $`S_2`$. By shrinking the area of the elliptic fibre, one moves to an F-theory fibration corresponding to a type IIB compactification on $`S_1\times B`$. This yields a four-dimensional $`N=2`$ compactification.
This four-dimensional theory can be compactified on a circle thus regaining the three-dimensional theory we had originally from the M-theory compactification. The relationship between the moduli spaces of the three-dimensional theory and four-dimensional theory can be understood from this fact. The moduli space of the three-dimensional theory is $`\mathrm{M}_1\times \mathrm{M}_2`$, where each $`\mathrm{M}_j`$ is quaternionic Kรคhler. The four-dimensional theory has a moduli space $`\mathrm{M}_H\times \mathrm{M}_V`$, where $`\mathrm{M}_H`$, the hypermultiplet moduli space, is exactly $`\mathrm{M}_1`$.
The vector multiplet moduli space $`\mathrm{M}_V`$, is special Kรคhler. The complex dimension of $`\mathrm{M}_V`$ is one less than the quaternionic dimension of $`\mathrm{M}_2`$. Quantum corrections make for a very complicated relationship between $`\mathrm{M}_V`$ and $`\mathrm{M}_2`$. Let us ignore these quantum corrections for now, which we may do since we are only making qualitative statements about the moduli space. In this case, ignoring any flux effects or M2-branes, we have, locally
$$\mathrm{M}_2=\frac{\mathrm{O}(4,20)}{\mathrm{O}(4)\times \mathrm{O}(20)},$$
(33)
and, from the $`c`$-map
$$\mathrm{M}_V=\frac{\mathrm{O}(2,18)}{\mathrm{O}(2)\times \mathrm{O}(18)}\times \frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}.$$
(34)
The first factor of (34) corresponds to the complex structure moduli space of $`S_2`$ if we declare $`S_2`$ to be an elliptic fibration with a section. In F-theory language, this corresponds to the moduli space of the location of 7-branes. The second factor of (34) would naรฏvely correspond to the complexified area of the base, $`B`$, of the elliptic fibration as this is the only modulus remaining once the fibre is shrunk to zero size. When moving between dimensions one must be careful with taking into account overall scalings of the metric. The result is that the second factor of (34) actually corresponds to the complexified volume of the K3 surface $`S_1`$. The area of the base becomes a parameter in the hypermultiplet moduli space $`\mathrm{M}_H`$. We refer to for more details.
Sen showed how type IIB orientifolds could be obtained from F-theory compactifications. Elliptic fibrations may contain โbad fibresโ, i.e., fibres which are not elliptic curves. These bad fibres have been classified by Kodaira. We refer to , for example, for a review. In Senโs analysis one takes a limit in the moduli space of complex structures of the elliptic fibration such that all the bad fibres become type $`\mathrm{I}_0^{}`$ in the Kodaira classification. We now have a type IIB string compactified on the orientifold $`S_1\times (C/_2)`$, where $`C`$ is an elliptic curve and the base of the elliptic fibration is $`BC/_2`$.
This limit freezes the location of the F-theory 7-branes making the moduli space locally
$$\mathrm{M}_V=\frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}\times \frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}\times \frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}.$$
(35)
These three complex moduli can be identified as
* The modulus of the F-theory elliptic fibre โ i.e., the axion-dilaton of the type IIB string.
* The modulus of the elliptic curve $`C`$, where the base of the elliptic fibration of $`S_2`$ is $`BC/_2`$.
* The complexified volume of $`S_1`$, as above.
The 16 moduli that we have โlostโ in passing from (34) to (35) are regained by allowing D7-branes to move away from the 4 O7-planes.
Now consider the effect of flux in the form of $`G_1`$ so as to yield an $`N=1`$ supersymmetric theory in four dimensions. This flux fixes the complex structure of $`S_1`$ and $`S_2`$ making both of these K3 surfaces attractive. First note that any attractive K3 surface is elliptic with a section so our condition for an F-theory limit is automatically satisfied.
The flux causes the dimension of $`\mathrm{M}_H\mathrm{M}_1`$ to be halved โ exactly as it was in the case of M-theory in section 2.2. For $`\mathrm{M}_V`$, the first factor of (34) is a complex structure moduli space and so disappears completely. In orientifold language, we fix the dilaton-axion, the complex structure of $`CT^2`$, and the location of all the D7-branes. All that remains unfixed in $`\mathrm{M}_V`$ is a single complex modulus corresponding to the complexified volume of $`S_1`$.
Senโs orientifold limit of F-theory is a limit of complex structure, but once we turn on flux, we have no deformations of the complex structure! The only way our M-theory compactification can correspond to an orientifold is if the elliptic fibration of the attractive $`S_2`$ has this fibration structure to begin with.
So let us suppose $`S_2`$ is an attractive K3 surface which is an elliptic fibration with only smooth, or type $`\mathrm{I}_0^{}`$ fibres. The base of such a fibration must be $`^1`$ and there must be exactly four $`\mathrm{I}_0^{}`$ fibres and no other singular fibres. In this case, the $`J`$-invariant of the fibre has no zeros or poles and is therefore a constant. This is exactly the same elliptic fibration data as one would obtain for a K3 surface which is a โKummer surfaceโ, i.e., a blow-up of a quotient $`A/_2`$, where $`A`$ is a 4-torus (or abelian surface to be more precise). It follows that $`S_2`$ is indeed a Kummer surface (following, for example, proposition 2.7 of ).
Any Kummer surface, which is attractive, must be a $`_2`$-quotient of an attractive abelian surface (see, for example, equation (5.8) of ). Such abelian surfaces are classified in much the same way as attractive K3 surfaces, that is, they are again in one-to-one correspondence with $`\mathrm{SL}(2,)`$-equivalence classes of positive-definite even integral binary quadratic forms. If $`Q`$ is the matrix associated with the binary quadratic form of our attractive Kummer surface $`S_2`$ and $`R`$ is the matrix associated with the attractive abelian surface $`A`$, then one can show that (see , for example)
$$Q=2R.$$
(36)
It follows that the attractive K3 surface $`S_2`$ is a Kummer surface if, and only if, the associated even binary quadratic form is twice another even binary quadratic form. Only the F-theory compactifications on $`\mathrm{K3}\times \mathrm{K3}`$ which satisfy this property will have orientifold interpretations.
Looking back at table 1, we see that 8 of our 13 possibilities admit an orientifold interpretation. The column headed โO?โ denotes whether an orientifold model exists.
One might be concerned that one should check that $`G`$ is compatible with the F-theory limit as spelt out in . That is, $`G`$ should have โone legโ in the fibre direction. This condition turns out to be automatically satisfied, at least in the case $`G_0=0`$, as we explain as follows.
The spectral sequence for the cohomology of a fibration yields
$$H^2(S,)=H^0(B,R^2\pi _{})H^1(B,R^1\pi _{})H^2(B,\pi _{}),$$
(37)
where $`H^p(B,R^q\pi _{})`$ may be schematically viewed as a form with $`p`$ legs in the base direction and $`q`$ legs in the fibre direction.
The term $`H^2(B,\pi _{})`$ is dual to the base $`B^1`$. The term $`H^0(B,R^2\pi _{})`$ is dual to the fibres including components of singular fibres. Both of these terms correspond to curves in $`S`$ and thus forms of type (1,1). Therefore any form of type (0,2) or (2,0) must be contained $`H^1(B,R^1\pi _{})`$. It follows that $`G`$ has one leg in the fibre direction assuming $`G=G_1`$ in (3).
Any attractive abelian surface $`A`$ must be of the form $`C\times C^{}`$, where $`C`$ and $`C^{}`$ are isogenous elliptic curves admitting complex multiplication . Here, โisogenousโ means that $`C^{}`$ is isomorphic, as an elliptic curve, to a free quotient of $`C`$ by any finite subgroup of $`\mathrm{U}(1)\times \mathrm{U}(1)`$. We refer to and references therein for a nice account of complex multiplication.
The elliptic fibration of the Kummer surface $`S_2`$ will therefore be an elliptic fibration with base $`C/_2`$ with fibre $`C^{}`$. It follows from Senโs argument that F-theory on $`S_1\times S_2`$ is equivalent to the type IIB orientifold on $`S_1\times (C/_2)`$ where the dilaton-axion of the type IIB theory is given by the $`\tau `$-parameter of the elliptic curve $`C^{}`$.
The fixing of the complex structures of $`C`$ and $`C^{}`$ account for the removal of the first two factors of (35) in the vector multiplet moduli space. The fact that $`C`$ and $`C^{}`$ are isogenous means that their $`\tau `$-parameters will be related by an $`\mathrm{GL}(2,)`$ transformation. In other words,
$$\tau _C^{}=\frac{a\tau _C+b}{c\tau _C+d},$$
(38)
where $`a,b,c,d`$ are integers not necessarily satisfying $`adbc=1`$.<sup>6</sup><sup>6</sup>6There is an example in which appears to violate this condition. This is because the basis defined in the appendix of is not a valid integral basis for $`H^2(S,)`$ and so the resulting $`G`$ is not actually in integral cohomology.
We should note the fact that an attractive abelian surface may, in general, be decomposed into $`C\times C^{}`$ in many inequivalent ways (other than the trivial exchange of $`C`$ and $`C^{}`$). Thus, a fixed $`S_1\times S_2`$ might be associated to none, or many inequivalent orientifold limits. An algorithm for determining a complete set of such factorizations was presented in . For example, if the abelian surface corresponds to
$$R=\left(\begin{array}{cc}12& 6\\ 6& 12\end{array}\right),$$
(39)
then one may factorize into a pair of elliptic curves with $`\tau _C=\omega `$ and $`\tau _C^{}=6\omega `$; or $`\tau _C=2\omega `$ and $`\tau _C^{}=3\omega `$, where $`\omega =\mathrm{exp}(2\pi i/3)`$. In our cases, listed in table 1, such an ambiguity never occurs.
## 3 Instanton Corrections
So far we have completely ignored any quantum corrections to the moduli space. Consider first the case of M-theory on $`S_1\times S_2`$ where the flux does not break any supersymmetry. This yields an $`N=4`$ theory in three dimensions. By the usual counting, any instanton solution that breaks half the supersymmetry will modify the prepotential and thus deform the metric on the moduli space. These instantons will not obstruct any moduli and the dimension of the moduli space will be unchanged by these quantum corrections.
The only source of such instanton corrections in M-theory will correspond to M5-brane instantons wrapping holomorphically embedded complex 3-folds within $`S_1\times S_2`$. Such divisors are clearly of the form $`S_1\times C_g`$, or $`C_g\times S_2`$, where $`C_g`$ is an algebraic curve of genus $`g`$.
Following, , one can show that these divisors will only contribute nontrivially to the prepotential if they have holomorphic Euler characteristic $`\chi _\mathrm{O}=2`$. Since $`\chi _\mathrm{O}(\mathrm{K3}\times C_g)=2(1g)`$, we see that our instantons must be of the form $`S_1\times ^1`$ or $`^1\times S_2`$.
Now suppose we turn flux on so as to break half the supersymmetry. The superpotential of the resulting low-energy effective theory will now receive instanton corrections from M5-branes wrapping divisors. A naรฏve interpretation of would lead one to believe that one would look for divisors with $`\chi _\mathrm{O}=1`$. There are no such divisors in $`S_1\times S_2`$ and so one would arrive at the conclusion that the Kรคhler moduli cannot be removed.
This is not the case however. It was shown in that some fermion zero modes on the M5-brane worldvolume are lost changing the counting argument of . The result is that, with the $`G`$-flux we are using, the desired instantons should have $`\chi _\mathrm{O}=2`$. That is, the instantons which contribute to the superpotential are precisely those wrapping $`S_1\times ^1`$ or $`^1\times S_2`$.
As discussed in the previous section, the complex structure on $`S_1`$ and $`S_2`$ is completely fixed by the choice of $`G`$-flux. Each K3 surface is attractive and, as such has Picard number equal to 20. This leaves each K3 surface with 20 complexified Kรคhler form moduli. If $`G`$ is purely of the form $`G=G_1=\mathrm{Re}(\mathrm{\Omega }_1\overline{\mathrm{\Omega }}_2)`$, then these 20 moduli are unfixed by the fluxes. Any terms from $`G_0`$ in (3) will fix some of these remaining 20 moduli.
In any case, at least in the supergravity approximation, one cannot remove all of these Kรคhler moduli by fluxes. It is possible to fix at least 10 of the Kรคhler moduli but in the F-theory limit one is restricted to fixing only 2 Kรคhler moduli using $`G_0`$ effects. It is conceivable that going beyond the supergravity approximation may change such statements as discussed in .
Let $`S`$ be an attractive K3 surface and let $`V=\mathrm{Pic}(S)=^{20}`$ be the subspace of $`H^2(S,)`$ spanned by the Kรคhler form. We wish to find a convenient basis for $`V`$. Let us consider an element of $`H^2(S,)`$ as a homomorphism from 2-chains in $`S`$ to $``$. If $`\alpha H^2(S,)`$ and $`x`$ is a 2-chain, we thus denote $`\alpha (x)`$.
The following proposition will be useful
###### Proposition 1
We may find a set $`\{e_1,\mathrm{},e_{20}\}`$ of holomorphically embedded $`^1`$โs in $`S`$ and a basis $`\{\xi _1,\mathrm{},\xi _{20}\}`$ of $`V`$ such that $`\xi _a(e_b)=\delta _{ab}`$.
To see this we use the fact that any attractive $`S`$ is an elliptic fibration $`\pi :SB`$ with at least one section as noted in section 2.3. Now take any rational curve (i.e., holomorphically embedded $`^1`$) $`CS`$. Since $`\pi `$ is a holomorphic map, the image of $`C`$ under $`\pi `$ is either a point or all of $`B`$. In the former case $`C`$ is a component of a singular fibre and in the latter case $`C`$ is a โsectionโ (or multisection) of the fibration.
In theorem 1.1 of it is shown that the complete Picard lattice of an elliptic surface with a section is generated by rational combinations of sections, smooth fibres and components of singular fibres. If there is at least one bad fibre which is reducible, the smooth elliptic fibre itself is homologous to a sum of smooth rational curves. This is indeed the case for attractive K3 surfaces as shown in . The proposition then follows.
An instanton correction to the superpotential from an M5-brane wrapping a divisor $`D`$ will be of the form $`f\mathrm{exp}(\mathrm{Vol}(D))`$, where $`\mathrm{Vol}(D)`$ is the complexified volume of $`D`$. The coefficient $`f`$ may depend on complex structure moduli but cannot depend on the Kรคhler moduli. This is because $`f`$ is computed perturbatively and the โaxionicโ shift symmetry of the complex partner to the Kรคhler form prevents any contribution to perturbation theory.
Using the bases $`\{e_1^{(1)},\mathrm{},e_{20}^{(1)}\}`$ for $`H_2(S_1)`$ and $`\{e_1^{(2)},\mathrm{},e_{20}^{(2)}\}`$ for $`H_2(S_2)`$ from our proposition we have volumes of the form
$$\begin{array}{cc}& \mathrm{Vol}(S_1)\mathrm{Area}(e_a^{(2)})\hfill \\ & \mathrm{Area}(e_a^{(1)})\mathrm{Vol}(S_2).\hfill \end{array}$$
(40)
The volume of $`S_j`$ is determined from the Kรคhler form which is determined by the areas of the $`^1`$โs. Proposition 1 then implies we have 40 independent functions on 40 variables. If the superpotential is a suitably generic function then we therefore expect classical vacua to be isolated in the Kรคhler moduli space. That is, we fix all the moduli.
There are two known effects that can spoil the genericity of an instanton contribution and make it vanish. Firstly, the instanton may have a moduli space of vanishing Euler characteristic in some sense. This is not true in our case as rational curves in K3 surface are always isolated. The second effect can be caused by fluxes as we now discuss.
### 3.1 Obstructed Instantons
Let $`D`$ be a threefold corresponding to a potential instanton $`S_1\times ^1`$ or $`^1\times S_2`$. Without loss of generality, we assume the instanton is of the form $`C_1\times S_2`$ from now on, with $`C_1^1`$. Let $`i:DS_1\times S_2`$ be the embedding. The term
$$_Db_2i^{}G,$$
(41)
in the M5-brane worldvolume action induces a tadpole for the anti-self-dual 2-form $`b_2`$ if $`i^{}G0`$. We therefore demand that $`i^{}G=0`$ is a necessary condition for any divisor $`D`$ to be considered an instanton.
How strong is the constraint $`i^{}G=0`$? Let us first consider the supersymmetry-breaking part of the flux $`G_1=\mathrm{Re}(\mathrm{\Omega }_1\overline{\mathrm{\Omega }}_2)`$. Viewing $`GH^4(S_1\times S_2,)`$ as a homomorphism from chains in $`S_1\times S_2`$ to $``$, we may write
$$i^{}G(x)=G(i(x)),$$
(42)
where $`x`$ is a 4-chain on $`D`$. Purely on dimensionality grounds, from (3), it is easy to see that, if $`G(i(x))0`$, then $`x`$ must be mapped under $`i`$ to a 2-chain on $`S_1`$ and a 2-chain on $`S_2`$. We therefore suppose that $`i(x)C_1\times C_2`$, for some 2-cycle $`C_2S_2`$. But then $`G(i(x))=0`$ since $`\mathrm{\Omega }_1`$ is of type $`(2,0)`$ and therefore must vanish on any $`^1`$ (as the latter is dual to a (1,1)-form). This means that none of our instantons are ruled out by this part of the $`G`$-flux.
Now let us consider the case where $`G_0`$ is nonzero and given by (4). These fluxes will fix some of the 20 Kรคhler moduli. The primitivity condition for $`G`$ means that $`J_j`$ will be a valid Kรคhler form for $`S_j`$ only if $`J_j`$ is perpendicular all the $`\omega _j^{(\alpha )}`$โs.<sup>7</sup><sup>7</sup>7Here we have mentioned only the real Kรคhler form. The complex partner of the Kรคhler form is similarly obstructed as discussed in , for example. Let is denote this space of Kรคhler forms $`V_j^0H^2(S_j,)`$. That is,
$$V_j^0=\underset{\alpha }{}\omega _{j}^{(\alpha )}{}_{}{}^{},$$
(43)
where the perpendicular complement is taken with respect to $`\omega _j^{(\alpha )}`$ in the 20 dimensional space $`\mathrm{Pic}(S_j)`$.
Such a nonzero $`G_0`$ will also rule out certain instantons. Consider a 4-cycle $`xC_1\times C_2`$, where both $`C_j`$โs are rational curves in $`S_j`$ and let $`\xi _j`$ denote the Poincarรฉ dual of $`C_j`$. Then
$$G(i(x))=\underset{\alpha }{}(\omega _1^\alpha .\xi _1)(\omega _2^\alpha .\xi _2)$$
(44)
The instanton $`C_1\times S_2`$ is therefore only valid (i.e., $`i^{}G=0`$) if $`\xi _1`$ is orthogonal to all the $`\omega _1^{(\alpha )}`$โs. That is,
$$\xi _1V_1^0.$$
(45)
Our instantons only contribute nontrivially to the superpotential if they correspond to $`^1\times \mathrm{K3}`$, where we assume the $`^1`$ is holomorphically and smoothly embedded in the K3 surface. That is, the $`^1`$ is a rational curve. Fortunately these rational curves can be categorized using properties of the lattice at hand. Any rational curve in a K3 surface has self-intersection $`2`$. This means it is Poincarรฉ dual to an element of the lattice $`H^2(S_j,)`$ of length squared $`2`$. Conversely, if $`\xi `$ is an element of length squared $`2`$ in $`H^2(S_j,)`$ then either $`\xi `$ or $`\xi `$ is Poincarรฉ dual to a rational curve.
This leads to the following:
###### Theorem 5
If $`G_0`$ is zero we generically fix all moduli. With a nonzero $`G_0`$, instanton effects will generically fix all moduli if, and only if, the spaces $`V_1^0`$ and $`V_2^0`$ defined in (43) are spanned by elements corresponding to rational curves. That is the $`V_j^0`$โs are spanned by elements in $`V_j^0H^2(S_j,)`$ of length squared $`2`$.
In simple cases, all the moduli are fixed. For example, suppose $`M=1`$ in (4) and $`(\omega _1^{(1)})^2=2`$. Suppose further that that Picard lattice contains a copy of the (negated) $`E_8`$ lattice as a summand and that $`\omega _1^{(1)}`$ is an element of this lattice. Then the orthogonal complement of this vector will be the $`E_7`$ lattice which is generated by vectors of length squared $`2`$.
It would be interesting to find examples where the moduli are not all fixed by instanton effects. This would involve analyzing sublattices in $`H^2(S,)`$ which are not generated by vectors of length squared $`2`$.
### 3.2 The Orientifold Limit
By going to the F-theory limit we may obtain the equivalent statement about instanton effects in the orientifold on $`\mathrm{K3}\times (T^2/_2)`$. Begin with M-theory on $`S_1\times S_2`$, where $`S_2`$ is an elliptic fibration. Let the area of the generic elliptic fibre be $`A`$. To take the F-theory limit we set $`A0`$.
The rescaling involved in this limit means that the volume of the M5-brane instanton must scale as $`A`$, as $`A0`$, in order that this instanton has a nontrivial effect . It follows that the instanton must either wrap an elliptic fibre, or a component of a bad fibre.
The instantons corresponding to $`^1\times S_2`$ indeed wrap the fibre and so descend to D3-brane instantons wrapped around $`^1\times (T^2/_2)`$ in the F-theory limit. The instantons corresponding to $`S_1\times ^1`$ will be trivial unless the $`^1`$ corresponds to a component of a bad fibre. In this case, the D3-brane instanton becomes wrapped on $`S_1\times \mathrm{pt}`$.
The moduli fixing then proceeds in the same way as it did for M-theory. Unless an inauspicious choice of $`G`$-flux is used, all the moduli should be fixed by instanton effects as follows. After flux was applied, the single remaining modulus in $`\mathrm{M}_V`$ corresponded to the volume of $`S_1`$. Clearly this is fixed by the instantons wrapping $`S_1\times \mathrm{pt}`$. The remaining moduli correspond to the areas of rational curves in $`S_1`$ and the area of $`T^2/_2`$. Given that the volume of $`S_1`$ has been fixed, we have precisely the right number of independent constraints from the $`^1\times (T^2/_2)`$ instantons to fix these latter moduli.
The fact that the single vector multiplet corresponding to the volume of $`S_1`$ is fixed was also observed in , where a more quantitative analysis was performed using duality.
## 4 Discussion
If one considers M-theory on $`\mathrm{K3}\times \mathrm{K3}`$ with no M2 branes and a flux chosen to break supersymmetry down to $`N=2`$ in three dimensions, then the complex structures of the two K3 surfaces are fixed. To be precise, the two K3 surfaces are both attractive K3 surfaces. There remain 20 complex moduli associated to each K3 surface which vary the Kรคhler form.
If we leave the 40 moduli unfixed by fluxes, then we have argued that generically one would expect instanton effects to fix all 40. If flux is used to fix further moduli then we showed that there is a possibility that some moduli can remain unfixed by instanton effects.
The obvious next step should be to compute these instanton effects more explicitly and determine the values of the moduli. This might be a difficult exercise for the following reasons.
Before the flux was turned on we have an $`N=4`$ supersymmetric theory in three dimensions. Corrections from M5-brane instantons will effect the metric on the moduli space. The moduli space of this theory is a product of quaternionic Kรคhler moduli spaces. It is a well-known difficult problem in string theory to determine the form of such quaternionic Kรคhler moduli spaces when there are nontrivial instanton corrections. The problem of studying M5-brane instantons corrections to the moduli space is exactly equivalent to studying worldsheet instanton corrections to the heterotic string on a K3 surface. Preliminary analysis in the latter was done in , for example, but few concrete results have been attained.
It should be emphasized that, even though there has been much interesting progress on quaternionic Kรคhler manifolds (such as ), these results tend to rely on the assumption that there is an isometry in the moduli space related to translations in the RR directions. This views the hypermultiplet moduli space as a fibration over some special Kรคhler base with a toroidal fibre given by the RR moduli. It is known (see , for example) that when non-perturbative corrections are taken into account, this fibration must have โbad fibresโ. These bad fibres will break these isometries in much the same way as an elliptic K3 surface has no isometries related to translation in the fibre direction. An interesting proposal for analyzing instanton effects on the hypermultiplet moduli space was given in but it appears to rely on the existence of these isometries.
Now when we turn the flux on, the M5-brane instantons contribute to a superpotential, rather than the moduli space metric. This does not mean that the metric remains uncorrected however. Now, with the decreased supersymmetry, quantum corrections to the metric are less constrained and even more difficult to determine than if they arose purely from instantons. We see, therefore, that computing the superpotential directly from instanton computations may be very difficult.
Even without this detailed knowledge, however, we have shown that one should expect a number of flux compactifications associated to M-theory on $`\mathrm{K3}\times \mathrm{K3}`$ (or its equivalent orientifold $`\mathrm{K3}\times (T^2/_2)`$) where all the moduli are fixed by the combined action of the flux and the instanton effects.
In the context of the F-theory limit, which is equivalent to an orientifold of the type IIB string on $`\mathrm{K3}\times (T^2/_2)`$ the result of this paper shows that the goal of fixing all moduli in this model is now accomplished.<sup>8</sup><sup>8</sup>8In the counting of fermionic zero modes on D3 brane is performed which leads to an analogous result. The first part, namely fixing the moduli by fluxes, was achieved in and a nice summary of this work was presented in . In absence of D3-branes, the 18+1 complex moduli โunfixableโ by fluxes span the scalar manifold
$$\mathrm{M}_{_{\text{unfixed}}}^{^{\text{min}}}=\frac{\mathrm{O}(2,18)}{\mathrm{O}(2)\times \mathrm{O}(18)}\times \frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}.$$
(46)
Here the first factor includes 18 complex fields, the remnant of the $`N=2`$ hypermultiplets. It includes the area of $`(T^2/_2)`$ and other hypermultiplets. The second factor is the remnant of the $`N=2`$ vector multiplet and it describes the volume of the K3 surface $`S_2`$ and its axionic partner. This case in our setting requires that both $`G_1`$ and $`G_0`$ are non-vanishing. We have to be careful therefore and comply with the conditions of the theorem 5, where it is explained that only certain choice of fluxes $`G_0`$ will allow us enough freedom (18+1 choice of proper 4-cycles) to fix by the instantons all remaining 18+1 complex moduli in (46).
An even simpler case, from the perspective of instantons, is when we introduce only $`G_1`$ flux (breaking $`N=2`$ into $`N=1`$ supersymmetry) will leaves us with the 20+1 complex moduli unfixed by fluxes. They span the scalar manifold
$$\mathrm{M}_{_{\text{unfixed}}}=\frac{\mathrm{O}(2,20)}{\mathrm{O}(2)\times \mathrm{O}(18)}\times \frac{\mathrm{SL}(2,)}{\mathrm{U}(1)}.$$
(47)
In such case we simply have 20+1 choices for the D3 instantons wrapping the 4-cycles in $`\mathrm{K3}\times (T^2/_2)`$ and all unfixed by fluxes moduli are fixed by instantons.
The whole story of fixing all moduli in the M-theory version of this model, compactified on $`\mathrm{K3}\times \mathrm{K3}`$ is incredibly simple and elegant. In the compactified three-dimensional model there are no vectors. Therefore without fluxes, we have two 80-dimensional quaternionic Kรคhler spaces, one for each $`\mathrm{K3}`$. With non-vanishing $`G_1`$ flux, each $`\mathrm{K3}`$ becomes an attractive $`\mathrm{K3}`$, one-half of all the moduli are fixed, but 40 in each $`\mathrm{K3}`$ still remain moduli and need to be fixed by instantons. There are 20 proper 4-cycles in each $`\mathrm{K3}`$ and they provide instanton corrections from M5-branes wrapped on these cycles: the moduli space is no more.
## Acknowledgments
We wish to thank N. Bliznashki, S. Ferrara, L. Fidkowski, C. Haase, S. Kachru, A.-K. Kashani-Poor, D. Kraines, D. Morrison, G. Moore, S. Sethi, A. Tomasiello and S. Trivedi for useful conversations. P.S.A. is supported in part by NSF grant DMS-0301476, Stanford University, SLAC and the Packard Foundation. R.K. is supported by NSF grant 0244728.
|
warning/0506/math0506309.html
|
ar5iv
|
text
|
# Injective Envelopes of Separable Cโ-algebras
## 1. Preliminary Results
### 1.1. Terminology and notation
As usual, we will denote by $`B(H)`$ and $`K(H)`$ the set of bounded and compact operators on a Hilbert space $`H`$. Because the algebras under study are not represented in any particular way as acting on a Hilbert space, we shall employ the following terminology. A C-algebra $`B`$ is said to be a $`W^{}`$-algebra if, as a Banach space, $`B`$ is the dual space $`X^{}`$ of some (in fact, unique) Banach space $`X`$. It is a classical fact \[23, Theorem III.3.5\] that a C-algebra $`B`$ is a W-algebra if and only if $`B`$ has a representation as a von Neumann algebra of operators acting on some Hilbert space. A C-algebra $`B`$ is called an AW-algebra if the left annihilator of each right ideal in $`A`$ is of the form $`Ap`$ for some projection $`pA`$. Although every W-algebra is an AW-algebra, the converse is not true: there exist AW-algebras that fail that have any faithful representation as a von Neumann algebra.
If $`B`$ is an AW-algebra, then $`pq`$ denotes the Murrayโvon Neumann equivalence of projections $`p`$ and $`q`$ in $`B`$. Thus, a projection $`pB`$ is finite if $`qp`$ and $`qp`$ only if $`q=p`$; otherwise $`p`$ is an infinite projection. If the identity $`1B`$ is a finite projection, then $`B`$ is said to be finite algebra. A projection $`pB`$ is abelian if the AW-algebra $`pBp`$ is commutative.
An AW-algebra $`B`$ is said to be: of type I if every direct summand of $`B`$ has an abelian projection; of type II if $`B`$ has no abelian projections but every direct summand has a finite projection; and of type III if all projections in $`B`$ are infinite. If the centre $`๐ต(B)`$ of an AW-algebra is $``$, then $`B`$ is a factor. Type I AW-algebras are of considerable interest herein. In particular, type I AW-algebras are injective C-algebras and type I AW-factors are of the form $`B(H)`$ .
AW-algebras differ from W-algebras in that the former can fail to have any normal states. An AW-algebra $`B`$ is wild if the only normal positive linear functional $`\phi `$ on $`B`$ is $`\phi =0`$. Every AW-factor is either a W-algebra or a wild AW-algebra .
A C-algebra $`A`$ is said to be postliminal (or type I, or GCR) if every representation of $`A`$ generates a type I von Neumann algebra, and $`A`$ is liminal (or CCR) if every irreducible representation $`\pi :AB(H)`$ satisfies $`\pi (A)=K(H)`$. An elementary C-algebra is one that $``$-isomorphic to $`K(H)`$ for some Hilbert space $`H`$.
We shall employ the following notation from . If $`\{E_\alpha \}_{\alpha \mathrm{\Lambda }}`$ is a family of operator systems, then
$$\begin{array}{ccc}\hfill \underset{\alpha \mathrm{\Lambda }}{}E_\alpha & =& \{(e_\alpha )_\alpha :e_\alpha E_\alpha \text{ and }sup_\alpha e_\alpha <\mathrm{}\};\hfill \\ \hfill \underset{\alpha \mathrm{\Lambda }}{}E_\alpha & =& \{(e_\alpha )_\alpha :e_\alpha E_\alpha \text{ and }\epsilon >0\hfill \\ & & \text{ only finitely many }e_\alpha \text{ satisfy }e_\alpha >\epsilon \}.\hfill \end{array}$$
Note that if $`\{A_\alpha \}_{\alpha \mathrm{\Lambda }}`$ is a family of C-algebras, then $`_\alpha A_\alpha `$ and $`_\alpha A_\alpha `$ are C-algebras and $`_\alpha A_\alpha `$ is an ideal of $`_\alpha A_\alpha `$. As operator systems can always be realised as $``$-closed, unital subspaces of unital C-algebras, $`_{\alpha \mathrm{\Lambda }}E_\alpha `$ is an operator system for every family $`\{E_\alpha \}_{\alpha \mathrm{\Lambda }}`$ of operator systems.
### 1.2. Injective envelopes
An operator system $`I`$ is injective if for every inclusion $`EF`$ of operator systems each completely positive linear map $`\omega :EI`$ has a completely positive extension to $`F`$. Arvesonโs extension theorem for completely positive linear maps with values in $`B(H)`$ demonstrates that $`B(H)`$ is injective. This fact can be used to show that if an operator system $`I`$ is represented as a unital, $``$-closed subspace of $`B(H)`$, then $`I`$ is injective if and only if $`I`$ is the range of some completely positive linear map $`\varphi :B(H)B(H)`$ for which $`\varphi ^2=\varphi `$. Such maps $`\varphi `$ are commonly referred to as projections, or conditional expectations. A theorem of Choi and Effros demonstrates that if $`I`$ is an injective operator system given by the range of a projection $`\varphi `$ on $`B(H)`$, then $`I`$ is completely order isomorphic to a C-algebra, obtained by changing the product of $`I`$ to $`xy=\varphi (xy)`$.
An injective envelope of an operator system $`E`$ is an injective operator system $`I`$ and a complete isometry $`\kappa :EI`$ such that, if $`I_0`$ is an injective operator system with $`\kappa (E)I_0I`$, then $`I_0=I`$. The existence and uniqueness (up to complete isometry) of the injective envelope was established by Hamana ; thus, it is a common practice to drop reference to $`\kappa `$ and assume that $`E`$ is already realised as an operator system in $`I`$. The following proposition of Hamana is a useful criterion for determining when an injective operator system $`I`$ containing $`E`$ is an injective envelope.
###### Proposition 1.1.
(\[11, Lemma 3.7\]) Consider an inclusion $`EI`$ of operator systems, where $`I`$ is injective. The following statements are equivalent.
1. $`I`$ is an injective envelope of $`E`$.
2. The only completely positive linear map $`\omega :II`$ for which $`\omega |_E=id_E`$ is the identity map $`\omega =id_I`$.
We note below a property that we shall make frequent use of.
###### Lemma 1.2.
If $`\{E_\alpha \}_{\alpha \mathrm{\Lambda }}`$ is a family of operator systems, then $`_\alpha E_\alpha `$ is injective if and only if $`E_\alpha `$ is injective for every $`\alpha \mathrm{\Lambda }`$.
Proof. Fix an inclusion $`EF`$ of operator systems.
Assume that $`_\alpha E_\alpha `$ is injective. If $`\phi :EE_\beta `$ is completely positive, define $`\stackrel{~}{\phi }:E_\alpha E_\alpha `$ by $`(\stackrel{~}{\phi }(x))_\beta =\phi (x)`$ and $`(\stackrel{~}{\phi }(x))_\alpha =0`$ if $`\alpha \beta `$. Then there exists $`\psi :F_\alpha E_\alpha `$ extending $`\stackrel{~}{\phi }`$. So $`\pi _\beta \psi `$ is a completely positive extension of $`\phi `$.
Conversely, if $`E_\alpha `$ is injective for every $`\alpha `$, and $`\phi :E_\alpha E_\alpha `$ is completely positive, then for each $`\alpha `$ the map $`\pi _\alpha \phi :EE_\alpha `$ is completely positive, and so there exists $`\psi _\alpha :FE_\alpha `$ completely positive extension. Thus the map $`_\alpha \psi _\alpha :F_\alpha E_\alpha `$ is a completely positive extension of $`\phi `$. $`\mathrm{}`$
### 1.3. Regular monotone completions
A C-algebra $`B`$ is monotone complete if every bounded increasing net $`\{h_\alpha \}_\alpha `$ in $`B_{\mathrm{sa}}`$ has a least upper bound in $`B_{sa}`$, where $`B_{\mathrm{sa}}`$ denotes the real vector space of hermitian elements of $`B`$. The least upper bound of a bounded increasing net $`\{h_\alpha \}_\alpha `$ in $`B_{\mathrm{sa}}`$ is denoted by $`sup_\alpha h_\alpha `$. A C-algebra $`B`$ is monotone $`\sigma `$-complete if every bounded increasing sequence $`\{h_n\}_n`$ in $`B_{\mathrm{sa}}`$ has a least upper bound in $`B_{sa}`$. (The terminology โmonotone completeโ is called โmonotone closedโ in some of the standard texts, such as and . We follow Hamana by using the term โmonotone closedโ in a sense different from and ; this is explained below.)
Monotone complete C-algebras are unital and if $`B`$ is monotone $`\sigma `$-complete and satisfies the countable chain condition (namely, for each for each $`SB_{\mathrm{sa}}`$ that is bounded above in $`B_{\mathrm{sa}}`$ there is a countable subset $`S_0S`$ such that any upper bound for $`S_0`$ is also an upper bound for $`S`$), then $`B`$ is monotone complete . Every W-algebra is monotone complete and a C-algebra $`B`$ is an AW-algebra if and only if each maximal abelian C-subalgebra $`DB`$ is monotone complete. However, it is not known whether every AW-algebra is monotone complete. A well-known theorem of Tomiyama for conditional expectations between C-algebras, which is proved below for operator systems, implies in particular that the injective envelope of an operator system is monotone closed.
###### Proposition 1.3.
Let $`EM`$ be operator systems, with $`M`$ monotone complete. If there exists a positive linear map $`\varphi :ME`$ such that $`\varphi _E=id_E`$, then $`E`$ is monotone complete.
Proof. Let $`\{h_\alpha \}_\alpha `$ be a bounded increasing net in $`E`$. It is in particular an increasing bounded net in $`M`$, so there exists $`\stackrel{~}{h}M`$, $`\stackrel{~}{h}=sup_\alpha h_\alpha `$. Let $`h=\varphi (\stackrel{~}{h})`$. Then $`hh_\alpha =\varphi (\stackrel{~}{h}h_\alpha )0`$, for every $`\alpha `$, so that $`h`$ is an upper bound for $`\{h_\alpha \}_\alpha `$. If $`kE`$ and $`h_\alpha k`$ for every $`\alpha `$, then because $`kM`$ we have that $`\stackrel{~}{h}k`$. Thus, $`kh=\varphi (k\stackrel{~}{h})0`$, which implies that $`h`$ is the supremum of $`\{h_\alpha \}_\alpha `$ in $`E`$. $`\mathrm{}`$
###### Corollary 1.4.
The injective envelope $`I(A)`$ of any C-algebra $`A`$ is monotone complete. In particular, $`I(A)`$ is an AW-algebra.
If $`B`$ is a monotone complete C-algebra, then a subset $`SB_{\mathrm{sa}}`$ is monotone closed in $`B`$ if, for every bounded increasing net $`\{s_\alpha \}_\alpha `$ in $`S`$, $`sup_\alpha s_\alpha `$ (which exists in $`B`$) is contained in $`S`$. In particular, if $`A`$ is a C-subalgebra of $`B`$ and if $`\text{m-cl}_BA_{\mathrm{sa}}`$ denotes the smallest subset of $`B_{\mathrm{sa}}`$ that contains $`A_{\mathrm{sa}}`$ and is monotone closed in $`B`$, then the monotone closure of $`A`$ in $`B`$ is defined to be the set
$$\text{m-cl}_BA=\text{m-cl}_BA_{\mathrm{sa}}+i\text{m-cl}_BA_{\mathrm{sa}}.$$
It so happens that $`\text{m-cl}_BA`$ is a monotone complete C-subalgebra of $`B`$ \[12, Lemma 1.4\].
A C-subalgebra $`C`$ of $`B`$ is called a monotone closed C-subalgebra of $`B`$ if $`\text{m-cl}_BC=C`$. Because the property of $`C`$ being monotone closed in $`B`$ involves both $`C`$ and $`B`$, it is possible for a C-subalgebra $`C`$ of $`B`$ to be monotone complete yet fail to be monotone closed in $`B`$. In fact, it is frequently the case that a von Neumann algebra $`MB(H)`$ is not monotone closed in $`B(H)`$.
A C-subalgebra $`A`$ of a C-algebra $`B`$ is said to be order dense in $`B`$ if
$$h=sup\{kA^+:kh\},hB^+.$$
For example, $`K(H)`$ is order dense in $`B(H)`$.
A regular monotone completion of a C-algebra $`A`$ is a C-algebra $`B`$ such that
1. $`A`$ is a C-subalgebra of $`B`$,
2. $`B`$ is monotone complete,
3. $`\text{m-cl}_BA=B`$, and
4. $`A`$ is order dense in $`B`$.
In , Hamana proved that a regular monotone completion exists for every C-algebra $`A`$ and any two regular monotone completions of $`A`$ are $``$-isomorphic. Thus, $`\overline{A}`$ is used to denote โtheโ regular monotone completion of $`A`$. Hamanaโs construction of $`\overline{A}`$ is via the injective envelope of $`A`$. Namely, $`\overline{A}`$ is the monotone closure of $`A`$ in $`I(A)`$.
The regular monotone $`\sigma `$-completion $`\overline{A}^\sigma `$ of a C-algebra $`A`$ was introduced by Wright . Hamana recovers $`\overline{A}^\sigma `$ via the injective envelope by considering monotone $`\sigma `$-closure of $`A`$ in $`I(A)`$ (the definitions are analogous to earlier ones, but with sequences in place of nets).
For each C-algebra $`A`$ there is a representation in which
$$A\overline{A}{}_{}{}^{\sigma }\overline{A}I(A),$$
where each containment is as a C-subalgebra. We shall assume this representation in our work herein. An important feature of this sequence of containments is:
$$\overline{A}\text{ is monotone closed in }I(A).$$
###### Theorem 1.5.
Assume that $`A`$ is a separable C-algebra.
1. (Wright) $`\overline{A}{}_{}{}^{\sigma }=\overline{A}`$.
2. (OzawaโSaitรด) The AW-algebra $`\overline{A}`$ has no type II direct summand.
3. (Hamana) If $`A`$ is postliminal, then $`\overline{A}`$ is of type I.
4. (Saitรด) If $`KA`$ is an essential ideal of $`A`$, then $`\overline{K}=\overline{A}`$.
5. If $`KA`$ is an essential ideal of $`A`$, then then $`I(K)=I(A)`$.
Proof. Only the proof of (5) need be given, as it is not explicitly stated in the literature. By (4), $`\overline{K}=\overline{A}`$ if $`KA`$ is an essential ideal of $`A`$. Furthermore, $`I(\overline{A})=I(A)`$, by \[12, Lemma 3.7\]. Hence, $`I(K)=I(\overline{K})=I(\overline{A})=I(A)`$. $`\mathrm{}`$
### 1.4. Local multiplier algebras
The multiplier algebra of a C-algebra $`A`$ is the C-subalgebra $`M(A)`$ of the enveloping von Neumann algebra $`A^{}`$ that consists of all $`xA^{}`$ for which $`xaA`$ and $`axA`$, for all $`aA`$. If $`JA`$ is an ideal, then $`J^{}`$ is identified with the closure of $`J`$ in $`A^{}`$ with respect to the strong operator topology. Thus, if $`J`$ and $`K`$ are ideals of $`A`$, and if $`JK`$, then $`M(J)M(K)M(A)`$.
An ideal $`K`$ of $`A`$ is said to be essential if $`KJ\{0\}`$ for every nonzero ideal $`JA`$. Any essential ideal is necessarily nonzero. Consider the multiplier algebra $`M(J)`$ of any essential ideal $`J`$ of $`A`$. If $`(A)`$ is the set of essential ideals of $`A`$, partially ordered by reverse inclusion, then the set $`๐(A)`$ of multiplier algebras $`M(K)`$ of $`K(A)`$ is a directed system of C-algebras. $`M_{\mathrm{loc}}(A)`$ is then defined to be the C direct limit of the directed system $`K(A)`$. In , Ara and Mathieu give a systematic account of the theory of local multiplier algebras of C-algebras. Their book is our basic reference on the topic.
There are various ways to realise $`M_{\mathrm{loc}}(A)`$ โconcretelyโ as a C-subalgebra of some other C-algebra:
1. as a C-subalgebra of a quotient of $`A^{}`$ ;
2. as a C-subalgebra of a quotient of $`A^{}`$, where the quotient is monotone $`\sigma `$-complete ;
3. as a C-subalgebra of $`I(A)`$ .
In this final case, $`M_{\mathrm{loc}}(A)`$ is realised by idealisers in $`I(A)`$ of essential ideals of $`A`$. Specifically, by \[10, Corollary 4.3\],
$$M_{\mathrm{loc}}(A)=\left(\underset{K(A)}{}\{xI(A):xK+KxK\}\right)^{},$$
where the closure is with respect to the norm topology of $`I(A)`$. Thus,
$$AM_{\mathrm{loc}}(A)I(A)$$
is an inclusion of C-subalgebras. In , Frank showed an additional sequence of inclusions as C-subalgebras:
$$AM_{\mathrm{loc}}(A)M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)\overline{A}I(A).$$
### 1.5. Injective envelopes of separable and prime C-algebras
###### Proposition 1.6.
If $`A`$ is a separable C-algebra, then $`I(A)`$ does not have a finite type II direct summand.
Proof. It is enough to show that if $`I(A)`$ has a finite direct summand, then this summand is of type I. Because $`I(A)e=I(Ae)`$ for any central projection $`eI(A)`$ \[12, Lemma 6.2\], and since the C-algebra $`Ae`$ is separable, we may assume without loss of generality that $`I(A)`$ itself is a finite algebra. Thus, the identity $`1I(A)`$ is a finite projection, and so $`1`$ is a finite projection in $`\overline{A}`$ as well. Therefore, $`\overline{A}`$ is of type I \[17, Theorem 2\]. But type I algebras are injective; hence $`\overline{A}=I(A)`$. $`\mathrm{}`$
The next proposition, which builds on work of Hamana, determines which C-algebras lead to factors.
###### Proposition 1.7.
The following statements are equivalent for any C-algebra $`A`$.
1. $`\overline{A}`$ is a factor.
2. $`I(A)`$ is a factor.
3. $`A`$ is prime.
Proof. The equivalence of (1) and (3) was established by Hamana \[12, Theorem 7.1\]. To prove that (1) and (2) are equivalent, note that $`๐ต(\overline{A})=๐ต\left(I(\overline{A})\right)`$, because $`\overline{A}`$ is monotone complete \[12, Theorem 6.3\]. Further, because $`I(\overline{A})=I(A)`$ \[12, Lemma 3.7\], we conclude that $`๐ต(\overline{A})=๐ต\left(I(A)\right)`$. Thus, $`\overline{A}`$ is a factor if and only if $`I(A)`$ is a factor. $`\mathrm{}`$
## 2. W-algebra Injective Envelopes
The injective envelope $`I(A)`$ of any C-algebra $`A`$ is an AW-algebra. However, in rare instances $`I(A)`$ is known in fact to be a W-algebra. This is so if $`A`$ can be represented as acting on a Hilbert space in such a way as to contain every compact operator . In this section we characterise those separable C-algebras $`A`$ for which $`I(A)`$ is a W-algebra.
###### Lemma 2.1.
If $`A`$ is a C-algebra for which $`I(A)`$ is a W-algebra, then $`\overline{A}`$ is a W-algebra.
Proof. Without loss of generality we may assume that $`I(A)`$ is represented as a von Neumann algebra acting on a Hilbert space. Let $`\{h_\alpha \}_\alpha `$ be any bounded increasing net in $`\overline{A}_{\mathrm{sa}}`$. Because $`I(A)`$ is a von Neumann algebra, $`\{h_\alpha \}_\alpha `$ has a least upper bound $`h`$ such that $`h=lim_\alpha h_\alpha `$ in the strong-operator-topology. Note that the supremum of $`\{h_\alpha \}_\alpha `$ in $`I(A)`$ necessarily coincides with $`h`$ and, because $`\overline{A}`$ is monotone closed in $`I(A)`$, $`h\overline{A}`$. Thus, $`\overline{A}`$ is a C-algebra of operators for which the strong-operator limit of every bounded increasing net of hermitian elements of $`\overline{A}`$ belongs to $`\overline{A}`$. By \[14, Lemma 1\], this implies that $`\overline{A}`$ is a von Neumann algebra. $`\mathrm{}`$
###### Lemma 2.2.
The following statements are equivalent for a von Neumann algebra $`M`$.
1. $`M`$ is a direct product of type I factors.
2. $`M`$ is generated by its minimal projections.
The lemma above is well known. However, as it is important for our work, the ideas that underlie the proof are worth noting here briefly. First of all, if $`M`$ is a direct product of type $`I`$ factors, then $`M`$ is generated by the family of all the minimal projections of all the factors. Conversely, if $`M`$ is generated by minimal projections, then it cannot have a type II nor type III direct summand. Indeed, if $`Me`$ is type II or type III, with $`e`$ a central projection in $`M`$, consider $`qM`$ a minimal projection such that $`qe0`$. Such a projection exists because otherwise $`e=0`$. Since $`q`$ is minimal in $`M`$, $`qe=qeq=q`$ and so $`qMe`$. But then $`Me`$ admits a minimal projection, which is a contradiction. Thus $`M`$ is type I, and it can be expressed as a direct integral over a type I factor-valued measure. The diffuse part of this measure has to be zero, because any projection in the diffuse part will not be minimal, and we can reason as before. Therefore, the measure is atomic and $`M`$ is a direct product of type I factors.
###### Corollary 2.3.
If $`M`$ is a von Neumann algebra generated by minimal projections, then $`M`$ is injective.
Proof. Type I factors are injective, by Arvesonโs theorem on the injectivity of $`B(H)`$. Lemma 2.2 asserts that $`M`$ is a direct product of type I factors; by Lemma 1.2, every direct product of injective C-algebras is injective. Hence, $`M`$ is injective. $`\mathrm{}`$
###### Lemma 2.4.
Suppose that $`A`$ is a C-subalgebra of a von Neumann algebra $`M`$ and that $`M=A^{\prime \prime }`$.
1. If $`M`$ is generated by its minimal projections, each of which is contained in $`A`$, then $`A`$ is order dense in $`M`$.
2. If $`A`$ is separable and if $`A`$ is order dense in $`M`$, then $`M`$ is generated by its minimal projections, each of which is contained in $`A`$.
Proof. For the proof of (1), choose a nonzero $`hM^+`$ and consider the set
$$=\{(k_i)A^+:\underset{\text{finite}}{}k_ih\}.$$
There is a strictly positive $`\lambda `$ in the spectrum $`\sigma (h)`$ of $`h`$. Let $`eM`$ be the spectral projection $`e=e^h\left([\lambda ,\mathrm{})\right)`$, where $`e^h`$ denotes the spectral resolution of $`h`$. Thus, $`0\lambda ehe`$. Moreover, $`e`$ majorises a minimal projection $`p`$ of $`M`$; by hypothesis, $`pA`$. Thus, $`0\lambda p=e(\lambda p)ee(\lambda )e=\lambda eheh`$, and so $`\lambda p`$, which proves that $`\mathrm{}`$. It is clear that $``$ is inductive under inclusions of those families and so, by Zornโs Lemma, $``$ has a maximal family $`W`$. Since every finite sum of this family is less than $`h`$,
$$y=sup\{\underset{kK}{}k:K\text{ is finite and }KW\}h.$$
If $`yh`$, then $`hyM^+`$, and so by the first paragraph there exists $`kA^+`$ such that $`khy`$. If it were true that $`kW`$, then for each net $`(h_i)`$ of those finite sums of elements in $`W`$ such that $`h_iy`$, the net $`(h_i+k)y+k`$, which contradicts the fact that $`y`$ is the supremum. Hence, $`kW`$. But if $`kW`$, then the family $`W`$ is not maximal, which is again a contradiction. Therefore, it must be that $`y=h`$, which proves that $`A`$ is order dense in $`M`$.
For the proof of (2), note that because $`A`$ is separable and $`A^{\prime \prime }=M`$, to prove that $`M`$ is generated by its minimal projections, each of which is contained in $`A`$, it is enough, by \[23, p. 139\], to prove that any normal state $`\omega M_{}`$ is faithful precisely when its restriction $`\omega |_A`$ to $`A`$ is faithful. Thus, let $`\omega `$ be a normal state on $`M`$ that is faithful on $`A`$. Assume that $`\omega (h)=0`$, where $`hM^+`$. Because $`h=sup\{kA^+:kh\}`$, we have that $`0\omega (k)\omega (h)=0`$ for each $`kA^+`$ with $`kh`$. Thus, $`\omega (k)=0`$, which implies that $`k=0`$ because $`\omega `$ is faithful on $`A`$. Hence, $`h=0`$ and so $`\omega `$ is faithful on $`M`$. $`\mathrm{}`$
The following theorem is the main result of the present section.
###### Theorem 2.5.
The following statements are equivalent for a separable C-algebra $`A`$.
1. $`I(A)`$ is a W-algebra.
2. $`I(A)`$ is a discrete type I W-algebra.
3. There exists a faithful representation $`\pi :AB(H)`$ such that the von Neumann algebra $`\pi (A)^{\prime \prime }`$ is generated by its minimal projections, each of which is contained in $`\pi (A)`$.
4. There exists an ideal $`K`$ of $`A`$ such that
1. $`K`$ is a minimal essential ideal and
2. $`K_{}_nK(H_n)`$, for some sequence of Hilbert spaces $`H_n`$.
Proof. Assume that $`I(A)`$ is a W-algebra. Then there is a faithful representation $`\stackrel{~}{\pi }:I(A)B(H)`$ such that $`\stackrel{~}{\pi }(I(A))`$ is a von Neumann algebra and $`\pi (A)`$ is a C-subalgebra of $`\stackrel{~}{\pi }(I(A))`$, where $`\pi =\stackrel{~}{\pi }_{|A}`$. Without loss of generality, we assume that $`I(A)`$ is a von Neumann algebra acting on a Hilbert space. Consider the regular monotone completion $`\overline{A}`$ of $`A`$, which can be realised as the monotone closure of $`A`$ in $`I(A)`$ by Hamanaโs theorem \[12, Theorem 3.1\]. Furthermore, because $`I(A)`$ is a von Neumann algebra, $`\overline{A}`$ is a von Neumann algebra, by Lemma 2.1. Thus, $`A^{\prime \prime }\overline{A}^{\prime \prime }=\overline{A}`$. As $`A`$ is separable and order dense in $`A^{\prime \prime }`$, the von Neumann algebra $`A^{\prime \prime }`$ is generated by its minimal projections, each of which is contained in $`A`$ (Lemma 2.4). Furthermore, by Lemma 2.2, $`A^{\prime \prime }`$ is a direct product of type I factors, which implies that $`A^{\prime \prime }`$ is injective by Corollary 2.3. Because $`AA^{\prime \prime }I(A)`$, we conclude that $`A^{\prime \prime }=\overline{A}=I(A)`$, by minimality of the injective envelope. This proves that (1) $``$ (2) $``$ (3).
We next show that (3) $``$ (4). Assume there exists a faithful representation $`\pi :AB(H)`$ such that the von Neumann algebra $`\pi (A)^{\prime \prime }`$ is generated by its minimal projections, each of which is contained in $`\pi (A)`$. Without loss of generality, assume that $`A`$ is already represented as a subalgebra of $`B(H)`$ and that $`M=A^{\prime \prime }`$ is generated by its minimal projections, each of which lie in $`A`$.
Let $`KA`$ be the ideal of $`A`$ generated by the minimal projections of $`M`$. We first show that $`K`$ is an essential ideal, minimal among all essential ideals of $`A`$. Suppose that $`JA`$ is a nonzero ideal. Choose any nonzero $`hJ^+`$. As shown in the proof of (1) of Lemma 2.4, there is a $`\lambda >0`$ and a spectral projection $`eM`$ of $`h`$ such that $`\lambda ehe`$, and there is a minimal projection $`p`$ of $`M`$ such that $`ep=pe=p`$ and $`0\lambda pphpJK`$. That is, $`JK\{0\}`$, which proves that $`K`$ is an essential ideal of $`A`$.
Because $`M=A^{\prime \prime }`$ is generated by its minimal projections, $`M`$ is a discrete type $`I`$ von Neumann algebra, by Lemma 2.2. Hence, there is a faithful normal $``$-representation $`\varrho `$ of $`M`$ on a Hilbert space $`H`$ of the form $`H=_nH_n`$ such that $`\varrho (K)\varrho (A)\varrho (M)=_nB(H_n)`$. Obviously, the minimal projections of any $`B(H_n)`$ are minimal projections of $`\varrho (M)`$ and are, hence, elements of $`\varrho (K)`$. On the other hand, if $`e`$ is a minimal projection of $`_nB(H_n)`$, then $`eB(H_n)`$ for some $`n`$ (for otherwise $`e`$ is cut by some minimal central projection). Therefore, $`_nK(H_n)\varrho (K)`$. However, $`\varrho (K)`$ is the smallest C-algebra that contains the minimal projections of $`\varrho (M)`$; hence $`\varrho (K)=_nK(H_n)`$. Since $`K_{}_nK(H_n)`$, it is a minimal essential ideal.
We now prove that (4) $``$ (1). Suppose that $`A`$ has a minimal essential ideal $`K`$ such that $`K_{}_nK(H_n)`$. Therefore, by \[1, Lemma 1.2.1\],
$$M(K)=M\left(\underset{n}{}K(H_n)\right)=\underset{n}{}M\left(K(H_n)\right)=\underset{n}{}B(H_n),$$
which shows that $`M(K)`$ is a (type I) W-algebra. Furthermore, because $`K`$ is a minimal essential ideal of $`A`$, $`M(K)=M_{\mathrm{loc}}(A)`$ by \[1, Remark 2.3.7\]. Hence, $`M_{\mathrm{loc}}(A)`$ is an injective W-algebra. However, $`AM_{\mathrm{loc}}(A)I(A)`$ as C-subalgebras, and so by definition of the injective envelope, it must be that $`M_{\mathrm{loc}}(A)=I(A)`$, which proves that $`I(A)`$ is a W-algebra. $`\mathrm{}`$
## 3. Type I Injective Envelopes
One extension of Arvesonโs fundamental theorem on the injectivity of $`B(H)`$ is a result of Hamana \[12, Proposition 5.2\] that states that every type I AW-algebra is injective. The following theorem describes those separable C-algebras that have type I injective envelopes.
###### Theorem 3.1.
If $`A`$ is a separable C-algebra $`A`$, then $`I(A)`$ is a type I AW-algebra if and only if $`A`$ has a liminal essential ideal. If this is the case, then $`\overline{A}=I(A)`$.
Proof. Assume that $`A`$ is separable and has a liminal essential ideal $`K`$. Because $`\overline{A}`$ and $`\overline{K}`$ are isomorphic \[20, Corollary 2.1\] and because $`K`$ is liminal, $`\overline{A}`$ is a type I AW-algebra \[12, Theorem 6.6\]. Hence, $`\overline{A}=I(A)`$ and $`I(A)`$ is a type I AW-algebra.
Conversely, assume that $`I(A)`$ is a type I AW-algebra. Because $`\overline{A}I(A)`$ and because $`\overline{A}`$ and $`I(A)`$ have the same type I direct summands \[12, Corollary 6.5\], we conclude that $`\overline{A}=I(A)`$. Thus, $`A`$ is order dense in $`I(A)`$.
Because $`I(A)`$ is of type I, the C-subalgebra $`JI(A)`$ generated by the abelian projections of $`I(A)`$ is a liminal ideal of $`I(A)`$ \[13, Theorem 2\]. We aim to prove that $`K=AJ`$ is a liminal essential ideal of $`A`$.
Suppose that $`\alpha _0`$ is an irreducible representation of $`J`$ on a Hilbert space $`H_{\alpha _0}`$. As $`J`$ is an ideal of $`I(A)`$, $`\alpha _0`$ extends uniquely to an irreducible representation $`\alpha `$ of $`I(A)`$ on the same Hilbert space $`H_{\alpha _0}`$. Thus, $`\alpha \left(I(A)\right)\alpha (J)=\alpha _0(J)=K(H_{\alpha _0})`$.
If $`\widehat{J}`$ denotes the spectrum of $`J`$ (unitary equivalence classes of irreducible representations of $`J`$) and if, for each $`\alpha _0\widehat{J}`$, $`\alpha `$ denotes the unique extension of $`\alpha _0`$ to an irreducible representation of $`I(A)`$, we consider the representation $`\rho `$ of $`I(A)`$ defined by
$$\rho =\underset{\alpha _o\widehat{J}}{}\alpha .$$
By construction, $`\rho _{|J}`$ is a faithful representation of $`J`$. We next show that $`\rho _{|A}`$ is a faithful representation of $`A`$. Suppose that $`aA^+`$ satisfies $`\rho (a)=0`$. If $`eI(A)`$ is any abelian projection, then $`eaeJ`$ and $`\rho (eae)=\rho (e)\rho (a)\rho (e)=0`$. Because $`\rho _{|J}`$ is a faithful representation of $`J`$, $`eae=0`$; so, $`a^{1/2}e=0`$. Thus, $`a^{1/2}e=0`$ for all abelian projections of $`I(A)`$. Because $`I(A)`$ is a type I AW-algebra,
$$1=sup\{e:eI(A)\text{ is an abelian projection}\}.$$
Therefore, by \[12, Lemma 1.9\],
$$a=a^{1/2}1a^{1/2}=sup\{a^{1/2}ea^{1/2}:eI(A)\text{ is an abelian projection}\}=\mathrm{\hspace{0.33em}0},$$
which proves that $`\rho _{|A}`$ is a faithful representation of $`A`$.
(Indeed $`\rho `$ is a faithful representation of $`I(A)`$ as well. To prove this, suppose that $`hI(A)^+=\overline{A}^+`$ satisfies $`\rho (h)=0`$. Thus, $`\rho (a)=0`$ for all $`aA^+`$ for which $`ah`$. Since $`\rho _{|A}`$ is a faithful representation of $`A`$, $`\rho (a)=0`$ only if $`a=0`$. Because $`h=sup\{aA^+:ah\}`$ and $`a=0`$ for every $`ah`$, we conclude that $`h=0`$, which proves that $`\rho `$ is faithful.)
Let $`sJ^+`$ be nonzero and choose any $`\alpha _0\widehat{J}`$. Then $`\alpha (a)`$ is compact for every $`aA^+`$ such that $`as`$. To verify this, fix $`aA^+`$ for which $`as`$; thus, $`\alpha (a)\alpha (s)=\alpha _0(s)`$. Let $`\{\xi _n\}_n`$ be a sequence in the unit sphere of the Hilbert space $`H_{\alpha _0}`$. By the compactness of $`\alpha (s)^{1/2}`$, there is a subsequence $`\{\xi _{n_k}\}_k`$ such that $`\{\alpha (s)^{1/2}\xi _{n_k}\}_k`$ is convergent. This implies that the sequence $`\{\alpha (a)^{1/2}\xi _{n_k}\}_k`$ is a Cauchy sequence, for
$$\begin{array}{ccc}\hfill \alpha (a)^{1/2}\xi _{n_j}\alpha (a)^{1/2}\xi _{n_m}^2& =& \alpha (a)\left(\xi _{n_j}\xi _{n_m}\right),\left(\xi _{n_j}\xi _{n_m}\right)\hfill \\ & & \\ & & \alpha (s)\left(\xi _{n_j}\xi _{n_m}\right),\left(\xi _{n_j}\xi _{n_m}\right)\hfill \\ & & \\ & =& \alpha (s)^{1/2}\xi _{n_j}\alpha (a)^{1/2}\xi _{n_m}^2.\hfill \end{array}$$
Hence, $`\{\alpha (a)^{1/2}\xi _{n_k}\}_k`$ is convergent, which yields $`\alpha (a)`$ compact. Since the choice of $`\alpha _0\widehat{J}`$ is arbitrary, this shows that $`\rho (a)\rho (J)`$ if $`aA^+`$ satisfies $`as`$. Because $`\rho `$ is faithful, this is to say that $`aJ`$ if $`aA^+`$ satisfies $`as`$. Furthermore, since $`s`$ is nonzero and $`A`$ is order dense in $`I(A)`$, there is a nonzero $`aA^+`$ such that $`as`$. In particular, this nonzero $`a`$ belongs to $`J`$, thereby proving that $`K=AJ\{0\}`$.
The previous paragraph establishes the following identity:
$$s=sup\{aK^+:as\},sJ^+.$$
This fact will now be used to prove that $`K`$ is an essential ideal of $`A`$. To this end, let $`L`$ be any ideal of $`A`$ for which $`LK=\{0\}`$. Thus if $`bL^+`$, then $`bab=0`$ for every $`aK^+`$. Now, if $`eI(A)`$ is any abelian projection, then $`eJ^+`$ and
$$e=sup\{aK^+:ae\}.$$
Therefore, again by \[12, Lemma 1.9\],
$$beb=sup\{babK^+:ae\}=\mathrm{\hspace{0.33em}0}.$$
Thus, $`eb=be=0`$ for every abelian projection $`eI(A)`$, which implies that $`b=0`$ (as demonstrated earlier in this proof). Hence, $`LK=\{0\}`$ only if $`L=\{0\}`$ and so $`K`$ is an essential ideal of $`A`$.
The final point to verify is that $`K`$ is liminal. But this follows from the fact that every C-subalgebra of a liminal C-algebra is liminal \[6, Proposition 4.2.4\], and by noting that $`K`$ is a C-subalgebra of the liminal ideal $`J`$ of $`I(A)`$. $`\mathrm{}`$
## 4. Applications
###### Theorem 4.1.
The following statements hold for every separable C-algebra $`A`$.
1. $`A`$ has a liminal essential ideal if and only if $`A`$ has postliminal essential ideal.
2. If $`A`$ is abelian, then $`I(A)`$ is a W-algebra if and only if there exists a finite or countably infinite set $`\mathrm{\Gamma }`$ such that $`I(A)=l^{\mathrm{}}(\mathrm{\Gamma })`$ and $`c_0(\mathrm{\Gamma })Al^{\mathrm{}}(\mathrm{\Gamma })`$.
3. If $`A`$ is simple and $`I(A)`$ is a W-algebra, then $`A=K(H)`$ for some Hilbert space $`H`$.
4. $`I(A)`$ admits a faithful state.
5. If $`A`$ is prime, then exactly one of the following two statements holds:
1. $`I(A)_{}B(H)`$, for some separable Hilbert space $`H`$;
2. $`I(A)`$ is a wild type III AW-factor.
In particular, if $`A`$ has no postliminal essential ideal, then $`I(A)`$ is a wild type III AW-factor.
Proof. For the proof of (1), every liminal ideal is postliminal, by definition. Thus, assume that $`A`$ has a postliminal essential ideal, say $`K`$. As $`A`$ and $`K`$ are separable and $`K`$ is an essential ideal, $`\overline{K}=\overline{A}`$ (Theorem 1.5). Because $`K`$ is liminal, $`\overline{K}`$ is type I, and so $`\overline{A}=I(A)`$ is of type I. By Theorem 3.1, $`A`$ has a liminal essential ideal, which proves (1).
To prove (2), suppose now that $`A`$ is abelian and $`I(A)`$ is a W-algebra. By Theorem 2.5, $`A`$ has a minimal essential ideal $`K`$ for which $`K_{}_{n\mathrm{\Gamma }}K(H_n)`$, for some finite or countable infinite set $`\mathrm{\Gamma }`$; however, as $`K`$ is abelian, $`K(H_n)=`$ for every $`n`$, whence $`K=c_0(\mathrm{\Gamma })`$. As $`K`$ contains all minimal projections in $`A`$, we deduce that $`Al^{\mathrm{}}(\mathrm{\Gamma })`$. Finally, $`I(c_0(\mathrm{\Gamma }))=l^{\mathrm{}}(\mathrm{\Gamma })`$ (because $`c_0(\mathrm{\Gamma })`$ is order dense in $`l^{\mathrm{}}(\mathrm{\Gamma })`$), so that $`I(A)=l^{\mathrm{}}(\mathrm{\Gamma })`$. The converse is a direct application of Theorem 2.5 where the minimal essential ideal of $`A`$ is $`c_0(\mathrm{\Gamma })`$.
To prove (3), assume that $`A`$ is simple and that $`I(A)`$ is a W-algebra. By (4) of Theorem 2.5, $`A`$ has a minimal essential ideal of the form $`K=_nK(H_n)`$. Being simple, $`A=K`$; and for $`K`$ to be simple, there can be only one summand. Thus, $`A=K(H_1)`$.
For the proof of (4), note that because $`A`$ is separable, $`A`$ has a faithful representation as a C-subalgebra of $`B(H)`$, where $`H`$ is a separable Hilbert space. Thus, by Hamanaโs construction of the injective envelope, there is a projection $`\varphi :B(H)B(H)`$ such that $`\varphi \left(B(H)\right)=I(A)`$. The separability of $`H`$ implies that $`B(H)`$ has a faithful state $`\omega `$. This state is also faithful on the C-algebra representation of $`I(A)`$. To prove this, recall that the product $``$ on $`I(A)`$ is given by $`xy=\varphi (xy)`$, for all $`x,yI(A)`$. Suppose $`xI(A)`$ is such that $`\omega (x^{}x)=0`$. Then $`\omega \left(\varphi (x^{}x)\right)=0`$ and so $`\varphi (x^{}x)=0`$, as $`\omega `$ is a faithful state on $`B(H)`$. Therefore, by the Schwarz inequality for completely positive maps, $`0\varphi (x)^{}\varphi (x)\varphi (x^{}x)=0`$. This implies that $`\varphi (x)=0`$. However, on $`I(A)`$ the map $`\varphi `$ acts as the identity. Thus, $`x=\varphi (x)=0`$, which proves that $`\omega `$ is a faithful state on the C-algebra representation of $`I(A)`$.
To prove (5), assume now that $`A`$ is prime. By Proposition 1.7, $`I(A)`$ is a factor. But this factor cannot be of type II for the following reasons. Proposition 1.6 already excludes the case of finite type II AW-factors. By , every type II AW-factor that admits a faithful state is a W-factor. Since $`I(A)`$ admits a faithful state and since $`I(A)`$ is a W-algebra only in the case where $`I(A)`$ is of type I (Theorem 2.5), it is impossible for $`I(A)`$ to be a type II AW-factor. Hence, $`I(A)`$ is a factor of either type I or type III.
In the case where $`I(A)`$ is of type I we have $`I(A)_{}B(H)`$ for some Hilbert space $`H`$, because all type I AW-factors have this form \[15, Theorem 2\]. Indeed, in this case, $`\overline{A}{}_{}{}^{\sigma }=I(A)_{}B(H)`$; since $`\overline{A}^\sigma `$ is countably decomposable, $`H`$ can be chosen to be separable.
If $`I(A)`$ is not of type I, then the type III AW-factor $`I(A)`$ is cannot be a W-algebra, by Theorem 2.5. Every AW-factor that is not W-algebra is wild ; hence, $`I(A)`$ is wild. $`\mathrm{}`$
We wish to remark that statement (4) of Theorem 4.1 above was previously noted (without proof) and employed in \[12, Corollary 3.8\].
Turning now to the local multiplier algebra, in most cases the precise determination of $`M_{\mathrm{loc}}(A)`$ is difficult, and so one is interested to know what properties $`M_{\mathrm{loc}}(A)`$ might exhibit. In particular, the following questions have been raised in the literature.
1. For which C-algebras $`A`$ is $`M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)=M_{\mathrm{loc}}(A)`$ ? ()
2. For which C-algebras $`A`$ is $`M_{\mathrm{loc}}(A)`$ injective ? ()
Partial answers to these questions are listed in the theorem below.
###### Theorem 4.2.
Assume that $`A`$ is a separable C-algebra.
1. If $`A`$ has a liminal essential ideal, then $`M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)`$ is an injective C-algebra of type I and
$$M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)=\overline{A}=I(A).$$
2. If $`A`$ has a minimal essential ideal that is $``$-isomorphic to a C-algebraic direct sum of elementary C-algebras, then $`M_{\mathrm{loc}}(A)`$ is an injective W-algebra of type I and
$$M_{\mathrm{loc}}(A)=M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)=\overline{A}=I(A);$$
Proof. To prove (1), let $`K`$ be a liminal essential ideal of $`A`$. As $`A`$ and $`K`$ are separable and $`K`$ is an essential ideal, $`\overline{K}=\overline{A}`$. Because $`K`$ is liminal, $`\overline{K}`$ is type I, and so $`\overline{A}=I(A)`$ is of type I. Again using that $`A`$ and $`K`$ are separable and that $`K`$ is an essential ideal, conclude that from \[21, Theorem 2.8\] that $`M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)=\overline{A}`$. Hence, $`M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)`$ is an injective C-algebra of type I.
For the proof of (2), note that Theorem 2.5 and its proof imply there is a minimal essential ideal $`K`$ of $`A`$ such that $`K_{}K(H_n)`$ and $`M(K)=M_{\mathrm{loc}}(A)=\overline{A}=I(A)`$. Every AW-algebra is its own local multiplier algebra \[1, Theorem 2.3.8\], and so
$$M_{\mathrm{loc}}(A)=M_{\mathrm{loc}}\left(M_{\mathrm{loc}}(A)\right)=I(A).$$
This completes the proof of (2). $`\mathrm{}`$
There is an unresolved issue: is $`M_{\mathrm{loc}}(A)`$ injective if $`A`$ is separable and has a liminal essential ideal ? Recall that if $`K`$ is an essential ideal of $`A`$, then $`\overline{K}=\overline{A}`$ . Thus, it is sufficient to ask: is $`M_{\mathrm{loc}}(A)`$ injective if $`A`$ is separable and liminal ? This question is at present open.
## 5. Nonseparable C-algebras
The focus of this paper has been on separable C-algebras. For example, Proposition 1.6 and Theorem 2.5 do not hold for nonseparable C-algebras. More specifically, if $`R`$ denotes the hyperfinite II<sub>1</sub> factor $`R`$, then $`R`$ is injective and, thus, $`R=I(R)`$ is a W-factor of type II. However, this leads to another question of interest: if $`M`$ is a nonhyperfinite II<sub>1</sub> factor, then what is the injective envelope of $`M`$ ? Because $`M`$ is simple, $`I(M)`$ is an AW-algebra factor; is $`I(M)`$ a finite AW-factor ? More generally, does the passage from $`M`$ to $`I(M)`$ preserve type if $`M`$ is a von Neumann algebra ?
Although Proposition 1.6 and Theorem 2.5 do not hold for nonseparable C-algebras, the necessity part of Theorem 3.1 was established without recourse to separability. Thus, the following theorem holds.
###### Theorem 5.1.
If the injective envelope of a C-algebra $`A`$ is of type I, then $`A`$ has a liminal essential ideal.
The original motivation for the concept of injectivity is Arvesonโs HahnโBanach Extension Theorem for completely positive linear maps, and the idea of an injective envelope stems from Arvesonโs theory of boundary representations . In the work on boundary representations, the algebras under consideration need not have been separable, but frequently the algebras were assumed to have nontrivial intersection with the compact operators. In this spirit we have the following result, which generalises one form the โboundary theoremโ from $`B(H)`$ to discrete type I von Neumann algebras and which shows that statement (3) of Theorem 2.5 holds for nonseparable C-algebras as well.
###### Theorem 5.2.
If $`\pi :AB(H)`$ is a faithful representation of a C-algebra $`A`$ on a Hilbert space $`H`$ such that $`\pi (A)^{\prime \prime }`$ is generated by its minimal projections, each of which is contained in $`\pi (A)`$, then $`\pi (A)^{\prime \prime }=I(A)`$.
Proof. Without loss of generality, we may assume that $`A`$ is already faithfully represented as a C-subalgebra of $`B(H)`$ such that $`M=A^{\prime \prime }`$ is generated by its minimal projections, each of which is contained in $`A`$. Because $`M`$ is generated by minimal projections, $`M`$ is an injective von Neumann algebra, by Corollary 2.3. To show that $`M`$ is the injective envelope of $`A`$, it is sufficient, by Proposition 1.1, to show that any completely positive linear map $`\phi :MM`$ that fixes $`A`$ must be the identity map on $`M`$. If this is indeed so, then $`M`$ is an injective envelope for $`A`$ and, by the uniqueness of the injective envelope, we deduce that $`M=I(A)`$. If $`\phi :MM`$ is a completely positive map such that $`\phi _{|A}=\text{id}_A`$, then we will show that $`\phi =\text{id}_M`$.
To this end, observe that because $`\phi :MM`$ is a unital completely positive map that preserves $`A`$, $`\phi `$ has the following property:
$$\phi (xk)=\phi (x)k,\text{ for every }kA.$$
This fact follows from the Cauchy-Schwarz inequality and from the fact that $`A`$ is in the multiplicative domain of $`\phi `$ (see \[22, 9.2\] or \[16, Corollary 2.6\]). Using this fact we shall deduce below that
(5.1)
$$x0\text{ if and only if }\phi (x)0.$$
Indeed, one implication is obvious from the positivity of $`\phi `$. To prove the other implication, assume that $`\phi (x)0`$. Thus, $`\phi (\text{Im}(x))=\text{Im}(\phi (x))=0`$. Let $`z=\text{Im}(x)`$ and write $`z=z^+z^{}`$, where $`z^+,z^{}M^+`$ are such that $`z^+z^{}=z^{}z^+=0`$.
Our first goal is to prove that $`z^+=0`$. Suppose, on the contrary, that $`z^+0`$. Thus, there is a strictly positive $`\lambda `$ in the spectrum of $`z^+`$; hence, there is a spectral projection $`pM`$ such that $`0\lambda ppz^+=z^+p`$. Note that $`z^{}p=0`$, as the projection $`p`$ is in the von Neumann algebra generated by $`z^+`$ and $`z^+z^{}=z^{}z^+=0`$. Let $`qA`$ be an arbitrary minimal projection of $`M`$ and consider the projection $`pqM`$. Because $`pqq`$ and $`q`$ is minimal, either $`pq=0`$ or $`pq=q`$. We will show that the latter case cannot occur (under the conventional assumption that minimal projections are defined to be nonzero). Assume that it is true that $`pq=q`$. Then $`0q=pqp`$. Pre- and post-multiply the inequality $`\lambda q\lambda pz^+p=zp`$ by $`q`$ to obtain $`\lambda qq(zp)qqzq`$. Note that $`\phi (zq)=\phi (z)q`$ (because $`A`$ is in the multiplicative domain of $`\phi `$) and that $`\phi (z)=0`$ (by hypothesis). Likewise, for any hermitian $`yM`$, $`\phi (qy)=\phi (yq)^{}=q\phi (y)`$. Thus, $`\phi (qzq)=q\phi (z)q=0`$ and $`0\lambda q=\phi (\lambda q)q\phi (z)q=0`$. This implies that $`q=0`$, which contradicts the fact that $`q`$ is minimal and, thus, nonzero. Therefore, it must be that $`pq=0`$, for every minimal projection $`q`$ of $`M`$. Because every nonzero projection in $`M`$ majorises a minimal projection, we conclude that $`p=0`$, in contradiction to the fact that $`p`$ is a nonzero spectral projection of $`z^+`$. Hence, it must be that $`z^+=0`$.
A similar argument shows that $`z^{}=0`$. We can find a nonzero $`\lambda ^+`$ and a minimal projection $`qA`$ such that $`qzq\lambda q`$; thus $`\lambda q=\phi (\lambda q)\phi (qzq)=q\phi (z)q=0`$, and again $`q=0`$.
We conclude that $`z=0`$, which implies that $`x`$ is selfadjoint. It remains to show that $`x`$ is positive. Assume that $`x`$ is not positive. Thus, there exists a nonzero spectral projection in the negative part of $`\sigma (x)`$; by taking once again a suitable minimal subprojection $`q`$, we can find $`\lambda >0`$ such that $`qxq\lambda q`$. But then $`\phi (qxq)\lambda q`$; and on the other hand, $`\phi (qxq)=q\phi (x)q0`$. The contradiction implies that no such $`q`$ can exist, and so $`x0`$.
From (5.1) and the fact the $`\phi `$ preserves $`A`$, we have that $`kA`$, $`kx`$ if and only if $`k\phi (x)`$. Statement (1) of Lemma 2.4 asserts that $`A`$ is order dense in $`M`$. Hence, $`\phi (x)=x`$ for every $`xM^+`$, which implies that $`\phi `$ is the identity map on $`M`$. $`\mathrm{}`$
## 6. Open Questions
Although this paper is mainly concerned with type I injective envelopes of separable C-algebras, there are a number of unresolved questions that underscore the limits of our current state of knowledge concerning injective envelopes in general. A few such questions are listed here.
1. Suppose that $`A`$ is a separable C-algebra.
1. Is $`M_{\mathrm{loc}}(A)`$ an AW-algebra ?
2. Is $`\overline{A}=I(A)`$ if $`I(A)`$ is of type III ?
2. Suppose that $`{\displaystyle \underset{1}{\overset{\mathrm{}}{}}}M_2`$ and $`{\displaystyle \underset{1}{\overset{\mathrm{}}{}}}M_3`$ denote the UHF C-algebras obtained through the tensor products of the matrix algebras $`M_2`$ and $`M_3`$ respectively. The injective envelope of each of these C-algebras is a wild type III AW-factor. Is it true that
$$I\left(\underset{1}{\overset{\mathrm{}}{}}M_2\right)=I\left(\underset{1}{\overset{\mathrm{}}{}}M_3\right)\mathrm{?}$$
3. Suppose that $`M`$ is a von Neumann algebra.
1. What is $`I(M)`$ if $`M`$ is not injective ?
2. If $`M`$ is a non-injective type II<sub>1</sub> factor, then is the AW-factor $`I(M)`$ also of type II ?
|
warning/0506/math0506107.html
|
ar5iv
|
text
|
# The Dynkin diagrams of Rational Double Points
## 1 Introduction
Rational double points<sup>1</sup><sup>1</sup>1In the literature, they are also called Du Val singularities, Kleinian singularities or simple critical points. are the simplest surface singularities and were first studied by Du Val . One may think of them as neglible singularities. They play an important rรดle in the classification of surfaces and occur in the theory of simultaneous resolutions of singularities.
In this essay we will be mainly concerned with the geometry of the exceptional set corresponding to a resolution of a rational double point. We will derive the classification of rational double points in terms of Dynkin diagrams. It should be noted, that the proof of this classification is rather lengthy. However, the author was unable to find the complete proof in a single source and therefore decided to present it in full detail. Most ideas are taken from two papers of Artin , , balanced with a slightly different approach in Reidโs draft . Further parts of the argument are taken from Durfee , Mumford and Brieskorn . The second article by Pinkham in treats the topic very nicely, although some difficult steps are omitted.
Finally, we will find a connection between the most simple objects in different fields of mathematics: Rational double points are linked with Platonic solids and simple Lie groups.
## 2 Basic facts on surface singularities
### 2.1 Definitions
We want to study surface singularities $`(X,x)`$; here $`X`$ is a normal, two-dimensional, projective variety over $``$ which is non-singular, except maybe at $`xX`$.
Two singularities are isomorphic, if there exist open neighbourhoods of the singular points which are isomorphic.
A resolution of $`(X,x)`$ is a birational, proper and surjective morphism
$$\pi :\stackrel{~}{X}X$$
where $`\stackrel{~}{X}`$ is a non-singular projective variety over $``$.
See section 9 for an example.
It is an important and difficult theorem, that resolutions always exist; for a general discussion we refer to , .
#### Immediate properties of the exceptional set
The exceptional set $`E:=\pi ^1(x)`$ is compact (since $`X`$ is proper) and one-dimensional (since $`\pi `$ is birational). Moreover it is connected by Zariskiโs connectedness theorem A.5. Therefore $`E`$ is a bunch of irreducible projective curves
$$E=\underset{i=1}{\overset{n}{}}E_i.$$
We say that a surface singularity is
rational, if for a resolution
$$\pi :\stackrel{~}{X}X$$
the first higher direct image sheaf of $`\stackrel{~}{X}\text{โs}`$ structure sheaf vanishes
$$R^1\pi _{}๐ช_{\stackrel{~}{X}}=0,$$
(1)
and
a double point, if the local ring $`๐ช_{X,x}`$ has multiplicity two, i.e. the leading coefficient of its Hilbert-Samuel polynomial is two ( III ยง23, vol. 2, VIII ยง10).
#### Remarks:
1. 1. The definition of a rational singularity is independent of the chosen resolution: Since $`R^1\pi _{}๐ช_X`$ is a coherent sheaf ( III.8.8.(b)) concentrated on $`x`$, all we are interested in is $`h^0(X,R^1\pi _{}๐ช_{\stackrel{~}{X}})`$. However, we will see soon in section 2.2 that
$$p_a(X)p_a(\stackrel{~}{X})=h^0(X,R^1\pi _{}๐ช_{\stackrel{~}{X}})$$
and the arithmetic genus of a non-singular projective variety is a birational invariant ( V.5.6).
2. The condition (1) may appear opaque at a first glance, but will hopefully become more transparent in the sequel. For example, it implies that the $`E_i`$ are rational curves.
2. Since we are in the normal case, the condition for a double point means, that two general curves on $`X`$ through $`x`$ have local intersection number two at $`x`$ ( 4.6). If $`X`$ is a hypersurface $`f^1(0)`$, yet another way to state this condition is
$$fm_x^2\text{ and }fm_x^3,$$
where $`m_x`$ is the ideal of functions vanishing at $`x`$ ( (7.48)).
### 2.2 A first consequence of the rationality condition (1)
Let us mention a simple consequence of the rationality condition (1).
###### Proposition 2.1
Let $`\pi :\stackrel{~}{X}X`$ be a resolution as above. If
$$R^1\pi _{}๐ช_{\stackrel{~}{X}}=0$$
then
$$p_a(X)=p_a(\stackrel{~}{X}).$$
We need a lemma.
###### Lemma 2.2
For any resolution $`\pi :\stackrel{~}{X}X`$, we have
$$\pi _{}๐ช_{\stackrel{~}{X}}=๐ช_X.$$
Proof ( p. 280): Since the question is local on $`X`$, we can assume $`X`$ is affine, say $`X=\text{Spec }A`$. By II.5.8.(b), $`\pi _{}๐ช_{\stackrel{~}{X}}`$ is a coherent sheaf of $`๐ช_X`$-algebras, hence $`B:=H^0(X,\pi _{}๐ช_{\stackrel{~}{X}})`$ is a finitely generated $`A`$-module. But $`A`$ and $`B`$ are integral domains with the same quotient field (since $`\pi `$ is birational) and $`A`$ is algebraically closed (since $`X`$ is normal), thus $`B=A`$, and $`\pi _{}๐ช_{\stackrel{~}{X}}=๐ช_X`$. $`\mathrm{}`$
Proof of the proposition ( Ex. III.8.1): Let
$$0I^0\stackrel{d^0}{}I^1\stackrel{d^1}{}I^2\mathrm{}$$
(2)
be an injective resolution for $`๐ช_{\stackrel{~}{X}}`$. We have not only $`R^1\pi _{}๐ช_{\stackrel{~}{X}}=0`$, but $`R^i\pi _{}๐ช_{\stackrel{~}{X}}=0`$ for $`i1`$, because the fibers of $`\pi `$ have dimension $`1`$ (A.7). Therefore, by applying $`\pi _{}`$ to (2), we obtain again an exact sequence
$$0\pi _{}๐ช_{\stackrel{~}{X}}\pi _{}I^0\stackrel{\pi _{}d^0}{}\pi _{}I^1\stackrel{\pi _{}d^1}{}\pi _{}I^2\mathrm{}.$$
(3)
Since injectives are flasque ( III.2.4), direct images of flasque sheaves are flasque, and flasque sheaves are acyclic for the global section functor ( III.2.5), we see that (3) is an acyclic resolution for $`๐ช_X=\pi _{}๐ช_{\stackrel{~}{X}}`$. Thus
$`H^i(X,๐ช_X)`$ $`=`$ $`{\displaystyle \frac{\text{ker }H^0(X,\pi _{}d^i)}{\text{im }H^0(X,\pi _{}d^{i1})}}`$
$`=`$ $`{\displaystyle \frac{\text{ker }H^0(\stackrel{~}{X},d^i)}{\text{im }H^0(\stackrel{~}{X},d^{i1})}}`$
$`=`$ $`H^i(\stackrel{~}{X},๐ช_{\stackrel{~}{X}})`$
and our claim follows. $`\mathrm{}`$
#### Remark:
This proof is just a degenerated case of the Leray spectral sequence
$`E_2^{p,q}=H^p(X,R^q\pi _{}๐ช_{\stackrel{~}{X}})`$
$``$
$`E_{\mathrm{}}^{p,q}=H^{p+q}(\stackrel{~}{X},๐ช_{\stackrel{~}{X}})`$
(cf. II.4.17.1, V; for general information on spectral sequences cf. , II ยง4), which takes in our setting the simple form (theorem A.6 and A.7)
$$E_2^{p,q}:\begin{array}{cccc}0& 0& 0& 0\\ H^0(X,R^1\pi _{}๐ช_{\stackrel{~}{X}})& \mathrm{?}& \mathrm{?}& 0\\ H^0(X,๐ช_X)& H^1(X,๐ช_X)& H^2(X,๐ช_X)& 0\end{array}$$
$`H^i(\stackrel{~}{X},๐ช_{\stackrel{~}{X}})=E_{\mathrm{}}^{i,0}=E_2^{i,0}=H^i(X,๐ช_X)\text{ for }i=0,1,`$
$`H^2(\stackrel{~}{X},๐ช_{\stackrel{~}{X}})=E_{\mathrm{}}^{2,0}=E_2^{2,0}/E_2^{0,1}=H^2(X,๐ช_X)/H^0(X,R^1\pi _{}๐ช_{\stackrel{~}{X}}).`$
From this we get
$$p_a(X)p_a(\stackrel{~}{X})=h^0(X,R^1\pi _{}๐ช_{\stackrel{~}{X}})$$
and our proposition (and its converse!) follow at once.
### 2.3 Further properties of the exceptional set $`E`$
It will be a great technical convenience to work with good resolutions; for them we require that
1. all $`E_i`$ are non-singular,
2. $`E_iE_jE_k=\mathrm{}`$ for mutually distinct $`i,j,k`$,
3. the intersection of $`E_i`$ and $`E_j`$ is transverse for $`ij`$.
Any resolution $`\pi :\stackrel{~}{X}X`$ of a surface $`X`$ can be brought in such a nice form by successively blowing up points of $`\stackrel{~}{X}`$ (cf. again , , and also V.3.8, V.3.9).
In the following we do always assume that $`\pi :\stackrel{~}{X}X`$ is good.
A fundamental fact about good resolutions is the following
###### Proposition 2.3
( p. 6) The intersection matrix of the resolution $`\left(E_iE_j\right)_{i,j=1\mathrm{}n}`$ is negative definite.
Proof: We take a meromorphic function $`fk(X)`$ with $`f(x)=0`$ and define two effective divisors
$$H_0:=f^1(0)\text{ and }H_{\mathrm{}}:=f^1(\mathrm{}).$$
Denote the proper transform of $`H_i`$ with $`\stackrel{~}{H}_i`$ for $`i=0,\mathrm{}`$ respectively. Then we have a linear equivalence of divisors
$$\stackrel{~}{H}_{\mathrm{}}\stackrel{~}{H}_0+\underset{i=1}{\overset{n}{}}m_iE_i$$
where $`m_i=\text{ord}_{E_i}f\pi >0`$.
It suffices to prove that the matrix
$$M:=\left(m_iE_im_jE_j\right)_{i,j=1,\mathrm{},n}$$
is negative definite. Now we have $`M_{i,j}0`$ if $`ij`$ (since the $`E_i`$ are irreducible) and
$`{\displaystyle \underset{i=1}{\overset{n}{}}}M_{i,j}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(m_iE_im_jE_j)`$
$`=`$ $`(\stackrel{~}{H}_{\mathrm{}}\stackrel{~}{H}_0)m_jE_j`$
$`=`$ $`0\stackrel{~}{H}_0m_jE_j0.`$
This implies that $`M`$ is negative semi-definite:
$`{\displaystyle \underset{i,j=1}{\overset{n}{}}}a_ia_jM_{i,j}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}a_i^2M_{i,i}+2{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,j=1}{i<j}}{\overset{n}{}}}a_ia_jM_{i,j}`$ (4)
$`=`$ $`{\displaystyle \underset{j=1}{\overset{n}{}}}\underset{0}{\underset{}{\left({\displaystyle \underset{i=1}{\overset{n}{}}}M_{i,j}\right)}}a_j^2{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,j=1}{i<j}}{\overset{n}{}}}\underset{0}{\underset{}{M_{i,j}(a_ia_j)^2}}0.`$
To show definiteness, we note that $`\stackrel{~}{H}_0`$ must pass through some $`E_i`$, hence
$$\underset{i=1}{\overset{n}{}}M_{i,j_0}<0\text{ for some }j_0.$$
Suppose we have equality in (4). Then $`a_{j_0}=0`$. Furthermore, we get $`a_i=a_j`$ if $`M_{i,j}>0`$, or inductively $`a_i=a_j`$ if $`E_i`$ and $`E_j`$ are connected in $`E`$. But $`E`$ is connected, hence $`a_i=0`$ for $`i=1,\mathrm{},n`$ in this case. $`\mathrm{}`$
In the proof of proposition 2.3, we encountered an effective exceptional divisor (i.e. an divisor supported on $`E`$)
$$Z:=\stackrel{~}{H}_{\mathrm{}}\stackrel{~}{H}_0=\underset{i=1}{\overset{n}{}}m_iE_i>0$$
which had the note-worthy property
$$ZE_i0\text{ for }i=1,\mathrm{},n.$$
(5)
Since $`E`$ is connected, any exceptional divisor $`Z`$ with this property (5) must satisfy $`ZE`$, by the arguments used in that proof. If two exceptional divisors $`Z^1=\underset{i=1}{\overset{n}{}}r_i^1E_i>0`$ and $`Z^2=\underset{i=1}{\overset{n}{}}r_i^2E_i>0`$ both satisfy (5) then obviously so does $`Z:=\mathrm{min}(Z^1,Z^2)=\underset{i=1}{\overset{n}{}}r_iE_i>0`$ where $`r_i:=\mathrm{min}(r_i^1,r_i^2)`$: $`ZE_iZ^jE_i0`$ whenever $`r_i=r_i^j`$. Hence there exists a minimal positive exceptional divisor, called the numerical divisor $`Z_{\mathrm{num}}`$ ( 4.5) (also called fundamental divisor ), for which (5) holds.
This divisor $`Z_{\mathrm{num}}`$ will provide a useful tool to describe the exceptional set of a rational singularity.
## 3 The geometry of the exceptional set $`E`$ of a resolution of a rational singularity
Throughout this section, we assume that $`\pi :\stackrel{~}{X}X`$ is good resolution of a rational singularity $`(X,x)`$ and $`E`$ its exceptional set.
We will prove that the $`E_i`$ are rational curves $`E_i^1`$. Moreover, we will be able to read off the multiplicity of $`(X,x)`$ as the self-intersection-number $`(Z_{\mathrm{num}})^2`$.
The idea is to study fatter and fatter infinitesimal neighbourhoods of $`E`$ in order to examine the embedding of $`E`$ in $`\stackrel{~}{X}`$.
We will identify an exceptional divisor $`Z=\underset{i=1}{\overset{n}{}}r_iE_i`$ with its associated positive cycle: this is the, generally non-reduced, scheme $`(\text{Supp }Z,๐ช_Z)`$. Recall that $`๐ช_Z=\text{coker}(๐ช_{\stackrel{~}{X}}(Z)๐ช_{\stackrel{~}{X}})`$, i.e. $`(\text{Supp }Z,๐ช_Z)`$ is the subscheme of $`\stackrel{~}{X}`$ defined by the coherent sheaf of ideals on $`\stackrel{~}{X}`$ whose sections on an open $`U\stackrel{~}{X}`$ are the rational functions $`f\mathrm{\Gamma }(U,๐ช_{\stackrel{~}{X}})`$ which have zeros of order at least $`r_i`$ along $`E_i`$ for all $`i`$ with $`E_iU\mathrm{}`$ . Note $`\text{Supp }Z=\underset{r_i>0}{}E_i`$.
### 3.1 The exceptional curves $`E_i`$ are rational
###### Theorem 3.1
( prop. 1, Lemma 1.3) The exceptional set of a good resolution of a rational singularity consists of rational projective curves $`E_i^1`$.
Proof (, ):
The proof relies on Grothendieckโs theorem on formal functions A.3, which takes in our case the form
$$0=\left(\left(R^1\pi _{}๐ช_{\stackrel{~}{X}}\right)_x\right)\widehat{}\underset{k=1}{\overset{\mathrm{}}{\underset{}{\mathrm{lim}}}}H^1(E,๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k})$$
where $`m_x๐ช_{X,x}`$ is the maximal ideal corresponding to $`x`$ and completion is taken with respect to the $`m_x`$-adic topology.
(We will see later in lemma 3.8, that $`๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k}=๐ช_{kZ_{\mathrm{num}}}`$.)
Since $`E`$ is one-dimensional, the natural map
$$๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k+1}๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k}$$
induces a surjection on cohomology
$$H^1(E,๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k+1})H^1(E,๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k})$$
by a vanishing theorem of Grothendieck A.6. Thus we see
$$H^1(E,๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k+1})=0\text{ for all }k.$$
We denote the sheaf of ideals of functions vanishing at $`x`$ by $`๐ช_x`$. Clearly every function in $`๐ช_x๐ช_{\stackrel{~}{X}}`$ vanishes on $`E`$; hence for every positive cycle $`Z`$ we can find an integer $`k`$ such that every function in $`๐ช_{x}^{}{}_{}{}^{k}๐ช_{\stackrel{~}{X}}`$ vanishes on $`Z`$. We now have
$$๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k}๐ช_Z$$
and (A.6)
$$0=H^1(E,๐ช_{\stackrel{~}{X}}_{๐ช_X}๐ช_{X,x}/m_{x}^{}{}_{}{}^{k+1})H^1(E,๐ช_Z).$$
In particular $`H^1(E,๐ช_{E_i})=0`$ for $`i=1,\mathrm{},n`$, from which we conclude $`p_a(E_i)=0`$, i.e. $`E_i^1`$. $`\mathrm{}`$
###### Corollary 3.2
In the proof of theorem 3.1 we have just seen $`H^1(E,๐ช_Z)=0`$ for every positive cycle $`Z`$.
We can make a more precise statement for the numerical divisor $`Z_{\mathrm{num}}`$.
###### Corollary 3.3
( thm. 3) With the assumptions of the theorem we have
$$p_a(Z_{\mathrm{num}})=0.$$
Proof ( 3.11): The statement follows immediately from corollary 3.3 and the general fact $`h^0(E,๐ช_{Z_{\mathrm{num}}})=1`$. This can be proved easily by induction: We know $`h^0(E)=1`$. Assume $`h^0(Y)=1`$ for a positive cycle $`EYZ_{\mathrm{num}}`$. We have
$$YE_i>0\text{ for some }i,$$
or equivalently $`\text{deg}_{E_i}๐ช_E(Y)1`$, by the very definition of $`Z_{\mathrm{num}}`$. Certainly $`Y+E_iZ_{\mathrm{num}}`$ in this situation. From
$$0=H^0(E_i,๐ช_{E_i}(Y))H^0(E,๐ช_{Y+E_i})H^0(E,๐ช_Y)$$
we conclude $`h^0(E,๐ช_{Y+E_i})=1`$. $`\mathrm{}`$
### 3.2 A criterion for rationality
###### Theorem 3.4
(, thm. 3)
Conversely, if we have
$$p_a(Z_{\mathrm{num}})=0$$
for the numerical cycle of a good resolution of a singularity $`(X,x)`$, then $`(X,x)`$ is rational.
We need the following lemma.
###### Lemma 3.5
Let $`Z=\underset{i=1}{\overset{n}{}}r_iE_i`$ be a positive cycle with the property that $`p_a(Y)0`$ for all positive cycles $`YZ`$. Then $`H^1(E,๐ช_Z)=0`$.
Proof: In particular $`p_a(E_i)=0`$ for all $`i`$ with $`r_i1`$, i.e. $`E_i^1`$. We use induction on $`\underset{i=1}{\overset{n}{}}r_i`$. Assume $`H^1(E,๐ช_Z)0`$. Let $`Z_i:=ZE_i`$ for $`r_i1`$. By induction hypothesis
$$H^1(E,๐ช_{Z_i})=0,$$
hence for the kernel $`M`$
$$0M๐ช_Z๐ช_{Z_i}0$$
we get (A.6)
$$H^1(E,M)H^1(E,๐ช_Z)0,$$
i.e. $`H^1(E,M)0`$. By the snake-lemma
$$\begin{array}{ccccccccc}& & & & 0& & M& & \\ & & & & & & & & & & \\ 0& & ๐ช_{\stackrel{~}{X}}(Z)& & ๐ช_{\stackrel{~}{X}}& & ๐ช_Z& & 0\\ & & & & & & & & & & \\ 0& & ๐ช_{\stackrel{~}{X}}(Z_i)& & ๐ช_{\stackrel{~}{X}}& & ๐ช_{Z_i}& & 0\\ & & & & & & & & & & \\ & & ๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i}& & 0& & & & \end{array}$$
we deduce $`M๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i}`$. Hence we can write
$$0H^1(E,๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i})=H^1(E_i,๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i}).$$
But $`E_i`$ is just $`^1`$, thus ( III.5.1)
$$\text{deg }๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i}2.$$
On the other hand
$$\text{deg }๐ช_{\stackrel{~}{X}}(Z_i)๐ช_{E_i}=Z_iE_i$$
and by the adjunction formula A.2 we get
$$ZE_i=(Z_i+E_i)E_i2+E_{i}^{}{}_{}{}^{2}=KE_i.$$
Summing up $`(Z+K)E_i0`$ yields with the adjunction formula A.2
$$2p_a(Z)2=(Z+K)Z0,$$
i.e. $`p_a(Z)1`$, a contradiction. $`\mathrm{}`$
Proof of the theorem ( prop. 1, thm. 3): We have seen in the proof of theorem 3.1, that by Grothendieckโs theorem on formal functions (A.3)
$$0=\left(\left(R^1\pi _{}๐ช_{\stackrel{~}{X}}\right)_x\right)\widehat{}\underset{k=1}{\overset{\mathrm{}}{\underset{}{\mathrm{lim}}}}H^1(E,๐ช_{kZ_{\mathrm{num}}}),$$
hence it is sufficient to prove
$$H^1(E,๐ช_{kZ_{\mathrm{num}}})=0\text{ for all }k.$$
We already know $`H^1(E,๐ช_{Z_{\mathrm{num}}})=0`$ (since $`p_a(Z_{\mathrm{num}})=0`$ and $`h^0(E,๐ช_{Z_{\mathrm{num}}})=1`$). From the surjection (A.6)
$$H^1(E,๐ช_{Z_{\mathrm{num}}})H^1(E,๐ช_{E_i})$$
we find that $`p_a(E_i)=0`$, i.e. $`E_i^1`$.
By the lemma, it is enough to show $`p_a(Y)0`$ for all positive cycles $`Y`$. Let $`Y_1:=Y`$ and define $`Y_{n+1}`$ inductively as follows
1. if $`Y_nZ_{\mathrm{num}}`$, then $`Y_{n+1}:=Y_nZ_{\mathrm{num}}0`$.
2. if $`Y_nZ_{\mathrm{num}}`$, then $`Y_nE_i>0`$ for some $`i`$ by the definition of $`Z_{\mathrm{num}}`$. Choose such an $`i`$ with smallest possible multiplicity in $`Y_n`$ and set $`Y_{n+1}:=Y_n+E_i`$.
Stop when $`Y_n=0`$. We use the equation ( Ex. V.1.3)
$$p_a(Z_1+Z_2)=p_a(Z_1)+p_a(Z_2)+Z_1Z_11$$
(6)
to calculate the arithmetic genus:
In case 1:
$`p_a(Y_n)`$ $`=`$ $`p_a(Y_{n+1}+Z_{\mathrm{num}})`$
$`=`$ $`p_a(Y_{n+1})+p_a(Z_{\mathrm{num}})+Y_{n+1}Z_{\mathrm{num}}1`$
$``$ $`p_a(Y_{n+1})1`$
In case 2:
$`p_a(Y_{n+1})`$ $`=`$ $`p_a(Y_n)+p_a(E_i)+Y_nE_i1`$
$``$ $`p_a(Y_n).`$
Steps of type $`(2)`$ cannot be repeated infinitely often without reaching a stage where $`Y_nZ_{\mathrm{num}}`$. Using equation (6) once again, we see that $`p_a(Y)`$ is a quadratic form in the coefficients $`s_i`$ of $`Y=\underset{i=1}{\overset{n}{}}s_iE_i`$ whose quadratic term is $`\frac{1}{2}\underset{i,j=1}{\overset{n}{}}s_is_jE_iE_j`$. But the matrix $`(E_iE_j)_{i,j=1,\mathrm{},n}`$ is negative definite, hence $`p_a(Y)`$ is bounded above. Consequently, there can be only a finite number of steps and the algorithm must terminate with $`Y_n=0`$. Then $`Y_{n1}=Z_{\mathrm{num}}`$ and we have
$$p_a(Y_1)\mathrm{}p_a(Y_{n1})=0.$$
$`\mathrm{}`$
### 3.3 Invertible sheaves on a positive cycle $`Z`$
It is a natural question to ask what are the invertible sheaves on a positive cycle $`Z`$, i.e. what is $`\text{Pic }Z`$?
In our particular case the answer is quite simple. It will provide an important tool in exploring the geometry of $`E`$ further.
For any invertible sheaf $``$ on the positive cycle $`ZE`$, we can define its multidegree
$$\text{deg}_Z:\text{Pic }Z^n$$
via the composite maps
$$\text{Pic }Z\text{Pic }E_i\stackrel{\text{deg}}{}\text{ for }i=1,\mathrm{},n.$$
Using local transverse cuts it is easily seen that this map is surjective $`\text{deg}_Z\text{Pic }Z^n`$: Choose a general point $`p`$ on any $`E_i`$ and construct a Cartier divisor $`\{(U_j,f_j)\}`$ with support $`p`$ and degree $`1`$ on $`E_i`$ whose local equation $`s๐ช_{Z,p}`$ restricts to a local equation of $`p`$ in $`๐ช_{E_i,p}`$. This gives
$$\text{deg}_Z\{(U_j,f_j)\}=(0,\mathrm{},0,\underset{i}{1},0,\mathrm{},0).$$
In fact, we will prove that $`\text{deg}_Z\text{Pic }Z^n`$ is an isomorphism.
It is a well-known fact that ( Ex. III.4.5)
$$\text{Pic }Z=H^1(E,๐ช_Z^{}).$$
(One may think of a ฤhech-$`1`$-cocycle $`\{(U_iU_j,g_{i,j})\}H^1(E,๐ช_Z^{})`$ as a set of transition functions $`g_{i,j}:๐ช_{U_iU_j}\stackrel{}{}๐ช_{U_iU_j}`$ which define a line bundle on $`Z`$.)
In the reduced case $`Z=E`$, it is easy to see what $`H^1(E,๐ช_E^{})`$ is, if we allow transcendental methods. Let $`_h`$ denote the functor form the category of schemes of finite type over $``$ to the category of complex analytic spaces. (cf. B and section 6).
Since $`E`$ is projective over $``$, a theorem by Serre ( B.2.1) tells us that
$$H^i(E,)H^i(E_h,_h)$$
for every coherent sheaf $``$ on $`E`$. The exponential sequence ( V.5)
$$0๐ช_{E_h}\stackrel{\text{exp}2\pi i}{}๐ช_{E_h}^{}0$$
yields (corollary 3.2 and theorem A.6)
$$0H^1(E_h,๐ช_{E_h}^{})H^2(E_h,)0,$$
i.e.
$$\text{Pic }EH^2(E_h,).$$
As we have already seen, E is built up out of $`n`$ spheres $`S^2^1`$, which intersect each other transversely. Hence by the Mayer-Vietoris sequence for, say singular cohomology
$$H^2(E_h,)^n.$$
(We will see later, that $`E`$ has the homotopy type of a bouquet of $`n`$ spheres $`E(S^2)^n`$.) Therefore, we obtain
$$\text{Pic }E^n.$$
Unfortunately, there is no analogue of the exponential sequence in the non-reduced case. Instead, we need the following proposition by Artin, whose proof uses a โfirst order exponentialโ.
###### Proposition 3.6
( lemma 1.4) We have
$$H^1(E,๐ช_Z)H^1(E,๐ช_E)^n$$
for every positive cycle $`ZE`$.
Proof: We will proceed by induction: The case $`Z=E`$ is trivial, so assume the proposition holds for $`Z^{}=ZE_iE`$. We fix our notation for the following kernels
$`0๐ฉ๐ช_Z๐ช_E0,`$
$`0๐ช_Z^{}๐ช_E^{}0,`$
$`0๐ฉ^{}๐ช_Z^{}๐ช_E0,`$
$`0^{}๐ช_Z^{}^{}๐ช_E^{}0,`$
$`0๐ฅ๐ช_Z๐ช_Z^{}0`$ and
$`0๐ฆ๐ช_Z^{}๐ช_Z^{}^{}0.`$
By A.6, it suffices to prove $`H^1(E,)=0`$.
Note that $`H^0(E,๐ช_E)=`$ (and also $`H^0(E,๐ช_E^{})=^{}`$), since $`E`$ is connected. In particular, we get a surjection
$$H^0(E,๐ช_Z)H^0(E,๐ช_E),$$
which implies (corollary 3.2)
$$H^1(E,๐ฉ)=0.$$
Similarly, $`H^1(E,^{})=0`$ (using the induction hypothesis). Now, these kernels are linked by the short exact sequences
$`0๐ฅ๐ฉ๐ฉ^{}0`$
$`0๐ฆ^{}0`$
and we obtain
$`H^0(E,๐ฉ^{})\stackrel{\delta }{}H^1(E,๐ฅ)0`$
$`H^0(E,^{})\stackrel{\delta ^{}}{}H^1(E,๐ฆ)H^1(E,)0.`$
Because of $`๐ฅ๐ฉ=0`$ (thus $`๐ฅ^2=0`$), we have an isomorphism
$$ฯต:๐ฅ\stackrel{}{}๐ฆ$$
via
$$s\mathrm{\Gamma }(U,๐ฅ)1+s\mathrm{\Gamma }(U,๐ฆ),$$
hence $`H^1(E,๐ฅ)H^1(E,๐ฆ)`$. Analogously, we have a bijection (not a morphism, in general!)
$`ฯต^{}:H^0(E,๐ฉ^{})`$ $``$ $`H^0(E,^{})`$
$`s^{}`$ $``$ $`1+s^{}.`$
Therefore, it suffices to show that the following diagram commutes
$$\begin{array}{ccc}H^0(E,๐ฉ^{})& \stackrel{\delta }{}& H^1(E,๐ฅ)\\ ฯต^{}& & ฯต& & \\ H^0(E,^{})& \stackrel{\delta ^{}}{}& H^1(E,๐ฆ).\end{array}$$
Pick an element $`s^{}H^0(E,๐ฉ^{})`$ and choose an open covering $`\{U_i\}`$ of $`E`$ such that $`s^{}`$ can be lifted to $`s_i\mathrm{\Gamma }(U_i,๐ฉ)`$. Now we can write $`\delta (s^{})`$ as the ฤech-$`1`$-cocycle
$$\{(U_iU_j,s_is_j)\}H^1(E,๐ฅ)$$
and get
$$ฯต(\delta (s^{}))=\{(U_iU_j,1+s_is_j)\}H^1(E,๐ฆ).$$
In the same way, we can lift $`ฯต^{}(s^{})=1+s^{}`$ to $`1+s_i\mathrm{\Gamma }(U_i,)`$ and obtain
$$ฯต^{}(\delta ^{}(s^{}))=\{(U_iU_j,\frac{1+s_i}{1+s_j})\}H^1(E,๐ฆ).$$
We use $`๐ฅ๐ฉ`$ in order to show
$$ฯต(\delta (s^{}))=ฯต^{}(\delta ^{}(s^{})).$$
Since
$$s_is_j\mathrm{\Gamma }(U_iU_j,๐ฅ)$$
we get
$$s_j(s_is_j)=0,$$
hence
$$(1+s_j)(1+s_is_j)=1+s_i,$$
i.e.
$$1+s_is_j=\frac{1+s_i}{1+s_j}.$$
This finishes the proof. $`\mathrm{}`$
### 3.4 The multiplicity of a rational singularity
The following theorem is the main result of this section.
###### Theorem 3.7
( cor. 6) The multiplicity of the rational singularity $`(X,x)`$ is equal to the negative of the self-intersection-number of the numerical cycle $`(Z_{\mathrm{num}})^2`$.
For the proof, we need two lemmas, which are interesting in their own rights.
###### Lemma 3.8
( thm. 4) We have
$$๐ช_x๐ช_{\stackrel{~}{X}}=๐ช_{\stackrel{~}{X}}(Z_{\mathrm{num}}).$$
Proof (, 4.17):
The inclusion $`๐ช_x๐ช_{\stackrel{~}{X}}๐ช_{\stackrel{~}{X}}(Z_{\mathrm{num}})`$ is easy : For $`f\mathrm{\Gamma }(U,๐ช_x๐ช_{\stackrel{~}{X}})`$ we can split the principal divisor $`(f)`$ in a part $`Z`$ supported on $`E`$ and a part $`D`$, which does not involve any of the $`E_i`$ at all: $`(f)=Z+D`$. Obviously $`Z>0`$. We have
$$(f)E_i=0\text{ and }DE_i0\text{ for }i=1,\mathrm{},n,$$
since $`f`$ is regular in a neighbourhood of $`E`$. Thus $`ZE_i0`$ for $`i=1,\mathrm{},n`$, that is $`ZZ_{\mathrm{num}}`$. Hence $`f\mathrm{\Gamma }(U,๐ช_{\stackrel{~}{X}}(Z_{\mathrm{num}}))`$.
For the other inclusion we have to show that for each point $`pE`$ there exists a local section $`f`$ of $`๐ช_x๐ช_{\stackrel{~}{X}}`$ such that $`(f)_{|U}=Z_{\mathrm{num}}^{}{}_{|U}{}^{}`$ for a neighbourhood $`U`$ of $`p`$ ( 4.17).
Let $`X^{}`$ be an affine neighbourhood of $`xX`$ and set $`\stackrel{~}{X}^{}:=\stackrel{~}{X}\times _XX^{}=\pi ^1(X^{})`$.
We will write for short $`๐ฅ:=๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}})`$.
We can construct a divisor $`A`$ on $`Z_{\mathrm{num}}`$ as a sum of local transverse cuts such that $`p\text{Supp}A`$ and $`\text{deg}_{Z_{\mathrm{num}}}A=\text{deg}_{Z_{\mathrm{num}}}๐ฅ_{|Z_{\mathrm{num}}}`$. The crucial point is, that proposition 3.6 implies now $`๐ช_{Z_{\mathrm{num}}}(A)๐ฅ_{|Z_{\mathrm{num}}}`$. Hence there exists a section $`sH^0(E,๐ฅ_{|Z_{\mathrm{num}}})`$ which does not vanish at $`p`$.
To finish the proof, all we have to do is to lift $`s`$ to a section on $`\stackrel{~}{X}^{}`$.
From the short exact sequence
$$0๐ฅ^2๐ฅ๐ฅ_{|Z_{\mathrm{num}}}0$$
(obtained from $`0๐ฅ๐ช_{\stackrel{~}{X}^{}}๐ช_{Z_{\mathrm{num}}}0`$ by tensoring with $`๐ฅ`$) we get
$$H^0(\stackrel{~}{X}^{},๐ฅ)H^0(E,๐ฅ_{|Z_{\mathrm{num}}})H^1(\stackrel{~}{X}^{},๐ฅ^2).$$
So it suffices to proof $`H^1(\stackrel{~}{X}^{},๐ฅ^2)=0`$. We will prove more generally:
##### Useful fact:
$`H^1(\stackrel{~}{X}^{},๐ฅ^k)=0\text{ for all }k`$.
Proof of the useful fact: Since $`X^{}`$ is affine
$$H^1(\stackrel{~}{X}^{},๐ฅ^k)=H^0(X^{},R^1\pi _{}๐ฅ^k)$$
by III.8.5.
The sheaf $`R^1\pi _{}๐ฅ^k`$ is concentrated in $`x`$; hence it is enough to prove $`\left(R^1\pi _{}๐ฅ^k\right)_x=0`$. By Grothendieckโs theorem on formal functions (A.3)
$$\left(\left(R^1\pi _{}๐ฅ^k\right)_x\right)\widehat{}=\underset{Z}{\underset{}{\mathrm{lim}}}H^1(E,๐ฅ_{}^{k}{}_{|Z}{}^{})$$
(cf. the proof of theorem 3.1), so we are left to show $`H^1(E,๐ฅ_{}^{k}{}_{|Z}{}^{})=0`$ for all positive cycles $`Z`$. Again, we can construct a divisor $`A`$ on $`Z`$ as a sum of local transverse cuts such that $`\text{deg}_ZA=\text{deg}_Z๐ฅ^k`$.
From the short exact sequence
$$0๐ช_Z๐ช_Z(A)๐ช_A(A)0$$
we get
$$H^1(E,๐ช_Z)H^1(E,๐ช_Z(A))H^1(A,๐ช_A(A)).$$
We have $`H^1(E,๐ช_Z)=0`$ by corollary 3.2 and $`H^1(A,๐ช_A(A))=0`$ for dimension reasons A.6, hence also $`H^1(E,๐ช_Z(A))=0`$. $`\mathrm{}`$
This shows
$$H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}}))H^0(E,๐ฅ_{|Z_{\mathrm{num}}})$$
and we can fetch a preimage $`s^{}H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}}))`$ of $`s`$. By construction, $`s_{}^{}{}_{p}{}^{}0`$, thus $`s_{}^{}{}_{q}{}^{}0`$ for all $`qU`$ for some neighbourhood $`U`$ of $`p`$. Or put differently,
$$(s^{})_{|U}=Z_{\mathrm{num}}^{}{}_{|U}{}^{}.$$
We have already seen $`\pi _{}๐ช_{\stackrel{~}{X}}=๐ช_X`$; therefore $`s^{}`$ gives rise to a section in $`\mathrm{\Gamma }(\pi (U),๐ช_X)`$, and thus in $`\mathrm{\Gamma }(\pi (U),๐ช_x)`$, since $`Z_{\mathrm{num}}E`$. $`\mathrm{}`$
###### Lemma 3.9
( 4.18) The ring $`_{k0}H^0(E,๐ฅ^k)`$ is generated in degree $`1`$, where $`๐ฅ:=๐ช_{Z_{\mathrm{num}}}(Z_{\mathrm{num}})`$.
Proof: We can use local transverse cuts to construct a divisor $`A`$ on $`Z_{\mathrm{num}}`$ with $`๐ช_{Z_{\mathrm{num}}}(A)๐ฅ`$. Therefore there exists a global section $`s_0H^0(E,๐ฅ)`$ whose divisor of zeros is precisely $`A`$. Since we had a lot of freedom in choosing $`A`$, we see that the linear system $`|A|=|๐ฅ|`$ is basepoint-free. Thus we can choose a $`sH^0(E,๐ฅ)`$ such that $`s`$ provides a local base at every point $`qA`$. We can use $`s^{k1}`$ to identify $`๐ฅ^{k1}๐ช_A๐ช_A`$. The short exact sequence
$$0๐ช_{Z_{\mathrm{num}}}๐ฅ๐ช_A0$$
yields (corollary 3.2)
$$0H^0(E,๐ช_{Z_{\mathrm{num}}})H^0(E,๐ฅ)H^0(A,๐ช_A)0$$
(7)
and (A.6)
$$H^1(E,๐ฅ)=0.$$
Let $`s_1,\mathrm{},s_dH^0(E,๐ฅ)`$ map to a basis of $`H^0(A,๐ช_A)`$, $`d:=h^0(A,๐ช_A)`$. Our lemma will follow from the following claim:
$`H^1(E,๐ฅ^k)=0,`$
$`H^0(E,๐ฅ^k)=\text{span}_{}\{s_{0}^{}{}_{}{}^{k},s_{0}^{}{}_{}{}^{kl}s^{l1}s_i:1lk,1id\}`$
Note that the sections $`s_{0}^{}{}_{}{}^{kl}s^{l1}s_i,l=1,\mathrm{},k,i=1,\mathrm{},d`$ and $`s_{0}^{}{}_{}{}^{k}`$ are linearly independent over $``$.
For $`k=1`$ our claim follows from (7) and $`h^0(E,๐ช_{Z_{\mathrm{num}}})=1`$:
$$h^0(E,๐ฅ)=h^0(E,๐ช_{Z_{\mathrm{num}}})+h^0(A,๐ช_A)=1+d.$$
For the induction step we argue similarly using
$$\begin{array}{ccccccc}0& & H^0(E,๐ฅ^{k1})& & H^0(E,๐ฅ^k)& & H^0(A,๐ช_A)\\ & & 0& & H^1(E,๐ฅ^k)& & 0\end{array}$$
to get
$$h^0(E,๐ฅ^k)=h^0(E,๐ฅ^{k1})+h^0(A,๐ช_A)=(k1)d+1+d$$
and
$$H^1(E,๐ฅ^k)=0.$$
$`\mathrm{}`$
Proof of the theorem ( 4.18): Let $`X^{}=\text{Spec }R`$ be an affine neighbourhood of $`xX`$ and set $`\stackrel{~}{X}^{}:=\stackrel{~}{X}\times _XX^{}`$. By lemma 3.8, we know
$$H^0(\stackrel{~}{X}^{},๐ช_x๐ช_{\stackrel{~}{X}})=H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(Z_{\mathrm{num}})).$$
(8)
We want to generalize (8) to
$$H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{k}๐ช_{\stackrel{~}{X}})=H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(kZ_{\mathrm{num}})).$$
(9)
With (9) the proof of our assertion is straightforward: Since
$$H^1(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}((k+1)Z_{\mathrm{num}}))=0$$
by corollary 3.2, we have
$$\frac{H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{k}๐ช_{\stackrel{~}{X}})}{H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{k+1}๐ช_{\stackrel{~}{X}})}=\frac{H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(kZ_{\mathrm{num}}))}{H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}((k+1)Z_{\mathrm{num}}))}=H^0(E,๐ช_{Z_{\mathrm{num}}}(kZ_{\mathrm{num}})).$$
The Riemann-Roch theorem for curves tells us
$`h^0(E,๐ช_{Z_{\mathrm{num}}}(kZ_{\mathrm{num}}))`$ $`=`$ $`1p_a(Z_{\mathrm{num}})+\text{deg }๐ช_{Z_{\mathrm{num}}}(kZ_{\mathrm{num}})`$
$`=`$ $`1k(Z_{\mathrm{num}})^2,`$
that is the leading coefficient of the Hilbert-Samuel polynomial of $`(๐ช_{X^{},x},m_x)`$ is $`(Z_{\mathrm{num}})^2`$.
We will prove (9) by induction, so assume (9) holds for $`k<l`$. Clearly (8) implies the inclusion
$$H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{l}๐ช_{\stackrel{~}{X}})H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(lZ_{\mathrm{num}})).$$
We want to show surjectivity. We take a $`gH^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(lZ_{\mathrm{num}}))`$ and restrict it to $`\overline{g}H^0(E,๐ช_{Z_{\mathrm{num}}}(lZ_{\mathrm{num}})).`$ By lemma 3.9 we have a surjection
$$H^0(E,๐ช_{Z_{\mathrm{num}}}(Z_{\mathrm{num}}))H^0(E,๐ช_{Z_{\mathrm{num}}}((l1)Z_{\mathrm{num}}))H^0(E,๐ช_{Z_{\mathrm{num}}}(lZ_{\mathrm{num}})),$$
i.e. we can write $`\overline{g}`$ in the form $`\overline{g}=\underset{j=1}{\overset{m}{}}\overline{x}_j\overline{y}_j`$ with
$`\overline{x}_jH^0(E,๐ช_{Z_{\mathrm{num}}}(Z_{\mathrm{num}}))\text{ and}`$
$`\overline{y}_jH^0(E,๐ช_{Z_{\mathrm{num}}}((l1)Z_{\mathrm{num}}))\text{ for }j=1,\mathrm{},m.`$
Lifting $`\overline{x}_j`$ and $`\overline{y}_j`$ to sections on $`\stackrel{~}{X}^{}`$
$`x_jH^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(Z_{\mathrm{num}}))\text{ and}`$
$`y_jH^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}((l1)Z_{\mathrm{num}}))\text{ for }j=1,\mathrm{},m`$
gives for $`f_2:=\underset{j=1}{\overset{m}{}}x_jy_j`$ by induction hypothesis
$`f_2H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{l}๐ช_{\stackrel{~}{X}})\text{ and}`$
$`gf_2H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}((l+1)Z_{\mathrm{num}})).`$
Continuing in this fashion gives
$`f_2,\mathrm{},f_pH^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{l}๐ช_{\stackrel{~}{X}})\text{ and}`$
$`gf_2\mathrm{}f_pH^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}((l+p1)Z_{\mathrm{num}})).`$
Hence it suffices to prove
$$H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(pZ_{\mathrm{num}}))H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{l}๐ช_{\stackrel{~}{X}})\text{ for }p0.$$
The point is that
$$\underset{k0}{}H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(kZ_{\mathrm{num}}))$$
is a finitely generated $`R`$-algebra. Assuming this, let $`M`$ be the maximal degree in a fixed set of generators. For $`p>lM`$, each element of $`H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(pZ_{\mathrm{num}}))`$ is a sum of products of at least $`l`$ generators, thus
$$H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}}(pZ_{\mathrm{num}}))H^0(\stackrel{~}{X}^{},๐ช_{x}^{}{}_{}{}^{l}๐ช_{\stackrel{~}{X}})\text{ for }p>lM.$$
For the proof of the assumption, note that the complete linear system $`|๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}})|`$ is free: By the useful fact, $`H^1(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}^{}}(2Z_{\mathrm{num}}))=0`$, i.e. we have a surjection
$$H^0(\stackrel{~}{X}^{},๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}}))H^0(E,๐ช_{Z_{\mathrm{num}}}(Z_{\mathrm{num}})).$$
We have seen in the proof of lemma 3.8 that $`|๐ช_{Z_{\mathrm{num}}}(Z_{\mathrm{num}})|`$ is free and hence so is $`|๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}})|`$. Thus we have a well-defined morphism
$$\varphi _{|Z_{\mathrm{num}}|}:\stackrel{~}{X}^{}X^{}\times ^N=_X^{}^N.$$
We denote its image by $`Y:=\text{im }\varphi _{|Z_{\mathrm{num}}|}`$. We want to show that $`Y`$ is closed. Since $`\stackrel{~}{X}X`$ is proper, so is $`\stackrel{~}{X}^{}X^{}`$ ( II.4.8.(c)). Because of the separatedness of $`_X^{}^NX^{}`$ ( II.4.9), we see that $`\varphi _{|Z_{\mathrm{num}}|}:\stackrel{~}{X}^{}_X^{}^N`$ is proper ( II.4.8.(e)), in particular $`\varphi _{|Z_{\mathrm{num}}|}`$ is closed. Hence $`Y`$ is closed.
The pullback under $`\varphi _{|Z_{\mathrm{num}}|}`$ of the relatively ample line bundle $`๐ช(1):=๐ช_X^{}_{}๐ช_^N(1)`$ is by definition of $`\varphi _{|Z_{\mathrm{num}}|}`$ simply $`๐ช_{\stackrel{~}{X}^{}}(Z_{\mathrm{num}})`$ ( 4.18).
Thus it suffices to show that
$$S^{}(Y):=\underset{k0}{}H^0(Y,๐ช_Y(k))$$
is finitely generated as a $`R`$-algebra. The homogenous coordinate ring $`S(Y)=A[x_0,\mathrm{},x_N]/I(Y)`$ of $`Y`$ is certainly a finitely generated $`R`$-algebra. By Ex. II.5.9, there exists a natural graded morphism
$$S(Y)S^{}(Y),$$
which is an isomorphism in high degrees, i.e.
$$S(Y)^d\stackrel{}{}S^{}(Y)^d\text{ for }d0,$$
and we are done. $`\mathrm{}`$
###### Corollary 3.10
For a rational double point $`(X,x)`$, we have $`(Z_{\mathrm{num}})^2=2`$.
## 4 The geometry of the exceptional set $`E`$ of a resolution of a rational double point
Once the hard work has been done in proving theorems 3.1 and 3.7, it is now easy to say explicitly what configurations can arise for $`E`$, if $`(X,x)`$ is a rational double point.
From now on, we will assume that $`\pi :\stackrel{~}{X}X`$ is a good resolution of a rational double point $`(X,x)`$ and $`E`$ its exceptional set.
By proposition 2.3, we have $`E_i^21`$ for $`i=1,\mathrm{},n`$. If $`E_{i_0}^2=1`$ for some $`i_0`$, then $`E_{i_0}^1`$ can be contracted by Castelnuovoโs criterion A.4 to give a resolution $`\pi ^{}:\stackrel{~}{X}^{}X`$ with fewer $`E_i`$. (In general, the resolution $`\pi ^{}:\stackrel{~}{X}^{}X`$ needs not to be good anymore, since the condition 2 in the definition of a good resolution might be violated. However, it is a simple consequence of the following theorem 4.1 and, again, lemma 2.3, that this cannot happen in our case. Note that we do not use this condition 2 in the proof of 4.1.)
Therefore, we can assume $`E_i^22`$ for $`i=1,\mathrm{},n`$ without loss of generality.
###### Theorem 4.1
() The $`E_i`$ have self-intersection-number $`2`$.
Proof: Let $`K`$ be a canonical divisor on $`\stackrel{~}{X}`$ ( V.1.4.4). The adjunction formula A.2 tells us
$$E_iK=E_i^2+2$$
(10)
and thus
$$E_iK0.$$
We apply the adjunction formula A.2 for $`Z_{\mathrm{num}}`$
$$2p_a(Z_{\mathrm{num}})2=(Z_{\mathrm{num}})^2+Z_{\mathrm{num}}K,$$
and get by the corollaries 3.3 and 3.10
$$0=Z_{\mathrm{num}}K=\underset{i=1}{\overset{n}{}}r_i(E_iK)0,\text{ i.e. }E_iK=0.$$
Using (10) again, we see $`E_i^2=2`$. $`\mathrm{}`$
We define the Dynkin diagram of the resolution $`\pi :\stackrel{~}{X}X`$ to be the weighted dual graph $`\mathrm{\Gamma }`$ associated to $`E`$: The vertices $`e_i`$ of $`\mathrm{\Gamma }`$ correspond to the $`E_i`$. Whenever $`E_i`$ and $`E_j`$ intersect for $`ij`$, the corresponding vertices are joined by an edge. Finally, we associate to every vertex $`e_i`$ of $`\mathrm{\Gamma }`$ the self-intersection-number $`E_i^2`$.
Every weighted graph $`\mathrm{\Gamma }`$ defines a bilinear form $`,`$ on the free module with the vertices $`e_i,i=1,\mathrm{},n`$ of $`\mathrm{\Gamma }`$ as basis in the following way: We take
$`e_i,e_i:=\text{the weight of }e_i\text{ and}`$
$`e_i,e_j:=\text{number of edges joining }e_i\text{ and }e_j.`$
The bilinear form of the Dynkin diagram of a resolution is obviously given by the matrix $`(E_iE_j)_{i,j=1,\mathrm{},n}`$ and hence negative definite by proposition 2.3. This puts very strong restrictions on the possible Dynkin diagrams $`\mathrm{\Gamma }`$.
###### Proposition 4.2
() Let $`\mathrm{\Gamma }`$ be a connected graph weighted by $`2`$ whose associated bilinear form is negative definite. Then $`\mathrm{\Gamma }`$ is a $`T`$-tree $`T_{p,q,r}`$
with $`\frac{1}{p}+\frac{1}{q}+\frac{1}{r}>1`$.
Proof: Every connected subgraph $`\mathrm{\Gamma }^{}`$ of $`\mathrm{\Gamma }`$ satisfies the hypothesis as well, hence can be
* neither a loop
,
since then $`(e_1+\mathrm{}+e_n)^2=0`$ contradicting the negative definiteness condition
* nor of the form
,
since then $`(2e_1+\mathrm{}+2e_n+f_1+\mathrm{}+f_4)^2=0`$.
Thus $`\mathrm{\Gamma }`$ must be of the form $`T_{p,q,r}`$.
The condition $`\frac{1}{p}+\frac{1}{q}+\frac{1}{r}>1`$ follows by an elementary argument: With respect to the standard basis given by the vertices of $`\mathrm{\Gamma }`$, the associated bilinear form is expressed by the matrix
$$\left(\begin{array}{ccccccccccc}2& 1& & & & & & & & & \\ 1& 2& 1& & & & & & & & \\ & 1& 2& & & & \underset{p,p+q}{\text{1}}& & & & \\ & & & \mathrm{}& & & & & & & \\ & & & & 2& 1& & & & & \\ & & & & 1& 2& \underset{p+q1,p+q}{\text{0}}& & & & \\ & & & \underset{p+q,p}{\text{1}}& & \underset{p+q,p+q1}{\text{0}}& 2& 1& & & \\ & & & & & & 1& 2& 1& & \\ & & & & & & & 1& 2& & \\ & & & & & & & & & \mathrm{}& 1\\ & & & & & & & & & 1& 2\end{array}\right).$$
But, up to congruence, this is equal to
$$\left(\begin{array}{cccccccccc}\frac{2}{1}& & & & & & & & & \\ & \frac{3}{2}& & & & & & & & \\ & & \mathrm{}& & & & & & & \\ & & & \frac{p+1}{p}& \underset{p,p+1}{\text{1}}& & & \underset{p,p+q}{\text{1}}& & \\ & & & \underset{p+1,p}{\text{1}}& \frac{q}{q1}& & & & & \\ & & & & & \mathrm{}& & & & \\ & & & & & & \frac{2}{1}& & & \\ & & & \underset{p+q,p}{\text{1}}& & & & \frac{r}{r1}& & \\ & & & & & & & & \mathrm{}& \\ & & & & & & & & & \frac{2}{1}\end{array}\right).$$
Now, this matrix is congruent to a diagonal matrix with negative main diagonal entries, except maybe a single one $`1p^1q^1r^1`$. $`\mathrm{}`$
###### Corollary 4.3
The Dynkin diagram associated to a rational double point $`(X,x)`$ must be one of the following diagrams
$`A_n`$ ($`n`$ vertices),
$`D_n`$ ($`n`$ vertices),
$`E_6`$ , $`E_7`$ and
$`E_8`$ .
We will see in section 5.2 that all Dynkin diagrams actually occur. We say that a rational double point is of type $`A_n`$, $`D_n`$ or $`E_n`$ according to its Dynkin diagram.
## 5 Example: The singularities $`/G`$ for finite $`G\text{SL}(2,)`$
After a bit of the theory of rational double points has been presented, we want to study an example.
### 5.1 Conjugacy classes of finite subgroups of $`\text{SL}(2,)`$
As a preliminary, we recall briefly the classification of conjugacy classes of finite subgroups of $`\text{SL}(2,)`$. We consider first $`\text{SO}(3,)`$. Up to conjugacy, the finite subgroups of $`\text{SO}(3,)`$ are the rotational symmetry groups of
* a pyramid (giving the cyclic subgroups $`C_n`$)
,
* an orange (corresponding to the dihedral subgroups $`D_n`$)
,
* and the Platonic solids, which give
+ the tetrahedral subgroup $`T=A_4`$
,
+ the octahedral subgroup $`O=S_4`$
,
+ and the icosahedral subgroup $`I=A_5`$
,
respectively .
If we identify $`S^2^1`$, we get an inclusion of the group of isometries of $`^1`$ (with respect to the usual metric) into the group of conformal transformations
$$\text{SO}(3,)\text{GL}(2,).$$
Under the double cover
$$\rho :\text{SL}(2,)\text{GL}(2,)=\text{SL}(2,)/\{\pm 1\}$$
this inclusion corresponds to
$$\rho ^1(\text{SO}(3,))=\text{SU}(2,).$$
Since for any finite subgroup $`G`$ of SL(2,$``$) we can find a $`G`$-invariant Hermitian metric by averaging an arbitrary one, every finite subgroup $`G`$ of $`\text{SL}(2,)`$ is conjugated to a subgroup of $`\text{SU}(2,)`$. Hence it corresponds to a finite subgroup of $`\text{SO}(3,)`$, unless it is a cyclic group of odd order. Thus we have derived the following classification of the conjugacy classes of finite subgroups of $`\text{SL}(2,)`$ ( II ยง1):
* the cyclic subgroup of order $`n`$ $`C_n`$,
* the binary dihedral subgroups $`\stackrel{~}{D}_n=\rho ^1(D_n)`$,
* the binary tetrahedral, octahedral and icosahedral subgroup $`\stackrel{~}{T}=\rho ^1(T)`$, $`\stackrel{~}{O}=\rho ^1(O)`$ and $`\stackrel{~}{I}=\rho ^1(I)`$ respectively.
### 5.2 The singularities $`^2/G`$
Now let $`G`$ be any of these subgroups; the affine orbit variety
$$^2/G=\text{Spec }[x_1,x_2]^G$$
has an isolated singularity at the origin. The singularities obtained in this fashion are all rational double points . It is a result from classical invariant theory that these singularities embed in codimension one
$$^2/G=\text{Spec }[x,y,z]/(f)\text{ where }f[x,y,z].$$
See for example Kleinโs influential book , or also 5.39. For a modern treatment, we refer to , p. 5. The following table 1 contains the basic information about these singularities.
### 5.3 The icosahedral case $`G=\stackrel{~}{I}`$
We will sketch the proof of the assertions made so far in the special case $`G=\stackrel{~}{I}`$. We see that
$$X:=^2/\stackrel{~}{I}=\text{Spec }[x,y,z]/(x^2+y^3+z^5)$$
has a singularity at $`x_0:=(0,0,0)`$, which must be a double point, since
$$\text{rank}_{}\frac{(x,y,z)^k+(x^2+y^3+z^5)}{(x,y,z)^{k+1}+(x^2+y^3+z^5)}2k+1.$$
We want to show that $`(X,x_0)`$ is a rational double point. Indeed, we will show how to resolve the singularity $`(X,x_0)`$ ( IV ยง9, p. 15). Blowing up $`^2`$ instead of $`^2/\stackrel{~}{I}`$ will reveal the relevant information more clearly. The blow-up of $`^2`$ at the origin is known to be $`๐ฏ(๐ช_^1(1))`$ ( V.3.1); here $`๐ฏ()`$ denotes the vector bundle determined by the locally trivial sheaf $``$. The quotient of $`๐ฏ(๐ช_^1(1))`$ by $`\{\pm 1\}`$ is $`๐ฏ(๐ช_^1(2))`$:
$`๐ฏ(๐ช_^1(1))/\{\pm 1\}=\{(w_1:w_2;z_1,z_2)^1\times ^2:w_1z_2=w_2z_1\}/\{\pm 1\}`$
$`\{(w_1:w_2;z_1,z_2)^1\times ^2:w_1^2z_2=w_2^2z_1\}=๐ฏ(๐ช_^1(2))`$
$`(w_1:w_2;z_1,z_2)(w_1:w_2;z_1^2,z_2^2).`$
Note that the $`\stackrel{~}{I}`$-action on $`^2`$ lifts to an action on $`๐ฏ(๐ช_^1(1))`$. Thus we have the following commutative diagram
$$\begin{array}{ccccc}๐ฏ(๐ช_^1(2))๐ฏ(๐ช_^1(1))/\{\pm 1\}& & ๐ฏ(๐ช_^1(1))& \stackrel{\sigma }{}& ^2\\ & & & & & & \\ ๐ฏ(๐ช_^1(2))/I& \stackrel{}{}& ๐ฏ(๐ช_^1(1))/\stackrel{~}{I}& \stackrel{\overline{\sigma }}{}& ^2/\stackrel{~}{I}\end{array}.$$
In particular, $`\overline{\sigma }^1(x_0)`$ is a copy of $`^1`$ with self-intersection-number $`2`$. A precise analysis of the $`I`$-action on $`๐ฏ(๐ช_^1(2))`$ shows, that $`I`$ acts on the zero-section $`S^2^1๐ฏ(๐ช_^1(2))`$ in the usual way as rotations which leave an inscribed icosahedron invariant. Furthermore the action of $`I`$ on $`T^{}^1๐ฏ(๐ช_^1(2))`$ is simply the cotangent action induced by the action of $`I`$ on $`^1`$ ( IV ยง7). Now, the group $`I`$ is acting free on $`S^2`$, except on three exceptional orbits, which consist of the vertices, the mid-edge-points and the mid-face-points of the inscribed icosahedron, respectively. Moreover, we see that these three orbits are also the only exceptional orbits for the action of $`I`$ on $`๐ฏ(๐ช_^1(2))`$. Therefore, the quotient variety $`๐ฏ(๐ช_^1(2))/I`$ is smooth except at the three points corresponding to these orbits. An explicit calculation using local coordinates shows that these three singular points are cyclic quotient singularities of type $`(5,4)`$, $`(3,2)`$ and $`(2,1)`$, respectively ( p. 17), this notion being defined as follows: A cyclic quotient singularity of type $`(n,q)`$ is a singularity, which is isomorphic to $`^2/\mu _{n,q}`$ where $`\mu _{n,q}`$ is the cyclic group generated by $`\left(\begin{array}{cc}\xi & 0\\ 0& \xi ^q\end{array}\right)`$ for a $`n`$-th root of unity $`\xi `$. Note that the numbers $`5`$, $`3`$, $`2`$ are the ramification indices at the ramification points of the map
$$S^2S^2/I$$
corresponding to the exceptional orbits.
A cyclic quotient singularity of type $`(n,q)`$ can be resolved by the Hirzebruch-Jung algorithm using successive blow-ups of points ( 2.6). The exceptional set of a resolution of a cyclic quotient singularity obtained in this way is a bunch of rational curves; the associated Dynkin diagram is of the form
,
where the $`b_i`$โs are calculated by a modified Euclidean algorithm
$$\frac{n}{q}=b_1\frac{1}{b_2\frac{1}{\mathrm{}\frac{1}{b_k}}}.$$
Applying the Hirzebruch-Jung algorithm three times for the three singular points we got, we obtain a resolution $`\pi :\stackrel{~}{X}X`$ with associated Dynkin diagram
The numerical divisor $`Z_{\mathrm{num}}`$ is easily verified to be
$$Z_{\mathrm{num}}=2E_1+4E_2+6E_3+3E_4+5E_5+4E_6++3E_7+2E_8.$$
A straightforward computation using equation (6) shows
$$p_a(Z_{\mathrm{num}})=0,$$
i.e. the singularity $`(X,x_0)`$ is rational by the criterion of theorem 3.4.
## 6 Changing to the complex analytic category
The examples introduced in the preceding section already exhaust all possibilities of rational double points up to isomorphism in the complex analytic category. Of course, such a statement cannot be true in the category of algebraic varieties, since rational double points can live on the various kinds of surfaces and the birationality class of the surface is encoded locally due to the coarseness of the Zariski topology.
We give the following (simplified) definition of the complex analytic category:
* Its objects are called complex analytic spaces and can be constructed as follows. Let $`U`$ be a simply connected open subset of $`^n`$, $`๐ช_U`$ the sheaf of complex analytic functions on $`U`$, and $`๐ฅ_X`$ a sheaf of ideals on $`(U,๐ช_U)`$. Denote by $`XU`$ the zeroset of $`๐ฅ_X`$ equipped with the standard topology and set $`๐ช_X:=๐ช_U/๐ฅ_X`$. The pair $`(X,๐ช_X)`$ is then a complex analytic space. (For the general definition one allows such simple building blocks to be glued together as in the definition of schemes.)
* a morphism of complex analytic spaces $`f:(X,๐ช_X)(X^{},๐ช_X^{})`$ is a continuous map $`f:XX^{}`$ such that $`f^{}:๐ช_X^{}๐ช_X`$ is well-defined.
To a great extent, the complex analytic category is similar to the category of algebraic varieties: For example, stalks $`๐ช_{X,x}`$ are noetherian local rings and for reduced complex analytic spaces $`(X,๐ช_X)`$ Rรผckertโs Nullstellensatz holds ( III ยง8).
## 7 Tautness of rational double points
### 7.1 Definition and theorem
To pick up the question of classifying rational double points in the complex analytic category, we introduce the notion of tautness.
Let $`(X,x)`$ be a two-dimensional normal singularity with a good resolution, whose exceptional set is a bunch of rational curves $`^1`$ and let $`\mathrm{\Gamma }`$ be its Dynkin diagram.
We say $`(X,x)`$ is taut, if up to analytic isomorphism, $`(X,x)`$ is the unique such singularity, that has a good resolution with a bunch of rational curves $`^1`$ as exceptional set and $`\mathrm{\Gamma }`$ as its Dynkin diagram .
We have the following theorem.
###### Theorem 7.1
The singularities listed in table 1 are taut.
This gives us a complete classification of rational double points up to analytic isomorphism. There are several proofs available for theorem 7.1.
### 7.2 Tjurinaโs proof
Maybe the most natural one is the proof by Tjurina : Suppose there were another singularity $`(X^{},x^{})`$ with an exceptional set consisting only of rational curves and the same Dynkin diagram, i.e. with an isomorphic exceptional variety $`(E^{},๐ช_E^{})(E,๐ช_E)`$. Then a sufficient condition for the existence of an isomorphism of neighbourhoods of $`E`$ and $`E^{}`$ is by Thm. 3, that $`(E,๐ช_{nE})`$ and $`(E^{},๐ช_{nE^{}})`$ are isomorphic for $`n`$ large enough. The proof given in proceeds by induction. Assuming we are given some isomorphism of $`(E,๐ช_Z)`$ and $`(E^{},๐ช_Z^{})`$ (here $`Z`$ and $`Z^{}`$ are exceptional divisors supported on $`E=_{i=1}^nE_i`$, $`E^{}=_{i=1}^nE_i^{}`$, respectively), then this can be extended to an isomorphism of $`(E,๐ช_{Z+E_i})`$ and $`(E^{},๐ช_{Z^{}+E_i^{}})`$, unless some obstruction occurs, which lies in some cohomology group . Grauert argues, that if all these cohomology groups vanish, then the singularity in question is taut. In general, these cohomology groups do not vanish and Tjurinaโs proof is more subtle. Essentially, he shows that the cohomology groups are too small to put obstructions on the lifting of every possible isomorphism of $`(E,๐ช_Z)`$ and $`(E^{},๐ช_Z^{})`$.
### 7.3 Brieskornโs first proof
For the sake of historical correctness, we mention that the first proof of theorem 7.1 was given by Brieskorn ( Satz 1). He showed that rational double points can be resolved by blowing up points alone, that is, it is not necessary to normalize or blow up curves. Such singularities are called absolutely isolated and were studied by Kirby ( 2.6, 2.7), who gave a classification of absolutely isolated double points: they are precisely those listed in table 1.
### 7.4 Brieskornโs second proof
However, we want to sketch another proof of theorem 7.1, which, also due to Brieskorn , is of compelling beauty and combines ideas from different fields of mathematics:
The local fundamental group of $`(X,x)`$ is defined as
$$\pi _{X,x}:=\underset{U}{\underset{}{\mathrm{lim}}}\pi _1(U\{x\})$$
where $`U`$ runs over all neighbourhoods of $`xX`$ ( ยง2). Equivalently, we can calculate $`\pi _{X,x}`$ as
$$\pi _{X,x}\underset{\stackrel{~}{U}}{\underset{}{\mathrm{lim}}}\pi _1(\stackrel{~}{U}\{x\})$$
where the limit is now taken over all neighbourhoods $`\stackrel{~}{U}`$ of $`E\stackrel{~}{X}`$. To actually compute $`\pi _{X,x}`$ it is sufficient to work out $`\pi _1`$ for a good neighbourhood $`U`$. According to , a neighbourhood $`U`$ of $`xX`$ is called good, if there exists a neighbourhood basis $`\{U_i\}`$ for $`x`$ such that $`U_i\{x\}`$ is a deformation retract of $`U\{x\}`$ for all $`i`$. Such a good neighbourhood has the homotopy type of a tubular neighbourhood $`M`$ of $`E`$ in the sense of Mumford . Intuitively spoken, a tubular neighbourhood is a levelset of the potential distribution due to a uniform charge on $`E`$. Mumford studied these tubular neighbourhoods $`M`$ and showed that they are built out of standard pieces $`S^1\times S^1\times [0,1]`$ โplumbedโ together in a certain fashion determined by $`(E_iE_j)_{i,j=1,\mathrm{},n}`$. This description and the Seifert-van-Kampen theorem enables him to give a presentation for $`\pi _1(M)`$ in terms of generators and relations. (The ideas of his proof can also be found in IV ยงยง10 - 14 on plumbed surfaces.) It turns out that for the intersection matrices $`(E_iE_j)_{i,j=1,\mathrm{},n}`$ of the resolutions of rational double points this group $`\pi _{X,x}=\pi _1(M)`$ is finite.
A rational double point has finite local fundamental group.
For the following see Satz 2.8, Thm. 3. From a merely topological point of view, $`xX`$ possesses a neighbourhood $`U`$ with $`U^{}:=U\{x\}`$ having a finite universal cover $`V^{}U^{}`$. This can be uniquely extended to a ramificated cover $`VU`$ by adding a point $`y`$ to $`V^{}`$. Moreover, we can equip $`V`$ with a normal analytic structure such that $`VU`$ becomes an analytically ramificated cover. Since $`V^{}`$ is simply connected, we see $`\pi _{V,y}=1`$. By another fundamental theorem in Mumfordโs paper ( p. 18), this shows the non-singularity of $`V`$ at $`y`$. Now $`\pi _{X,x}`$ is operating via cover transformations on $`V^{}`$, hence also on $`V`$ with fixed point $`y`$. We need another definition to state our results so far.
###### Definition and Proposition 7.2
(Two-dimensional quotient singularities) For a neighbourhood $`V`$ of the origin $`O`$ in $`^2`$ and a finite group $`G`$ of analytic automorphisms of $`V`$ fixing $`O`$, the quotient space $`V/G`$ has the structure of a normal complex analytic surface and the projection $`VV/G`$ is analytic . We say that $`U`$ is a two-dimensional quotient singularity, if $`U`$ is isomorphic to a singularity of the form $`V/G`$.
We have just seen:
A rational double point is a quotient singularity.
By a simple linearization argument ( Lemma 2.2), we can restrict ourselves to the study of quotient singularities of the form $`^2/G`$ where $`G`$ is a finite subgroup of $`\text{GL}(2,)`$. Obviously, conjugated subgroups yield isomorphic quotient spaces. The conjugacy classes of finite subgroups of $`\text{GL}(2,)`$ have been listed by Du Val ( ยง21). Later, Prill showed that only a particular class of finite subgroups has to be studied: the so-called small subgroups . He also classified them ( Satz 2.3). Using Prillโs results, Brieskorn gave a complete classification of the quotient spaces that can arise in terms of the Dynkin diagram of their resolution ( p. 348). This shows that two-dimensional quotient singularities are taut and finishes Brieskornโs proof.
## 8 Seven characterizations of rational double points
The following remark allows us to use the intermediate results in the above discussion 7.4 to give alternative characterizations of rational double points in the analytic category.
###### Remark 8.1
It can be shown that $`^2/G`$, for $`G\text{GL}(2,)`$ finite, embeds in codimension one if and only if $`G`$ is a subgroup of $`\text{SL}(2,)`$ ( cor. 5.3).
###### Theorem 8.2
()
Let $`(X,x)`$ be a normal surface singularity that embeds in codimension one. Then the following conditions are equivalent in the complex analytic category.
1. $`(X,x)`$ is a rational double point.
2. $`(X,x)`$ has a good resolution with an exceptional set consisting of rational curves with self-intersection-number $`2`$.
3. $`(X,x)`$ has a good resolution with an exceptional set consisting of rational curves and a Dynkin diagram listed in table 1.
4. The local fundamental group of $`(X,x)`$ is finite.
5. $`(X,x)`$ is a two-dimensional quotient singularity.
6. $`(X,x)`$ is isomorphic to $`^2/G`$ for finite $`G\text{GL}(2,)`$.
7. $`(X,x)`$ is isomorphic to one of the affine varieties studied in section 5.
Remark on the proof: We have already seen a proof of the implications $`(1)(2)(3)`$ and presented some ideas for $`(3)(4)(5)(6)`$. In the example 5.3, we studied a special case of $`(7)(1)`$. The last implication $`(6)(7)`$ is precisely remark 8.1.
In his survey article , Durfee presents further ten characterizations of rational double points. The characterizations he gives provide a connection of rational double points for example with weighted homogeneous polynomials, vanishing cycles, a certain limit involving volumes, monodromy groups and Morse functions. A more number-theoretical characterization in terms of almost factorial rings (fast-faktoriell) rings is due to Brieskorn ( Satz 1.5). Finally, a link with elementary catastrophes is discussed in a survey article by Slodowy ( 9).
## 9 Example: The conical double point
Although we did not give a proof of theorem 8.2, we shall at least study an example to illustrate the phenomena encountered there. We will consider the conical double point $`X:=V(f)^3`$ where $`f=xzy^2[x,y,z]`$. Note that after a change of variables $`xzy^2`$ becomes $`x_1^2+x_2^2+x_3^2`$; so $`X`$ is just the surface singularity labeled $`A_1`$ from table 1. $`X`$ is a double cone with vertex a rational double point:
.
Obviously, $`X`$ has a normal singularity at $`x=(0,0,0)`$ and $`X`$ is embedded in codimension one. $`(X,x)`$ is a double point, since
$$\text{rank}_{}\frac{(x,y,z)^k+(f)}{(x,y,z)^{k+1}+(f)}=2k+1,$$
i.e. the leading coefficient of the Hilbert-Samuel polynomial of the local ring $`๐ช_{X,x}`$ at $`x`$ it two.
Furthermore, $`(X,x)`$ is absolutely isolated, because the singularity can be resolved by a single blow-up at $`x`$, as can be easily seen by the toric description of $`X`$ :
.
We can work out the blow-up explicitly and obtain
$$\stackrel{~}{X}^{}\underset{(x,y,z;p:q:r)}{๐ธ^3\times ^2}$$
given by equations
$$xzy^2,prq^2,py=qx,pz=rx,qz=ry.$$
$`\stackrel{~}{X}^{}`$ is isomorphic to
$$\stackrel{~}{X}\underset{(x,z;u:v)}{๐ธ^2\times ^1}$$
cut out by
$$xv^2=zu^2$$
via
$`\stackrel{~}{X}`$ $`\stackrel{}{}`$ $`\stackrel{~}{X}^{}`$
$`(x,z;u:v)`$ $``$ $`(x,{\displaystyle \frac{v}{u}}x\text{ or }{\displaystyle \frac{u}{v}}z,z;u^2:uv:v^2)`$
Now $`\stackrel{~}{X}`$ is a line bundle on $`^1`$ whose zero-section has self-intersection-number $`2`$. Thus $`\stackrel{~}{X}`$ is just the line bundle on the projective line associated to the sheaf $`๐ช_^1(2)`$, i.e. the cotangent bundle $`T^{}^1`$. See also IV ยง7.
Since $`\stackrel{~}{X}`$ is a smooth variety, we have a resolution of $`(X,x)`$
$$\pi :\stackrel{~}{X}\stackrel{~}{X}^{}X.$$
The exceptional set $`E`$ of $`\pi `$ is precisely the zero-section of $`\stackrel{~}{X}`$, hence isomorphic to $`^1`$. Moreover $`E^2=2`$. The real picture reflects the situation very nicely
.
For the numerical cycle we get $`Z_{\mathrm{num}}=E`$, i.e. $`(X,x)`$ is rational by theorem 3.4, and characterization $`(1)`$ is verified.
As already mentioned in section 5.2, $`X`$ is isomorphic to the affine orbit variety $`^2/\{\pm 1\}`$, where we write $`1`$ for the reflection in the origin of the complex plane $`^2`$. This corresponds to characterizations $`(5)`$, $`(6)`$ and $`(7)`$.
Finally, let us calculate the local fundamental group $`\pi _{X,x}`$. We have a covering map
$$^2\{O\}X\{x\}$$
with covering transformation group $`\{\pm 1\}`$. For every $`\{\pm 1\}`$-invariant simply connected neighbourhood $`U`$ of $`O^2`$, we observe that $`U\{O\}`$ is also simply connected, hence
$$\pi _1((U\{O\})/\{\pm 1\})=\{\pm 1\}.$$
But such an $`U`$ can be chosen arbitrarily small, thus
$$\pi _{X,x}=\{\pm 1\},$$
which is finite and shows characterization $`(4)`$.
## 10 Lie groups and rational double points
The โ$`A_n`$-$`D_n`$-$`E_n`$โ - labeling of the various types of rational double points was actually borrowed from the classification theory of Lie groups. In this last section we will sketch some of the deep connections between Lie groups and rational double points. Essentially, we shall give a summary of 10.
### 10.1 Dynkin diagrams of simple Lie groups
A connected complex Lie group is called (almost) simple, if it contains no normal subgroup of positive dimension. In their classification theory, the simply connected simple Lie groups play a special rรดle as their universal coverings (which are finite ( 10)). These groups are classified by their corresponding Dynkin diagrams . Surprisingly, the Dynkin diagrams of table 1 occur again.
We recall briefly the relevant part of this classification; note that most of the following facts hold in a more general context (, 3.1).
Let $`G`$ be a simply connected simple Lie group of rank $`r`$ and $`๐ค`$ its Lie algebra. We fix a maximal torus $`T(^{})^r`$ of $`G`$ with character group
$$X^{}(T)=\text{Hom}(T,^{})=^r.$$
We denote the normalizer of $`T`$ in $`G`$ by $`N_G(T)`$. The group $`W:=N_G(T)/T`$ is finite and is called the Weyl group of $`G`$ with respect to $`T`$.
The restriction of the adjoint representation of $`G`$ on $`๐ค`$ to $`T`$ has eigenspaces $`๐ค_\alpha `$ on which $`T`$ acts by the character $`\alpha X^{}(T)`$ and we obtain the Cartan decomposition of $`๐ค`$
$$๐ค=\underset{\alpha X^{}(T)}{}๐ค_\alpha .$$
The finite set $`\mathrm{\Sigma }:=\{\alpha X^{}(T):\alpha 0,๐ค_\alpha \{0\}\}`$ is called the root space. Clearly, $`\mathrm{\Sigma }`$ is invariant under the action of $`W`$.
We can define a $`W`$-invariant scalar product $`,`$ on $`X^{}(T)`$ (called the Killing form) such that the elements of $`W`$ become reflections in the hyperplane perpendicular to a root $`\alpha `$
$$\beta \beta \frac{2\alpha ,\beta }{\alpha ,\alpha }\alpha .$$
Here $`\frac{2\alpha ,\beta }{\alpha ,\alpha }`$ must be integer. (The main tool in proving this and similar facts is identifying the subalgebra $`๐ค_\alpha ๐ค_\alpha [๐ค_\alpha ,๐ค_\alpha ]`$ with $`๐ฐ๐ฉ(2,)`$ and applying the representation theory of $`๐ฐ๐ฉ(2,)`$.)
The geometry of how $`\mathrm{\Sigma }`$ sits in the Euclidean lattice $`(X^{}(T),,)`$ is very rigid. For example
$$4\mathrm{cos}^2\mathrm{}(\alpha ,\beta )=\frac{4\alpha ,\beta ^2}{\alpha ,\alpha \beta ,\beta }$$
must be an integer between zero and four, i.e. there are just a few possibilities for the angle between two roots $`\alpha `$ and $`\beta `$.
By choosing a direction $`lT`$ (in general position with respect to $`\mathrm{\Sigma }`$) we can specify the positive roots $`\alpha \mathrm{\Sigma }`$ to be those with $`\alpha (l)>0`$. In particular, we can focus on simple roots: these are positive roots that are not the sum of two other positive roots. The system of simple roots gives rise to a Dynkin diagram, where we take a vertex for each simple root and join two vertices by exactly $`4\mathrm{cos}^2\mathrm{}(\alpha ,\beta )`$ lines. If we insist on all roots $`\alpha `$ having the same length $`\alpha ,\alpha `$, the only possibilities for the Dynkin diagram are
$`A_n`$ (n vertices),
$`D_n`$ (n vertices),
$`E_6`$ ,
$`E_7`$ and
$`E_8`$ .
These diagrams $`A_n`$, $`D_n`$, $`E_6`$, $`E_7`$ and $`E_8`$ actually occur for the simply connected simple Lie groups corresponding to the classical Lie algebras $`๐ฐ๐ฉ(n+1,)`$, $`๐ฐ๐ฌ(2n,)`$ and the exceptional Lie algebras $`๐ข_6`$, $`๐ข_7`$ and $`๐ข_8`$, respectively.
It can be shown that a simply connected simple Lie group can be recovered from its Dynkin diagram.
### 10.2 A theorem of Brieskorn
Let $`G`$ be a simply connected simple Lie group. We consider the quotient $`H`$ of $`G`$ by its adjoint action in the category of algebraic varieties $`p:GH`$. There is an explicit way to describe $`p`$. Let $`r`$ be the rank of $`G`$ and $`\rho _i:G\text{GL}(V_i),i=1,\mathrm{},r`$ be the $`r`$ fundamental irreducible representations of $`G`$ on finite-dimensional vector spaces. Then the character map
$`\chi :G`$ $``$ $`^r`$
$`g`$ $``$ $`(\mathrm{},\text{trace}_{V_i}\rho _i(g),\mathrm{})`$
coincides with $`p`$.
#### Example:
For $`G=\text{SL}(n,)`$ we have $`\text{rank }G=n1`$ and the $`n1`$ fundamental irreducible representations are given by the exterior powers
$$V_i=\stackrel{i}{}^n.$$
The corresponding characters are, up to sign, just the non-trivial coefficients of the characteristic polynomial
$$\text{char}(g)=\text{det}(\lambda g)=\lambda ^n\text{trace}(g)\lambda ^{n1}+\text{trace}(^2g)\lambda ^{n2}+\mathrm{}.$$
Thus we can regard $`\chi `$ as associating to $`g\text{SL}(2,)`$ its characteristic polynomial.
The point is now to study the unipotent variety
$$\text{Uni}(G):=p^1(p(e)).$$
#### Example:
For $`G=\text{SL}(n,)`$ the unipotent variety consists precisely of the unipotent matrices.
The variety $`\text{Uni}(G)`$ is a finite union of conjugacy classes and contains a unique conjugacy class of dimension $`d:=\text{dim }Gr`$ (since $`p`$ is flat) โ the regular class. The complement of the regular class in $`\text{Uni}(G)`$ is the closure of a unique conjugacy class of dimension $`d2`$ โ the subregular class $`\text{Sub}(G)`$.
We choose a $`SG`$ such that
$`S\text{ is smooth of dimension }\text{dim }Gd+2,`$
$`S\text{Sub}(G)=\{x\}\text{ and}`$
$`T_xS+T_x\text{Sub}(G)=T_xG,`$
i.e. we require that $`S`$ is a slice of codimension $`d2`$ transversal to $`\text{Sub}(G)`$ at an element $`x\text{Sub}(G)`$.
Let $`X:=S\text{Uni}(G)`$.
The following theorem was conjectured by Grothendieck and proved by Brieskorn.
###### Theorem 10.1
(, 10) If $`G`$ is a simply connected simple Lie group of type $`A_n`$, $`D_n`$ or $`E_n`$, then $`(X,x)`$ is a rational double point.
#### Example:
We want to illustrate this theorem in the simplest possible case $`G=\text{SL}(2,)`$. All regular unipotent elements are conjugate to the matrix
$$\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)$$
and there exists just a single subregular unipotent element
$$x:=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
As a transversal slice we can simply take $`S:=\text{SL}(2,)`$. Now $`p=\chi `$ is given by the trace
$`\chi :\text{SL}(2,)`$ $``$ $``$
$`\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)`$ $``$ $`a+d`$
and we get
$`X`$ $`=`$ $`S\text{Uni}(G)`$
$`=`$ $`\text{Uni}(G)=\chi ^1\left(\chi \left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\right)`$
$`=`$ $`\{\left(\begin{array}{cc}1+x& y\\ z& 1+u\end{array}\right):x+u=0\text{ and }xuyz=0\}`$
$`=`$ $`\{(x,y,z):x^2+yz=0\},`$
i.e. $`X`$ has a conical double point at $`x=(0,0,0)`$.
### 10.3 Resolutions of rational double points in the Lie group context
A closed subgroup $`PG`$ is called parabolic, if the quotient space $`G/P`$ is a projective variety. The minimal parabolic subgroups are the Borel subgroups. All Borel subgroups are conjugate to each other in $`G`$ and the normalizer $`N_G(P)`$ of a parabolic subgroup coincides with P. Thus the set of all Borel subgroups $``$ becomes a projective variety
$$=G/B$$
where $`B`$ is any Borel subgroup of $`G`$. More generally, $`๐ซ:=G/P`$ may be identified with the set of subgroups conjugate to the parabolic subgroup $`P`$.
#### Example:
The parabolic subgroups of $`\text{SL}(n,)`$ are exactly the stabilizer of the flags
$$0V_{i_1}\mathrm{}V_{i_k}^n\text{ with }\text{rank}_{}V_{i_j}=i_j,j=1,\mathrm{},k.$$
Hence the Borel subgroups correspond to maximal flags
$$0V_1\mathrm{}V_{n1}^n,$$
i.e. are conjugate to the subgroup of upper triangular matrices.
It was Springer who showed that the natural projection from the incidence variety
$$I:=\{(x,B)\text{Uni}(G)\times :xB\}$$
to $`\text{Uni}(G)`$ is a $`G`$-equivariant resolution of the singularities of $`\text{Uni}(G)`$
$$\pi :I\text{Uni}(G).$$
Let $`G,x,S,X`$ be as in theorem 10.1. It is a consequence of the $`G`$-equivariance of $`\pi `$ that the restriction
$$\pi :\stackrel{~}{X}:=\pi ^1(X)X$$
is again a resolution. In fact, it is a minimal one, that is, we cannot apply theorem A.4 to obtain a resolution with smaller exceptional set.
We can interpret the exceptional set
$$E:=\pi ^1(x)$$
in two different ways.
On one hand, we know from theorem 10.1, that $`(X,x)`$ is a rational double point. Hence $`E`$ must be a bunch of projective lines $`^1`$ intersecting each other as prescribed by the Dynkin diagram $`\mathrm{\Gamma }`$ of $`(X,x)`$.
On the other hand, we can write
$$E=\{(x,B)\{x\}\times :xB\}.$$
The vertices of the Dynkin diagram $`\mathrm{\Gamma }_G`$ of $`G`$ correspond to the simple roots of $`G`$ (after a maximal torus $`T_0`$ and a direction $`lT_0`$, or equivalently, a Borel subgroup $`B_0T_0`$ have been specified). Let $`P_\alpha `$ be the minimal proper (i.e. non-Borel) parabolic subgroup generated by $`B_0`$ and the root subgroup $`U_\alpha `$, where $`\alpha `$ is a simple root. Because of $`N_G(P_\alpha )=P_\alpha `$, we can identify the set of subgroups conjugate to $`P_\alpha `$ with the projective variety
$$๐ซ_\alpha :=G/P_\alpha .$$
The natural map
$$f_\alpha :G/B_0๐ซ_\alpha G/P_\alpha $$
maps each Borel subgroup $`B`$ to the unique parabolic subgroup $`P๐ซ_\alpha `$ containing $`B`$. Since each $`P๐ซ_\alpha `$ has semisimple rank 1, this map has projective lines as fibres.
Steinberg and Tits showed that $`E`$ is a bunch of projective lines โ one line of the form $`f_\alpha ^1(P),P๐ซ_\alpha `$ for every simple root $`\alpha `$ โ which intersect as prescribed by the edges of $`\mathrm{\Gamma }_G`$.
#### Example:
We verify these statements by explicit calculation in the simplest non-trivial case $`G=\text{SL}(3,)`$.
As maximal torus $`T_0`$ we may take the diagonal matrices and as Borel subgroup the upper triangular matrices.
The root space of $`\text{SL}(3,)`$ is
where the character $`\alpha _{i,j}X^{}(T)`$ is defined by
$$\alpha _{i,j}\left(\begin{array}{ccc}a_1& & \\ & a_2& \\ & & a_3\end{array}\right)=a_ia_j.$$
The eigenspace $`๐ฐ๐ฉ(3,)_{\alpha _{i,j}}`$, for example, consists of those matrices $`(a_{i,j})_{i,j=1,\mathrm{},3}๐ฐ๐ฉ(3,)`$ whose single non-zero entry is $`a_{2,1}`$. Using the exponential map
$$\text{exp}:๐ฐ๐ฉ(3,)\text{SL}(3,)$$
we see that the root subgroup $`U_{\alpha _{21}}`$ is
$$U_{\alpha _{21}}=\{\left(\begin{array}{ccc}1& 0& 0\\ \lambda & 1& 0\\ 0& 0& 1\end{array}\right):\lambda \}.$$
The simple roots are $`\alpha _{1,2}`$ and $`\alpha _{2,3}`$ and we obtain
as Dynkin diagram of $`\text{SL}(3,)`$ ( 12). The unipotent variety $`\text{Uni}(\text{SL}(3,))`$ is given by the unipotent matrices, all regular unipotent elements are conjugate to
$$\left(\begin{array}{ccc}1& 1& 0\\ 0& 1& 1\\ 0& 0& 1\end{array}\right)$$
and all subregular unipotent elements are conjugate to
$$x:=\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 1\\ 0& 0& 1\end{array}\right).$$
We have the minimal proper parabolic subgroups
$$P_{\alpha _{1,2}}=B_0,U_{\alpha _{2,1}}$$
and
$$P_{\alpha _{2,3}}=B_0,U_{\alpha _{3,2}}$$
which are the stabilizer of the flags
$$0\text{span}_{}\{v_1,v_2\}\text{span}_{}\{v_1,v_2,v_3\}$$
and
$$0\text{span}_{}\{v_1\}\text{span}_{}\{v_1,v_2,v_3\},$$
respectively. Clearly, $`P_{\alpha _{2,3}}`$ is conjugate to the parabolic subgroup $`P_{\alpha _{2,3}}^{}`$ stabilizing the flag
$$0\text{span}_{}\{v_2\}\text{span}_{}\{v_1,v_2,v_3\}.$$
The set of Borel subgroups containing $`x`$, which we had identified with $`E`$, is given by
$$f_{\alpha _{1,2}}^1(P_{\alpha _{1,2}})f_{\alpha _{2,3}}^1(P_{\alpha _{2,3}}^{}).$$
(All the other fibres of the maps $`f_{\alpha _{1,2}}`$ and $`f_{\alpha _{2,3}}`$ contain only a finite number of Borel subgroups which contain $`x`$.)
The two fibres intersect in a single Borel subgroup: the subgroup stabilizing the maximal flag
$$0\text{span}_{}\{v_2\}\text{span}_{}\{v_1,v_2\}\text{span}_{}\{v_1,v_2,v_3\}.$$
Hence the Dynkin diagram of $`E`$ looks like
.
## Appendix A Appendix - results from Algebraic Geometry
###### Theorem A.1
(Riemann-Roch on a surface)( V.1.6) If $`D`$ is any divisor on the non-singular surface $`X`$, then
$$\chi (๐ช_X(D))=\frac{1}{2}D(DK)+1+p_a(X)$$
where $`K`$ is a canonical divisor on $`X`$ ( V.1.4.4).
A simple corollary is the following.
###### Theorem A.2
(General adjunction formula)( Ex. V.1.3.(a)) If $`D`$ is an effective divisor on the non-singular surface $`X`$, then
$$2p_a(D)2=D(D+K).$$
###### Theorem A.3
(Grothendieckโs theorem on formal functions)( 4.2.1 or for projective morphisms III.11.1) Let $`f:XY`$ be a proper morphism of noetherian schemes and $``$ a coherent sheaf on $`X`$. For $`yY`$ denote by $`m_y๐ช_{Y,y}`$ the maximal ideal of the stalk at $`y`$. Then we have a natural isomorphism
$$0=\left(\left(R^if_{}\right)_y\right)\widehat{}\underset{k=1}{\overset{\mathrm{}}{\underset{}{\mathrm{lim}}}}H^i(f^1(y),_{๐ช_Y}\frac{๐ช_{Y,y}}{m_{y}^{}{}_{}{}^{k}})\text{ for all }i$$
where the completion is taken with respect to the $`m_y`$-adic topology.
###### Theorem A.4
(Castelnuovoโs criterion for contracting a curve)( IV ยง15, V.5.7) If $`C`$ is a curve on a non-singular surface $`X`$ with $`C^1`$ and $`C^2=1`$, then there exists a morphism $`f:XX^{}`$ to a non-singular surface $`X^{}`$ which contracts $`C`$ to a point $`p`$, such that $`X`$ is isomorphic via $`f`$ to the blow-up of $`X^{}`$ with center $`p`$, and $`C`$ is the exceptional curve.
###### Theorem A.5
(Zariskiโs connectedness theorem)( 4.3.1, III.11.4) Let $`f:XY`$ be a birational morphism between projective varieties and assume that $`Y`$ is normal. Then $`f`$ has connected fibers.
###### Theorem A.6
(A vanishing theorem of Grothendieck)( III.2.7) For any sheaf of abelian groups $``$ on a noetherian scheme $`X`$ of dimension $`n`$, we have
$$H^i(X,)=0\text{ for }i>n.$$
###### Theorem A.7
(A vanishing theorem for higher direct image sheaves)( III.11.2) Let $`f:XY`$ be a projective morphism of noetherian schemes and denote by $`r`$ the maximal dimension of its fibers. Then for all coherent sheaves $``$ on $`X`$, we have
$$R^if_{}=0\text{ for }i>r.$$
|
warning/0506/math0506265.html
|
ar5iv
|
text
|
# Acyclicity versus total acyclicity for complexes over noetherian rings
## Introduction
Let $`R`$ be a commutative noetherian ring with a dualizing complex $`D`$; in this article, this means, in particular, that $`D`$ is a bounded complex of injective $`R`$-modules; see Section 3 for a detailed definition. The starting point of the work described below was a realization that $`๐(\mathrm{Prj}R)`$ and $`๐(\mathrm{Inj}R)`$, the homotopy categories of complexes of projective $`R`$-modules and of injective $`R`$-modules, respectively, are equivalent. This equivalence comes about as follows: $`D`$ consists of injective modules and, $`R`$ being noetherian, direct sums of injectives are injective, so $`D_R`$ defines a functor from $`๐(\mathrm{Prj}R)`$ to $`๐(\mathrm{Inj}R)`$. This functor factors through $`๐(\mathrm{Flat}R)`$, the homotopy category of flat $`R`$-modules, and provides the lower row in the following diagram:
The triangulated structures on the homotopy categories are preserved by $`\mathrm{๐๐๐ผ}`$ and $`D_R`$. The functors in the upper row of the diagram are the corresponding right adjoints; the existence of $`๐`$ is proved in Proposition (2.4). Theorem (4.2) then asserts:
###### Theorem I.
The functor $`D_R:๐(\mathrm{Prj}R)๐(\mathrm{Inj}R)`$ is an equivalence of triangulated categories, with quasi-inverse $`๐\mathrm{Hom}_R(D,)`$.
This equivalence is closely related to, and may be viewed as an extension of, Grothendieckโs duality theorem for $`๐^f(R)`$, the derived category of complexes whose homology is bounded and finitely generated. To see this connection, one has to consider the classes of compact objects โ the definition is recalled in (1.2) โ in $`๐(\mathrm{Prj}R)`$ and in $`๐(\mathrm{Inj}R)`$. These classes fit into a commutative diagram of functors:
The functor $`๐ฏ`$ is induced by the composite
$$๐(\mathrm{Prj}R)\stackrel{\mathrm{Hom}_R(,R)}{}๐(R)\stackrel{๐ผ๐บ๐}{}๐(R),$$
and it is a theorem of Jรธrgensen that $`๐ฏ`$ is an equivalence of categories. The equivalence $`๐จ`$ is induced by the canonical functor $`๐(R)๐(R)`$; see . Given these descriptions it is not hard to verify that $`D_R`$ preserves compactness; this explains the top row of the diagram. Now, Theorem I implies that $`D_R`$ restricts to an equivalence between compact objects, so the diagram above implies $`๐\mathrm{Hom}_R(,D)`$ is an equivalence; this is one version of the duality theorem; see Hartshorne . Conversely, given that $`๐\mathrm{Hom}_R(,D)`$ is an equivalence, so is the top row of the diagram; this is the crux of the proof of Theorem I.
Theorem I appears in Section 4. The relevant definitions and the machinery used in the proof of this result, and in the rest of the paper, are recalled in Sections 1 and 2. In the remainder of the paper we develop Theorem (4.2) in two directions. The first one deals with the difference between the category of acyclic complexes in $`๐(\mathrm{Prj}R)`$, denoted $`๐_{\mathrm{ac}}(\mathrm{Prj}R)`$, and its subcategory consisting of totally acyclic complexes, denoted $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$. We consider also the injective counterparts. Theorems (5.3) and (5.4) are the main new results in this context; here is an extract:
###### Theorem II.
The quotients $`๐_{\mathrm{ac}}(\mathrm{Prj}R)/๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ and $`๐_{\mathrm{ac}}(\mathrm{Inj}R)/๐_{\mathrm{tac}}(\mathrm{Inj}R)`$ are compactly generated, and there are, up to direct factors, equivalences
$$\mathrm{Thick}(R,D)/\mathrm{Thick}(R)\stackrel{}{}\left[\left(๐_{\mathrm{ac}}(\mathrm{Prj}R)/๐_{\mathrm{tac}}(\mathrm{Prj}R)\right)^c\right]^{\mathrm{op}}$$
$$\mathrm{Thick}(R,D)/\mathrm{Thick}(R)\stackrel{}{}\left(๐_{\mathrm{ac}}(\mathrm{Inj}R)/๐_{\mathrm{tac}}(\mathrm{Inj}R)\right)^c.$$
In this result, $`\mathrm{Thick}(R,D)`$ is the thick subcategory of $`๐^f(R)`$ generated by $`R`$ and $`D`$, while $`\mathrm{Thick}(R)`$ is the thick subcategory generated by $`R`$; that is to say, the subcategory of complexes of finite projective dimension. The quotient $`\mathrm{Thick}(R,D)/\mathrm{Thick}(R)`$ is a subcategory of the category $`๐^f(R)/\mathrm{Thick}(R)`$, which is sometimes referred to as the stable category of $`R`$. Since a dualizing complex has finite projective dimension if and only if $`R`$ is Gorenstein, one corollary of the preceding theorem is that $`R`$ is Gorenstein if and only if every acyclic complex of projectives is totally acyclic, if and only if every acyclic complex of injectives is totally acyclic.
Theorem II draws attention to the category $`\mathrm{Thick}(R,D)/\mathrm{Thick}(R)`$ as a measure of the failure of a ring $`R`$ from being Gorenstein. Its role is thus analogous to that of the full stable category with regards to regularity: $`๐^f(R)/\mathrm{Thick}(R)`$ is trivial if and only if $`R`$ is regular. See (5.6) for another piece of evidence that suggests that $`\mathrm{Thick}(R,D)/\mathrm{Thick}(R)`$ is an object worth investigating further.
In Section 6 we illustrate the results from Section 5 on local rings whose maximal ideal is square-zero. Their properties are of interest also from the point of view of Tate cohomology; see (6.5).
Sections 7 and 8 are a detailed study of the functors induced on $`๐(R)`$ by those in Theorem I. This involves two different realizations of the derived category as a subcategory of $`๐(R)`$, both obtained from the localization functor $`๐(R)๐(R)`$ to $`๐_{\mathrm{prj}}(R)`$: one by restricting it to the subcategory of K-projective complexes, and the other by restricting it to $`๐_{\mathrm{inj}}(R)`$, the subcategory of K-injective complexes. The inclusion $`๐_{\mathrm{prj}}(R)๐(\mathrm{Prj}R)`$ admits a right adjoint $`๐`$; for a complex $`X`$ of projective modules the morphism $`๐(X)X`$ is a K-projective resolution. In the same way, the inclusion $`๐_{\mathrm{inj}}(R)๐(\mathrm{Inj}R)`$ admits a left adjoint $`๐`$, and for a complex $`Y`$ of injectives the morphism $`Y๐(Y)`$ is a K-injective resolution. Consider the functors $`๐ฆ=๐(D_R)`$ restricted to $`๐_{\mathrm{prj}}(R)`$, and $`๐ฅ=๐๐\mathrm{Hom}_R(D,)`$ restricted to $`๐_{\mathrm{inj}}(R)`$. These functors better visualized as part of the diagram below:
It is clear that $`(๐ฆ,๐ฅ)`$ is an adjoint pair of functors. However, the equivalence in the upper row of the diagram does not imply an equivalence in the lower one. Indeed, given Theorem I and the results in Section 5 it is not hard to prove:
*The natural morphism $`X\mathrm{๐ฅ๐ฆ}(X)`$ is an isomorphism if and only if the mapping cone of the morphism $`(D_RX)๐(D_RX)`$ is totally acyclic.*
The point of this statement is that the mapping cones of resolutions are, in general, only acyclic. Complexes in $`๐_{\mathrm{inj}}(R)`$ for which the morphism $`\mathrm{๐ฆ๐ฅ}(Y)Y`$ is an isomorphism can be characterized in a similar fashion; see Propositions (7.3) and (7.4). This is the key observation that allows us to describe, in Theorems (7.10) and (7.11), the subcategories of $`๐_{\mathrm{prj}}(R)`$ and $`๐_{\mathrm{inj}}(R)`$ where the functors $`G`$ and $`F`$ restrict to equivalences.
Building on these results, and translating to the derived category, we arrive at:
###### Theorem III.
A complex $`X`$ of $`R`$-modules has finite G-projective dimension if and only if the morphism $`X๐\mathrm{Hom}_R(D,D_R^๐X)`$ in $`๐(R)`$ is an isomorphism and $`H(D_R^๐X)`$ is bounded on the left.
The notion of finite G-projective dimension, and finite G-injective dimension, is recalled in Section 8. The result above is part of Theorem (8.1); its counterpart for G-injective dimensions is Theorem (8.2). Given these, it is clear that Theorem I restricts to an equivalence between the category of complexes of finite G-projective dimension and the category of complexes of finite G-injective dimension.
Theorems (8.1) and (8.2) recover recent results of Christensen, Frankild, and Holm , who arrived at them from a different perspective. The approach presented here clarifies the connection between finiteness of G-dimension and (total) acyclicity, and uncovers a connection between Grothendieck duality and the equivalence between the categories of complexes of finite G-projective dimension and of finite G-injective dimension by realizing them as different shadows of the same equivalence: that given by Theorem I.
So far we have focused on the case where the ring $`R`$ is commutative. However, the results carry over, with suitable modifications in the statements and with nearly identical proofs, to non-commutative rings that possess dualizing complexes; the appropriate comments are collected towards the end of each section. We have chosen to present the main body of the work, Sections 48, in the commutative context in order to keep the underlying ideas transparent, and unobscured by notational complexity.
### Notation
The following symbols are used to label arrows representing functors or morphisms: $``$ indicates an equivalence (between categories), $``$ an isomorphism (between objects), and $``$ a quasi-isomorphism (between complexes).
## 1. Triangulated categories
This section is primarily a summary of basic notions and results about triangulated categories used frequently in this article. For us, the relevant examples of triangulated categories are homotopy categories of complexes over noetherian rings; they are the focus of the next section. Our basic references are Weibel , Neeman , and Verdier .
###### 1.1.
Triangulated categories. Let $`๐ฏ`$ be a triangulated category. We refer the reader to and for the axioms that define a triangulated category. When we speak of subcategories, it is implicit that they are full.
A non-empty subcategory $`๐ฎ`$ of $`๐ฏ`$ is said to be *thick* if it is a triangulated subcategory of $`๐ฏ`$ that is closed under retracts. If, in addition, $`๐ฎ`$ is closed under all coproducts allowed in $`๐ฏ`$, then it is *localizing*; if it is closed under all products in $`๐ฏ`$ it is *colocalizing*.
Let $`๐`$ be a class of objects in $`๐ฏ`$. The intersection of the thick subcategories of $`๐ฏ`$ containing $`๐`$ is a thick subcategory, denoted $`\mathrm{Thick}(๐)`$. We write $`\mathrm{Loc}(๐)`$, respectively, $`\mathrm{Coloc}(๐)`$, for the intersection of the localizing, respectively, colocalizing, subcategories containing $`๐`$. Note that $`\mathrm{Loc}(๐)`$ is itself localizing, while $`\mathrm{Coloc}(๐)`$ is colocalizing.
###### 1.2.
Compact objects and generators. Let $`๐ฏ`$ be a triangulated category admitting arbitrary coproducts. An object $`X`$ of $`๐ฏ`$ is *compact* if $`\mathrm{Hom}_๐ฏ(X,)`$ commutes with coproducts; that is to say, for each coproduct $`_iY_i`$ of objects in $`๐ฏ`$, the natural morphism of abelian groups
$$\underset{i}{}\mathrm{Hom}_๐ฏ(X,Y_i)\mathrm{Hom}_๐ฏ(X,\underset{i}{}Y_i)$$
is bijective. The compact objects form a thick subcategory that we denote $`๐ฏ^c`$. We say that a class of objects $`๐ฎ`$ *generates* $`๐ฏ`$ if $`\mathrm{Loc}(๐ฎ)=๐ฏ`$, and that $`๐ฏ`$ is *compactly generated* if there exists a generating set consisting of compact objects.
Let $`๐ฎ`$ be a class of compact objects in $`๐ฏ`$. Then $`๐ฎ`$ generates $`๐ฏ`$ if and only if for any object $`Y`$ of $`๐ฏ`$, we have $`Y=0`$ provided that $`\mathrm{Hom}_๐ฏ(\mathsf{\Sigma }^nS,Y)=0`$ for all $`S`$ in $`๐ฎ`$ and $`n`$; see \[18, (2.1)\].
Adjoint functors play a useful, if technical, role in this work, and pertinent results on these are collected in the following paragraphs. MacLaneโs book \[15, Chapter IV\] is the basic reference for this topic; see also \[23, (A.6)\].
###### 1.3.
Adjoint functors. Given categories $`๐`$ and $``$, a diagram
indicates that $`๐ฅ`$ and $`๐ฆ`$ are adjoint functors, with $`๐ฅ`$ left adjoint to $`๐ฆ`$; that is to say, there is a natural isomorphism $`\mathrm{Hom}_{}(๐ฅ(A),B)\mathrm{Hom}_๐(A,๐ฆ(B))`$ for $`A๐`$ and $`B`$.
###### 1.4.
Let $`๐ฏ`$ be a category, $`๐ฎ`$ a full subcategory of $`๐ฏ`$, and $`๐:๐ฏ๐ฎ`$ a right adjoint of the inclusion $`\mathrm{๐๐๐ผ}:๐ฎ๐ฏ`$. Then $`๐\mathrm{๐๐๐ผ}\mathrm{๐๐ฝ}_๐ฎ`$. Moreover, for each $`T`$ in $`๐ฏ`$, an object $`P`$ in $`๐ฎ`$ is isomorphic to $`๐(T)`$ if and only if there is a morphism $`PT`$ with the property that the induced map $`\mathrm{Hom}_๐ฏ(S,P)\mathrm{Hom}_๐ฏ(S,T)`$ is bijective for each $`S๐ฎ`$.
###### 1.5.
Let $`๐ฅ:๐ฎ๐ฏ`$ be an exact functor between triangulated categories such that $`๐ฎ`$ is compactly generated.
1. The functor $`๐ฅ`$ admits a right adjoint if and only if it preserves coproducts.
2. The functor $`๐ฅ`$ admits a left adjoint if and only if it preserves products.
3. If $`๐ฅ`$ admits a right adjoint $`๐ฆ`$, then $`๐ฅ`$ preserves compactness if and only if $`๐ฆ`$ preserves coproducts.
For (1), we refer to \[18, (4.1)\]; for (2), see \[19, (8.6.1)\]; for (3), see \[18, (5.1)\].
###### 1.6.
Orthogonal classes. Given a class $`๐`$ of objects in a triangulated category $`๐ฏ`$, the full subcategories
$`๐^{}`$ $`=\{Y๐ฏ\mathrm{Hom}_๐ฏ(\mathsf{\Sigma }^nX,Y)=0\text{for all }X๐\text{ and }n\},`$
$`{}_{}{}^{}๐`$ $`=\{X๐ฏ\mathrm{Hom}_๐ฏ(X,\mathsf{\Sigma }^nY)=0\text{for all }Y๐\text{ and }n\}.`$
are called the classes *right orthogonal* and *left orthogonal* to $`๐`$, respectively. It is elementary to verify that $`๐^{}`$ is a colocalizing subcategory of $`๐ฏ`$, and equals $`\mathrm{Thick}(๐)^{}`$. In the same vein, $`{}_{}{}^{}๐`$ is a localizing subcategory of $`๐ฏ`$, and equals $`{}_{}{}^{}\mathrm{Thick}(๐)`$.
Caveat: Our notation for orthogonal classes conflicts with the one in .
An additive functor $`๐ฅ:๐`$ between additive categories is an *equivalence up to direct factors* if $`๐ฅ`$ is full and faithful, and every object in $``$ is a direct factor of some object in the image of $`๐ฅ`$.
###### Proposition 1.7.
Let $`๐ฏ`$ be a compactly generated triangulated category and let $`๐๐ฏ`$ be a class of compact objects.
1. The triangulated category $`๐^{}`$ is compactly generated. The inclusion $`๐^{}๐ฏ`$ admits a left adjoint which induces, up to direct factors, an equivalence
$$๐ฏ^c/\mathrm{Thick}(๐)\stackrel{}{}(๐^{})^c.$$
2. For each class $`๐`$, the triangulated category $`^{}/๐^{}`$ is compactly generated. The canonical functor $`^{}^{}/๐^{}`$ induces, up to direct factors, an equivalence
$$\mathrm{Thick}(๐)/\mathrm{Thick}()\stackrel{}{}(^{}/๐^{})^c.$$
###### Proof.
First observe that $`๐`$ can be replaced by a set of objects because the isomorphism classes of compact objects in $`๐ฏ`$ form a set. Neeman gives in \[17, (2.1)\] a proof of (1); see also \[17, p. 553 ff\]. For (2), consider the following diagram
where $`๐บ`$ and $`๐ป`$ denote adjoints of the corresponding inclusion functors and unlabeled functors are induced by $`๐บ`$ and $`๐ป`$ respectively. The localizing subcategory $`\mathrm{Loc}(๐)`$ of $`๐ฏ`$ is generated by $`๐`$ and hence it is compactly generated and its full subcategory of compact objects is precisely $`\mathrm{Thick}(๐)`$; see \[17, (2.2)\]. Moreover, the composite
$$\mathrm{Loc}(๐)\stackrel{๐๐๐ผ}{}๐ฏ\stackrel{๐ผ๐บ๐}{}๐ฏ/๐^{}$$
is an equivalence. From the right hand square one obtains an analogous description of $`^{}/๐^{}`$, namely: the objects of $`๐`$ in $`๐ฏ^c/\mathrm{Thick}()`$ generate a localizing subcategory of $`^{}`$, and this subcategory is compactly generated and equivalent to $`^{}/๐^{}`$. Moreover, the full subcategory of compact objects in $`^{}/๐^{}`$ is equivalent to the thick subcategory generated by $`๐`$ which is, up to direct factors, equivalent to $`\mathrm{Thick}(๐)/\mathrm{Thick}()`$. โ
## 2. Homotopy categories
We begin this section with a recapitulation on the homotopy category of an additive category. Then we introduce the main objects of our study: the homotopy categories of projective modules, and of injective modules, over a noetherian ring, and establish results which prepare us for the development in the ensuing sections.
Let $`๐`$ be an additive category; see \[23, (A.4)\]. We grade complexes cohomologically, thus a complex $`X`$ over $`๐`$ is a diagram
$$\mathrm{}X^n\stackrel{^n}{}X^{n+1}\stackrel{^{n+1}}{}X^{n+2}\mathrm{}$$
with $`X^n`$ in $`๐`$ and $`^{n+1}^n=0`$ for each integer $`n`$. For such a complex $`X`$, we write $`\mathsf{\Sigma }X`$ for its suspension: $`(\mathsf{\Sigma }X)^n=X^{n+1}`$ and $`_{\mathsf{\Sigma }X}=_X`$.
Let $`๐(๐)`$ be the homotopy category of complexes over $`๐`$; its objects are complexes over $`๐`$, and its morphisms are morphisms of complexes modulo homotopy equivalence. The category $`๐(๐)`$ has a natural structure of a triangulated category; see or .
Let $`R`$ be a ring. Unless stated otherwise, modules are left modules; right modules are sometimes referred to as modules over $`R^{\mathrm{op}}`$, the opposite ring of $`R`$. This proclivity for the left carries over to properties of the ring as well: when we say noetherian without any further specification, we mean left noetherian, etc. We write $`๐(R)`$ for the homotopy category of complexes over $`R`$; it is $`๐(๐)`$ with $`๐`$ the category of $`R`$-modules. The paragraphs below contain basic facts on homotopy categories required in the sequel.
###### 2.1.
Let $`๐`$ be an additive category, and let $`X`$ and $`Y`$ complexes over $`๐`$. Set $`๐=๐(๐)`$. Let $`d`$ be an integer. We write $`X^d`$ for the subcomplex
$$\mathrm{}0X^dX^{d+1}\mathrm{}$$
of $`X`$, and $`X^{d1}`$ for the quotient complex $`X/X^d`$. In $`๐`$ these fit into an exact triangle
$$X^dXX^{d1}\mathsf{\Sigma }X^d$$
This induces homomorphisms of abelian groups $`\mathrm{Hom}_๐(X,Y)\mathrm{Hom}_๐(X^d,Y)`$ and $`\mathrm{Hom}_๐(X^{d1},Y)\mathrm{Hom}_๐(X,Y)`$. These have the following properties.
1. One has isomorphisms of abelian groups:
$$H^d(\mathrm{Hom}_๐(X,Y))\mathrm{Hom}_๐(X,\mathsf{\Sigma }^dY)\mathrm{Hom}_๐(\mathsf{\Sigma }^dX,Y).$$
2. If $`Y^n=0`$ for $`nd`$, then the map $`\mathrm{Hom}_๐(X^d,Y)\mathrm{Hom}_๐(X,Y)`$ is bijective.
3. If $`Y^n=0`$ for $`nd`$, then the map $`\mathrm{Hom}_๐(X,Y)\mathrm{Hom}_๐(X^d,Y)`$ is bijective.
There are also versions of (2) and (3), where the hypothesis is on $`X`$.
Indeed, these remarks are all well-known, but perhaps (2) and (3) less so than (1). To verify (2), note that (1) implies
$$H^0(\mathrm{Hom}_๐(X^{d+1},Y))=0=H^1(\mathrm{Hom}_๐(X^{d+1},Y)),$$
so applying $`\mathrm{Hom}_๐(,Y)`$ to the exact triangle ($``$) yields that the induced homomorphism of abelian groups
$$H^0(\mathrm{Hom}_๐(X^d,Y))H^0(\mathrm{Hom}_๐(X,Y))$$
is bijective, which is as desired. The argument for (3) is similar.
Now we recall, with proof, a crucial observation from \[14, (2.1)\]:
###### 2.2.
Let $`R`$ be a ring, $`M`$ an $`R`$-module, and let $`๐M`$ be an injective resolution of $`M`$. Set $`๐=๐(R)`$. If $`Y`$ is a complex of injective $`R`$-modules, the induced map
$$\mathrm{Hom}_๐(๐M,Y)\mathrm{Hom}_๐(M,Y)$$
is bijective. In particular, $`\mathrm{Hom}_๐(๐R,Y)H^0(Y)`$.
Indeed, one may assume $`(๐M)^n=0`$ for $`n1`$, since all injective resolutions of $`M`$ are isomorphic in $`๐`$. The inclusion $`M๐M`$ leads to an exact sequence of complexes
$$0M๐MX0$$
with $`X^n=0`$ for $`n1`$ and $`H(X)=0`$. Therefore for $`d=1,0`$ one has isomorphisms
$$\mathrm{Hom}_๐(\mathsf{\Sigma }^dX,Y)\mathrm{Hom}_๐(\mathsf{\Sigma }^dX,Y^1)=0,$$
where the first one holds by an analogue of (2.1.2), and the second holds because $`Y^1`$ is a complex of injectives bounded on the left. It now follows from the exact sequence above that the induced map $`\mathrm{Hom}_๐(๐M,Y)\mathrm{Hom}_๐(M,Y)`$ is bijective.
The results below are critical ingredients in many of our arguments. We write $`๐^{,b}(\mathrm{prj}R)`$ for the subcategory of $`๐(R)`$ consisting of complexes $`X`$ of finitely generated projective modules with $`H(X)`$ bounded and $`X^n=0`$ for $`n0`$, and $`๐^f(R)`$ for its image in $`๐(R)`$, the derived category of $`R`$-modules.
###### 2.3.
Let $`R`$ be a (not necessarily commutative) ring.
1. When $`R`$ is coherent on both sides and flat $`R`$-modules have finite projective dimension, the triangulated category $`๐(\mathrm{Prj}R)`$ is compactly generated and the functors $`\mathrm{Hom}_R(,R):๐(\mathrm{Prj}R)๐(R^{\mathrm{op}})`$ and $`๐(R^{\mathrm{op}})๐(R^{\mathrm{op}})`$ induce equivalences
$$๐^c(\mathrm{Prj}R)\stackrel{}{}๐^{,b}(\mathrm{prj}R^{\mathrm{op}})^{\mathrm{op}}\stackrel{}{}๐^f(R^{\mathrm{op}})^{\mathrm{op}}.$$
2. When $`R`$ is noetherian, the triangulated category $`๐(\mathrm{Inj}R)`$ is compactly generated, and the canonical functor $`๐(\mathrm{Inj}R)๐(R)`$ induces an equivalence
$$๐^c(\mathrm{Inj}R)\stackrel{}{}๐^f(R)$$
Indeed, (1) is a result of Jรธrgensen \[11, (2.4)\] and (2) is a result of Krause \[14, (2.3)\].
In the propositions below $`d(R)`$ denotes the supremum of the projective dimensions of all flat $`R`$-modules.
###### Proposition 2.4.
Let $`R`$ be a two-sided coherent ring such that $`d(R)`$ is finite. The inclusion $`๐(\mathrm{Prj}R)๐(\mathrm{Flat}R)`$ admits a right adjoint:
Moreover, the category $`๐(\mathrm{Prj}R)`$ admits arbitrary products.
###### Proof.
By Proposition (2.3.1), the category $`๐(\mathrm{Prj}R)`$ is compactly generated. The inclusion $`\mathrm{๐๐๐ผ}`$ evidently preserves coproducts, so (1.5.1) yields the desired right adjoint $`๐`$. The ring $`R`$ is right coherent, so the (set-theoretic) product of flat modules is flat, and furnishes $`๐(\mathrm{Flat}R)`$ with a product. Since $`\mathrm{๐๐๐ผ}`$ is an inclusion, the right adjoint $`๐`$ induces a product on $`๐(\mathrm{Prj}R)`$: the product of a set of complexes $`\{P_\lambda \}_{\lambda \mathrm{\Lambda }}`$ in $`๐(\mathrm{Prj}R)`$ is the complex $`๐\left(_\lambda P_\lambda \right)`$. โ
The proof of Theorem 2.7 below uses homotopy limits in the homotopy category of complexes; its definition is recalled below.
###### 2.5.
Homotopy limits. Let $`R`$ be a ring and let $`\mathrm{}X(r+1)X(r)`$ be a sequence of morphisms in $`๐(R)`$. The *homotopy limit* of the sequence $`\{X(i)\}`$, denoted $`\mathrm{holim}X(i)`$, is defined by an exact triangle
The homotopy limit is uniquely defined, up to an isomorphism in $`๐(R)`$; see for details.
The result below identifies, in some cases, a homotopy limit in the homotopy category with a limit in the category of complexes.
###### Lemma 2.6.
Let $`R`$ be a ring. Consider a sequence of complexes of $`R`$-modules:
$$\mathrm{}X(i)\stackrel{\epsilon (i)}{}X(i1)\mathrm{}X(r+1)\stackrel{\epsilon (r+1)}{}X(r).$$
If for each degree $`n`$, there exists an integer $`s_n`$ such that $`\epsilon (i)^n`$ is an isomorphism for $`is_n+1`$, then there exists a degree-wise split-exact sequence of complexes
In particular, it induces in $`๐(R)`$ an isomorphism $`\mathrm{holim}X(i)\underset{}{\mathrm{lim}}X(i)`$.
###### Proof.
To prove the desired degree-wise split exactness of the sequence, it suffices to note that if $`\mathrm{}M(r+1)\stackrel{\delta (r+1)}{}M(r)`$ is a sequence of $`R`$-modules such that $`\delta (i)`$ is an isomorphism for $`is+1`$, for some integer $`s`$, then one has a split exact sequence of $`R`$-modules:
where the morphism $`\eta `$ is induced by $`\eta _i:M(s)M(i)`$ with
$$\eta _i=\{\begin{array}{cc}\delta (i+1)\mathrm{}\delta (s)\hfill & \text{if }is1\hfill \\ \mathrm{id}\hfill & \text{if }i=s\hfill \\ \delta (i)^1\mathrm{}\delta (s+1)^1\hfill & \text{if }is+1.\hfill \end{array}$$
Indeed, in the sequence above, the map ($`\mathrm{id}\mathrm{shift}`$) is surjective since the system $`\{M_i\}`$ evidently satisfies the Mittag-Leffler condition, see \[23, (3.5.7)\]. Moreover, a direct calculation shows that $`\mathrm{Im}(\eta )=\mathrm{Ker}(\mathrm{id}\mathrm{shift})`$. It remains to note that the morphism $`\pi :M(i)M(s)`$ defined by $`\pi (a_i)=a_s`$ is such that $`\pi \eta =\mathrm{id}`$.
Finally, it is easy to verify that degree-wise split exact sequences of complexes induce exact triangles in the homotopy category. Thus, by the definition of homotopy limits, see (2.5), and the already established part of the lemma, we deduce: $`\mathrm{holim}X(i)\underset{}{\mathrm{lim}}X(i)`$ in $`๐(R)`$, as desired. โ
The result below collects some properties of the functor $`๐:๐(\mathrm{Flat}R)๐(\mathrm{Prj}R)`$. It is noteworthy that the proof of part (3) describes an explicit method for computing the value of $`๐`$ on complexes bounded on the left. As usual, a morphism of complexes is called a *quasi-isomorphism* if the induced map in homology is bijective.
###### Theorem 2.7.
Let $`R`$ be a two-sided coherent ring with $`d(R)`$ finite, and let $`F`$ be a complex of flat $`R`$-modules.
1. The morphism $`๐(F)F`$ is a quasi-isomorphism.
2. If $`F^n=0`$ for $`n0`$, then $`๐(F)`$ is a projective resolution of $`F`$.
3. If $`F^n=0`$ for $`nr`$, then $`๐(F)`$ is isomorphic to a complex $`P`$ with $`P^n=0`$ for $`nrd(R)`$.
###### Proof.
(1) For each integer $`n`$, the map $`\mathrm{Hom}_๐(\mathsf{\Sigma }^nR,๐(F))\mathrm{Hom}_๐(\mathsf{\Sigma }^nR,F)`$, induced by the morphism $`๐(F)F`$, is bijective; this is because $`R`$ is in $`๐(\mathrm{Prj}R)`$. Therefore (2.1.1) yields $`H^n(๐(F))H^n(F)`$, which proves (1).
(2) When $`F^n=0`$ for $`nr`$, one can construct a projective resolution $`PF`$ with $`P^n=0`$ for $`nr`$. Thus, for each $`X๐(\mathrm{Prj}R)`$ one has the diagram below
$$\mathrm{Hom}_๐(X^r,P)=\mathrm{Hom}_๐(X,P)\mathrm{Hom}_๐(X,F)=\mathrm{Hom}_๐(X^r,F).$$
where equalities hold by (2.1.2). The complex $`X^r`$ is K-projective, so the composed map is an isomorphism; hence the same is true of the one in the middle. This proves that $`๐(F)P`$; see (1.4).
(3) We may assume $`d(R)`$ is finite. The construction of the complex $`P`$ takes place in the category of complexes of $`R`$-modules. Note that $`F^{>i}`$ is a subcomplex of $`F`$ for each integer $`ir`$ ; denote $`F(i)`$ the quotient complex $`F/F^{>i}`$. One has surjective morphisms of complexes of $`R`$-modules
$$\mathrm{}F(i)\stackrel{\epsilon (i)}{}F(i1)\mathrm{}F(r+1)\stackrel{\epsilon (r+1)}{}F(r)=0$$
with $`\mathrm{Ker}(\epsilon (i))=\mathsf{\Sigma }^iF^i`$. The surjections $`FF(i)`$ are compatible with the $`\epsilon (i)`$, and the induced map $`F\underset{}{\mathrm{lim}}F(i)`$ is an isomorphism. The plan is to construct a commutative diagram in the category of complexes of $`R`$-modules
$$\begin{array}{c}\hfill \text{}\end{array}$$
with the following properties: for each integer $`ir+1`$ one has that
1. $`P(i)`$ consists of projectives $`R`$-modules and $`P(i)^n=0`$ for $`n(rd(R),i]`$;
2. $`\delta (i)`$ is surjective, and $`\mathrm{Ker}\delta (i)^n=0`$ for $`n<id(R)`$;
3. $`\kappa (i)`$ is a surjective quasi-isomorphism.
The complexes $`P(i)`$ and the attendant morphisms are constructed iteratively, starting with $`\kappa (r+1):P(r+1)F(r+1)=\mathsf{\Sigma }^{r+1}F^{r+1}`$ a surjective projective resolution, and $`\delta (r+1)=0`$. One may ensure $`P(r+1)^n=0`$ for $`nr+2`$, and also for $`nrd(R)`$, because the projective dimension of the flat $`R`$-module $`F^{r+1}`$ is at most $`d(R)`$. Note that $`P(r+1)`$, $`\delta (r+1)`$, and $`\kappa (r+1)`$ satisfy conditions (a)โ(c).
Let $`ir+2`$ be an integer, and let $`\kappa (i1):P(i1)F(i1)`$ be a homomorphism with the desired properties. Build a diagram of solid arrows
where $`\iota `$ is the canonical injection, and $`\theta :Q\mathsf{\Sigma }^iF^i`$ is a surjective projective resolution, chosen such that $`Q^n=0`$ for $`n<id(R)`$. The Horseshoe Lemma now yields a complex $`P(i)`$, with underlying graded $`R`$-module $`QP(i1)`$, and dotted morphisms that form the commutative diagram above; see \[23, (2.2.8)\]. It is clear that $`P(i)`$ and $`\delta (i)`$ satisfy conditions (a) and (b). As to (c): since both $`\theta `$ and $`\kappa (i1)`$ are surjective quasi-isomorphisms, so is $`\kappa (i)`$. This completes the construction of the diagram ($``$).
Set $`P=\underset{}{\mathrm{lim}}P(i)`$; the limit is taken in the category of complexes. We claim that $`P`$ is a complex of projectives and that $`๐(F)P`$ in $`๐(\mathrm{Prj}R)`$.
Indeed, by property (b), for each integer $`n`$ the map $`P(i+1)^nP(i)^n`$ is bijective for $`i>n+d(R)`$, so $`P^n=P(n+d(R))^n`$, and hence the $`R`$-module $`P^n`$ is projective. Moreover $`P^n=0`$ for $`nrd(R)`$, by (a).
The sequences of complexes $`\{P(i)\}`$ and $`\{F(i)\}`$ satisfy the hypotheses of Lemma (2.6); the former by construction, see property (b), and the latter by definition. Thus, Lemma (2.6) yields the following isomorphisms in $`๐(R)`$:
$$\mathrm{holim}P(i)P\text{and}\mathrm{holim}F(i)F.$$
Moreover, the $`\kappa (i)`$ induce a morphism $`\kappa :\mathrm{holim}P(i)\mathrm{holim}F(i)`$ in $`๐(R)`$. Let $`X`$ be a complex of projective $`R`$-modules. To complete the proof of (3), it suffices to prove that for each integer $`i`$ the induced map
$$\mathrm{Hom}_๐(X,\kappa (i)):\mathrm{Hom}_๐(X,P(i))\mathrm{Hom}_๐(X,F(i))$$
is bijective. Then, a standard argument yields that $`\mathrm{Hom}_๐(X,\kappa )`$ is bijective, and in turn this implies $`P\mathrm{holim}P(i)๐(\mathrm{holim}F(i))๐(F)`$, see (1.4).
Note that, since $`\kappa (i)`$ is a quasi-isomorphism and $`P(i)^n=0=F(i)^n`$ for $`ni+1`$, the morphism $`\kappa (i):P(i)F(i)`$ is a projective resolution. Since projective resolutions are isomorphic in the homotopy category, it follows from (2) that $`P(i)๐(F(i))`$, and hence that the map $`\mathrm{Hom}_๐(X,\kappa (i))`$ is bijective, as desired. Thus, (3) is proved. โ
## 3. Dualizing complexes
Let $`R`$ be a commutative noetherian ring. In this article, a *dualizing complex* for $`R`$ is a complex $`D`$ of $`R`$-modules with the following properties:
1. the complex $`D`$ is bounded and consists of injective $`R`$-modules;
2. the $`R`$-module $`H^n(D)`$ is finitely generated for each $`n`$;
3. the canonical map $`R\mathrm{Hom}_R(D,D)`$ is a quasi-isomorphism.
See Hartshorne \[9, Chapter V\] for basic properties of dualizing complexes. The presence of a dualizing complex for $`R`$ implies that its Krull dimension is finite. As to the existence of dualizing complexes: when $`R`$ is a quotient of a Gorenstein ring $`Q`$ of finite Krull dimension, it has a dualizing complex: a suitable representative of the complex $`๐\mathrm{Hom}_Q(R,Q)`$ does the job. On the other hand, Kawasaki has proved that if $`R`$ has a dualizing complex, then it is a quotient of a Gorenstein ring.
###### 3.1.
A dualizing complex induces a contravariant equivalence of categories:
This property characterizes dualizing complexes: if $`C`$ is a complex of $`R`$-modules such that $`๐\mathrm{Hom}_R(,C)`$ induces a contravariant self-equivalence of $`๐^f(R)`$, then $`C`$ is isomorphic in $`๐(R)`$ to a dualizing complex for $`R`$; see \[9, (V.2)\]. Moreover, if $`D`$ and $`E`$ are dualizing complexes for $`R`$, then $`E`$ is quasi-isomorphic to $`P_RD`$ for some complex $`P`$ which is locally free of rank one; that is to say, for each prime ideal $`๐ญ`$ in $`R`$, the complex $`P_๐ญ`$ is quasi-isomorphic $`\mathsf{\Sigma }^nR_๐ญ`$ for some integer $`n`$; see \[9, (V.3)\].
###### Remark 3.2.
Let $`R`$ be a ring with a dualizing complex. Then, as noted above, the Krull dimension of $`R`$ is finite, so a result of Gruson and Raynaud \[20, (II.3.2.7)\] yields that the projective dimension of each flat $`R`$-module is at most the Krull dimension of $`R`$. The upshot is that Proposition (2.4) yields an adjoint functor
and this has properties described in Theorem (2.7). In the remainder of the article, this remark will be used often, and usually without comment.
In , Christensen, Frankild, and Holm have introduced a notion of a dualizing complex for a pair of, possibly non-commutative, rings:
###### 3.3.
Non-commutative rings. In what follows $`S,R`$ denotes a pair of rings, where $`S`$ is left noetherian and $`R`$ is left coherent and right noetherian. This context is more restrictive than that considered in \[6, Section 1\], where it is not assumed that $`R`$ is left coherent. We make this additional hypothesis on $`R`$ in order to invoke (2.3.1).
###### 3.3.1.
A *dualizing complex* for the pair $`S,R`$ is complex $`D`$ of $`S`$-$`R`$ bimodules with the following properties:
1. $`D`$ is bounded and each $`D^n`$ is an $`S`$-$`R`$ bimodule that is injective both as an $`S`$-module and as an $`R^{\mathrm{op}}`$-module;
2. $`H^n(D)`$ is finitely generated as an $`S`$-module and as an $`R^{\mathrm{op}}`$-module for each $`n`$;
3. the following canonical maps are quasi-isomorphisms:
$$R\mathrm{Hom}_S(D,D)\text{and}S\mathrm{Hom}_{R^{\mathrm{op}}}(D,D)$$
When $`R`$ is commutative and $`R=S`$ this notion of a dualizing complex coincides with the one recalled in the beginning of this section. The appendix in contains a detailed comparison with other notions of dualizing complexes in the non-commutative context.
The result below implies that the conclusion of Remark (3.2): existence of a functor $`๐`$ with suitable properties, applies also in the situation considered in (3.3).
###### Proposition 3.4.
Let $`D`$ be a dualizing complex for the pair of rings $`S,R`$, where $`S`$ is left noetherian and $`R`$ is left coherent and right noetherian.
1. The projective dimension of each flat $`R`$-module is finite.
2. The complex $`D`$ induces a contravariant equivalence:
Indeed, (1) is contained in \[6, (1.5)\]. Moreover, (2) may be proved as in the commutative case, see \[9, (V.2.1)\], so we provide only a
###### Sketch of a proof of (2).
By symmetry, it suffices to prove that for each complex $`X`$ of right $`R`$-modules if $`H(X)`$ is bounded and finitely generated in each degree, then so is $`H(\mathrm{Hom}_{R^{\mathrm{op}}}(X,D))`$, as an $`S`$-module, and that the biduality morphism
$$\theta (X):X\mathrm{Hom}_S(\mathrm{Hom}_{R^{\mathrm{op}}}(X,D),D))$$
is a quasi-isomorphism. To begin with, since $`H(X)`$ is bounded, we may pass to a quasi-isomorphic complex and assume $`X`$ is itself bounded, in which case the complex $`\mathrm{Hom}_{R^{\mathrm{op}}}(X,D)`$, and hence its homology, is bounded.
For the remainder of the proof, by replacing $`X`$ by a suitable projective resolution, we assume that each $`X^i`$ is a finitely generated projective module, with $`X^i=0`$ for $`i0`$. In this case, for any bounded complex $`Y`$ of $`S`$-$`R`$ bimodules, if the $`S`$-module $`H(Y)`$ is finitely generated in each degree, then so is the $`S`$-module $`H(\mathrm{Hom}_{R^{\mathrm{op}}}(X,Y))`$; this can be proved by an elementary induction argument, based on the number
$$sup\{iH^i(Y)0\}inf\{iH^i(Y)0\},$$
keeping in mind that $`S`$ is noetherian. Applied with $`Y=D`$, one obtains that each $`H^i(\mathrm{Hom}_{R^{\mathrm{op}}}(X,D))`$ is finitely generated, as desired.
As to the biduality morphism: fix an integer $`n`$, and pick an integer $`dn`$ such that the morphism of complexes
$$\mathrm{Hom}_S(\mathrm{Hom}_{R^{\mathrm{op}}}(X^d,D),D))\mathrm{Hom}_S(\mathrm{Hom}_{R^{\mathrm{op}}}(X,D),D))$$
is bijective in degrees $`n1`$; such a $`d`$ exists because $`D`$ is bounded. Therefore, $`H^n(\theta (X))`$ is bijective if and only if $`H^n(\theta (X^d))`$ is bijective. Thus, passing to $`X^d`$, we may assume that $`X^i=0`$ when $`|i|0`$. One has then a commutative diagram of morphisms of complexes
The isomorphism on the right holds because $`X`$ is a finite complex of finitely generated projectives; for the same reason, since $`\theta (R)`$ is a quasi-isomorphism, see (3.3.1.c), so is $`X_R\theta (R)`$. Thus, $`\theta (X)`$ is a quasi-isomorphism. This completes the proof. โ
## 4. An equivalence of homotopy categories
The standing assumption in the rest of this article is that $`R`$ is a *commutative* noetherian ring. Towards the end of each section we collect remarks on the extensions of our results to the non-commutative context described in (3.3).
The main theorem in this section is an equivalence between the homotopy categories of complexes of projectives and complexes of injectives. As explained in the discussion following Theorem I in the introduction, it may be viewed as an extension of the Grothendieck duality theorem, recalled in (3.1). Theorem (4.2) is the basis for most results in this work.
###### Remark 4.1.
Let $`D`$ be a dualizing complex for $`R`$; see Section 3.
For any flat module $`F`$ and injective module $`I`$, the $`R`$-module $`I_RF`$ is injective; this is readily verified using Baerโs criterion. Thus, $`D_R`$ is a functor between $`๐(\mathrm{Prj}R)`$ and $`๐(\mathrm{Inj}R)`$, and it factors through $`๐(\mathrm{Flat}R)`$. If $`I`$ and $`J`$ are injective modules, the $`R`$-module $`\mathrm{Hom}_R(I,J)`$ is flat, so $`\mathrm{Hom}_R(D,)`$ defines a functor from $`๐(\mathrm{Inj}R)`$ to $`๐(\mathrm{Flat}R)`$; evidently it is right adjoint to $`D_R:๐(\mathrm{Flat}R)๐(\mathrm{Inj}R)`$.
Here is the announced equivalence of categories. The existence of $`๐`$ in the statement below is explained in Remark (3.2), and the claims implicit in the right hand side of the diagram are justified by the preceding remark.
###### Theorem 4.2.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$. The functor $`D_R:๐(\mathrm{Prj}R)๐(\mathrm{Inj}R)`$ is an equivalence. A quasi-inverse is $`๐\mathrm{Hom}_R(D,)`$:
where $`๐`$ denotes the right adjoint of the inclusion $`๐(\mathrm{Prj}R)๐(\mathrm{Flat}R)`$.
###### 4.3.
The functors that appear in the theorem are everywhere dense in the remainder of this article, so it is expedient to abbreviate them: set
$`๐ณ=D_R:๐(\mathrm{Prj}R)๐(\mathrm{Inj}R)\text{and}`$
$`๐ฒ=๐\mathrm{Hom}_R(D,):๐(\mathrm{Inj}R)๐(\mathrm{Prj}R).`$
The notation โ$`๐ณ`$โ should remind one that this functor is given by a tensor product. The same rule would call for an โ$`๐ง`$โ to denote the other functor; unfortunately, this letter is bound to be confounded with an โ$`H`$โ, so we settle for an โ$`๐ฒ`$โ.
###### Proof.
By construction, $`(\mathrm{๐๐๐ผ},๐)`$ and $`(D_R,\mathrm{Hom}_R(D,))`$ are adjoint pairs of functors. It follows that their composition $`(๐ณ,๐ฒ)`$ is an adjoint pair of functors as well. Thus, it suffices to prove that $`๐ณ`$ is an equivalence: this would imply that $`S`$ is its quasi-inverse, and hence also an equivalence.
Both $`๐(\mathrm{Prj}R)`$ and $`๐(\mathrm{Inj}R)`$ are compactly generated, by Proposition (2.3), and $`๐ณ`$ preserves coproducts. It follows, using a standard argument, that it suffices to verify that $`๐ณ`$ induces an equivalence $`๐^c(\mathrm{Prj}R)๐^c(\mathrm{Inj}R)`$. Observe that each complex $`P`$ of finitely generated projective $`R`$-modules satisfies
$$\mathrm{Hom}_R(P,D)D_R\mathrm{Hom}_R(P,R).$$
Thus one has the following commutative diagram
By (2.3.2), the equivalence $`๐^+(\mathrm{Inj}R)๐^+(R)`$ identifies $`๐^c(\mathrm{Inj}R)`$ with $`๐^f(R)`$, while by (3.1), the functor $`\mathrm{Hom}_R(,D)`$ induces an auto-equivalence of $`๐^f(R)`$. Hence, by the commutative diagram above, $`๐ณ`$ induces an equivalence $`๐^c(\mathrm{Prj}R)๐^c(\mathrm{Inj}R)`$. This completes the proof. โ
In the proof above we utilized the fact that $`๐(\mathrm{Prj}R)`$ and $`๐(\mathrm{Inj}R)`$ admit coproducts compatible with $`๐ณ`$. The categories in question also have products; this is obvious for $`๐(\mathrm{Inj}R)`$, and contained in Proposition (2.4) for $`๐(\mathrm{Prj}R)`$. The equivalence of categories established above implies:
###### Corollary 4.4.
The functors $`๐ณ`$ and $`๐ฒ`$ preserve coproducts and products.
###### Remark 4.5.
Let $`๐R`$ be an injective resolution of $`R`$, and set $`D^{}=๐ฒ(๐R)`$. Injective resolutions of $`R`$ are uniquely isomorphic in $`๐(\mathrm{Inj}R)`$, so the complex $`๐ฒ(๐R)`$ is independent up to isomorphism of the choice of $`๐R`$, so one may speak of $`D^{}`$ without referring to $`๐R`$.
###### Lemma 4.6.
The complex $`D^{}`$ is isomorphic to the image of $`D`$ under the composition
$$๐^f(R)\stackrel{}{}๐^{,b}(\mathrm{prj}R)\stackrel{\mathrm{Hom}_R(,R)}{}๐(\mathrm{Prj}R).$$
###### Proof.
The complex $`D`$ is bounded and has finitely generated homology modules, so we may choose a projective resolution $`P`$ of $`D`$ with each $`R`$-module $`P^n`$ finitely generated, and zero for $`n0`$. In view of Theorem (4.2), it suffices to verify that $`๐ณ(\mathrm{Hom}_R(P,R))`$ is isomorphic to $`๐R`$. The complex $`๐ณ(\mathrm{Hom}_R(P,R))`$, that is to say, $`D_R\mathrm{Hom}_R(P,R)`$ is isomorphic to the complex $`\mathrm{Hom}_R(P,D)`$, which consists of injective $`R`$-modules and is bounded on the left. Therefore $`\mathrm{Hom}_R(P,D)`$ is K-injective. Moreover, the composite
$$R\mathrm{Hom}_R(D,D)\mathrm{Hom}_R(P,D)$$
is a quasi-isomorphism, and one obtains that in $`๐(\mathrm{Inj}R)`$ the complex $`\mathrm{Hom}_R(P,D)`$ is an injective resolution of $`R`$. โ
The objects in the subcategory $`\mathrm{Thick}(\mathrm{Prj}R)`$ of $`๐(\mathrm{Prj}R)`$ are exactly the complexes of finite projective dimension; those in the subcategory $`\mathrm{Thick}(\mathrm{Inj}R)`$ of $`๐(\mathrm{Inj}R)`$ are the complexes of finite injective dimension. It is known that the functor $`D_R`$ induces an equivalence between these categories; see, for instance, \[1, (1.5)\]. The result below may be read as the statement that this equivalence extends to the full homotopy categories.
###### Proposition 4.7.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$. The equivalence $`๐ณ:๐(\mathrm{Prj}R)๐(\mathrm{Inj}R)`$ restricts to an equivalence between $`\mathrm{Thick}(\mathrm{Prj}R)`$ and $`\mathrm{Thick}(\mathrm{Inj}R)`$. In particular, $`\mathrm{Thick}(\mathrm{Inj}R)`$ equals $`\mathrm{Thick}(\mathrm{Add}D)`$.
###### Proof.
It suffices to prove that the adjoint pair of functors $`(๐ณ,๐ฒ)`$ in Theorem (4.2) restrict to functors between $`\mathrm{Thick}(\mathrm{Prj}R)`$ and $`\mathrm{Thick}(\mathrm{Inj}R)`$.
The functor $`๐ณ`$ maps $`R`$ to $`D`$, which is a bounded complex of injectives and hence in $`\mathrm{Thick}(\mathrm{Inj}R)`$. Therefore $`๐ณ`$ maps $`\mathrm{Thick}(\mathrm{Prj}R)`$ into $`\mathrm{Thick}(\mathrm{Inj}R)`$.
Conversely, given injective $`R`$-modules $`I`$ and $`J`$, the $`R`$-module $`\mathrm{Hom}_R(I,J)`$ is flat. Therefore $`\mathrm{Hom}_R(D,)`$ maps $`\mathrm{Thick}(\mathrm{Inj}R)`$ into $`\mathrm{Thick}(\mathrm{Flat}R)`$, since $`D`$ is a bounded complex of injectives. By Theorem (2.7.2), for each flat $`R`$-module $`F`$, the complex $`๐(F)`$ is a projective resolution of $`F`$. The projective dimension of $`F`$ is finite since $`R`$ has a dualizing complex; see (3.2). Hence $`๐`$ maps $`\mathrm{Thick}(\mathrm{Flat}R)`$ to $`\mathrm{Thick}(\mathrm{Prj}R)`$. โ
###### 4.8.
Non-commutative rings. Consider a pair of rings $`S,R`$ as in (3.3), with a dualizing complex $`D`$. Given Proposition (3.4), the proof of Theorem (4.2) carries over verbatim to yield:
###### Theorem.
The functor $`D_R:๐(\mathrm{Prj}R)๐(\mathrm{Inj}S)`$ is an equivalence, and the functor $`๐\mathrm{Hom}_S(D,)`$ is a quasi-inverse. โ
This basic step accomplished, one can readily transcribe the remaining results in this section, and their proofs, to apply to the pair $`S,R`$; it is clear what the corresponding statements should be.
## 5. Acyclicity versus total acyclicity
This section contains various results concerning the classes of (totally) acyclic complexes of projectives, and of injectives. We start by recalling appropriate definitions.
###### 5.1.
Acyclic complexes. A complex $`X`$ of $`R`$-modules is *acyclic* if $`H^nX=0`$ for each integer $`n`$. We denote $`๐_{\mathrm{ac}}(R)`$ the full subcategory of $`๐(R)`$ formed by acyclic complexes of $`R`$-modules. Set
$$๐_{\mathrm{ac}}(\mathrm{Prj}R)=๐(\mathrm{Prj}R)๐_{\mathrm{ac}}(R)\text{and}๐_{\mathrm{ac}}(\mathrm{Inj}R)=๐(\mathrm{Inj}R)๐_{\mathrm{ac}}(R).$$
Evidently acyclicity is a property intrinsic to the complex under consideration. Next we introduce a related notion which depends on a suitable subcategory of $`\mathrm{Mod}R`$.
###### 5.2.
Total acyclicity. Let $`๐`$ be an additive category. A complex $`X`$ over $`๐`$ is *totally acyclic* if for each object $`A๐`$ the following complexes of abelian groups are acyclic.
$$\mathrm{Hom}_๐(A,X)\text{and}\mathrm{Hom}_๐(X,A)$$
We denote by $`๐_{\mathrm{tac}}(๐)`$ the full subcategory of $`๐(๐)`$ consisting of totally acyclic complexes. Specializing to $`๐=\mathrm{Prj}R`$ and $`A=\mathrm{Inj}R`$ one gets the notion of a *totally acyclic complex of projectives* and a *totally acyclic complex of injectives*, respectively.
Theorems (5.3) and (5.4) below describe various properties of (totally) acyclic complexes. In what follows, we write $`๐_{\mathrm{ac}}^c(\mathrm{Prj}R)`$ and $`๐_{\mathrm{ac}}^c(\mathrm{Inj}R)`$ for the class of compact objects in $`๐_{\mathrm{ac}}(\mathrm{Prj}R)`$ and $`๐_{\mathrm{ac}}(\mathrm{Inj}R)`$, respectively; in the same way, $`๐_{\mathrm{tac}}^c(\mathrm{Prj}R)`$ and $`๐_{\mathrm{tac}}^c(\mathrm{Inj}R)`$ denote compacts among the corresponding totally acyclic objects.
###### Theorem 5.3.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$.
1. The categories $`๐_{\mathrm{ac}}(\mathrm{Prj}R)`$ and $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ are compactly generated.
2. The equivalence $`๐^f(R)๐^c(\mathrm{Prj}R)^{\mathrm{op}}`$ induces, up to direct factors, equivalences
$$๐^f(R)/\mathrm{Thick}(R)\stackrel{}{}๐_{\mathrm{ac}}^c(\mathrm{Prj}R)^{\mathrm{op}}$$
$$๐^f(R)/\mathrm{Thick}(R,D)\stackrel{}{}๐_{\mathrm{tac}}^c(\mathrm{Prj}R)^{\mathrm{op}}.$$
3. The quotient $`๐_{\mathrm{ac}}(\mathrm{Prj}R)/๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ is compactly generated, and one has, up to direct factors, an equivalence
$$\mathrm{Thick}(R,D)/\mathrm{Thick}(R)\stackrel{}{}\left[\left(๐_{\mathrm{ac}}(\mathrm{Prj}R)/๐_{\mathrm{tac}}(\mathrm{Prj}R)\right)^c\right]^{\mathrm{op}}.$$
The proof of this result, and also of the one below, which is an analogue for complexes of injectives, is given in (5.10). It should be noted that, in both cases, part (1) is not new: for the one above, see the proof of \[12, (1.9)\], and for the one below, see \[14, (7.3)\].
###### Theorem 5.4.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$.
1. The categories $`๐_{\mathrm{ac}}(\mathrm{Inj}R)`$ and $`๐_{\mathrm{tac}}(\mathrm{Inj}R)`$ are compactly generated.
2. The equivalence $`๐^f(R)๐^c(\mathrm{Inj}R)`$ induces, up to direct factors, equivalences
$$๐^f(R)/\mathrm{Thick}(R)\stackrel{}{}๐_{\mathrm{ac}}^c(\mathrm{Inj}R)$$
$$๐^f(R)/\mathrm{Thick}(R,D)\stackrel{}{}๐_{\mathrm{tac}}^c(\mathrm{Inj}R).$$
3. The quotient $`๐_{\mathrm{ac}}(\mathrm{Inj}R)/๐_{\mathrm{tac}}(\mathrm{Inj}R)`$ is compactly generated, and we have, up to direct factors, an equivalence
$$\mathrm{Thick}(R,D)/\mathrm{Thick}(R)\stackrel{}{}\left(๐_{\mathrm{ac}}(\mathrm{Inj}R)/๐_{\mathrm{tac}}(\mathrm{Inj}R)\right)^c.$$
Here is one consequence of the preceding results. In it, one cannot restrict to complexes (of projectives or of injectives) of finite modules; see the example in Section 6.
###### Corollary 5.5.
Let $`R`$ be a noetherian ring with a dualizing complex. The following conditions are equivalent.
1. The ring $`R`$ is Gorenstein.
2. Every acyclic complex of projective $`R`$-modules is totally acyclic.
3. Every acyclic complex of injective $`R`$-modules is totally acyclic.
###### Proof.
Theorems (5.3.3) and (5.4.3) imply that (b) and (c) are equivalent, and that they hold if and only if $`D`$ lies in $`\mathrm{Thick}(R)`$, that is to say, if and only if $`D`$ has finite projective dimension. This last condition is equivalent to $`R`$ being Gorenstein; see \[9, (V.7.1)\]. โ
###### Remark 5.6.
One way to interpret Theorems (5.3.3) and (5.4.3) is that the category $`\mathrm{Thick}(R,D)/\mathrm{Thick}(R)`$ measures the failure of the Gorenstein property for $`R`$. This invariant of $`R`$ appears to possess good functorial properties. For instance, let $`R`$ and $`S`$ be local rings with dualizing complexes $`D_R`$ and $`D_S`$, respectively. If a local homomorphism $`RS`$ is quasi-Gorenstein, in the sense of Avramov and Foxby \[1, Section 7\], then tensoring with $`S`$ induces an equivalence of categories, up to direct factors:
$$_R^๐S:\mathrm{Thick}(R,D_R)/\mathrm{Thick}(R)\stackrel{}{}\mathrm{Thick}(S,D_S)/\mathrm{Thick}(S)$$
This is a quantitative enhancement of the ascent and descent of the Gorenstein property along such homomorphisms.
The notion of total acyclicity has a useful expression in the notation of (1.6).
###### Lemma 5.7.
Let $`๐`$ be an additive category. One has $`๐_{\mathrm{tac}}(๐)=๐^{}{}_{}{}^{}๐`$, where $`๐`$ is identified with complexes concentrated in degree zero.
###### Proof.
By (2.1.1), for each $`A`$ in $`๐`$ the complex $`\mathrm{Hom}_๐(X,A)`$ is acyclic if and only if $`\mathrm{Hom}_{๐(๐)}(X,\mathsf{\Sigma }^nA)=0`$ for every integer $`n`$; in other words, if and only if $`X`$ is in $`{}_{}{}^{}๐`$. By the same token, $`\mathrm{Hom}_๐(A,X)`$ is acyclic if and only if $`X`$ is in $`๐^{}`$. โ
###### 5.8.
Let $`R`$ be a ring. The following identifications hold:
$`๐_{\mathrm{tac}}(\mathrm{Prj}R)=๐_{\mathrm{ac}}(\mathrm{Prj}R){}_{}{}^{}(\mathrm{Prj}R)`$
$`๐_{\mathrm{tac}}(\mathrm{Inj}R)=(\mathrm{Inj}R)^{}๐_{\mathrm{ac}}(\mathrm{Inj}R).`$
Indeed, both equalities are due to (5.7), once it is observed that for any complex $`X`$ of $`R`$-modules, the following conditions are equivalent: $`X`$ is acyclic; $`\mathrm{Hom}_R(P,X)`$ is acyclic for each projective $`R`$-module $`P`$; $`\mathrm{Hom}_R(X,I)`$ is acyclic for each injective $`R`$-module $`I`$.
In the presence of a dualizing complex total acyclicity can be tested against a pair of objects, rather than against the entire class of projectives, or of injectives, as called for by the definition. This is one of the imports of the result below. Recall that $`๐R`$ denotes an injective resolution of $`R`$, and that $`D^{}=๐ฒ(๐R)`$; see (4.5).
###### Proposition 5.9.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$.
1. The functor $`๐ณ`$ restricts to an equivalence of $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ with $`๐_{\mathrm{tac}}(\mathrm{Inj}R)`$.
2. $`๐_{\mathrm{ac}}(\mathrm{Prj}R)=\{R\}^{}`$ and $`๐_{\mathrm{tac}}(\mathrm{Prj}R)=\{R,D^{}\}^{}`$.
3. $`๐_{\mathrm{ac}}(\mathrm{Inj}R)=\{๐R\}^{}`$ and $`๐_{\mathrm{tac}}(\mathrm{Inj}R)=\{๐R,D\}^{}`$.
###### Proof.
(1) By Proposition (4.7), the equivalence induced by $`๐ณ`$ identifies $`\mathrm{Thick}(\mathrm{Prj}R)`$ with $`\mathrm{Thick}(\mathrm{Inj}R)`$. This yields the equivalence below:
$$\begin{array}{c}๐_{\mathrm{tac}}(\mathrm{Prj}R)=\mathrm{Thick}(\mathrm{Prj}R)^{}{}_{}{}^{}\mathrm{Thick}(\mathrm{Prj}R)\hfill \\ \hfill \stackrel{}{}\mathrm{Thick}(\mathrm{Inj}R)^{}{}_{}{}^{}\mathrm{Thick}(\mathrm{Inj}R)=๐_{\mathrm{tac}}(\mathrm{Inj}R)\end{array}$$
The equalities are by Lemma (5.7).
(3) That $`๐_{\mathrm{ac}}(\mathrm{Inj}R)`$ equals $`\{๐R\}^{}`$ follows from (2.2). Given this, the claim on $`๐_{\mathrm{tac}}(\mathrm{Inj}R)`$ is a consequence of (5.8) and the identifications
$$\{D\}^{}=\mathrm{Thick}(\mathrm{Add}D)^{}=\mathrm{Thick}(\mathrm{Inj}R)^{}=(\mathrm{Inj}R)^{},$$
where the second one is due to Proposition (4.7).
(2) The equality involving $`๐_{\mathrm{ac}}(\mathrm{Prj}R)`$ is immediate from (2.1.1). Since $`R_RDD`$ and $`D^{}_RD๐R`$, the second claim follows from (1) and (3). โ
###### 5.10.
Proof of Theorems (5.4) and (5.3). The category $`๐ฏ=๐(\mathrm{Inj}R)`$ is compactly generated, the complexes $`๐R`$ and $`D`$ are compact, and one has a canonical equivalence $`๐ฏ^c\stackrel{}{}๐^f(R)`$; see (2.3.2). Therefore, Theorem (5.4) is immediate from Proposition (5.9.3), and Proposition (1.7) applied with $`=\{๐R\}`$ and $`๐=\{๐R,D\}`$.
To prove Theorem (5.3), set $`๐ฏ=๐(\mathrm{Prj}R)`$. By (2.3.1), this category is compactly generated, and in it $`R`$ and $`D^{}`$ are compact; for $`D^{}`$ one requires also the identification in (4.5). Thus, in view of Proposition (5.9.2), Proposition (1.7) applied with $`=\{R\}`$ and $`๐=\{R,D^{}\}`$ yields that the categories $`๐_{\mathrm{ac}}(\mathrm{Prj}R)`$ and $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$, and their quotient, are compactly generated. Furthermore, it provides equivalences up to direct factors
$$๐^c(\mathrm{Prj}R)/\mathrm{Thick}(R)\stackrel{}{}๐_{\mathrm{ac}}^c(\mathrm{Prj}R)$$
$$๐^c(\mathrm{Prj}R)/\mathrm{Thick}(R,D^{})\stackrel{}{}๐_{\mathrm{tac}}^c(\mathrm{Prj}R)$$
$$\mathrm{Thick}(R,D^{})/\mathrm{Thick}(R)\stackrel{}{}\left(๐_{\mathrm{ac}}(\mathrm{Prj}R)/๐_{\mathrm{tac}}(\mathrm{Prj}R)\right)^c.$$
Combining these with the equivalence $`๐^f(R)๐^c(\mathrm{Prj}R)^{\mathrm{op}}`$ in (2.3.1) yields the desired equivalences. โ
###### Remark 5.11.
Proposition (5.9.3) contains the following result: a complex of injectives $`X`$ is totally acyclic if and only if both $`X`$ and $`\mathrm{Hom}_R(D,X)`$ are acyclic. We should like to raise the question: if both $`\mathrm{Hom}_R(X,D)`$ and $`\mathrm{Hom}_R(D,X)`$ are acyclic, is then $`X`$ acyclic, and hence totally acylic? An equivalent formulation is: if $`X`$ is a complex of projectives and $`X`$ and $`\mathrm{Hom}_R(X,R)`$ are acyclic, is then $`X`$ totally acyclic?
In an earlier version of this article, we had claimed an affirmative answer to this question, based on a assertion that if $`X`$ is a complex of $`R`$-modules such that $`\mathrm{Hom}_R(X,D)`$ is acyclic, then $`X`$ is acyclic. This assertion is false. Indeed, let $`R`$ be a complete local domain, with field of fractions $`Q`$. A result of Jensen \[10, Theorem 1\] yields $`\mathrm{Ext}_R^i(Q,R)=0`$ for $`i1`$, and it is easy to check that $`\mathrm{Hom}_R(Q,R)=0`$ as well. Thus, $`\mathrm{Hom}_R(Q,๐R)`$ is acyclic. It remains to recall that when $`R`$ is Gorenstein, $`๐R`$ is a dualizing complex for $`R`$.
###### 5.12.
Non-commutative rings. Theorems (5.3) and (5.4), and Proposition (5.9), all carry over, again with suitable modifications in the statements, to the pair of rings $`S,R`$ from (3.3). The analogue of Corollary (5.5) is especially interesting:
###### Corollary.
The following conditions are equivalent.
1. The projective dimension of $`D`$ is finite over $`R^{\mathrm{op}}`$.
2. The projective dimension of $`D`$ is finite over $`S`$.
3. Every acyclic complex of projective $`R`$-modules is totally acyclic.
4. Every acyclic complex of injective $`S`$-modules is totally acyclic. โ
## 6. An example
Let $`A`$ be a commutative noetherian local ring, with maximal ideal $`๐ช`$, and residue field $`k=A/๐ช`$. Assume that $`๐ช^2=0`$, and that $`\mathrm{rank}_k(๐ช)2`$. Observe that $`A`$ is *not* Gorenstein; for instance, its socle is $`๐ช`$, and hence of rank at least $`2`$. Let $`E`$ denote the injective hull of the $`R`$-module $`k`$; this is a dualizing complex for $`A`$.
###### Proposition 6.1.
Set $`๐=๐(\mathrm{Prj}A)`$ and let $`X`$ be a complex of projective $`A`$-modules.
1. If $`X`$ is acyclic and the $`A`$-module $`X^d`$ is finite for some $`d`$, then $`X0`$ in $`๐`$.
2. If $`X`$ is totally acyclic, then $`X0`$ in $`๐`$.
3. The cone of the homothety $`A\mathrm{Hom}_A(P,P)`$, where $`P`$ is a projective resolution of $`D`$, is an acyclic complex of projectives, but it is not totally acyclic.
4. In the derived category of $`A`$, one has $`\mathrm{Thick}(A,D)=๐^f(A)`$, and hence
$$\mathrm{Thick}(A,D)/\mathrm{Thick}(A)=๐^f(A)/\mathrm{Thick}(A).$$
The proof is given in (6.4). It hinges on some properties of minimal resolutions over $`A`$, which we now recall. Since $`A`$ is local, each projective $`A`$-module is free. The Jacobson radical $`๐ช`$ of $`A`$ is square-zero, and in particular, nilpotent. Thus, Nakayamaโs lemma applies to each $`A`$-module $`M`$, hence it has a projective cover $`PM`$, and hence a minimal projective resolution; see \[7, Propositions 3 and 15\]. Moreover, $`\mathrm{\Omega }=\mathrm{Ker}(PM)`$, the first syzygy of $`M`$, satisfies $`\mathrm{\Omega }๐ชP`$, so that $`๐ช\mathrm{\Omega }๐ช^2P=0`$, so $`๐ช\mathrm{\Omega }=0`$.
###### Lemma 6.2.
Let $`M`$ be an $`A`$-module; set $`b=\mathrm{}_A(M)`$, $`c=\mathrm{}_A(\mathrm{\Omega })`$.
1. If $`M`$ is finite, then its Poincarรฉ series is
$$P_M^A(t)=b+\frac{ct}{1et}$$
In particular, $`\beta _n^A(M)`$, the $`n`$th Betti number of $`M`$, equals $`ce^{n1}`$, for $`n1`$.
2. If $`\mathrm{Ext}_A^n(M,A)=0`$ for some $`n2`$, then $`M`$ is free.
###### Proof.
(1) This is a standard calculation, derived from the exact sequences
$$0๐ชAk0\text{and}0\mathrm{\Omega }PM0$$
The one on the left implies $`P_k^A(t)=1+etP_k^A(t)`$, so $`P_k^A(t)=(1et)^1`$, while the one on the right yields $`P_M^A(t)=b+ctP_k^A(t)`$, since $`๐ช\mathrm{\Omega }=0`$.
(2) If $`M`$ is not free, then $`\mathrm{\Omega }0`$ and hence has $`k`$ as a direct summand. In this case, since $`\mathrm{Ext}_A^{n1}(\mathrm{\Omega },A)\mathrm{Ext}_A^n(M,A)=0`$, one has $`\mathrm{Ext}_A^{n1}(k,A)=0`$, which in turn implies that $`A`$ is Gorenstein; a contradiction. โ
The following test to determine when an acyclic complex is homotopically trivial is surely known. Note that it applies to any (commutative) noetherian ring of finite Krull dimension, and, in particular, to the ring $`A`$ that is the focus of this section.
###### Lemma 6.3.
Let $`R`$ be a ring whose finitistic global dimension is finite. An acyclic complex $`X`$ of projective $`R`$-modules is homotopically trivial if and only if for some integer $`s`$ the $`R`$-module $`\mathrm{Coker}(X^{s1}X^s)`$ is projective.
###### Proof.
For each integer $`n`$ set $`M(n)=\mathrm{Coker}(X^{n1}X^n)`$. It suffices to prove that the $`R`$-module $`M(n)`$ is projective for each $`n`$. This is immediate for $`ns`$ because $`M(s)`$ is projective so that the sequence $`\mathrm{}X^{s1}X^sM(s)0`$ is split exact.
We may now assume that $`ns+1`$. By hypothesis, there exists an integer $`d`$ with the following property: for any $`R`$-module $`M`$, if its projective dimension, $`\mathrm{pd}_RM`$ is finite, then $`\mathrm{pd}_RMd`$. It follows from the exact complex
$$0M(s)X^{s+1}\mathrm{}X^{n+d}M(n+d)0$$
that $`\mathrm{pd}_RM(n+d)`$ is finite. Thus, $`\mathrm{pd}_RM(n+d)d`$, and another glance at the exact complex above reveals that $`M(n)`$ must be projective, as desired. โ
Now we are ready for the
###### 6.4.
Proof of Proposition (6.1). In what follows, set $`M(s)=\mathrm{Coker}(X^{s1}X^s)`$.
(1) Pick an integer $`n1`$ with $`e^{n1}\mathrm{rank}_A(X^d)+1`$. Since $`X`$ is acyclic, $`\mathsf{\Sigma }^{dn}X^{d+n}`$ is a free resolution of the $`A`$-module $`M(n+d)`$. Let $`\mathrm{\Omega }`$ be the first syzygy of $`M(n+d)`$. One then obtains the first one of the following equalities:
$$\mathrm{rank}_A(X^d)\beta _n^A(M(n+d))\mathrm{}_A(\mathrm{\Omega })e^{n1}\mathrm{}_A(\mathrm{\Omega })(\mathrm{rank}_A(X^d)+1)$$
The second equality is Lemma (6.2.1) applied to $`M(n+d)`$ while the last one is by the choice of $`n`$. Thus $`\mathrm{}_A(\mathrm{\Omega })=0`$, so $`\mathrm{\Omega }=0`$ and $`M(n+d)`$ is free. Now Lemma (6.3) yields that $`X`$ is homotopically trivial.
(2) Fix an integer $`d`$. Since $`\mathsf{\Sigma }^dX^d`$ is a projective resolution of $`M(d)`$, total acyclicity of $`X`$ implies that the homology of $`\mathrm{Hom}_A(\mathsf{\Sigma }^dX^d,A)`$ is zero in degrees $`1`$, so $`\mathrm{Ext}_A^n(M(d),A)=0`$ for $`n1`$. Lemma (6.2.2) established above implies $`M(d)`$ is free. Once again, Lemma (6.3) completes the proof.
(3) Suppose that the cone of $`A\mathrm{Hom}_A(P,P)`$ is totally acyclic. This leads to a contradiction: (2) implies that the cone is homotopic to zero, so $`A\mathrm{Hom}_A(P,P)`$ in $`๐`$. This entails the first of the following isomorphisms in $`๐(A)`$; the others are standard.
$`\mathrm{Hom}_A(k,A)`$ $`\mathrm{Hom}_A(k,\mathrm{Hom}_A(P,P))`$
$`\mathrm{Hom}_A(P_Ak,P)`$
$`\mathrm{Hom}_k(P_Ak,\mathrm{Hom}_A(k,P))`$
$`\mathrm{Hom}_k(P_Ak,\mathrm{Hom}_A(k,A)_AP)`$
$`\mathrm{Hom}_k(P_Ak,\mathrm{Hom}_A(k,A)_k(k_AP))`$
Passing to homology and computing ranks yields $`H(k_AP)k`$, and this implies $`DA`$. This cannot be for $`\mathrm{rank}_k\mathrm{soc}(D)=1`$, while $`\mathrm{rank}_k\mathrm{soc}(A)=e`$ and $`e2`$.
(4) Combining Theorem (5.3.2) and (3) gives the first part. The second part then follows from the first. A direct and elementary argument is also available: As noted above the $`A`$-module $`D`$ is not free; thus, the first syzygy module $`\mathrm{\Omega }`$ of $`D`$ is non-zero, so has $`k`$ as a direct summand. Since $`\mathrm{\Omega }`$ is in $`\mathrm{Thick}(A,D)`$, we deduce that $`k`$, and hence every homologically finite complex of $`A`$-modules, is in $`\mathrm{Thick}(A,D)`$.
###### Remark 6.5.
Let $`A`$ be the ring introduced at the beginning of this section, and let $`X`$ and $`Y`$ be complexes of $`A`$-modules.
The Tate cohomology of $`X`$ and $`Y`$, in the sense of Jรธrgensen , is the homology of the complex $`\mathrm{Hom}_A(T,Y)`$, where $`T`$ is a complete projective resolution of $`X`$; see (7.6). By Proposition (6.1.2) any such $`T`$, being totally acyclic, is homotopically trivial, so the Tate cohomology modules of $`X`$ and $`Y`$ are all zero. The same is true also of the version of Tate cohomology introduced by Krause \[14, (7.5)\] via complete injective resolutions. This is because $`A`$ has no non-trivial totally acyclic complexes of injectives either, as can be verified either directly, or by appeal to Proposition (5.9.1).
These contrast drastically with another generalization of Tate cohomology over the ring $`A`$, introduced by Vogel and described by Goichot . Indeed, Avramov and Veliche \[3, (3.3.3)\] prove that for an arbitrary commutative local ring $`R`$ with residue field $`k`$, if the Vogel cohomology with $`X=k=Y`$ has finite rank even in a *single* degree, then $`R`$ is Gorenstein.
## 7. Auslander categories and Bass categories
Let $`R`$ be a commutative noetherian ring with a dualizing complex $`D`$. We write $`๐_{\mathrm{prj}}(R)`$ for the subcategory of $`๐(\mathrm{Prj}R)`$ consisting of K-projective complexes, and $`๐_{\mathrm{inj}}(R)`$ for the subcategory of $`๐(\mathrm{Inj}R)`$ consisting of K-injective complexes. This section is motivated by the following considerations: One has adjoint pairs of functors
and
and composing these functors with those in Theorem (4.2) gives functors
$$๐ฆ=(๐๐ณ):๐_{\mathrm{prj}}(R)๐_{\mathrm{inj}}(R)\text{and}๐ฅ=(๐๐ฒ):๐_{\mathrm{inj}}(R)๐_{\mathrm{prj}}(R).$$
These functors fit into the upper half of the picture below:
The vertical arrows in the lower half are obtained by factoring the canonical functor $`๐(\mathrm{Prj}R)๐(R)`$ through $`๐`$, and similarly $`๐(\mathrm{Inj}R)๐(R)`$ through $`๐`$. A straightforward calculation shows that the functors in the last row of the diagram are induced by those in the middle. Now, while $`๐ณ`$ and $`๐ฒ`$ are equivalences โ by Theorem (4.2) โ the functors $`๐ฆ`$ and $`๐ฅ`$ need not be; indeed, they are equivalences if and only if $`R`$ is Gorenstein; see Corollary (7.5) ahead. The results in this section address the natural:
###### Question.
Identify subcategories of $`๐_{\mathrm{prj}}(R)`$ and $`๐_{\mathrm{inj}}(R)`$ on which $`๐ฆ`$ and $`๐ฅ`$ restrict to equivalences.
Given the equivalences in the lower square of the diagram an equivalent problem is to characterize subcategories of $`๐(R)`$ on which the functors $`D_R^๐`$ and $`๐\mathrm{Hom}_R(D,)`$ induce equivalences. This leads us to the following definitions:
###### 7.1.
Auslander category and Bass category. Consider the categories
$`\widehat{๐}(R)=\{X๐(R)\text{the natural map }X๐\mathrm{Hom}_R(D,D_R^๐X)\text{ is an isomorphism.}\}`$
$`\widehat{}(R)=\{Y๐(R)\text{the natural map }D_R^๐๐\mathrm{Hom}_R(D,Y)Y\text{ is an isomorphism.}\}`$
The notation is intended to be reminiscent of the ones for the *Auslander category* $`๐(R)`$ and the *Bass category* $`(R)`$, introduced by Avramov and Foxby , which are the following subcategories of the derived category:
$`๐(R)=\{X\widehat{๐}(R)X\text{ and }D_R^๐X\text{ are homologically bounded.}\}`$
$`(R)=\{Y\widehat{}(R)Y\text{ and }๐\mathrm{Hom}_R(D,Y)\text{ are homologically bounded.}\}`$
The definitions are engineered to lead immediately to the following
###### Proposition 7.2.
The adjoint pair of functors $`(๐ฆ,๐ฅ)`$ restrict to equivalences of categories between $`\widehat{๐}(R)`$ and $`\widehat{}(R)`$, and between $`๐(R)`$ and $`(R)`$.โ
In what follows, we identify $`\widehat{๐}(R)`$ and $`\widehat{}(R)`$ with the subcategories of $`๐_{\mathrm{prj}}(R)`$ and $`๐_{\mathrm{inj}}(R)`$ on which $`๐ฒ๐ณ`$ and $`๐ณ๐ฒ`$, respectively, restrict to equivalences. The Auslander category and the Bass category are identified with appropriate subcategories.
The main task then is describe the complexes in the categories being considered. In this section we provide an answer in terms of the categories of K-projectives and K-injectives; in the next one, it is translated to the derived category. Propositions (7.3) and (7.4) below are the first step towards this end. In them, the *cone* of a morphism $`UV`$ in a triangulated category refers to an object $`W`$ obtained by completing the morphism to an exact triangle: $`UVW\mathsf{\Sigma }U`$. We may speak of *the* cone because they exist and are all isomorphic.
###### Proposition 7.3.
Let $`X`$ be a complex of projective $`R`$-modules. If $`X`$ is K-projective, then it is in $`\widehat{๐}(R)`$ if and only if the cone of the morphism $`๐ณ(X)\mathrm{๐๐ณ}(X)`$ in $`๐(\mathrm{Inj}R)`$ is totally acyclic.
###### Remark.
The cone in question is always acyclic, because $`๐ณ(X)\mathrm{๐๐ณ}(X)`$ is an injective resolution; the issue thus is the difference between acyclicity and total acyclicity.
###### Proof.
Let $`\eta :๐ณ(X)\mathrm{๐๐ณ}(X)`$ be a K-injective resolution. In $`๐(\mathrm{Prj}R)`$ one has then a commutative diagram
of adjunction morphisms, where the isomorphism is by Theorem (4.2). It is clear from the diagram above that
$`X`$ is in $`\widehat{๐}(R)`$ $`\kappa \text{ is a quasi-isomorphism}`$
$`๐ฒ(\eta )\text{ is a quasi-isomorphism}`$
It thus remains to prove that the last condition is equivalent to total acyclicity of the cone of $`\eta `$. In $`๐(\mathrm{Inj}R)`$ complete $`\eta `$ to an exact triangle:
From this triangle one obtains that $`๐ฒ(\eta )`$ is a quasi-isomorphism if and only if $`๐ฒ(C)`$ is acyclic. Now $`๐ฒ(C)`$ is quasi-isomorphic to $`\mathrm{Hom}_R(D,C)`$, see Theorem (2.7.1), and the acyclicity of $`\mathrm{Hom}_R(D,C)`$ is equivalent to $`C`$ being in $`\{D\}^{}`$, in $`๐(\mathrm{Inj}R)`$. However, $`C`$ is already acyclic, and hence in $`\{๐R\}^{}`$. Therefore Proposition (5.9.3) implies that $`๐ฒ(C)`$ is acyclic if and only if $`C`$ is totally acyclic, as desired. โ
An analogous proof yields:
###### Proposition 7.4.
Let $`Y`$ be a complex of injective $`R`$-modules. If $`Y`$ is K-injective, then it is in $`\widehat{}(R)`$ if and only if the cone of the morphism $`\mathrm{๐๐ฒ}(Y)๐ฒ(Y)`$ in $`๐(\mathrm{Prj}R)`$ is totally acyclic. โ
###### Corollary 7.5.
Let $`R`$ be a noetherian ring with a dualizing complex. The ring $`R`$ is Gorenstein if and only if $`\widehat{๐}(R)=๐_{\mathrm{prj}}(R)`$, if and only if $`\widehat{}(R)=๐_{\mathrm{inj}}(R)`$.
###### Proof.
Combine Propositions (7.3) and (7.4) with Corollary (5.5). โ
One shortcoming in Propositions (7.3) and (7.4) is they do not provide a structural description of objects in the Auslander and Bass categories. Addressing this issue requires a notion of complete resolutions.
###### 7.6.
Complete resolutions. The subcategory $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ of $`๐(\mathrm{Prj}R)`$ is closed under coproducts; moreover, it is compactly generated, by Theorem (5.3.1). Thus, the inclusion $`๐_{\mathrm{tac}}(\mathrm{Prj}R)๐(\mathrm{Prj}R)`$ admits a right adjoint:
For each complex $`X`$ in $`๐(\mathrm{Prj}R)`$ we call $`๐(X)`$ the *complete projective resolution* of $`X`$. In $`๐(\mathrm{Prj}R)`$, complete the natural morphism $`๐(X)X`$ to an exact triangle:
$$๐(X)X๐(X)\mathsf{\Sigma }๐(X)$$
Up to an isomorphism, this triangle depends only on $`X`$.
Similar considerations show that the inclusion $`๐_{\mathrm{tac}}(\mathrm{Inj}R)๐(\mathrm{Inj}R)`$ admits a left adjoint. We denote it $`๐`$, and for each complex $`Y`$ of injectives call $`๐(Y)`$ the *complete injective resolution* of $`Y`$. This leads to an exact triangle in $`๐(\mathrm{Inj}R)`$:
$$๐(Y)Y๐(Y)\mathsf{\Sigma }๐(Y)$$
Relevant properties of complete resolutions and the corresponding exact triangles are summed up in the next two result; the arguments are standard, and details are given for completeness.
###### Lemma 7.7.
Let $`X`$ be a complex of projectives $`R`$-modules.
1. The morphism $`X๐(X)`$ is a quasi-isomorphism and $`๐(X)`$ is in $`๐_{\mathrm{tac}}(\mathrm{Prj}R)^{}`$.
2. Any exact triangle $`TXU\mathsf{\Sigma }T`$ in $`๐(\mathrm{Prj}R)`$ where $`T`$ is totally acyclic and $`U`$ is in $`๐_{\mathrm{tac}}(\mathrm{Prj}R)^{}`$ is isomorphic to $`๐(X)X๐(X)\mathsf{\Sigma }๐(X)`$.
###### Proof.
(1) By definition, one has an exact triangle
$$๐(X)X๐(X)\mathsf{\Sigma }๐(X).$$
Since the complex $`๐(X)`$ is acyclic, the homology long exact sequence arising from this triangle proves that $`X๐(X)`$ is an quasi-isomorphism, as claimed. Moreover, for each totally acyclic complex $`T`$ the induced map below is bijective:
($``$)
$$\mathrm{Hom}_๐(T,๐(X))\mathrm{Hom}_๐(T,X)$$
This holds because $`๐`$ is a right adjoint to the inclusion $`๐_{\mathrm{tac}}(\mathrm{Prj}R)๐(\mathrm{Prj}R)`$. Since $`๐()`$ commutes with translations, the morphism $`\mathsf{\Sigma }^n๐(X)\mathsf{\Sigma }^nX`$ coincides with the morphism $`๐(\mathsf{\Sigma }^nX)\mathsf{\Sigma }^nX`$. Thus, from ($``$) one deduces that the induced map
$$\mathrm{Hom}_๐(T,๐(\mathsf{\Sigma }^nX))\mathrm{Hom}_๐(T,\mathsf{\Sigma }^nX)$$
is bijective for each integer $`n`$. It is now immediate from the exact triangle above that $`\mathrm{Hom}_๐(T,๐(X))=0`$; this settles (1), since $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$ is stable under translations.
(2) Given such an exact triangle, the induced map $`\mathrm{Hom}_๐(,T)\mathrm{Hom}_๐(,X)`$ is bijective on $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$, since $`\mathrm{Hom}_๐(,U)`$ vanishes on $`๐_{\mathrm{tac}}(\mathrm{Prj}R)`$. Thus, there is an isomorphism $`\alpha :T๐(X)`$, by (1.4), and one obtains a commutative diagram
of morphisms in $`๐(\mathrm{Prj}R)`$. Since the rows are exact triangles, and we are in a triangulated category, there exists a $`\beta `$ as above that makes the diagram commute. Moreover, since $`\alpha `$ is an isomorphism, so is $`\beta `$; this is the desired result. โ
One has also a version of Lemma (7.7) for complexes of injectives; proving it calls for a new ingredient, provided by the next result. Recall that $`๐R`$ denotes an injective resolution of $`R`$ and $`D^{}=๐ฒ(๐R)`$; see (4.5).
###### Lemma 7.8.
$`{}_{}{}^{}๐_{\mathrm{tac}}^{}(\mathrm{Inj}R)=\mathrm{Loc}(๐R,D)`$
###### Proof.
Proposition (5.9.3) implies that $`๐R`$ and $`D`$ are contained in $`{}_{}{}^{}๐_{\mathrm{tac}}^{}(\mathrm{Inj}R)`$, and hence so is $`\mathrm{Loc}(๐R,D)`$. To see that the reverse inclusion also holds note that $`\mathrm{Loc}(๐R,D)`$ is compactly generated (by $`๐R`$ and $`D`$) and closed under coproducts. Thus, by (1.5.1), the inclusion $`\mathrm{Loc}(๐R,D)๐(\mathrm{Inj}R)`$ admits a right adjoint, say $`๐`$. Let $`X`$ be a complex of injectives. Complete the canonical morphism $`๐(X)X`$ to an exact triangle
$$๐(X)XC\mathsf{\Sigma }๐(X)$$
For each integer $`n`$ the induced map $`\mathrm{Hom}_๐(,\mathsf{\Sigma }^n๐(X))\mathrm{Hom}_๐(,\mathsf{\Sigma }^nX)`$ is bijective on $`\{๐R,D\}`$, so the exact triangle above yields that $`\mathrm{Hom}_๐(๐R,\mathsf{\Sigma }^nC)=0=\mathrm{Hom}_๐(D,\mathsf{\Sigma }^nC)`$. Therefore, $`C`$ is totally acyclic, by Proposition (5.9.3). In particular, when $`X`$ is in $`{}_{}{}^{}๐_{\mathrm{tac}}^{}(\mathrm{Inj}R)`$, one has $`\mathrm{Hom}_๐(X,C)=0`$, so the exact triangle above is split, that is to say, $`X`$ is a direct summand of $`๐(X)`$, and hence in $`\mathrm{Loc}(๐R,D)`$, as claimed. โ
Here is the analogue of Lemma (7.7) for complexes of injectives; it is a better result for it provides a structural description of $`๐(Y)`$.
###### Lemma 7.9.
Let $`Y`$ be a complex of injective $`R`$-modules.
1. The morphism $`๐(Y)Y`$ is a quasi-isomorphism and $`๐(Y)`$ is in $`\mathrm{Loc}(๐R,D)`$.
2. Any exact triangle $`VXT\mathsf{\Sigma }V`$ in $`๐(\mathrm{Inj}R)`$ where $`T`$ is totally acyclic and $`V`$ is in $`\mathrm{Loc}(๐R,D)`$ is isomorphic to $`๐(Y)Y๐(Y)\mathsf{\Sigma }๐(Y)`$.
###### Proof.
An argument akin to the proof of Lemma (7.7.1) yields that $`๐(Y)Y`$ is a quasi-isomorphism and that $`๐(Y)`$ is in $`{}_{}{}^{}๐_{\mathrm{tac}}^{}(\mathrm{Inj}R)`$, which equals $`\mathrm{Loc}(๐R,D)`$, by Lemma (7.8). Given this, the proof of part (2) is similar to that of Lemma (7.7.2). โ
Our interest in complete resolutions is due to Theorems (7.11) and (7.10), which provide one answer to the question raised at the beginning of this section.
###### Theorem 7.10.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$, and let $`X`$ be a complex of projective $`R`$-modules. If $`X`$ is K-projective, then the following conditions are equivalent.
1. The complex $`X`$ is in $`\widehat{๐}(R)`$.
2. The complex $`๐(X)`$ is in $`\mathrm{Coloc}(\mathrm{Prj}R)`$.
3. In $`๐(\mathrm{Prj}R)`$, there exists an exact triangle $`TXU\mathsf{\Sigma }U`$ where $`T`$ is totally acyclic and $`U`$ is in $`\mathrm{Coloc}(\mathrm{Prj}R)`$.
###### Proof.
Let $`๐(X)X๐(X)\mathsf{\Sigma }๐(X)`$ be the exact triangle associated to the complete projective resolution of $`X`$; see (7.6). Let $`\eta :๐ณ(X)\mathrm{๐๐ณ}(X)`$ be a K-injective resolution, and consider the commutative diagram
arising as follows: the vertical map on the left is a quasi-isomorphism because it sits in the exact triangle with third vertex $`\mathrm{๐ณ๐}(X)`$, which is acyclic since $`๐(X)`$ is totally acyclic; see Proposition (5.9.1). Since $`\mathrm{๐๐ณ}(X)`$ is K-injective, $`\eta `$ extends to yield $`\kappa `$, which is a quasi-isomorphism because the other maps in the square are.
Note that the cone of the morphism $`๐ณ(X)\mathrm{๐ณ๐}(X)`$ is $`\mathsf{\Sigma }\mathrm{๐ณ๐}(X)`$, so applying the octahedral axiom to the commutative square above gives us an exact triangle
where $`\mathrm{Cone}()`$ refers to the cone of the morphism in parenthesis. Since $`๐(X)`$ is totally acyclic, so is $`\mathrm{๐ณ๐}(X)`$, by Proposition (5.9.1). Hence the exact triangle above yields:
($``$) $`\mathrm{Cone}(\eta )`$ is totally acyclic if and only if $`\mathrm{Cone}(\kappa )`$ is totally acyclic.
This observation is at the heart of the equivalence one has set out to establish.
(a) $``$ (b): Proposition (7.3) yields that $`\mathrm{Cone}(\eta )`$ is totally acyclic, and hence so is $`\mathrm{Cone}(\kappa )`$, by ($``$). Consider the exact triangle
According to Lemma (7.7.1) the complex $`๐(X)`$ is in $`๐_{\mathrm{tac}}(\mathrm{Prj}R)^{}`$, so Proposition (5.9) yields that $`\mathrm{๐ณ๐}(X)`$ is in $`๐_{\mathrm{tac}}(\mathrm{Inj}R)^{}`$, and hence the total acyclicity of $`\mathrm{Cone}(\kappa )`$ implies
$$\mathrm{Hom}_๐(\mathrm{Cone}(\kappa ),\mathrm{๐ณ๐}(X))=0$$
Thus the triangle above is split exact, and $`\mathrm{๐ณ๐}(X)`$ is a direct summand of $`\mathrm{๐๐ณ}(X)`$. Consequently $`\mathrm{๐ณ๐}(X)`$ is in $`\mathrm{Coloc}(\mathrm{Inj}R)`$, so, by Theorem (4.2) and Corollary (4.4), one obtains that $`๐(X)`$ is in $`\mathrm{Coloc}(\mathrm{Prj}R)`$, as desired.
(b) $``$ (a): By Corollary (4.4), as $`๐(X)`$ is in $`\mathrm{Coloc}(\mathrm{Prj}R)`$ the complex $`\mathrm{๐ณ๐}(X)`$ is in $`\mathrm{Coloc}(\mathrm{Inj}R)`$, that is to say, it is K-injective. The map $`\kappa :\mathrm{๐ณ๐}(X)\mathrm{๐๐ณ}(X)`$, being a quasi-isomorphism between K-injectives, is an isomorphism. Therefore $`\mathrm{Cone}(\kappa )0`$ so ($``$) implies that $`\mathrm{Cone}(\eta )`$ is totally acyclic. It remains to recall Proposition (7.3).
That (b) implies (c) is patent, and (c) $``$ (b) follows from Lemma (7.7), because $`๐_{\mathrm{tac}}(\mathrm{Prj}R)^{}\mathrm{Coloc}(\mathrm{Prj}R)`$. The completes the proof of the theorem. โ
An analogous argument yields a companion result for complexes of injectives:
###### Theorem 7.11.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$, and let $`Y`$ be a complex of injective $`R`$-modules. If $`Y`$ is K-injective, then the following conditions are equivalent.
1. The complex $`Y`$ is in $`\widehat{}(R)`$.
2. The complex $`๐(Y)`$ is in $`\mathrm{Loc}(D)`$.
3. In $`๐(\mathrm{Inj}R)`$, there exists an exact triangle $`VYT\mathsf{\Sigma }V`$ where $`V`$ is in $`\mathrm{Loc}(D)`$ and $`T`$ is totally acyclic.โ
Section 8 translates Theorems (7.11) and (7.10) to the derived category of $`R`$.
###### 7.12.
Non-commutative rings. Consider a pair of rings $`S,R`$ with a dualizing complex $`D`$, defined in (3.3). As in (7.1), one can define the Auslander category of $`R`$ and the Bass category of $`S`$; these are equivalent via the adjoint pair of functors $`(D_R,๐\mathrm{Hom}_S(D,))`$. The analogues of Theorems (7.10) and (7.11) extend to the pair $`S,R`$, and they describe the complexes in $`\widehat{๐}(R)`$ and $`\widehat{}(S)`$.
## 8. Gorenstein dimensions
Let $`R`$ be a commutative noetherian ring, and let $`X`$ be a complex of $`R`$-modules. We say that $`X`$ has *finite Gorenstein projective dimension*, or, in short: *finite G-projective dimension*, if there exists an exact sequence of complexes of projective $`R`$-modules
$$0UT๐X0$$
where $`T`$ is totally acyclic, $`๐X`$ is a K-projective resolution of $`X`$, and $`U^n=0`$ for $`n0`$.
Similarly, a complex $`Y`$ of $`R`$-modules has *finite G-injective dimension* if there exists an exact sequence of complexes of injective $`R`$-modules
$$0๐YTV0$$
where $`T`$ is totally acyclic, $`๐Y`$ is a K-injective resolution of $`Y`$, and $`V^n=0`$ for $`n0`$.
The preceding definitions are equivalent to the usual ones, in terms of G-projective and G-injective resolutions; see Veliche , and Avramov and Martsinkovsky .
The theorem below contains a recent result of Christensen, Frankild, and Holm; more precisely, the equivalence of (i) and (ii) in \[6, (4.1)\], albeit in the case when $`R`$ is commutative; however, see (8.3).
###### Theorem 8.1.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$, and $`X`$ a complex of $`R`$-modules. The following conditions are equivalent:
1. $`X`$ has finite G-projective dimension.
2. $`๐X`$ is in $`\widehat{๐}(R)`$ and $`D_R^๐X`$ is homologically bounded on the left.
3. $`๐(๐X)`$ is isomorphic, in $`๐(\mathrm{Prj}R)`$, to a complex $`U`$ with $`U^n=0`$ for $`n0`$.
When $`H(X)`$ is bounded, these conditions are equivalent to: $`X`$ is in $`๐(R)`$.
###### Proof.
Substituting $`X`$ with $`๐X`$, one may assume that $`X`$ is K-projective and that $`D_R^๐X`$ is quasi-isomorphic to $`D_RX`$, that is to say, to $`๐ณ(X)`$.
(a) $``$ (b): By definition, there is an exact sequence of complexes of projectives $`0UTX0`$ where $`T`$ is totally acyclic and $`U^n=0`$ for $`n0`$. Passing to $`๐(\mathrm{Prj}R)`$ gives rise to an exact triangle
$$UTX\mathsf{\Sigma }U$$
Since $`T`$ is totally acyclic, $`๐ณ(X)`$ is quasi-isomorphic to $`๐ณ(\mathsf{\Sigma }U)`$; the latter is bounded on the left as a complex, hence the former is homologically bounded on the left, as claimed. This last conclusion yields also that $`๐ณ(\mathsf{\Sigma }U)`$ is in $`\mathrm{Coloc}(\mathrm{Inj}R)`$. Thus, by Theorem (4.2) and Corollary (4.4), the complex $`\mathsf{\Sigma }U`$ is in $`\mathrm{Coloc}(\mathrm{Prj}R)`$, so the exact triangle above and Theorem (7.10) imply that $`X`$ is in $`\widehat{๐}(R)`$.
(b) $``$ (c): By Theorem (7.10), there is an exact triangle
$$TXU\mathsf{\Sigma }T$$
with $`T`$ totally acyclic and $`U`$ in $`\mathrm{Coloc}(\mathrm{Prj}R)`$. The first condition implies that $`๐ณ(U)`$ is quasi-isomorphic to $`๐ณ(X)`$, and hence homologically bounded on the left, while the second implies, thanks to Corollary (4.4), that it is in $`\mathrm{Coloc}(\mathrm{Inj}R)`$, that is to say, it is K-injective. Consequently $`๐ณ(U)`$ is isomorphic to a complex of injectives $`I`$ with $`I^n=0`$ for $`n0`$. This implies that the complex of flat $`R`$-modules $`\mathrm{Hom}_R(D,๐ณ(U))`$ is bounded on the left. Theorem (2.7.3) now yields that the complex $`๐(\mathrm{Hom}_R(D,๐ณ(U)))`$, that is to say, $`\mathrm{๐ฒ๐ณ}(U)`$, is bounded on the left; thus, the same is true of $`U`$ as it is isomorphic to $`\mathrm{๐ฒ๐ณ}(U)`$, by Theorem (4.2). It remains to note that $`\mathrm{Coloc}(\mathrm{Prj}R)๐_{\mathrm{tac}}(\mathrm{Prj}R)^{}`$, so $`๐(X)U`$ by Lemma (7.7).
(c) $``$ (a): Lift the morphism $`X๐(X)U`$ in $`๐(\mathrm{Prj}R)`$ to a morphism $`\alpha :XU`$ of complexes of $`R`$-modules. In the mapping cone exact sequence
$$0U\mathrm{Cone}(\alpha )\mathsf{\Sigma }X0$$
$`\mathrm{Cone}(\alpha )`$ is homotopic to $`๐(X)`$, and hence totally acyclic, while $`U^n=0`$ for $`n0`$, by hypothesis. Thus, the G-projective dimension of $`\mathsf{\Sigma }X`$, and hence of $`X`$, is finite.
Finally, when $`H(X)`$ is bounded, $`D_R^๐X`$ is always bounded on the right. It is now clear from definitions that the condition that $`X`$ is in $`๐(R)`$ is equivalent to (b). โ
Here is a characterization of complexes in $`๐(R)`$ that are in the Bass category. For commutative rings, it recovers \[6, (4.4)\]; see (8.3). The basic idea of the proof is akin the one for the theorem above, but the details are dissimilar enough to warrant exposition.
###### Theorem 8.2.
Let $`R`$ be a noetherian ring with a dualizing complex $`D`$, and $`Y`$ a complex of $`R`$-modules. The following conditions are equivalent:
1. $`Y`$ has finite G-injective dimension.
2. $`๐Y`$ is in $`\widehat{}(R)`$ and $`๐\mathrm{Hom}_R(D,Y)`$ is homologically bounded on the right.
3. $`๐(๐Y)`$ is isomorphic, in $`๐(\mathrm{Inj}R)`$, to a complex $`V`$ with $`V^n=0`$ for $`n0`$.
When $`H(Y)`$ is bounded, these conditions are equivalent to: $`Y`$ is in $`(R)`$.
###### Proof.
Replacing $`Y`$ with $`๐Y`$ we assume that $`Y`$ is K-injective, so $`๐\mathrm{Hom}_R(D,Y)`$ is quasi-isomorphic to $`\mathrm{Hom}_R(D,Y)`$. In the argument below the following remark is used without comment: in $`๐(\mathrm{Inj}R)`$, given an exact triangle
$$Y_1Y_2T\mathsf{\Sigma }Y_1$$
if $`T`$ is totally acyclic, then one has a sequence
$$\mathrm{Hom}_R(D,Y_1)\stackrel{}{}๐ฒ(Y_1)\stackrel{}{}๐ฒ(Y_2)\stackrel{}{}\mathrm{Hom}_R(D,Y_2).$$
of quasi-isomorphisms. Indeed, the first and the last quasi-isomorphism hold by Theorem (2.7.1), while the middle one holds because $`๐ฒ(T)`$ is totally acyclic, by Theorem (4.2).
(a) $``$ (b): The defining property of complexes of finite G-injective dimension provides an exact sequence of complexes of injectives $`0YTV0`$ where $`T`$ is totally acyclic and $`V^n=0`$ for $`n0`$. Passing to $`๐(\mathrm{Inj}R)`$ gives rise to an exact triangle
$$\mathsf{\Sigma }^1VYTV$$
Since $`T`$ is totally acyclic, $`\mathrm{Hom}_R(D,\mathsf{\Sigma }^1V)`$ is quasi-isomorphic to $`\mathrm{Hom}_R(D,Y)`$; the former is bounded on the right as a complex, so the latter is homologically bounded on the right, as claimed. Furthermore, since $`V`$ is bounded on the right, so is $`\mathrm{Hom}_R(D,\mathsf{\Sigma }^1V)`$. Theorem (2.7.2) then yields that $`๐ฒ(\mathsf{\Sigma }^1V)`$ is its projective resolution, and hence it is in $`\mathrm{Loc}(R)`$. Thus, by Theorem (4.2), the complex $`\mathsf{\Sigma }^1V`$ is in $`\mathrm{Loc}(D)`$, so the exact triangle above and Theorem (7.11) imply that $`Y`$ is in $`\widehat{}(R)`$.
(b) $``$ (c): By hypothesis and Theorem (7.11) there exists and exact triangle
$$VYT\mathsf{\Sigma }V$$
in $`๐(\mathrm{Inj}R)`$, where $`V`$ lies in $`\mathrm{Loc}(D)`$ and $`T`$ is totally acyclic. Thus $`๐ฒ(V)`$ is in $`\mathrm{Loc}(R)`$, that is to say, it is K-projective, and it is quasi-isomorphic to $`\mathrm{Hom}_R(D,Y)`$, and hence it is homologically bounded on the right. Therefore, $`๐ฒ(V)`$ is isomorphic to a complex of projectives $`P`$ with $`P^n=0`$ for $`n0`$. By Theorem (4.2), this implies that $`V`$ is isomorphic to $`๐ณ(P)`$, which is bounded on the right.
(c) $``$ (a): Lift the morphism $`V๐(Y)Y`$ in $`๐(\mathrm{Inj}R)`$ to a morphism $`\alpha :VY`$ of complexes of $`R`$-modules. In the mapping cone exact sequence
$$0Y\mathrm{Cone}(\alpha )\mathsf{\Sigma }V0$$
the complex $`\mathrm{Cone}(\alpha )`$ is homotopic to $`๐(Y)`$, and hence totally acyclic, while $`V^n=0`$ for $`n0`$, by hypothesis. Thus, the G-injective dimension of $`Y`$ is finite.
Finally, when $`Y`$ is homologically bounded, $`๐\mathrm{Hom}_R(D,Y)`$ is bounded on the left, so $`Y`$ is in $`(R)`$ if and only if it satisfies condition (b). โ
###### 8.3.
Non-commutative rings. Following the thread in (3.3), (4.8), (5.12), and (7.12), the development of this section also carries over to the context of a pair of rings $`S,R`$ with a dualizing complex $`D`$. In this case, the analogues of Theorems (8.1) and (8.2) identify complexes of finite G-projective dimension over $`R`$ and of finite G-injective dimension over $`S`$ as those in the Auslander category of $`R`$ and the Bass category of $`S`$, respectively. These results contain \[6, (4.1),(4.4)\], but only when one assumes that the ring $`R`$ is left coherent as well; the reason for this has already been given in (3.3).
## Acknowledgments
This project was initiated in June 2004, when Srikanth Iyengar was visiting the University of Paderborn. He thanks the Department of Mathematics at Paderborn and its algebra group for hospitality. The authors thank Lucho Avramov for his comments and suggestions, Lars W. Christensen for numerous discussions on which motivates the results in Section 8, and Amnon Neeman for a careful reading of the manuscript.
|
warning/0506/math0506528.html
|
ar5iv
|
text
|
# Generalizations of Agolโs inequality and nonexistence of tight laminations
## 1 Results
Agolโs inequality (, Theorem 2.1.) is the following:
Agolโs inequality: If $`M`$ is a hyperbolic 3-manifold containing an incompressible, properly embedded surface $`F`$, then
$$Vol\left(M\right)2V_3\chi \left(Guts\left(\overline{MF}\right)\right),$$
where $`V_3`$ is the volume of a regular ideal tetrahedron in hyperbolic 3-space.
In , this inequality has been improved to
$$Vol\left(M\right)Vol\left(Guts\left(\overline{MF}\right)\right)V_{oct}\chi \left(Guts\left(\overline{MF}\right)\right),$$
where $`V_{oct}`$ is the volume of a regular ideal octahedron in hyperbolic 3-space.
In this paper we will, building on ideas from , prove a general inequality for the (transversal) Gromov norm $`M_{}`$ and the normal Gromov norm $`M_{}^{norm}`$ of laminations.
To state the result in its general form we first need two definitions.
Definition (Pared acylindrical): Let $`Q`$ be a manifold with a given decomposition
$$Q=_0Q_1Q.$$
The pair $`(Q,_1Q)`$ is called a pared acylindrical manifold, if any continuous mapping of pairs $`f:(๐^1\times [0,1],๐^1\times \{0,1\})(Q,_1Q)`$, which is $`\pi _1`$-injective as a map of pairs, must be homotopic, as a map of pairs
$$(๐^1\times [0,1],๐^1\times \{0,1\})(Q,_1Q),$$
into $`Q.`$
Definition (Essential Decomposition): Let $`(N,N)`$ be a pair of topological spaces such that $`N=QR`$ for two subspaces $`Q,R`$. Let
$$_0Q=QR,_1Q=QN,_1R=RN,Q=_0Q_1Q,R=_0Q_1R.$$
We say that the decomposition $`N=QR`$ is an essential decomposition of $`(N,N)`$ if the inclusions
$$_1QQN,_1RRN,NN,_0QQ,_0QR$$
are each $`\pi _1`$-injective (for each path-component).
###### Theorem 1.
Let $`M`$ be a compact, orientable, connected n-manifold and $``$ a lamination (of codimension one) of $`M`$.
Assume that $`N:=\overline{M}`$ has a decomposition $`N=QR`$ into orientable n-manifolds (with boundary) $`Q,R`$ such that the following assumptions are satisfied for $`_0Q=QR,_1Q=QN,_1R=RN`$:
i) each path-component of $`_0Q`$ has amenable fundamental group,
ii) $`(Q,_1Q)`$ is pared acylindrical, $`_1Q`$ is acylindrical
iii) $`Q,N,_1Q,_1R,_0Q`$ are aspherical,
iv) the decomposition $`N=QR`$ is an essential decomposition of $`(N,N)`$.
Then
$$M,M_{}^{norm}\frac{1}{n+1}Q.$$
In the case of 3-manifolds $`M`$ carrying an essential lamination $``$, considering $`Q=Guts\left(\overline{M}\right)`$ yields then as a special case:
###### Theorem 2.
Let $`M`$ be a compact 3-manifold with (possibly empty) boundary consisting of incompressible tori, and let $``$ be an essential lamination of $`M`$. Then
$$M,M_{}^{norm}\chi \left(Guts\left(\overline{M}\right)\right).$$
More generally, if $`P`$ is a polyhedron with $`f`$ faces, then
$$M,M_{,P}^{norm}\frac{2}{f2}\chi \left(Guts\left(\overline{M}\right)\right).$$
The following corollary applies, for example, to all hyperbolic manifolds $`M`$ obtained by Dehn-filling the complement of the figure-eight knot in $`๐^3`$. (It is known that each of these $`M`$ contains tight laminations. By the following corollary, all these tight laminations have empty guts.)
Corollary 4: If $`M`$ is a finite-volume hyperbolic 3-manifold with $`Vol\left(M\right)<2V_3=2.02\mathrm{}`$, then $`M`$ carries no essential lamination $``$ with $`M_{,P}^{norm}=M_P`$ for all polyhedra P, and nonempty guts. In particular, there is no tight essential lamination with nonempty guts.
It was observed by Calegari-Dunfield in that a generalization of Agolโs inequality to the case of tight laminations, together with the results in about tight laminations with empty guts, would imply the following corollary.
Corollary 5 (, Conjecture 9.7.): The Weeks manifold admits no tight lamination $``$.
Putting this together with the main result of a recent paper by Tao Li (), one can even improve this result as follows.
Corollary 6: The Weeks manifold admits no transversely orientable essential lamination.
Finally, we also have an application of Theorem 1 to higher-dimensional manifolds.
Corollary 7: Let $`M`$ be a compact Riemannian $`n`$-manifold of negative sectional curvature and finite volume. Let $`FM`$ be a geodesic $`n1`$-dimensional hypersurface of finite volume. Then $`F\frac{n+1}{2}M`$.
The basic idea of Theorem 1, say for simplicity in the special situation of Corollary 7, is the following: a simplex which contributes to a normalized fundamental cycle of $`M`$ should intersect $`Q=2F`$ in at most $`n+1`$ codimension one simplices. This is of course not true in general: simplices can wrap around $`M`$ many times and intersect $`F`$ arbitrarily often, and even a homotopy rel. vertices will not change this. As an obvious examle, look at the following situation: let $`\gamma `$ be a closed geodesic transverse to $`F`$, and for some large $`N`$ let $`\sigma `$ be a straight simplex contained in a small neighborhood of $`\gamma ^N`$. Then $`\sigma `$ intersects $`F`$ $`N`$ times and, since $`\sigma `$ is already straight, this number of intersections can of course not be reduced by straightening. This shows that some more involved straightening must take place, and that the acylindricity of $`F`$ is an essential condition. The way to use acylindricity will be to find a normalization such that many subsets of simplices are mapped to cylinders, which degenerate and thus can be removed without changing the homology class.
We remark that many technical points, in particular the use of multicomplexes, can be omitted if (in the setting of Theorem 2) one does not consider incompressible surfaces or essential laminations, but just geodesic surfaces in hyperbolic manifolds. In this case, all essential parts of the proof of Theorem 1 enter without the notational complications caused by the use of multicomplexes. Therefore we have given a fairly detailed outline of the proof for this special case in Section 6.1. This should help to motivate the general proof in Section 6.2. (We mention that Theorem 1 is not true without assuming amenability of $`\pi _1_0Q`$. This indicates that the proof of multicomplexes in the proof of Theorem 1 seems unavoidable.)
Acknowledgements: It is probably obvious that this paper is strongly influenced by Agolโs preprint . Moreover, the argument that a generalization of Agolโs inequality would imply Corollary 5 is due to .
## 2 Preliminaries
### 2.1 Laminations
Let $`M`$ be an n-manifold, possibly with boundary. In this paper all manifolds will be smooth and orientable. (Hence they are triangulable by Whiteheadโs theorem and possess a locally finite fundamental class.) A (codimension 1) lamination $``$ of $`M`$ is a foliation of a closed subset $``$ of $`M`$, i.e., a decomposition of a closed subset $`M`$ into immersed codimension 1 submanifolds (leaves) so that $`M`$ is covered by charts $`\varphi _j:๐^{n1}\times ๐M`$, the intersection of any leaf with the image of any chart $`\varphi _j`$ being a union of plaques of the form $`\varphi _j\left(๐^{n1}\times \{\}\right)`$. (We will denote by $``$ both the lamination and the laminated subset of $`M`$, i.e. the union of leaves.) If $`M`$ has boundary, we will always assume without further mentioning that $``$ is either transverse to $`M`$ (that is, every leaf is transverse to $``$) or tangential to $`M`$ (that is, $`M`$ is a leaf of $``$). If neither of these two conditions were true, then the transverse and normal Gromov norm would be infinite, therefore all lower bounds will be trivially true.
To construct the leaf space $`T`$ of $``$, one considers the pull-back lamination $`\stackrel{~}{}`$ on the universal covering $`\stackrel{~}{M}`$. The space of leaves $`T`$ is defined as the quotient of $`\stackrel{~}{M}`$ under the following equivalence relation $``$. Two points $`x,y\stackrel{~}{M}`$ are equivalent if either they belong to the same leaf of $`\stackrel{~}{}`$, or they belong to the same connected component of the metric completion $`\overline{\stackrel{~}{M}\stackrel{~}{}}`$ (for the path metric inherited by $`\stackrel{~}{M}\stackrel{~}{}`$ from an arbitrary Riemannian metric on $`\stackrel{~}{M}`$).
Laminations of 3-manifolds. A lamination $``$ of a 3-manifold $`M`$ is called essential if no leaf is a sphere or a torus bounding a solid torus, $`\overline{M}`$ is irreducible, and $`\left(\overline{M}\right)`$ is incompressible and end-incompressible in $`\overline{M}`$, where again the metric completion $`\overline{M}`$ of $`M`$ is taken w.r.t. the path metric inherited from any Riemannian metric on $`M`$, see , ch.1. (Note that $`\overline{M}`$ is immersed in $`M`$, the leaves of $``$ in the image of the immersion are called boundary leaves.)
Examples of essential laminations are taut foliations or compact, incompressible, boundary-incompressible surfaces in compact 3-manifolds. (We always consider laminations without isolated leaves. If a lamination has isolated leaves, then it can be converted into a lamination without isolated leaves by replacing each two-sided isolated leaf $`S_i`$ with the trivially foliated product $`S_i\times [0,1]`$, resp. each one-sided isolated leaf with the canonically foliated normal $`I`$-bundle, without changing the topological type of $`M`$.)
If $``$ is an essential lamination, then the leaf space $`T`$ is an order tree, with segments corresponding to directed, transverse, efficient arcs. (An order tree $`T`$ is a set $`T`$ with a collection of linearly ordered subsets, called segments, such that the axioms of , Def. 6.9., are satisfied.) Moreover, $`T`$ is an $`๐`$-order tree, that is, it is a countable union of segments and each segment is order isomorphic to a closed interval in $`๐`$. $`T`$ can be topologized by the order topology on segments (and declaring that a set is closed if the intersection with each segment is closed). For this topology, $`\pi _0T`$ and $`\pi _1T`$ are trivial (see, for example, , Chapter 5, and its references).
The order tree $`T`$ comes with a fixed-point free action of $`\pi _1M`$. Fenley () has exhibited hyperbolic 3-manifolds whose fundamental groups do not admit any fixed-point free action on $`๐`$-order trees. Thus there are hyperbolic 3-manifolds not carrying any essential lamination.
If $`M`$ is hyperbolic and $``$ an essential lamination, then $`\overline{M}`$ has a characteristic submanifold which is the maximal submanifold that can be decomposed into $`I`$-bundles and solid tori, respecting boundary patterns (see , for precise definitions). The complement of this characteristic submanifold is denoted by $`Guts\left(\right)`$. It admits a hyperbolic metric with geodesic boundary and cusps. (Be aware that some authors, like , include the solid tori into the guts.) If $`=F`$ is a properly embedded, incompressible, boundary-incompressible surface, then Agolโs inequality states that $`Vol\left(M\right)2V_3\chi \left(Guts\left(F\right)\right)`$. This implies, for example, that a hyperbolic manifold of volume $`<2V_3`$ can not contain any geodesic surface of finite area. Recently, this inequality has been improved to $`Vol\left(M\right)Vol\left(Guts\left(F\right)\right)V_{oct}\chi \left(Guts\left(F\right)\right)`$ in , using estimates coming from Perelmanโs work on the Ricci flow.
Assume that $``$ is a codimension one lamination of an n-manifold $`M`$ such that its leaf space $`T`$ is an $`๐`$-order tree. (For example this is the case if $`n=3`$ and $``$ is essential.) An essential lamination is called tight if $`T`$ is Hausdorff. It is called unbranched if $`T`$ is homeomorphic to $`๐`$. It is said to have two-sided branching (, Definition 2.5.2) if there are leaves $`\lambda ,\lambda _1,\lambda _2,\mu ,\mu _1,\mu _2`$ such that the corresponding points in the $`T`$ satisfy $`\lambda <\lambda _1,\lambda <\lambda _2,\mu >\mu _1,\mu >\mu _2`$ but $`\lambda _1,\lambda _2`$ are incomparable and $`\mu _1,\mu _2`$ are incomparable. It is said to have one-sided branching if it is neither unbranched nor has two-sided branching.
If $`M`$ is a hyperbolic 3-manifold and carries a tight lamination with empty guts, then Calegari and Dunfield have shown (, Theorem 3.2.) that $`\pi _1M`$ acts effectively on the circle, i.e., there is an injective homomorphism $`\pi _1MHomeo\left(๐^1\right)`$. This implies that the Weeks manifold (the closed hyperbolic manifold of smallest volume) can not carry a tight lamination with empty guts (, Corollary 9.4.). The aim of this paper is to find obstructions to the existence of laminations with nonempty guts.
### 2.2 Simplicial volume and refinements
Let $`M`$ be a compact, orientable, connected n-manifold, possibly with boundary. Its top integer (singular) homology group $`H_n(M,M;๐)`$ is cyclic. The image of a generator under the change-of-coefficients homomorphism $`H_n(M,M;๐)H_n(M,M;๐)`$ is called a fundamental class and is denoted $`[M,M]`$. If $`M`$ is not connected, we define $`[M,M]`$ to be the formal sum of the fundamental classes of its connected components.
The simplicial volume $`M,M`$ is defined as $`M,M=inf\{_{i=1}^ra_i\}`$ where the infimum is taken over all singular chains $`_{i=1}^ra_i\sigma _i`$ (with real coefficients) representing the fundamental class in $`H_n(M,M;๐)`$.
If $`MM`$ carries a complete hyperbolic metric of finite volume $`Vol\left(M\right)`$, then $`M,M=\frac{1}{V_n}Vol\left(M\right)`$ with $`V_n=sup\{Vol\left(\mathrm{\Delta }\right):\mathrm{\Delta }๐^n\text{geodesic simplex}\}`$ (see ,,, ).
More generally, let $`P`$ be any polyhedron. Then the invariant $`M,M_P`$ is defined in as follows: denoting by $`C_{}(M,M;P;๐)`$ the complex of $`P`$-chains with real coefficients, and by $`H_{}(M,M;P;๐)`$ its homology, there is a canonical chain homomorphism $`\psi :C_{}(M,M;P;๐)C_{}(M,M;๐)`$, given by some triangulations of $`P`$ which is to be chosen such that all possible cancellations of boundary faces are preserved. $`M,M_P`$ is defined as the infimum of $`_{i=1}^ra_i`$ over all $`P`$-chains $`_{i=1}^ra_iP_i`$ such that $`\psi \left(_{i=1}^ra_iP_i\right)`$ represents the fundamental class $`[M,M]`$. Let $`V_P:=sup\left\{Vol\left(\mathrm{\Delta }\right)\right\}`$, where the supremum is taken over all straight $`P`$-polyhedra $`\mathrm{\Delta }๐^3`$. Proposition 1 is Lemma 4.1. in . (The proof in is quite short, and it does not give details for the cusped case. However, the proof in the cusped case can be completed using the arguments in sections 5 and 6 of Francavigliaโs paper .)
###### Proposition 1.
If $`MM`$ admits a hyperbolic metric of finite volume $`Vol\left(M\right)`$, then
$$M,M_P=\frac{1}{V_P}Vol\left(M\right).$$
Let $`M`$ be a manifold and $``$ a codimension one lamination of $`M`$. Let $`\mathrm{\Delta }^n`$ be the standard simplex in $`๐^{n+1}`$, and $`\sigma :\mathrm{\Delta }^nM`$ some continuous singular simplex. The lamination $``$ induces an equivalence relation on $`\mathrm{\Delta }^n`$ by: $`xy\sigma \left(x\right)`$ and $`\sigma \left(y\right)`$ belong to the same connected component of $`L\sigma \left(\mathrm{\Delta }^n\right)`$ for some leaf $`L`$ of $``$. We say that a singular simplex $`\sigma :\mathrm{\Delta }^nM`$ is laminated if the equivalence relation $``$ is induced by a lamination $`_\sigma `$ of $`\mathrm{\Delta }^n`$. We call a lamination $``$ of $`\mathrm{\Delta }^n`$ affine if there is an affine mapping $`f:\mathrm{\Delta }^n๐`$ such that $`x,y\mathrm{\Delta }^n`$ belong to the same leaf if and only if $`f\left(x\right)=f\left(y\right)`$. We say that a lamination $`๐ข`$ of $`\mathrm{\Delta }^n`$ is conjugate to an affine lamination if there is a simplicial homeomorphism $`H:\mathrm{\Delta }^n\mathrm{\Delta }^n`$ such that $`H^{}๐ข`$ is an affine lamination.
We say that a singular $`n`$-simplex $`\sigma :\mathrm{\Delta }^nM`$, $`n2`$, is transverse to $``$ if it is laminated and it is either contained in a leaf, or $`_\sigma `$ is conjugate to an affine lamination $`๐ข`$ of $`\mathrm{\Delta }^n`$.
For $`n=1`$, we say that a singular 1-simplex $`\sigma :\mathrm{\Delta }^1M`$ is transverse to $``$ if it is either contained in a leaf, or for each lamination chart $`\varphi :U๐^{m1}\times ๐^1`$ (with m-th coordinate map $`\varphi _m:U๐^1`$) one has that $`\varphi _m\sigma _{\sigma ^1\left(U\right)}:\sigma ^1\left(U\right)๐^1`$ is locally surjective at all points of $`int\left(\mathrm{\Delta }^1\right)`$, i.e. for all $`pint\left(\mathrm{\Delta }^1\right)\sigma ^1\left(U\right)`$, the image of $`\varphi _m\sigma _{\sigma ^1\left(U\right)}`$ contains a neighborhood of $`\varphi _m\sigma \left(p\right)`$.
We say that the simplex $`\sigma :\mathrm{\Delta }^nM`$ is normal to $``$ if, for each leaf $`F`$, $`\sigma ^1\left(F\right)`$ consists of normal disks, i.e. disks meeting each edge of $`\mathrm{\Delta }^n`$ at most once. (If $`F=M`$ is a leaf of $``$ we also allow that $`\sigma ^1\left(F\right)`$ can be a face of $`\mathrm{\Delta }^n`$). In particular, any transverse simplex is normal.
In the special case of foliations $``$ one has that the transversality of a singular simplex $`\sigma `$ is implied by (hence equivalent to) the normality of $`\sigma `$, as can be shown along the lines of , section 1.3.
More generally, let $`P`$ be any polyhedron. Then we say that a singular polyhedron $`\sigma :PM`$ is normal to $``$ if, for each leaf $`F`$, $`\sigma ^1\left(F\right)`$ consists of normal disks, i.e. disks meeting each edge of $`P`$ at most once (or being equal to a face of $`P`$, if $`F`$ is a boundary leaf).
###### Definition 1.
Let $`M`$ be a compact, oriented, connected n-manifold, possibly with boundary, and $``$ a foliation or lamination on $`M`$. Let $`\mathrm{\Delta }^n`$ be the standard simplex and $`P`$ any polyhedron. Then
$$M,M_{}:=inf\{\underset{i=1}{\overset{r}{}}a_i:\psi \left(\underset{i=1}{\overset{r}{}}a_i\sigma _i\right)\text{ represents }[M,M],\sigma _i:\mathrm{\Delta }^nM\text{ transverse to }\}$$
and
$$M,M_{,P}^{norm}:=inf\{\underset{i=1}{\overset{r}{}}a_i:\psi \left(\underset{i=1}{\overset{r}{}}a_i\sigma _i\right)\text{ represents }[M,M],\sigma _i:PM\text{ normal to }\}.$$
In particular, we define $`M,M_{}^{norm}=M,M_{,\mathrm{\Delta }^n}^{norm}`$.
All norms are finite, under the assumption that $``$ is transverse or tangential to $`M`$.
There is an obvious inequality
$$M,MM,M_{}^{norm}M,M_{}.$$
In the case of foliations, equality $`M,M_{}^{norm}=M,M_{}`$ holds.
(We remark that all definitions extend in an obvious way to disconnected manifolds by summing over the connected components.)
Proposition 2 and Lemma 1 are a straightforward generalisation of , Theorem 2.5.9, and of arguments in .
###### Proposition 2.
Let $`M`$ be a compact, oriented 3-manifold.
a) If $``$ is an essential lamination which is either unbranched or has one-sided branching such that the induced lamination of $`M`$ is unbranched, then
$$M,M_{,P}^{norm}=M,M_P$$
for each polyhedron $`P`$.
b) If $``$ is a tight essential lamination, then
$$M,M_{,P}^{norm}=M,M_P$$
for each polyhedron $`P`$.
###### Proof.
Since $``$ is an essential lamination, we know from , Theorem 6.1., that the leaves are $`\pi _1`$-injective, the universal covering $`\stackrel{~}{M}`$ is homeomorphic to $`๐^3`$ and that the leaves of the pull-back lamination are planes, in particular aspherical. Therefore Proposition 2 is a special case of Lemma 1.โ
###### Lemma 1.
Let $`M`$ be a compact, oriented, aspherical manifold, and $``$ a lamination of codimension one.
Assume that the leaves are $`\pi _1`$-injective and aspherical, and that the leaf space $`T`$ is an $`๐`$-order tree.
a) If the leaf space $`T`$ is either $`๐`$ or branches in only one direction, such that the induced lamination of $`M`$ has leaf space $`๐`$, then
$$M,M_{,P}^{norm}=M,M_P$$
for each polyhedron $`P`$.
b) If the leaf space is a Hausdorff tree, then
$$M,M_{,P}^{norm}=M,M_P$$
for each polyhedron $`P`$.
###### Proof.
To prove the wanted equalities, it suffices in each case to show that any (relative) cycle can be homotoped to a cycle consisting of normal polyhedra.
We denote by $`\stackrel{~}{}`$ the pull-back lamination of $`\stackrel{~}{M}`$ and $`p:\stackrel{~}{M}T=\stackrel{~}{M}/\stackrel{~}{}`$ the projection to the leaf space.
a) First we consider the case that $`P`$=simplex (, Section 4.1) and $``$ unbranched. For this case, we can repeat the argument in , Lemma 2.2.8. Namely, let us be given a (relative) cycle $`_{i=1}^ra_i\sigma _i`$, lift it to a $`\pi _1M`$-equivariant (relative) cycle on $`\stackrel{~}{M}`$ and then perform an (equivariant) straightening, by induction on the dimension of subsimplices of the lifts $`\stackrel{~}{\sigma _i}`$ as follows: for each edge $`\stackrel{~}{e}`$ of any lift $`\stackrel{~}{\sigma _i}`$, its projection $`p\left(\stackrel{~}{e}\right)`$ to the leaf space $`T`$ is homotopic to a unique straight arc $`str\left(p\left(\stackrel{~}{e}\right)\right)`$ in $`T๐`$. It is easy to see (covering the arc by foliation charts and then extending the lifted arc stepwise) that $`str\left(p\left(\stackrel{~}{e}\right)\right)`$ can be lifted to an arc $`str\left(\stackrel{~}{e}\right)`$ with the same endpoints as $`\stackrel{~}{e}`$, and that the homotopy between $`str\left(p\left(\stackrel{~}{e}\right)\right)`$ and $`p\left(\stackrel{~}{e}\right)`$ can be lifted to a homotopy between $`str\left(\stackrel{~}{e}\right)`$ and $`\stackrel{~}{e}`$. $`str\left(\stackrel{~}{e}\right)`$ is transverse to $``$, because its projection is a straight arc in $`T`$. These homotopies of edges can be extended to a homotopy of the whole (relative) cycle. Thus we have straightened the 1-skeleton of the given (relative) cycle.
Now let us be given a 2-simplex $`\stackrel{~}{f}:\mathrm{\Delta }^2\stackrel{~}{M}`$ with transverse edges. There is an obvious straightening $`str\left(p\left(\stackrel{~}{f}\right)\right)`$ of $`p\left(\stackrel{~}{f}\right):\mathrm{\Delta }^2T`$ as follows: if, for $`tT`$, $`\left(p\stackrel{~}{f}\right)^1\left(t\right)`$ has two preimages $`x_1,x_2`$ on edges of $`\mathrm{\Delta }^2`$ (which are necessarily unique), then $`str\left(p\left(\stackrel{~}{f}\right)\right)`$ maps the line which connects $`x_1`$ and $`x_2`$ in $`\mathrm{\Delta }^2`$ constantly to $`t`$. It is clear that this defines a continuous map $`str\left(p\left(\stackrel{~}{f}\right)\right):\mathrm{\Delta }^2T`$.
Since leaves $`\stackrel{~}{F}`$ of $`\stackrel{~}{}`$ are connected ($`\pi _0\stackrel{~}{F}=0`$), $`str\left(p\left(\stackrel{~}{f}\right)\right)`$ can be lifted to a map $`str\left(\stackrel{~}{f}\right):\mathrm{\Delta }^2\stackrel{~}{M}`$ with $`p\left(str\left(\stackrel{~}{f}\right)\right)=str\left(p\left(\stackrel{~}{f}\right)\right)`$. $`str\left(\stackrel{~}{f}\right)`$ is transverse to $``$, because its projection is a straight simplex in $`T`$.
There is an obvious homotopy between $`p\left(\stackrel{~}{f}\right)`$ and $`str\left(p\left(\stackrel{~}{f}\right)\right)`$. For each $`tT`$, the restriction of the homotopy to $`\left(p\stackrel{~}{f}\right)^1\left(t\right)`$ can be lifted to a homotopy in $`\stackrel{~}{M}`$, because $`\pi _1\stackrel{~}{M}=0`$. Since $`\pi _2\stackrel{~}{M}=0`$, these homotopies for various $`tT`$ fit together continuously to give a homotopy between $`\stackrel{~}{f}`$ and $`str\left(\stackrel{~}{f}\right)`$.
These homotopies of 2-simplices leave the (already transverse) boundaries pointwise fixed, thus they can be extended to a homotopy of the whole (relative) cycle. Hence we have straightened the 2-skeleton of the given (relative) cycle.
Assume that we have already straightened the $`k`$-skeleton, for some $`k๐`$. The analogous procedure, using $`\pi _{k1}\stackrel{~}{F}=0`$ for all leaves, and $`\pi _k\stackrel{~}{M}=0,\pi _{k+1}\stackrel{~}{M}=0`$, allows to straighten the $`\left(k+1\right)`$-skeleton of the (relative) cycle. This finishes the proof in the case that $``$ is unbranched.
The generalization to the case that $``$ has one-sided branching such that the induced lamination of $`M`$ is unbranched works as in , Theorem 2.6.6.
We remark that in the case $`P=simplex`$ we get not only a normal cycle, but even a transverse cycle.
Now we consider the case of arbitrary polyhedra $`P`$. Let $`_{i=1}^ra_i\sigma _i`$ be a P-cycle. It can be subtriangulated to a simplicial cycle $`_{i=1}^ra_i_{j=1}^s\tau _{i,j}`$. Again the argument in , Lemma 2.2.8 (resp. its version for manifolds with boundary), shows that this simplicial cycle can be homotoped such that each $`\tau _{i,j}`$ is transverse (and such that boundary cancellations are preserved). But transversality of each $`\tau _{i,j}`$ implies by definition that $`\sigma _i=_{j=1}^s\tau _{i,j}`$ is normal (though in general not transverse) to $``$.
b) By assumption $`\stackrel{~}{M}/\stackrel{~}{}`$ is a Hausdorff tree. We observe that its branching points are the projections of complementary regions. Indeed, let $`F`$ be a leaf of $``$, then $`\stackrel{~}{F}`$ is a submanifold of the contractible manifold $`\stackrel{~}{M}`$. By asphericity and $`\pi _1`$-injectivity of $`F`$, $`\stackrel{~}{F}`$ must be contractible. By Alexander duality it follows that $`\stackrel{~}{M}\stackrel{~}{F}`$ has two connected components. Therefore the complement of the point $`p\left(\stackrel{~}{F}\right)`$ in the leaf space has (at most) two connected components, thus $`p\left(\stackrel{~}{F}\right)`$ can not be a branch point.
Again, to define a straightening of $`P`$-chains it suffices to define a canonical straightening of singular polyhedra $`P`$ such that straightenings of common boundary faces will agree. Let $`\stackrel{~}{v}_0,\mathrm{},\stackrel{~}{v}_n`$ be the vertices of the image of $`P`$. For each pair $`\{\stackrel{~}{v}_i,\stackrel{~}{v}_j\}`$ there exists at most one edge $`\stackrel{~}{e}_{ij}`$ with vertices $`\stackrel{~}{v}_i,\stackrel{~}{v}_j`$ in the image of $`P`$. Since the leaf space is a tree, we have a unique straight arc $`str\left(p\left(\stackrel{~}{e}_{ij}\right)\right)`$ connecting the points $`p\left(\stackrel{~}{v}_i\right)`$ and $`p\left(\stackrel{~}{v}_j\right)`$ in the leaf space. As in a), one can lift this straight arc $`str\left(p\left(\stackrel{~}{e}_{ij}\right)\right)`$ to an arc $`str\left(\stackrel{~}{e}_{ij}\right)`$ in $`\stackrel{~}{M}`$, connecting $`\stackrel{~}{v}_i`$ and $`\stackrel{~}{v}_j`$, which is transverse to $``$. We define this arc $`str\left(\stackrel{~}{e}_{ij}\right)`$ to be the straightening of $`\stackrel{~}{e}_{ij}`$. As in a), we have homotopies of 1-simplices, which extend to a homotopy of the whole (relative) cycle. Thus we have straightened the 1-skeleton.
Now let us be given the 3 vertices $`\stackrel{~}{v}_0,\stackrel{~}{v}_1,\stackrel{~}{v}_2`$ of a 2-simplex $`\stackrel{~}{f}`$ with straight edges. If the projections $`p\left(\stackrel{~}{v}_0\right),p\left(\stackrel{~}{v}_1\right),p\left(\stackrel{~}{v}_2\right)`$ belong to a subtree isomorphic to a connected subset of $`๐`$, then we can straighten $`\stackrel{~}{f}`$ as in a). If not, we have that the projection of the 1-skeleton of this simplex has exactly one branch point, which corresponds to a complementary region. (The projection may of course meet many branch points of the tree, but the image of the projection, considered as a subtree, can have at most one branch point. In general, a subtree with $`n`$ vertices can have at most $`n2`$ branch points.) The preimage of the complement of this complementary region consists of three connected subsets of the 2-simplex (โcorners around the verticesโ). We can straighten each of these subsets and do not need to care about the complementary region corresponding to the branch point. Thus we have straightened the 2-skeleton.
Assume that we have already straightened the $`k`$-skeleton, for some $`k๐`$. Let us be given the $`k+2`$ vertices $`\stackrel{~}{v}_0,\stackrel{~}{v}_1,\mathrm{},\stackrel{~}{v}_{k+1}`$ of a $`\left(k+1\right)`$-simplex with straight faces. Then we have (at most $`k`$) branch points in the projection of the simplex, which correspond to complementary regions. Again we can straighten the parts of the simplex which do not belong to these complementary regions as in a), since they are projected to linearly ordered subsets of the tree. Thus we have straightened the $`\left(k+1\right)`$-skeleton.
Since, by the recursive construction, we have defined straightenings of simplices with common faces by first defining (the same) straightenings of their common faces, the straightening of a (relative) cycle will be again a (relative) cycle, in the same (relative) homology class. โ
Remark: For $`M_{}`$ instead of $`M_{}^{norm}`$, equality b) is in general wrong, and equality a) is unknown (but presumably wrong).
If $``$ is essential but not tight, one may still try to homotope cycles to be transverse, by possibly changing the lamination. In the special case that the cycle is coming from a triangulation, this has been done in and by Brittenham resp. Gabai. It is not obvious how to generalize their arguments to cycles with overlapping simplices.
## 3 Retracting chains to codimension zero submanifolds
### 3.1 Definitions
The results of this section are essentially all due to Gromov, but we follow mainly our exposition in . We start with some recollections about multicomplexes (cf. , Section 3, or , Section 1).
A multicomplex $`K`$ is a topological space $`K`$ with a decomposition into simplices, where each $`n`$-simplex is attached to the $`n1`$-skeleton $`K_{n1}`$ by a simplicial homeomorphism $`f:\mathrm{\Delta }^nK_{n1}`$. (In particular, each $`n`$-simplex has $`n+1`$ distinct vertices.)
As opposed to simplicial complexes, in a multicomplex there may be $`n`$-simplices with the same $`n1`$-skeleton.
We call a multicomplex minimally complete if the following holds: whenever $`\sigma :\mathrm{\Delta }^nK`$ is a singular n-simplex, such that $`_0\sigma ,\mathrm{},_n\sigma `$ are distinct simplices of $`K`$, then $`\sigma `$ is homotopic relative $`\mathrm{\Delta }^n`$ to a unique simplex in $`K`$.
We call a minimally complete multicomplex $`K`$ aspherical if all simplices $`\sigma \tau `$ in $`K`$ satisfy $`\sigma _1\tau _1`$. That means, simplices are uniquely determined by their 1-skeleton.
Orientations of multicomplexes are defined as usual in the simplicial theory. For a simplex $`\sigma `$, $`\overline{\sigma }`$ will denote the simplex with the opposite orientation.
A submulticomplex $`L`$ of a multicomplex $`K`$ consists of a subset of the set of simplices closed under face maps. $`(K,L)`$ is a pair of multicomplexes if $`K`$ is a multicomplex and $`L`$ is a submulticomplex of $`K`$.
A group $`G`$ acts simplicially on a pair of multicomplexes $`(K,L)`$ if it acts on the set of simplices of $`K`$, mapping simplices in $`L`$ to simplices in $`L`$, such that the action commutes with all face maps. For $`gG`$ and $`\sigma `$ a simplex in $`K`$, we denote by $`g\sigma `$ the simplex obtained by this action.
### 3.2 Construction of $`K\left(X\right)`$
We recall the construction from , section 1.3 (originally due to , page 45-46).
For a topological space $`X`$, we denote by $`S_{}\left(X\right)`$ the simplicial set of all singular simplices in $`X`$ and $`S_{}\left(X\right)`$ its geometric realization.
For a topological space $`X`$, a multicomplex $`\widehat{K}\left(X\right)S_{}\left(X\right)`$ is constructed as follows. The 0-skeleton $`\widehat{K}_0\left(X\right)`$ equals $`S_0\left(X\right)`$. The 1-skeleton $`\widehat{K}_1\left(X\right)`$ contains one element in each homotopy class (rel. $`\{0,1\}`$) of singular 1-simplices $`f:[0,1]X`$ with $`f\left(0\right)f\left(1\right)`$. For $`n2`$, assuming by recursion that the n-1-skeleton is defined, the n-skeleton $`\widehat{K}_n\left(X\right)`$ contains one singular n-simplex in each homotopy class (rel. boundary) of singular n-simplices $`f:\mathrm{\Delta }^nX`$ with $`f\widehat{K}_{n1}\left(X\right)`$. We can choose simplices in $`\widehat{K}\left(X\right)`$ such that $`\sigma \widehat{K}\left(X\right)\overline{\sigma }\widehat{K}\left(X\right)`$, where $`\overline{\sigma }`$ denotes the simplex with the opposite orientation. We will henceforth assume that $`\widehat{K}\left(X\right)`$ is constructed according to this condition.
According to , $`\widehat{K}\left(X\right)`$ is weakly homotopy equivalent to $`X`$.
The multicomplex $`K\left(X\right)`$ is defined as the quotient
$$K\left(X\right):=\widehat{K}\left(X\right)/$$
where simplices in $`\widehat{K}\left(X\right)`$ are identified if and only if they have the same 1-skeleton. Let $`p`$ be the canonical projection $`p:\widehat{K}\left(X\right)K\left(X\right).`$
$`K\left(X\right)`$ is minimally complete and aspherical.
If $`X^{}X`$ is a subspace, then we have (not necessarily injective) simplicial mappings $`\widehat{j}:\widehat{K}\left(X^{}\right)\widehat{K}\left(X\right)`$ and $`j:K\left(X^{}\right)K\left(X\right)`$.
If $`\pi _1X^{}\pi _1X`$ is injective (for each path-connected component of $`X^{}`$), then $`j`$ is injective (, Section 1.3) and we can (and will) consider $`K\left(X^{}\right)`$ as a submulticomplex of $`K\left(X\right)`$. (Since simplices in $`\widehat{K}\left(X^{}\right)`$ have image in $`X^{}`$, this means that we assume to have constructed $`\widehat{K}\left(X\right)`$ such that simplices in $`\widehat{K}\left(X\right)`$ have image in $`X^{}`$ whenever this is possible.) If moreover $`\pi _nX^{}\pi _nX`$ is injective for all $`n2`$ (e.g. if $`X^{}`$ is aspherical), then also $`\widehat{j}`$ is injective and $`\widehat{K}\left(X^{}\right)`$ can be considered as a submulticomplex of $`\widehat{K}\left(X\right)`$.
In particular, if $`X`$ and $`X^{}`$ are aspherical and $`\pi _1X^{}\pi _1X`$ is injective, then there is an inclusion
$$i_{}:C_{}^{simp}(K\left(X\right),K\left(X^{}\right))=C_{}^{simp}(\widehat{K}\left(X\right),\widehat{K}\left(X^{}\right))C_{}^{sing}(X,X^{})$$
into the relative singular chain complex of $`(X,X^{})`$.
Infinite and locally finite chains. In this paper we will also work with infinite chains, and in particular with locally finite chains on noncompact manifolds, as introduced in , section 0.2.
For a topological space $`X`$, a formal sum $`_{iI}a_i\sigma _i`$ of singular k-simplices with real coefficients (with a possibly infinite index set $`I`$, and the convention $`a_i0`$ for $`iI`$) is an infinite singular k-chain. It is said to be a locally finite chain if each point of $`X`$ is contained in the image of at most finitely many $`\sigma _i`$. Infinite resp. locally finite k-chains form real vector spaces $`C_k^{inf}\left(X\right)`$ resp. $`C_k^{lf}\left(X\right)`$. The boundary operator maps locally finite k-chains to locally finite k-1-chains, hence, for a pair of spaces $`(X,X^{})`$ the homology $`H_{}^{lf}(X,X^{})`$ of the complex of locally finite chains can be defined.
For a noncompact, orientable $`n`$-manifold $`X`$ with (possibly noncompact) boundary $`X`$, one has a fundamental class $`[X,X]H_n^{lf}(X,X)`$. We will say that an infinite chain $`_{iI}a_i\sigma _i`$ represents $`[X,X]`$ if it is homologous to a locally finite chain representing $`[X,X]H_n^{lf}(X,X)`$.
For a simplicial complex $`K`$, we denote by $`C_k^{simp,inf}\left(K\right)`$ the $`๐`$-vector space of (possibly infinite) formal sums $`_{iI}a_i\sigma _i`$ with $`a_i๐`$ and $`\sigma _i`$ k-simplices in $`K`$. If $`\pi _nX^{}\pi _nX`$ is injective for $`n1`$, we have again the obvious inclusion $`i_{}:C_{}^{simp,inf}(\widehat{K}\left(X\right),\widehat{K}\left(X^{}\right))C_{}^{inf}(X,X^{})`$.
The following observation is of course a well-known application of the homotopy extension property, but we will use it so often that we state it here for reference.
###### Observation 1.
: Let $`X`$ be a topological space and $`\sigma _0:\mathrm{\Delta }^nX`$ a singular simplex. Let $`H:\mathrm{\Delta }^n\times IX`$ be a homotopy with $`H(x,0)=\sigma _0\left(x\right)`$ for all $`x\mathrm{\Delta }^n`$. Then there exists a homotopy $`\overline{H}:\mathrm{\Delta }^n\times IX`$ with $`\overline{H}_{\mathrm{\Delta }^n\times I}=H`$ and $`\overline{H}_{\mathrm{\Delta }^n\times \left\{0\right\}}=\sigma _0`$.
If $`X^{}X`$ is a subspace and the images of $`\sigma _0`$ and $`H`$ belong to $`X^{}`$, then we can choose $`\overline{H}`$ such that its image belongs to $`X^{}`$.
###### Lemma 2.
Let $`(X,X^{})`$ be a pair of topological spaces. Assume $`\pi _nX^{}\pi _nX`$ is injective for each path-component of $`X^{}`$ and each $`n1`$.
a) Let $`_{iI}a_i\tau _iC_n^{inf}(X,X^{})`$ be a (possibly infinite) singular n-chain. Assume that $`I`$ is countable, and that each path-component of $`X`$ and each non-empty path-component of $`X^{}`$ contain uncountably many points. Then $`_{iI}a_i\tau _iC_n^{inf}(X,X^{})`$ is homotopic to a (possibly infinite) simplicial chain $`_ia_i\tau _i^{}C_n^{simp,inf}(\widehat{K}\left(X\right),\widehat{K}\left(X^{}\right))`$. In particular,
$$\underset{i}{}a_i\tau _i^{}C_n^{simp,inf}(\widehat{K}\left(X\right),\widehat{K}\left(X^{}\right))C_{}^{inf}(X,X^{})$$
is homologous to $`_{iI}a_i\tau _i`$.
b) Let $`\sigma _0\widehat{K}\left(X\right)`$ and $`H:\mathrm{\Delta }^n\times IX`$ a homotopy with $`H(.,0)=\sigma _0`$. Consider a minimal triangulation $`\mathrm{\Delta }^n\times I=\mathrm{\Delta }_0\mathrm{}\mathrm{\Delta }_n`$ of $`\mathrm{\Delta }^n\times I`$ into n+1 n+1-simplices. Assume that $`H\left(\mathrm{\Delta }^n\times I\right)`$ consists of simplices in $`\widehat{K}\left(X\right)`$. Then $`H`$ is homotopic (rel. $`\mathrm{\Delta }^n\times \left\{0\right\}\mathrm{\Delta }^n\times I`$) to a map $`\overline{H}:\mathrm{\Delta }^n\times IX`$ such that $`\overline{H}_{\mathrm{\Delta }_i}\widehat{K}\left(X\right)`$, in particular $`\sigma _1:=\overline{H}(.,1)\widehat{K}\left(X\right)`$.
###### Proof.
a) From the assumptions it follows that there exists a homotopy of the 0-skeleton such that each vertex is moved into a distinct point of $`X`$, and such that vertices in $`X^{}`$ remain in $`X^{}`$ during the homotopy. By Observation 1, this homotopy can by induction be extended to a homotopy of the whole chain.
Now we prove the claim by induction on $`k`$ ($`0k<n`$). We assume that the $`k`$-skeleton of $`_{iI}a_i\tau _i`$ consists of simplices in $`\widehat{K}\left(X\right)`$ and we want to homotope $`_{iI}a_i\tau _i`$ such that the homotoped k+1-skeleton consists of simplices in $`\widehat{K}\left(X\right)`$.
By construction, each singular k+1-simplex $`\sigma `$ in $`X`$ with boundary a simplex in $`\widehat{K}\left(X\right)`$ is homotopic (rel. boundary) to a unique k+1-simplex in $`\widehat{K}\left(X\right)`$. Since the homotopy keeps the boundary fixed, the homotopies of different k+1-simplices are compatible. By Observation 1, the homotopy of the k+1-skeleton can by induction be extended to a homotopy of the whole chain.
If the image of the k+1-simplex $`\sigma `$ is contained in $`X^{}`$, then it is homotopic rel. boundary to a simplex in $`\widehat{K}\left(X^{}\right)`$, for a homotopy with image in $`X^{}`$. Thus we can realise the homotopy such that all simplices with image in $`X^{}`$ are homotoped inside $`X^{}`$.
b) follows by the same argument as a), succesively applied to $`\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n`$.
We remark that there exists a canonical simplicial map
$$p:C_{}^{simp,inf}(\widehat{K}\left(X\right),\widehat{K}\left(X^{}\right))C_{}^{simp,inf}(K\left(X\right),K\left(X^{}\right)).$$
$`p`$ is defined by induction. It is defined to be the identity on the 1-skeleton. If it is defined on the n-1-skeleton, for $`n2`$, then, for an n-simplex $`\tau `$, $`p\left(\tau \right)K\left(X\right)`$ is the unique simplex with $`_ip\left(\tau \right)=p\left(_i\tau \right)`$ for $`i=0,\mathrm{},n`$.
### 3.3 Action of $`G=\mathrm{\Pi }\left(A\right)`$
We repeat the definitions from , section 1.5. (originally due to ), as they will be frequently used in the remainder of the paper.
Let $`(P,A)`$ be a pair of minimally complete multicomplexes.
We define its nontrivial-loops space $`\mathrm{\Omega }^{}A`$ as the set of homotopy classes (rel. $`\{0,1\}`$) of continuous maps $`\gamma :[0,1]A`$ with $`\gamma \left(0\right)=\gamma \left(1\right)`$ and not homotopic (rel. $`\{0,1\}`$) to a constant map.
We define
$$\mathrm{\Pi }\left(A\right):=\left\{\begin{array}{c}\{\gamma _1,\mathrm{},\gamma _n\}:nN,\gamma _1,\mathrm{},\gamma _nA_1\mathrm{\Omega }^{}A\\ \gamma _i\left(0\right)\gamma _j\left(0\right),\gamma _i\left(1\right)\gamma _j\left(1\right)\text{ for }ij,\\ \{\gamma _1\left(0\right),\mathrm{},\gamma _n\left(0\right)\}=\{\gamma _1\left(1\right),\mathrm{},\gamma _n\left(1\right)\}.\end{array}\right\}.$$
If $`\gamma ,\gamma ^{}`$ are elements of $`A_1`$ with $`\gamma ^{}\overline{\gamma }`$ and $`\gamma \left(0\right)=\gamma ^{}\left(1\right)`$, we denote<sup>1</sup><sup>1</sup>1We follow the usual convention to define the concatenation of paths by $`\gamma \gamma ^{}\left(t\right)=\gamma \left(2t\right)`$ if $`t\frac{1}{2}`$ and $`\gamma \gamma ^{}\left(t\right)=\gamma ^{}\left(2t1\right)`$ if $`t\frac{1}{2}`$. Unfortunately this implies that, in order to let $`\mathrm{\Pi }\left(A\right)`$ act on $`P`$, we will have the multiplication in $`\mathrm{\Pi }\left(A\right)`$ such that, for example, $`\left\{\gamma \right\}\left\{\gamma ^{}\right\}=\left\{\gamma ^{}\gamma \right\}`$. We hope that this does not lead to confusion. $`\gamma \gamma ^{}A_1`$ to be the unique edge of $`A`$ in the homotopy class of the concatenation. If $`\gamma A_1`$ and $`\gamma ^{}\mathrm{\Omega }^{}A`$ (or vice versa), with $`\gamma \left(1\right)\gamma \left(0\right)=\gamma ^{}\left(1\right)=\gamma ^{}\left(0\right)`$, we also denote $`\gamma \gamma ^{}A_1`$ the unique edge in the homotopy class of the concatenation. If $`\gamma ,\gamma ^{}\mathrm{\Omega }^{}A`$ with $`\gamma \left(1\right)=\gamma \left(0\right)=\gamma ^{}\left(1\right)=\gamma ^{}\left(0\right)`$, we denote $`\gamma \gamma ^{}\mathrm{\Omega }^{}A`$ the concatenation of homotopy classes of loops.
This can be used to define a multiplication on $`\mathrm{\Pi }\left(A\right)`$ as follows:
given $`\{\gamma _1,\mathrm{},\gamma _m\}`$ and $`\{\gamma _1^{},\mathrm{},\gamma _n^{}\}`$, we chose a reindexing of the unordered sets $`\{\gamma _1,\mathrm{},\gamma _m\}`$ and $`\{\gamma _1^{},\mathrm{},\gamma _n^{}\}`$ such that we have: $`\gamma _j\left(1\right)=\gamma _j^{}\left(0\right)`$ for $`1ji`$ and $`\gamma _j\left(1\right)\gamma _k^{}\left(0\right)`$ for $`ji+1,ki+1`$. (Since we are assuming that all $`\gamma _j\left(1\right)`$ are pairwise distinct, and also all $`\gamma _j^{}\left(0\right)`$ are pairwise distinct, such a reindexing exists for some $`i0`$, and it is unique up to permuting the indices $`i`$ and permuting separately the indices of $`\gamma _j`$โs and $`\gamma _k^{}`$โs with $`ji+1,ki+1`$.)
Moreover we permute the indices $`\{1,\mathrm{},i\}`$ such that there exists some $`h`$ with $`0hi`$ satisfying the following conditions:
\- for $`1jh`$ we have either $`\gamma _j^{}\overline{\gamma _j}A_1`$ or $`\gamma _j^{}\gamma _j^1\mathrm{\Omega }^{}A`$,
\- for $`h<ji`$ we have either $`\gamma _j^{}=\overline{\gamma _j}A_1`$ or $`\gamma _j^{}=\gamma _j^1\mathrm{\Omega }^{}A`$.
With this fixed reindexing we define
$$\{\gamma _1,\mathrm{},\gamma _m\}\{\gamma _1^{},\mathrm{},\gamma _n^{}\}:=\{\gamma _1^{}\gamma _1,\mathrm{},\gamma _h^{}\gamma _h,\gamma _{i+1},\mathrm{},\gamma _m,\gamma _{i+1}^{},\mathrm{},\gamma _n^{}\}.$$
(Note that we have omitted all $`\gamma _j^{}\gamma _j`$ with $`j>h`$. The choice of $`\gamma _j^{}\gamma _j`$ rather than $`\gamma _j\gamma _j^{}`$ is just because we want to define a left action on $`(P,A)`$.)
We have shown in (footnote to Section 1.5.1) that the product belongs to $`\mathrm{\Pi }\left(A\right)`$. Moreover, the so-defined multiplication is independent of the chosen reindexing. It is clearly associative. A neutral element is given by the empty set. The inverse to $`\{\gamma _1,\mathrm{},\gamma _n\}`$ is given by $`\{\gamma _1^{},\mathrm{},\gamma _n^{}\}`$ with $`\gamma _i^{}=\overline{\gamma _i}`$ if $`\gamma _iA_1`$ resp. $`\gamma _i^{}=\gamma _i^1`$ if $`\gamma _i\mathrm{\Omega }^{}A`$. (Indeed, in this case $`h=0`$, thus $`\{\gamma _1,\mathrm{},\gamma _n\}\{\gamma _1^{},\mathrm{},\gamma _n^{}\}`$ is the empty set.) Thus we have defined a group law on $`\mathrm{\Pi }\left(A\right)`$.
We remark that there is an inclusion $`\mathrm{\Pi }\left(A\right)map_0(A_0,[[0,1],A]_P)`$, where $`[[0,1],A]_P`$ is the set of homotopy classes (in $`P`$) rel. $`\{0,1\}`$ of maps from $`[0,1]`$ to $`A`$, and $`map_0(A_0,[[0,1],A]_P)`$ is the set of maps $`f:A_0[[0,1],A]_P`$ with
\- $`f\left(y\right)\left(0\right)=y`$ for all $`yA_0`$ and
\- $`f(.)\left(1\right):A_0A_0`$ is a bijection.
This inclusion is given by sending $`\{\gamma _1,\mathrm{},\gamma _n\}`$ to the map $`f`$ defined by $`f\left(\gamma _i\left(0\right)\right)=\left[\gamma _i\right]`$ for $`i=1,\mathrm{},n`$, and $`f\left(y\right)=\left[c_y\right]`$ (the constant path) for $`y\{\gamma _1\left(0\right),\mathrm{},\gamma _n\left(0\right)\}`$.
The inclusion is a homomorphism with respect to the group law defined on $`map_0(A_0,[[0,1],A]_P)`$ by $`\left[gf\left(y\right)\right]:=\left[f\left(y\right)\right]\left[g\left(f\left(y\right)\left(1\right)\right)\right]`$.
Action of $`\mathrm{\Pi }\left(A\right)`$ on $`P`$:
From now on we assume: $`P`$ is aspherical. We define an action of $`map_0(A_0,[[0,1],A]_P)`$ on $`P`$. This gives, in particular, an action of $`\mathrm{\Pi }\left(A\right)`$ on $`P`$.
Let $`gmap_0(A_0,[(0,1),A]_P)`$. Define $`gy=g\left(y\right)\left(1\right)`$ for $`yA_0`$ and $`gx=x`$ for $`xP_0A_0`$. This defines the action on the 0-skeleton of $`P`$.
We extend this to an action on the 1-skeleton of $`P`$: Recall that, by minimal completeness of $`P`$, $`1`$-simplices $`\sigma `$ are in 1-1-correspondence with homotopy classes (rel. $`\{0,1\}`$) of (nonclosed) singular 1-simplices in $`P`$ with vertices in $`P_0`$. Using this correspondence, define $`g\sigma :=\left[\overline{g\left(\sigma \left(0\right)\right)}\right]\left[\sigma \right]\left[g\left(\sigma \left(1\right)\right)\right]`$, where $``$ denotes concatenation of (homotopy classes of) paths.
In , Section 1.5.1, we proved that this defines an action on $`P_1`$ and that there is an extension of ths action to an action on $`P`$. (The extension is unique because $`P`$ is aspherical.)
We remark, because this will be one of the assumptions to apply Lemma 7, that the action of any element $`g\mathrm{\Pi }\left(A\right)`$ is homotopic to the identity. The homotopy between the action of the identity and the action of $`\{\gamma _1,\mathrm{},\gamma _r\}`$ given by the action of $`\{\gamma _1^t,\mathrm{},\gamma _r^t\},0t1`$, with $`\gamma _i^t\left(s\right)=\gamma _i\left(st\right)`$.
The next Lemma follows directly from the construction, but we will use it so often that we want to explicitly state it.
###### Lemma 3.
Let $`(P,A)`$ be a pair of aspherical, minimally complete multicomplexes, with the action of $`G=\mathrm{\Pi }\left(A\right)`$. If $`\sigma P`$ is a simplex, all of whose vertices are not in $`A`$, then $`g\sigma =\sigma `$ for all $`gG`$.
For a topological space and a subset $`PS_{}\left(X\right)`$, closed under face maps, the (antisymmetric) bounded cohomology $`H_b^{}\left(P\right)`$ and its pseudonorm are defined literally like for multicomplexes in , Section 3.2. The following well-known fact will be needed for applications of Lemma 7 (to the setting of Theorem 1) with $`P=K^{str}\left(Q\right),G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$.
###### Lemma 4.
a) Let $`(P,A)`$ be a pair of minimally complete multicomplexes. If each connected component of $`A`$ has amenable fundamental group, then $`\mathrm{\Pi }\left(A\right)`$ is amenable.
b) Let $`X`$ be a topological space, $`PS_{}\left(X\right)`$ a subset closed under face maps, and $`G`$ an amenable group acting on $`P`$. Then the canonical homomorphism
$$id1:C_{}^{simp}\left(P\right)C_{}^{simp}\left(P\right)_{๐G}๐$$
induces an isometric monomorphism in bounded cohomology.
###### Proof.
a) The proof is an obvious adaptation of the proof of , Lemma 4.
b) This is proved by averaging bounded cochains, see .โ
### 3.4 Retraction to central simplices
###### Lemma 5.
Let $`(N,N)`$ be a pair of topological spaces with $`N=QR`$ for two subspaces $`Q,R`$. Let
$$_0Q=QR,_1Q=QN,_1R=RN,Q=_0Q_1Q,R=_0Q_1R.$$
Assume that $`_1QQN,_1RRN,NN,_0QQ,_0QR`$ are $`\pi _1`$-injective, and that $`N,_1Q,_1R,_0Q`$ are aspherical (and thus the corresponding $`K(.)`$ can be considered as submulticomplexes of $`K\left(N\right)`$.)
Consider the simplicial action of $`G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$ on $`K\left(N\right)`$.
Then there is a chain homomorphism
$$r:C_{}^{simp,inf}\left(K\left(N\right)\right)_{๐G}๐C_{}^{simp,inf}\left(K\left(Q\right)\right)_{๐G}๐$$
in degrees $`2`$, mapping $`C_{}^{simp,inf}\left(GK\left(N\right)\right)_{๐G}๐`$ to $`C_{}^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐`$ such that
\- if $`\sigma `$ is a simplex in $`K\left(N\right)`$, then $`r\left(\sigma 1\right)=\kappa 1`$, where either $`\kappa `$ is a simplex in $`K\left(Q\right)`$ or $`\kappa =0`$,
\- if $`\sigma `$ is a simplex in $`K\left(Q\right)`$, then $`r\left(\sigma 1\right)=\sigma 1`$,
\- if $`\sigma `$ is a simplex in $`K\left(R\right)`$, then $`r\left(\sigma 1\right)=0`$.
###### Proof.
This is , Proposition 6. (We have replaced the assumption $`ker\left(\pi _1_0Q\pi _1Q\right)=ker\left(\pi _1_0Q\pi _1R\right)`$ from by the stronger assumption of $`\pi _1`$-injectivity, since this will be true in all our applications and we have no need for the more general assumption.) The Conclusion is stated in for locally finite chains, but of course $`r`$ extends linearly to infinite chains. โ
Remark: If some edge of $`\sigma `$ is contained in $`K\left(_0Q\right)=K\left(Q\right)K\left(R\right)`$, then
$$\sigma 1=0C_{}^{simp,inf}\left(K\left(N\right)\right)_{๐G}๐,$$
see , Section 1.5.2. (The proof is essentially the same as that of Lemma 15 below.) In particular, if $`\sigma `$ is contained in both $`K\left(Q\right)`$ and $`K\left(R\right)`$, then $`r\left(\sigma 1\right)=r\left(0\right)=0`$.
Fundamental cycles in $`K\left(N\right)`$ and $`K\left(Q\right)`$. Let $`N`$ be a (possibly noncompact) connected, orientable n-manifold with (possibly noncompact) boundary $`N`$. Then $`H_n^{lf}(N,N)๐`$ by Whiteheadโs theorem and a generator is called $`[N,N]`$. (It is only defined up to sign, but this will not concern our arguments.) Recall that an infinite chain is said to represent $`[N,N]`$ if it is homologous to a locally finite chain representing $`[N,N]`$.
If $`NN`$ is $`\pi _1`$-injective and $`N`$ is aspherical, then $`C_{}^{simp,inf}(\widehat{K}\left(N\right),\widehat{K}\left(N\right))C_{}^{sing,inf}(N,N)`$, see Section 3.2. Thus it makes sense to say that some chain $`zC_{}^{simp,inf}(\widehat{K}\left(N\right),\widehat{K}\left(N\right))`$ represents the fundamental class $`[N,N]`$.
If $`_1QQ`$ is $`\pi _1`$-injective and $`Q`$ and $`_1Q`$ are aspherical, and if $`G:=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$, then $`C_{}^{simp,inf}\left(GK\left(_1Q\right)\right)=C_{}^{simp,inf}\left(G\widehat{K}\left(_1Q\right)\right)C_{}^{sing,inf}\left(Q\right)`$, because $`G`$ maps simplices in $`im\left(K\left(Q\right)K\left(Q\right)\right)`$ to simplices in $`im\left(K\left(Q\right)K\left(Q\right)\right)`$. Thus it makes sense to say that some chain $`zC_{}^{simp,inf}(K\left(Q\right),GK\left(_1Q\right))`$ represents the fundamental class $`[Q,Q]`$.
The projection $`p:\widehat{K}\left(N\right)K\left(N\right)`$ is defined at the end of 3.2.
###### Lemma 6.
Let $`N^{n2}`$ be an orientable n-manifold with boundary, and $`Q,RN`$ orientable n-manifolds with boundary, such that $`N=QR`$ satisfies the assumptions of Lemma 5 and $`_0Q,_1Q,_1R`$ are n-1-dimensional submanifolds (with boundary) of $`Q`$ resp. $`R`$. Assume, in addition, that $`Q`$ is aspherical.
If $`_ia_i\sigma _iC_n^{simp,inf}(\widehat{K}\left(N\right),\widehat{K}\left(N\right))`$ represents $`[N,N]`$, then
$$\underset{i}{}a_ir\left(p\left(\sigma _i\right)\right)1C_n^{simp,inf}(K\left(Q\right),GK\left(_1Q\right))_{๐G}๐$$
represents <sup>2</sup><sup>2</sup>2This means that it represents the image of $`h1`$ under the canonical homomorphism $`H_n^{sing,inf}(Q,Q)_{๐G}๐H_n\left(C_{}^{sing,inf}(Q,Q)_{๐G}๐\right)`$, where $`hH_n^{simp,inf}(K\left(Q\right),GK\left(_1Q\right))`$ represents $`[Q,Q]H_n^{sing}(Q,Q)`$ $`[Q,Q]1`$
and
$$\underset{i}{}a_ir\left(p\left(\sigma _i\right)\right)1C_n^{simp,inf}\left(GK\left(Q\right)\right)_{๐G}๐$$
represents <sup>3</sup><sup>3</sup>3This means that it represents the image of $`h1`$ under the canonical homomorphism $`H_n^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐H_n\left(C_{}^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐\right)`$, where $`hH_n^{simp,inf}\left(GK\left(_1Q\right)\right)`$ represents $`\left[Q\right]H_n^{sing}\left(Q\right).`$ $`\left[Q\right]1`$.
###### Proof.
Since $`p`$ and $`r`$ are chain maps, it suffices to check the claim for some chosen representative of $`[N,N]`$. So let $`zC_{}^{simp,inf}(\widehat{K}\left(N\right),\widehat{K}\left(N\right))`$ be a representative of $`[N,N]`$ chosen such that
$$p\left(z\right)=z_Q+z_R$$
where $`z_Q`$ represents $`[Q,Q]`$ and $`z_R`$ represents $`[R,R]`$ and such that
$$z_Q=w_1+w_2,z_R=w_2+w_3$$
with $`w_1C_{n1}^{simp,inf}\left(K\left(_1Q\right)\right),w_2C_{n1}^{simp,inf}\left(K\left(_0Q\right)\right),w_3C_{n1}^{simp,inf}\left(K\left(_1R\right)\right)`$ representing $`\left[_1Q\right],\left[_0Q\right],\left[_1R\right]`$, respectively.
From Lemma 5:
$$r\left(p\left(z\right)1\right)=z_Q1,$$
which implies the first claim, and
$$r\left(p\left(z\right)1\right)=z_Q1=w_11+w_21.$$
Since $`w_1+w_2`$ represents $`\left[Q\right]`$, this implies the second claim.
(Remark: From the Remark after Lemma 5 we have $`w_21=0`$. This implies $`r\left(p\left(z\right)1\right)=z_Q1=w_11`$, that is, $`r\left(p\left(z\right)1\right)`$ represents at the same time $`\left[Q\right]1`$ and $`\left[_1Q\right]1`$.)
### 3.5 Using amenability
Lemma 7 is well-known in slightly different formulations and we reprove it here only for completeness. We will apply<sup>4</sup><sup>4</sup>4If a group $`G`$ acts simplicially on a multicomplex $`M`$, then $`C_{}\left(M\right)_{๐G}๐`$ are abelian groups with well-defined boundary operator $`_{}1`$, even though $`M/G`$ may not be a multicomplex, like for the action of $`G=\mathrm{\Pi }_X\left(X\right)`$ on $`K\left(X\right)`$, for a topological space $`X`$. We remark that $`C_{}\left(M\right)_{๐G}๐C_{}\left(M\right)_{๐G}๐`$ is just the quotient chain complex for the $`G`$-action. In particular, even though $`C_{}\left(M\right)`$ is an $`๐G`$-module, it does not make any difference whether we tensor over $`๐G`$ or $`๐G`$. Lemma 7 in the proof of Theorem 1 with $`X=Q,G=q_{}\left(\mathrm{\Pi }\left(K\left(_0Q\right)\right)\right)`$ and $`K=GK^{str}\left(_1Q\right)`$. (The following lemma has of course also a relative version, but we will not need that for our argument.)
###### Lemma 7.
: Let $`X`$ be a closed, orientable manifold and $`KS_{}\left(X\right)`$ closed under face maps. Assume that
\- there is an amenable group $`G`$ acting on $`K`$, such that the action of each $`gG`$ on $`K`$ is homotopic to the identity
\- there is a fundamental cycle $`zC_{}^{simp}\left(K\right)`$ such that $`z1`$ is homologous to a cycle $`h=_{j=1}^sb_j\tau _j1C_{}^{simp}\left(K\right)_{๐G}๐`$.
Then
$$X\underset{j=1}{\overset{s}{}}b_j.$$
###### Proof.
If $`X=0`$, there is nothing to prove.
Thus we may assume $`X0`$, which implies (, p.17) that there is $`\beta H_b^n\left(X\right)`$, a bounded cohomology class dual to $`\left[X\right]H_n\left(X\right)`$, with $`\beta =\frac{1}{X}`$.
Let $`p:C_{}^{simp}\left(K\right)C_{}^{simp}\left(K\right)_{๐G}๐`$ be the homomorphism defined by $`p\left(\sigma \right)=\sigma 1`$. Since $`G`$ is amenable we have, by the proof of Lemma 4b) in , an โaveraging homomorphismโ $`Av:H_b^{}\left(K\right)H_b^{}\left(C_{}\left(K\right)_{๐G}๐\right)`$ such that $`Av`$ is left-inverse to $`p^{}`$ and $`Av`$ is an isometry. Hence we have
$$Av\left(\beta \right)=\beta =\frac{1}{X}.$$
Moreover, denoting by $`\left[_{j=1}^sb_j\tau _j1\right]`$ the homology class of $`_{j=1}^sb_j\tau _j1`$, we have obviously
$$Av\left(\beta \right)\left[\underset{j=1}{\overset{s}{}}b_j\tau _j1\right]Av\left(\beta \right)\underset{j=1}{\overset{s}{}}b_j$$
and therefore
$$X=\frac{1}{Av\left(\beta \right)}\frac{_{j=1}^sb_j}{Av\left(\beta \right)\left[_{j=1}^sb_j\tau _j1\right]}.$$
It remains to prove $`Av\left(\beta \right)\left[_{j=1}^sb_j\tau _j1\right]=1`$.
For this we have to look at the definition of $`Av`$, which is as follows: Let $`\gamma C_b^{}\left(K\right)`$ be a bounded cochain. By amenability there exists a bi-invariant mean $`av:B\left(G\right)๐`$ on the bounded functions on $`G`$ with $`inf_{gG}\delta \left(g\right)av\left(\delta \right)sup_{gG}\delta \left(g\right)`$ for all $`\delta B\left(G\right)`$. Then, given any $`p\left(\sigma \right)C_{}\left(K\right)_{๐G}๐`$ one can fix an identification between $`G`$ and $`G\sigma `$, the set of all $`\sigma ^{}`$ with $`p\left(\sigma ^{}\right)=p\left(\sigma \right)`$, and thus consider the restriction of $`\gamma `$ to $`G\sigma `$ as a bounded cochain on $`G`$. Define $`Av\left(\gamma \right)\left(p\left(\sigma \right)\right)`$ to be the average $`av`$ of this bounded cochain on $`GG\sigma `$. (This definition is independent of all choices, see .)
Now, if $`z=_{j=1}^sb_j\tau _j`$ is a fundamental cycle, then we have $`\beta \left(z\right)=1`$.
If $`gG`$ is arbitrary, then left multiplication with $`g`$ is a chain map on $`C_{}^{simp}\left(K\right)`$, as well as on $`C_{}^{sing}\left(X\right)`$. Since the action of $`g`$ on $`K`$ is homotopic to the identity, it induces the identity on the image of $`C_{}^{simp}\left(K\right)C_{}^{sing}\left(X\right)`$. Thus, for each cycle $`zC_{}^{simp}\left(K\right)`$ representing $`\left[X\right]H_{}^{sing}\left(X\right)`$, the cycle $`gzC_{}^{simp}\left(K\right)`$ must also represent $`\left[X\right]`$.
If $`gz`$ represents $`\left[X\right]`$, then $`\beta \left(gz\right)=\beta \left(\left[X\right]\right)=1`$. In conclusion, we have $`\beta \left(p\left(z^{}\right)\right)=1`$ for each $`z^{}`$ with $`p\left(z^{}\right)=p\left(z\right)`$. By definition of $`Av`$, this implies $`Av\left(\beta \right)\left(p\left(z\right)\right)=1`$ for each fundamental cycle $`z`$.
In particular, $`Av\left(\beta \right)\left[_{j=1}^sb_j\tau _j1\right]=1`$, finishing the proof of the lemma. โ
Remark: In the proof of Theorem 1, we will work with $`C_{}^{simp}\left(K\right)_{๐G}๐`$ rather than $`C_{}^{simp}\left(K\right)`$. This is analogous to Agolโs construction of โcrushing the cusps to pointsโ in . However $`C_{}^{simp}\left(K\left(Q\right)\right)_{๐\mathrm{\Pi }\left(_0Q\right)}๐C_{}^{simp}\left(K\left(Q/_0Q\right)\right)`$, thus one can not simplify our arguments by working directly with $`Q/_0Q`$.
## 4 Disjoint planes in a simplex
In this section, we will discuss the possibilities how a simplex can be cut by planes without producing parallel arcs in the boundary. (More precisely, we pose the additional condition that the components of the complement can be coloured by black and white such that all vertices belong to black components, and we actually want to avoid only parallel arcs in the boundary of white components.) For example, for the 3-simplex, it will follow that there is essentially only the possibility in Case 1, pictured below, meanwhile in Case 2 each triangle has a parallel arc with another triangle, regardless how the quadrangle is triangulated.
Let $`\mathrm{\Delta }^n๐^{n+1}`$ be the standard simplex<sup>5</sup><sup>5</sup>5As usual, $`v_i`$ is the vertex with all coordinates, except the i-th, equal to zero, and $`_i\mathrm{\Delta }^n`$ denotes the subsimplex spanned by all vertices except $`v_i`$. We will occasionally identify singular 1-simplices $`\sigma :\mathrm{\Delta }^1M`$ with paths $`e:[0,1]M`$ by the rule $`e\left(t\right)=\sigma (t,1t)`$. In particular, $`e\left(0\right)=\sigma \left(v_0\right)=_1\sigma `$ and $`e\left(1\right)=\sigma \left(v_1\right)=_0\sigma `$. with vertices $`v_0,\mathrm{},v_n`$. It is contained in the plane $`E=\{(x_1,\mathrm{},x_{n+1})๐^{n+1}:x_1+\mathrm{}+x_{n+1}=1\}`$.
In this section we will be interested in n-1-dimensional affine planes $`PE`$ whose intersection with $`\mathrm{\Delta }^n`$ either contains no vertex, consists of exactly one vertex, or consists of a face of $`\mathrm{\Delta }^n`$. For such planes we define their type as follows.
###### Definition 2.
Let $`PE`$ be an n-1-dimensional affine plane such that $`P\mathrm{\Delta }^n`$ either contains no vertex, consists of exactly one vertex, or consists of a face of $`\mathrm{\Delta }^n`$.
If $`P\mathrm{\Delta }^n=_0\mathrm{\Delta }^n`$, then we say that $`P`$ is of type $`\left\{0\right\}`$.
If $`P\mathrm{\Delta }^n=_j\mathrm{\Delta }^n`$ with $`j1`$, then we say that $`P`$ is of type $`\left\{01\mathrm{}\widehat{j}\mathrm{}n\right\}`$.
If $`P\{v_0,\mathrm{},v_n\}=\left\{v_0\right\}`$, then we say that $`P`$ is of type $`\left\{0\right\}`$.
If $`P\{v_0,\mathrm{},v_n\}=\mathrm{}`$ or $`P\{v_0,\mathrm{},v_n\}=\left\{v_j\right\}`$ with $`j1`$, then we say that $`P`$ is of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`a_1,\mathrm{},a_k\{1,\mathrm{},n\}`$ if:
$`v_i`$ belongs to the same connected component of $`\mathrm{\Delta }^n\left(P\mathrm{\Delta }^n\right)`$ as $`v_0`$
if and only if $`i\{a_1,\mathrm{},a_k\}`$.
###### Observation 2.
Let $`P_1,P_2`$ be two planes of type $`\left\{0a_1\mathrm{}a_k\right\}`$ resp. $`\left\{0b_1\mathrm{}b_l\right\}`$ and let $`Q_1=P_1\mathrm{\Delta }^n\mathrm{},Q_2=P_2\mathrm{\Delta }^n\mathrm{}`$. Then $`Q_1Q_2=\mathrm{}`$ implies that either $`\{a_1,\mathrm{},a_k\}=\{b_1,\mathrm{},b_l\}`$ or exactly one of the following conditions holds:
\- $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}`$,
\- $`\{b_1,\mathrm{},b_l\}\{a_1,\mathrm{},a_k\}`$,
\- $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}=\{1,\mathrm{},n\}`$.
###### Proof.
$`\mathrm{\Delta }^nQ_1`$ consists of two connected components, $`C_1`$ and $`C_2`$. W.l.o.g. assume that $`v_0C_1`$. $`\mathrm{\Delta }^nQ_2`$ consists of two connected components, $`D_1`$ and $`D_2`$. W.l.o.g. assume that $`v_0D_1`$. In particular, $`C_1D_1\mathrm{}`$.
Since $`Q_1Q_2=\mathrm{}`$, it follows that $`Q_2`$ is contained in one of $`C_1`$ or $`C_2`$, and $`Q_1`$ is contained in one of $`D_1`$ or $`D_2`$.
Case 1: $`Q_1D_1`$. Then either we have $`C_1D_1`$, which implies $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}`$, or we have $`C_2D_1`$, which implies $`\{1,\mathrm{},n\}\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}`$, hence $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}=\{1,\mathrm{},n\}`$.
Case 2: $`Q_1D_2`$. This implies $`Q_2C_1`$ and after interchanging $`Q_1`$ and $`Q_2`$ we are in Case 1.โ
Notational remark: โarcโ will mean the intersection of an n-1-dimensional affine plane $`PE`$ (such that $`P\mathrm{\Delta }^n\mathrm{}`$ either contains no vertex, consists of exactly one vertex or consists of a face) with a 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$. If an arc consists of only one vertex, we call it a degenerate arc.
###### Definition 3.
(Parallel arcs) Let $`P_1,P_2E`$ be n-1-dimensional affine planes. Let $`\tau ^2`$ be a 2-dimensional subsimplex of $`\mathrm{\Delta }^n`$ with vertices $`v_r,v_s,v_t`$. We say that disjoint arcs $`e_1,e_2`$ obtained as intersections of $`P_1`$ resp. $`P_2`$ with (the same) $`\tau ^2`$ are parallel arcs if one of the following holds:
\- both are nondegenerate and any two of $`\{v_r,v_s,v_t\}`$ belong to the same connected component of $`\tau ^2e_1`$ if and only if they belong to the same connected component of $`\tau ^2e_2`$,
\- one, say $`e_1`$ is nondegenerate, the other, say with vertices $`v_s,v_t`$ is contained in a face, and $`v_r`$ belongs to another connected component of $`\tau ^2e_1`$ as both $`v_s`$ and $`v_t`$,
\- one, say $`e_1`$, is nondegenerate, the other is degenerate, say equal to $`v_r`$, and both $`v_s,v_t`$ belong to another connected component of $`\tau ^2e_1`$ as $`v_r`$,
\- both are degenerate and equal,
\- both are contained in a face and equal,
\- one is degenerate, the other is contained in a face.
###### Lemma 8.
Let $`\mathrm{\Delta }^n๐^{n+1}`$ be the standard simplex. Let $`P_1,P_2E`$ be n-1-dimensional affine planes with $`Q_i=P_i\mathrm{\Delta }^n\mathrm{}`$ for $`i=1,2`$.
Let $`P_1`$ be of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`1kn2`$ and $`P_2`$ of type $`\left\{0b_1\mathrm{}b_l\right\}`$ with $`l`$ arbitrary.
Then either $`Q_1Q_2\mathrm{}`$, or $`Q_1`$ and $`Q_2`$ have a parallel arc.
###### Proof.
Assume that $`Q_1Q_2=\mathrm{}`$.
By Observation 2, there are 4 possible cases if $`Q_1Q_2=\mathrm{}`$.
Case 1: $`\left\{0a_1\mathrm{}a_k\right\}=\left\{0b_1\mathrm{}b_l\right\}`$. Then we clearly have parallel arcs.
Case 2: $`\left\{0a_1\mathrm{}a_k\right\}`$ is a proper subset of $`\left\{0b_1\mathrm{}b_l\right\}`$, i.e. $`1k<ln1`$ and $`a_1=b_1,\mathrm{},a_k=b_k`$. There is at least one index, say $`i`$, not contained in $`\left\{0b_1\mathrm{}b_l\right\}`$. Consider the 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ with vertices $`v_0,v_{a_1},v_i`$. It intersects $`P_1`$ and $`P_2`$ in parallel arcs, because $`P_1`$ and $`P_2`$ both separate $`v_0`$ and $`v_{a_k}`$ from $`v_i`$.
Case 3: $`\left\{0b_1\mathrm{}b_l\right\}`$ is a proper subset of $`\left\{0a_1\mathrm{}a_k\right\}`$, i.e. $`0l<kn2`$ and $`a_1=b_1,\mathrm{},a_l=b_l`$. There are two indices $`i,j`$ not contained in $`\left\{0a_1\mathrm{}a_k\right\}`$. Consider the 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ with vertices $`v_0,v_i,v_j`$. It intersects $`P_1`$ and $`P_2`$ in parallel arcs, because $`P_1`$ and $`P_2`$ both separate $`v_0`$ from $`v_i`$ and $`v_j`$.
Case 4: $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}=\{1,\mathrm{},n\}`$. By $`kn2`$, there are two indices $`i,j`$ with $`i,j\left\{0a_1\mathrm{}a_k\right\}`$. Hence $`i,j\{b_1,\mathrm{},b_l\}`$. Moreover, there exists an index $`h`$ such that $`h\{a_1,\mathrm{},a_k\}`$ but $`h\{b_1,\mathrm{},b_l\}`$. (If not, we would have $`\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}`$, hence $`\{1,\mathrm{},n\}=\{a_1,\mathrm{},a_k\}\{b_1,\mathrm{},b_l\}\{b_1,\mathrm{},b_l\}`$, contradicting $`Q_2\mathrm{}`$.) Consider the 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ with vertices $`v_i,v_j,v_h`$. It intersects $`P_1`$ and $`P_2`$ in parallel arcs, because both $`P_1`$ and $`P_2`$ separate $`v_i`$ and $`v_j`$ from $`v_h`$..
###### Definition 4.
(Canonical colouring of complementary regions)
Let $`P_1,P_2,\mathrm{}E`$ be a (possibly infinite) set of n-1-dimensional affine planes with $`Q_i:=P_i\mathrm{\Delta }^n\mathrm{}`$ and $`Q_iQ_j=\mathrm{}`$ for all $`ij`$. Assume that each $`Q_i`$ either contains no vertices or consists of exactly one vertex.
A colouring of
\- the connected components of $`\mathrm{\Delta }^n_iQ_i`$ by colours black and white, and
\- of all $`Q_i`$ by black,
is called a canonical colouring (associated to $`P_1,P_2,\mathrm{}`$) if:
\- all vertices of $`\mathrm{\Delta }^n`$ are coloured black,
\- each $`Q_i`$ is incident to at least one white component.
###### Definition 5.
(White-parallel arcs) Let $`\{P_i:iI\}`$ be a set of of $`n1`$-dimensional affine planes $`P_iE`$, with $`Q_i:=P_i\mathrm{\Delta }^n\mathrm{}`$ for $`iI`$. Assume that $`Q_iQ_j=\mathrm{}`$ for all $`ijI`$, and that we have a canonical colouring associated to $`\{P_i:iI\}`$. We say that arcs $`e_i,e_j`$ obtained as intersections of $`P_i,P_j`$ ($`i,jI`$) with some 2-dimensional subsimplex $`\tau ^2`$ of $`\mathrm{\Delta }^n`$ are white-parallel arcs if they are parallel arcs and, moreover, belong to the boundary of the closure of the same white component.
We mention two consequences of Lemma 8. These will not be needed for the proof of Lemma 10, but they will be necessary for the proof of Theorem 1.
###### Corollary 1.
Let $`\mathrm{\Delta }^n๐^{n+1}`$ be the standard simplex. Let $`P_1,\mathrm{},P_mE`$ be a finite set of n-1-dimensional affine planes and let $`Q_i=P_i\mathrm{\Delta }^n`$ for $`i=1,\mathrm{},m`$.
Assume that $`Q_iQ_j=\mathrm{}`$ for all $`ij`$, and that we have an associated canonical colouring, such that $`Q_i`$ and $`Q_j`$ do not have a white-parallel arc for $`ij`$.
Then either $`m=0`$, or
$`m=n+1`$ and $`P_1`$ is of type $`\left\{0\right\}`$, $`P_{n+1}`$ is of type $`\left\{\mathrm{0\; 1}\mathrm{}n1\right\}`$, and $`P_i`$ is of type $`\left\{01\mathrm{}\widehat{i1}\mathrm{}n\right\}`$ for $`i=2,\mathrm{},n`$.
###### Proof.
If the conclusion were not true, there would exist a plane $`P_1`$ of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`1kn2`$. Let $`W`$ be the white component of the canonical colouring, which is incident to $`P_1`$. Because, for a canonical colouring, no vertex belongs to a white component, there must be at least one more plane $`P_2`$ incident to $`W`$. Since $`Q_1Q_2=\mathrm{}`$, from Lemma 8 we get that $`Q_1`$ and $`Q_2`$ have a parallel arc. Because $`Q_1`$ and $`Q_2`$ are incident to $`W`$, the arc is white-parallel.โ
###### Corollary 2.
Let $`\mathrm{\Delta }^n๐^{n+1}`$ be the standard simplex. Let $`P_1,P_2,\mathrm{}E`$ be a (possibly infinite) set of n-1-dimensional affine planes and let $`Q_i=P_i\mathrm{\Delta }^n`$ for $`i=1,2,\mathrm{}`$. Assume that we have an associated canonical colouring.
Let $`P_i`$ be of type $`\left\{0a_1^i\mathrm{}a_{c\left(i\right)}^i\right\}`$, for $`i=1,2,\mathrm{}`$. Then
\- either $`c\left(1\right)\{0,n1\}`$,
\- or whenever, for some $`i\{2,3,\mathrm{}\}`$, $`P_1`$ and $`P_i`$ bound a white component of $`\mathrm{\Delta }^n_jQ_j`$, then they must have a white-parallel arc.
###### Proof.
Assume that $`c\left(1\right)\{0,n1\}`$. Let $`W`$ be the white component bounded by $`P_1`$. $`W`$ is bounded by a finite number of planes, thus we can apply Corollary 1, and conclude that $`P_1`$ has a white-parallel arc with each other plane adjacent to $`W`$. โ
###### Definition 6.
Let $`PE`$ be an n-1-dimensional affine plane, and $`T`$ a triangulation of the polytope $`Q:=P\mathrm{\Delta }^n`$. We say that $`T`$ is minimal, if all vertices of $`T`$ are vertices of $`Q`$. We say that an edge of some simplex in $`T`$ is an exterior edge if it is an edge of $`Q`$.
###### Observation 3.
Let $`PE`$ be an n-1-dimensional affine plane, and $`T`$ a triangulation of the polytope $`Q:=P\mathrm{\Delta }^n`$. If $`T`$ is minimal, then each edge of $`Q`$ is an (exterior) edge of (exactly one) simplex in $`T`$.
###### Proof.
By minimality, the triangulation does not introduce new vertices. Thus every edge of $`Q`$ is an edge of some simplex.โ
###### Observation 4.
Let $`PE`$ be an n-1-dimensional affine plane with $`Q:=P\mathrm{\Delta }^n\mathrm{}`$. Assume that $`P`$ is of type $`\left\{0a_1\mathrm{}a_k\right\}`$.
a) Each vertex of $`Q`$ arises as the intersection of $`P`$ with an edge $`e`$ of $`\mathrm{\Delta }^n`$. The vertices of $`e`$ are $`v_i`$ and $`v_j`$ with $`i\{0,a_1,\mathrm{},a_k\}`$ and $`j\{0,a_1,\mathrm{},a_k\}`$. (We will denote such a vertex by $`\left(v_iv_j\right)`$.)
b) Two vertices $`\left(v_{i_1}v_{j_1}\right)`$ and $`\left(v_{i_2}v_{j_2}\right)`$ of $`Q`$ are connected by an edge of $`Q`$ (i.e. an exterior edge of any triangulation) if either $`i_1=i_2`$ or $`j_1=j_2`$.
###### Proof.
a) holds because $`e`$ has to connect vertices in distinct components of $`\mathrm{\Delta }^nQ`$. b) holds because the edge of $`Q`$ has to belong to some 2-dimensional subsimplex of $`\mathrm{\Delta }^n`$, with vertices either $`v_{i_1},v_{j_1},v_{j_2}`$ or $`v_{i_1},v_{i_2},v_{j_1}`$.โ
Remark: if, for an affine hyperplane $`PE`$, $`Q=P\mathrm{\Delta }^n`$ consists of exactly one vertex, then we will consider the minimal triangulation of $`Q`$ to consist of one (degenerate) n-1-simplex. This convention helps to avoid needless case distinctions.
###### Lemma 9.
Let $`\{P_iE:iI\}`$ be a set of n-1-dimensional affine planes and let $`Q_i:=P_i\mathrm{\Delta }^n`$ for $`iI`$. Assume that $`Q_iQ_j=\mathrm{}`$ for all $`ij`$ and that we have an associated canonical colouring. Assume that we have fixed, for each $`iI`$, a minimal triangulation $`Q_i=_a\tau _{ia}`$ of $`Q_i`$.
If $`P_1`$ is of type $`\left\{0a_1^1\mathrm{}a_{c\left(1\right)}^1\right\}`$ with $`1c\left(1\right)n2`$, then for each simplex $`\tau _{1a}Q_1`$ there exists some $`jI`$ and some simplex $`\tau _{jb}Q_j`$ (of the fixed triangulation of $`Q_j`$) such that $`\tau _{ia}`$ and $`\tau _{jb}`$ have a white-parallel arc.
###### Proof.
Let $`w_1,\mathrm{},w_n`$ be the $`n`$ vertices of the n-1-simplex $`\tau _{1k}`$. By Observation 4a), each $`w_l`$ arises as intersection of $`Q_1`$ with some edge $`\left(v_{r_l}v_{s_l}\right)`$ of $`\mathrm{\Delta }^n`$, and the vertices $`v_{r_l},v_{s_l}`$ satisfy $`r_l\{0,a_1^1,\mathrm{},a_{c\left(1\right)}^1\}`$ and $`s_l\{0,a_1^1,\mathrm{},a_{c\left(1\right)}^1\}`$.
For the canonical colouring, there must be a white component $`W`$ bounded by $`P_1`$. We distinguish the cases whether $`W`$ and $`v_0`$ belong to the same connected component of $`\mathrm{\Delta }^nQ_1`$ or not.
Case 1: $`W`$ and $`v_0`$ belong to the same connected component of $`\mathrm{\Delta }^nQ_1`$.
Since $`c\left(1\right)n2`$, there exist at most $`n1`$ possible values for $`r_l`$. Hence there exists $`lm\{1,\mathrm{},n\}`$ such that $`v_{r_l}=v_{r_m}`$.
Let $`e`$ be the edge of $`\tau _{1k}Q_1`$ connecting $`w_l`$ and $`w_m`$. By Observation 4b), $`e`$ is an exterior edge. Consider the 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ with vertices $`v_{r_l},v_{s_l},v_{s_m}`$. We have that $`P_1`$ intersects $`\tau ^2`$ in $`e`$, i.e. in an arc separating $`v_{r_l}`$ from the other two vertices of $`\tau ^2`$.
Note that $`r_l\{0,a_1^1,\mathrm{},a_{c\left(1\right)}^1\}`$, hence $`v_{r_l}`$ belongs to the same component of $`\mathrm{\Delta }^nQ_1`$ as $`v_0`$. In particular, $`v_{r_l}`$ belongs to the same component of $`\mathrm{\Delta }^nQ_1`$ as $`W`$. On the other hand, since the colouring is canonical, all vertices are coloured black and $`v_{r_l}`$ can not belong to the white component $`W`$. Thus there must be some plane $`P_j`$ such that $`Q_j`$ bounds $`W`$ and separates $`v_{r_l}`$ from $`Q_1`$. (The possiblity $`P_j\mathrm{\Delta }^n=\left\{v_{r_l}\right\}`$ is allowed.) In particular, some (possibly degenerate) exterior edge $`f`$ of $`Q_j`$ separates $`v_{r_l}`$ from $`v_{s_l},v_{s_m}`$. Thus $`e`$ and $`f`$ are white-parallel arcs. By Observation 3, $`f`$ is an edge of some $`\tau _{jl}`$.
Case 2: $`W`$ and $`v_0`$ donโt belong to the same connected component of $`\mathrm{\Delta }^nQ_1`$.
Since $`nc\left(1\right)n1`$, there exist some $`lm\{1,\mathrm{},n\}`$ such that $`v_{s_l}=v_{s_m}`$.
Let $`e`$ be the edge of $`\tau _{1k}Q_1`$ connecting $`w_l`$ and $`w_m`$. $`e`$ is an exterior edge by Observation 4b). Consider the 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ with vertices $`v_{r_l},v_{r_m},v_{s_l}`$. $`P_1`$ intersects $`\tau ^2`$ in $`e`$, i.e. in an arc separating $`v_{s_l}`$ from the other two vertices of $`\tau ^2`$.
We have that $`s_l\{0,a_1^1,\mathrm{},a_{c\left(1\right)}^1\}`$, hence $`v_{s_l}`$ does not belong to the same component of $`\mathrm{\Delta }^nQ_1`$ as $`v_0`$. This implies that $`v_{s_l}`$ belongs to the same component of $`\mathrm{\Delta }^nQ_1`$ as $`W`$. On the other hand, since the colouring is canonical, $`v_{s_l}`$ can not belong to the white component $`W`$ and there must be some plane $`P_j`$ such that $`Q_j`$ bounds $`W`$ and separates $`v_{s_l}`$ from $`Q_1`$. In particular, some exterior edge $`f`$ of $`Q_j`$ separates $`v_{s_l}`$ from $`v_{r_l},v_{r_m}`$. Thus $`e`$ and $`f`$ are white-parallel arcs. By Observation 3, $`f`$ is an edge of some $`\tau _{jl}`$.โ
###### Lemma 10.
Let $`\{P_i:iI\}`$ be a set of n-1-dimensional affine planes with $`Q_i:=P_i\mathrm{\Delta }^n\mathrm{}`$ for $`iI`$. Let $`P_i`$ be of type $`\left\{0a_1^{\left(i\right)}\mathrm{}a_{k_i}^{\left(i\right)}\right\}`$ for $`iI`$. Assume that $`Q_iQ_j=\mathrm{}`$ for $`ijI`$, and that we have an associated canonical colouring. Assume that for each $`Q_i`$ one has fixed a minimal triangulation $`Q_i=_{k=1}^{t\left(i\right)}\tau _{ik}`$.
For each $`iI`$, let
$$D_i=\mathrm{}\{\tau _{ik}Q_i:\text{ there is no }\tau _{jl}Q_j\text{ such that }\tau _{ik},\tau _{jl}\text{ have a white-parallel arc}\}.$$
Then
$$\underset{iI}{}D_i=0or\underset{iI}{}D_i=n+1.$$
###### Proof.
First we remark that the number of planes may be infinite, but we may of course remove pairs of planes $`P_i,P_j`$ whenever they are of the same type and bound the same white component. This removal (of $`P_i,P_j`$ and the common white component) does not affect $`_{iI}D_i`$. Since there are only finitely many different types of planes, we may w.l.o.g. assume that we start with a finite number $`P_1,\mathrm{},P_m`$ of planes. (It may happen that after this removal no planes and no white components remain. In this case $`_{iI}D_{iI}=0`$.) So we assume now that we have a finite number of planes $`P_1,\mathrm{},P_m`$, and no two planes of the same type bound a white region.
The first case to consider is that all planes are of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`k=0`$ or $`k=n1`$. Since all vertices are coloured black, this means that $`m=n+1`$ and (upon renumbering) $`P_1`$ is of type $`\left\{0\right\}`$, $`P_{n+1}`$ is of type $`\left\{\mathrm{0\; 1}\mathrm{}n1\right\}`$, and $`P_i`$ is of type $`\left\{01\mathrm{}\widehat{i1}\mathrm{}n\right\}`$ for $`i=2,\mathrm{},n`$. Hence $`D_1=\mathrm{}=D_{n+1}=1`$ and $`_{i=1}^{n+1}D_i=n+1`$.
Now we assume that there exists $`P_i`$, w.l.o.g. $`P_1`$, of type $`\left\{0a_1^{\left(1\right)}\mathrm{}a_{k_1}^{\left(1\right)}\right\}`$ with $`1c\left(1\right)n2`$. Let $`W`$ be the white component bounded by $`P_1`$ and let w.l.o.g. $`P_2,\mathrm{},P_l`$ be the other planes bounding $`W`$. Then Lemma 9 says that each simplex in the chosen triangulation of $`Q_1`$ has a parallel arc with some simplex in the chosen triangulation of each of $`Q_2,\mathrm{},Q_l`$. In particular, $`D_1=0`$. For $`j\{2,\mathrm{},l\}`$, if $`1c\left(j\right)n2`$, the same argument shows that $`D_j=0`$. If $`j\{2,\mathrm{},l\}`$ and $`c\left(j\right)=0`$ or $`c\left(j\right)=n1`$, then $`Q_j`$ consists of only one simplex. By Corollary 2, this simplex has a parallel arc with (some exterior edge of) $`Q_1`$ and thus (by Observation 3) with (some) simplex of the chosen triangulation of $`Q_1`$. This shows $`D_j=0`$ also in this case. Altogether we conclude $`_{j=1}^lD_j=0`$ and thus $`_{i=1}^mD_i=_{i=l+1}^mD_i`$. Hence we can remove<sup>6</sup><sup>6</sup>6To remove a white component means that this component together with the neighbouring black components will form one new black component. the white component $`W`$ and its bounding planes $`P_1,\mathrm{},P_l`$ to obtain a smaller number of planes and a new canonical colouring without changing $`_{i=1}^mD_i`$. Since we start with finitely many planes, we can repeat this reduction finitely many times and will end up either with an empty set of planes or with a set of planes of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`k=0`$ or $`k=n1`$. Thus either $`_{i=1}^mD_i=0`$ or $`_{i=1}^mD_i=n+1`$. โ
We have thus proved that, in presence of a canonical colouring, the number of n-1-simplices without white-parallel arcs in a minimal triangulation of the $`Q_i`$โs is $`0`$ or $`n+1`$. We remark that in the proof of Theorem 1 we will actually count only those triangles which neither have a white-parallel arc nor a degenerate arc. Thus, in general, we may remain with even less than $`n+1`$ n-1-simplices.
## 5 A straightening procedure
In this section we will always work with the following set of assumptions.
Assumption I: $`Q`$ is an aspherical n-dimensional manifold with aspherical boundary $`Q`$. We have n-1-dimensional submanifolds $`_0Q,_1QQ`$ such that $`Q=_0Q_1Q,_0Q=_1Q`$ and $`_1Q\mathrm{}`$ is aspherical.
The example that one should have in mind is a nonpositively curved manifold $`Q`$ with totally geodesic boundary $`_1Q`$ and cusps corresponding to $`_0Q`$.
In the case of nonpositively curved manifolds with totally geodesic boundary, there is a well-known straightening procedure (explained for closed hyperbolic manifolds in , Lemma C.4.3.), which homotopes each relative cycle into a straight relative cycle.
However, we will need a more subtle straightening procedure, which considers relative cycles with a certain 0-1-labeling of their edges and straightens the 1-labeled edges into certain distinguished 1-simplices. This straightening procedure will be explained in Section 5.3. Before, we explain a construction which will morally (although not literally) โreduceโ the proof of Theorem 1 to the case that $`_0QC`$ is path-connected, for each path-component $`C`$ of $`Q`$.
### 5.1 Making $`_0QC`$ connected
###### Construction 1.
Let Assumption I be satisfied. Then there exists a continuous map of triples $`q:(Q,Q,_1Q)(Q,Q,_1Q)`$ which is (as a map of triples) homotopic to the identity and such that, for each path-component $`C`$ of $`Q`$, the image $`A:=q\left(_0QC\right)`$ is path-connected.
Moreover, for each path-component $`F`$ of $`_1Q`$, the path-components of $`F_0Q_1Q`$ can be numbered by $`E_0^F,\mathrm{},E_s^F`$ and one can choose points $`x_{E_i^F}E_i^F`$ such that $`q\left(x_{E_i^F}\right)x_{E_0^F}`$ for $`i=0,\mathrm{},s`$.
###### Proof.
For each path-component $`F`$ of $`_1Q`$, number the path-components of $`F_0Q_1Q`$ by $`E_0^F,\mathrm{},E_s^F`$, where $`s`$ depends on $`F`$. Choose one point $`x_E^FE`$ for each path-component $`EF`$ of $`_0Q_1Q`$. Whenever $`E_0,E_i`$ is a pair of path-components of $`_0Q_1Q`$ adjacent to the same path-component $`F`$ of $`_1Q`$, choose a 1-dimensional submanifold $`l_{E_0^FE_i^F}_1Q`$ with $`l_{E_0^FE_i^F}=\left\{x_{E_0^F}\right\}\left\{x_{E_i^F}\right\}`$. The $`l_{E_0^FE_i^F}`$ may be chosen succesively such that they are disjoint from each other (apart from the common vertex $`x_{E_0^F}`$) and disjoint from $`_0Q`$ (apart from the vertices $`x_{E_0^F}`$ and $`x_{E_i^F}`$).
For each pair $`\{E_0^F,E_i^F\}`$ let $`h:l_{E_0^FE_i^F}\left\{x_{E_0^F}\right\}`$ be the constant map from $`l_{E_0^FE_i^F}`$ to $`x_{E_0^F}`$. For each path-component $`F`$ of $`_1Q`$, the union
$$\underset{i=1}{\overset{s}{}}l_{E_0^FE_i^F}$$
is an embedded wedge of arcs in $`_1Q`$, hence it is contractible. In particular, $`h`$ is homotopic to the identity. By the homotopy extension property exists $`g:FF`$ with $`g_{l_{E_0^FE_i^F}}=hx_{E_0}`$ for all $`l_{E_0^FE_i^F}`$, and $`gid`$ by a homotopy extending the homotopies between $`h`$ and $`id`$.
Thus we defined $`g`$ on each path-component $`F`$ of $`_1Q`$ with $`F_0Q\mathrm{}`$. On path-components $`F`$ of $`_1Q`$ with $`F_0Q=\mathrm{}`$ we define $`g=id`$. Hence we have defined $`g`$ on all of $`_1Q`$.
On path-components $`C`$ of $`_0Q`$ with $`C_1Q=\mathrm{}`$, we define $`f=id`$. Again by the homotopy extension property exists $`f:QQ`$ with $`f_{_1Q}=g`$, $`f_C=id`$ for path-components $`C`$ of $`_0Q`$ with $`C_1Q=\mathrm{}`$, and $`fid`$ by a homotopy extending the homotopy of $`g`$. (Of course, $`f`$ does not preserve those path-components of $`_0Q`$ which intersect $`_1Q`$.)
Once again by the homotopy extension property exists $`q:QQ`$ with $`qid`$ such that $`q`$ extends $`f`$ and the homotopy between $`q`$ and $`id`$ extends the homotopy between $`f`$ and $`id`$.
Due to the stepwise construction, $`q`$ is a map of triples, homotopic to the identity by a homotopy of triples. Moreover, $`A:=q\left(_0QC\right)`$ is path-connected for each component $`C`$ of $`Q`$. Indeed, any two points in $`_0QC`$ can be connected by a sequence of paths which either have image in $`_0Q`$ or belong to $`_{i=1}^sl_{E_0^FE_i^F}`$ for some path-component $`F`$ of $`_1QC`$. The image of these paths under $`q`$, in both cases, is in $`A`$. โ
Remark: $`q`$ induces a simplicial map $`q:K\left(Q\right)K\left(Q\right)`$ and a homomorphism $`q_{}:\mathrm{\Pi }\left(K\left(_0Q\right)\right)\mathrm{\Pi }\left(K\left(A\right)\right)`$ defined by $`q_{}\left(\{\gamma _1,\mathrm{},\gamma _n\}\right):=\{q\left(\gamma _1\right),\mathrm{},q\left(\gamma _n\right)\}`$ such that
$$q_{}\left(g\right)q\left(\sigma \right)=q\left(g\sigma \right)$$
holds for each $`\sigma K\left(Q\right),g\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$.
###### Proof.
Continuous maps $`q:QQ`$ induce simplicial maps $`q:K\left(Q\right)K\left(Q\right)`$. (The simplicial map agrees with $`q`$ on the 0-skeleton, and it maps each 1-simplex $`eK_1\left(Q\right)`$ to the unique 1-simplex of $`K_1\left(Q\right)`$ that is in the homotopy class rel. $`\{0,1\}`$ of $`q\left(e\right)`$.)
Let $`eK_1\left(Q\right)`$. By construction $`\{\gamma _1,\mathrm{},\gamma _n\}e=\left[\alpha e\overline{\beta }\right]`$ for some $`\alpha ,\beta \{\gamma _1,\mathrm{},\gamma _n\}\{c_{e\left(0\right)},c_{e\left(1\right)}\}`$. Thus
$$\{q\left(\gamma _1\right),\mathrm{},q\left(\gamma _n\right)\}q\left(e\right)=[q\left(\alpha \right)q\left(e\right)q\left(\overline{\beta }\right)\}=q\left(\{\gamma _1,\mathrm{},\gamma _n\}e\right).$$
This implies the claim for the 1-skeleton and thus, by asphericity of $`K\left(Q\right)`$, for all $`\sigma K\left(Q\right)`$.โ
### 5.2 Definition of $`K^{str}\left(Q\right)`$
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I.
Recall that we have defined in Section 3.2 an aspherical multicomplex $`K\left(Q\right)S_{}\left(Q\right)`$ with the property that (for aspherical $`Q`$) each singular simplex in $`Q`$, with boundary in $`K\left(Q\right)`$ and pairwise distinct vertices, is homotopic rel. boundary to a unique simplex in $`K\left(Q\right)`$.
The aim of this subsection is to describe a selection procedure yielding a subset $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$. The final purpose of the straightening procedure will be to produce a large number of (weakly) degenerate simplices, in the sense of the following definition.
###### Definition 7.
Let $`Q`$ be an compact manifold with boundary $`Q`$. We say that a simplex in $`S_{}\left(Q\right)`$ is degenerate if one of its edges is a constant loop. We say that it is weakly degenerate if it is degenerate or its image is contained in $`Q`$.
Notational remark: for subsets $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ we will denote $`K_{}^{str}\left(_0Q\right):=K_{}^{str}\left(Q\right)S_{}\left(_0Q\right),K_{}^{str}\left(_1Q\right):=K_{}^{str}\left(Q\right)S_{}\left(_1Q\right),K_{}^{str}\left(_0QQ\right):=K_{}^{str}\left(Q\right)S_{}\left(_0Q\right).`$
###### Lemma 11.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I. Let $`K\left(Q\right)S_{}\left(Q\right)`$ be as defined in Section 3.2. Let $`q:QQ`$ and $`\{x_{E_i^F}_0Q_1Q:0is\}`$ be given by Construction 1.
Then there exists a subset $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$, closed under face maps, such that:
i) If $`C`$ is a path-component of $`_0Q`$ with $`C_1Q=\mathrm{}`$, then $`K_0^{str}\left(Q\right)`$ contains each point in $`C`$,
ii) for a path-component $`F`$ of $`_1Q`$ with $`F_0Q=\mathrm{}`$, there is exactly one point $`x_FK_0^{str}\left(Q\right)F`$,
for a path-component $`F`$ of $`_1Q`$ with $`F_0Q\mathrm{}`$, we have $`K_0^{str}\left(Q\right)F=\{x_{E_0^F},\mathrm{},x_{E_s^F}\}`$,
iii) $`K_0^{str}\left(Q\right)=K_0^{str}\left(Q\right)`$,
iv) $`K_1^{str}\left(Q\right)`$ consists of
\- all 1-simplices $`eK\left(Q\right)`$ with $`eK_0^{str}\left(Q\right)`$, and
\- exactly one 1-simplex for each nontrivial homotopy class (rel. boundary) of loops $`e`$ with $`_0e=_1eK_0^{str}\left(Q\right)`$,
\- the constant loop for the homotopy class of the constant loop at $`x`$, if $`xK_0^{str}\left(Q\right)`$,
v) for $`n2`$, if $`\sigma S_n\left(Q\right)`$ is an n-simplex with $`\sigma K_{n1}^{str}\left(Q\right)`$, then $`\sigma `$ is homotopic rel. boundary to a unique $`\tau K_n^{str}\left(Q\right)`$,
vi) if $`\sigma K_n^{str}\left(Q\right)`$ is homotopic rel. boundary to some $`\tau K_n\left(Q\right)`$, then $`\sigma =\tau `$,
vii) if $`\sigma K_n^{str}\left(Q\right)`$ is homotopic rel. boundary to a simplex $`\tau S_n\left(_1Q\right)`$, then $`\sigma K_n^{str}\left(_1Q\right)`$; if $`\sigma K_1^{str}\left(Q\right)`$ is homotopic rel. boundary to a simplex $`\tau S_1\left(_0Q\right)`$, then $`\sigma K_1^{str}\left(_0Q\right)`$,
viii) $`K_{}^{str}\left(Q\right)`$ is aspherical, i.e. if $`\sigma ,\tau K_{}^{str}\left(Q\right)`$ have the same 1-skeleton, then $`\sigma =\tau `$.
###### Proof.
$`K_{}^{str}\left(Q\right)`$ is defined by induction on the dimension of simplices as follows.
Definition of $`K_0^{str}\left(Q\right)`$:
Choose $`K_0^{str}\left(Q\right)`$ such that conditions i),ii),iii) are satisfied. Note that we have chosen a nonempty set of 0-simplices since we are assuming $`_1Q\mathrm{}`$.
Definition of $`K_1^{str}\left(Q\right)`$:
For an ordered pair
$$(x,y)K_0^{str}\left(Q\right)\times K_0^{str}\left(Q\right)$$
with $`xy`$, there exists in each homotopy class (rel. boundary) of arcs $`e`$ with
$$e\left(0\right)=x,e\left(1\right)=y$$
a unique simplex in $`K_1\left(Q\right)`$. Choose these 1-simplices to belong to $`K_1^{str}\left(Q\right)`$. (Uniqueness implies that vi) ) is true for $`n=1`$.) Moreover, for pairs
$$(x,x)K_0^{str}\left(Q\right)\times K_0^{str}\left(Q\right)$$
choose one simplex in each homotopy class (rel. boundary) of loops $`e`$ with
$$e\left(0\right)=e\left(1\right)=x.$$
For the homotopy class of the constant loop choose the constant loop.
Choose the 1-simplices in $`_0Q`$ and/or $`_1Q`$ whenever this is possible. (If a 1-simplex is homotopic into both $`_0Q`$ and $`_1Q`$, then it is necessarily homotopic into $`_0Q_1Q`$. Indeed, a disk realizing a homotopy between 1-simplices in $`_0Q`$ and $`_1Q`$ can be made transversal to $`_0Q_1Q`$ and then intersects $`_0Q_1Q`$ in an arc resp. loop.) Hence vii) is satisfied for $`n=1`$.
Definition of $`K_n^{str}\left(Q\right)`$ for $`n2`$, assuming that $`K_{n1}^{str}\left(Q\right)`$ is defined:
For an $`n+1`$-tuple $`\kappa _0,\mathrm{},\kappa _n`$ of $`n1`$-simplices in $`K_{n1}^{str}\left(Q\right)`$, satisfying
$$_i\kappa _j=_{j1}\kappa _i$$
for all $`i,j`$, there are two possibilities:
\- if no edge of any $`\kappa _i`$ is a loop, then, by asphericity of $`Q`$, there is a unique n-simplex
$$\sigma K_n\left(Q\right)$$
with
$$_i\sigma =\kappa _i$$
for $`i=0,\mathrm{},n`$. In this case set $`\kappa :=\sigma `$. Uniqueness implies that vi) is satisfied for $`n`$. (By the construction in Section 3.2 $`\kappa K_n\left(_1Q\right)`$ if $`\kappa `$ is homotopic rel. boundary into $`_1Q`$.)
\- otherwise, choose an $`n`$-simplex
$$\kappa S_n\left(Q\right)$$
with
$$_i\kappa =\kappa _i$$
for $`i=0,\mathrm{},n`$. By asphericity of $`Q`$, $`\kappa `$ exists and is unique up to homotopy rel. boundary. Choose the simplices in $`_1Q`$ whenever this is possible.
By construction, $`K_{}^{str}\left(Q\right)`$ is closed under face maps and satisfies the conditions i)-vii). Condition viii) follows by induction on the dimension of subsimplices of $`\sigma `$ and $`\tau `$ from condition v). โ
The simplices in $`K_{}^{str}\left(Q\right)`$ will be called the straight simplices.
We remark that $`K_{}^{str}\left(Q\right)`$ is not a multicomplex because simplices in $`K_{}^{str}\left(Q\right)`$ need not have pairwise distinct vertices. (Note also that simplices in $`K\left(Q\right)`$ belong to $`K^{str}\left(Q\right)`$ if and only if all their vertices belong to $`K_0^{str}\left(Q\right)`$, by construction.)
###### Observation 5.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I. Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy the conditions i)-viii) from Lemma 11. Then
$`q:QQ`$ induces a simplicial map $`q:K^{str}\left(Q\right)K^{str}\left(Q\right)`$, compatible with the simplicial map $`q:K\left(Q\right)K\left(Q\right)`$ from Section 5.1.
###### Proof.
By construction, $`q`$ maps $`K_0^{str}\left(Q\right)`$ to itself. Indeed:
\- if $`C`$ is a path-component of $`_0Q`$ with $`C_1Q=\mathrm{}`$, then $`q\left(v\right)=v`$ for each $`vC`$,
\- if $`F`$ is a path-component $`F`$ of $`_1Q`$ with $`F_0Q=\mathrm{}`$, then $`q\left(v\right)=v`$ for each $`vF`$ (in particular for the unique $`vFK_0^{str}\left(Q\right)`$),
\- if $`F`$ is a path-component of $`_1Q`$ with $`F_0Q\mathrm{}`$, then we have $`K_0^{str}\left(Q\right)F=\{x_{E_0^F},\mathrm{},x_{E_s^F}\}`$, and $`q\left(x_{E_i^F}\right)=x_{E_0^F}`$ for $`i=0,\mathrm{},s`$ by Construction 1.
Hence $`q`$ induces a simplicial map on $`K^{str}\left(Q\right)`$. (The simplicial map agrees with $`q`$ on the 0-skeleton, and it maps each 1-simplex $`eK_1^{str}\left(Q\right)`$ to the unique 1-simplex of $`K_1^{str}\left(Q\right)`$ that is in the homotopy class rel. $`\{0,1\}`$ of $`q\left(e\right)`$. Since $`K^{str}\left(Q\right)`$ is aspherical, this determines the simplicial map $`q`$ uniquely.) โ
### 5.3 Definition of the straightening
###### Definition 8.
Let $`(Q,_1Q)`$ be a pair of topological spaces and let $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$ a (possibly infinite) singular chain.
a) A set of cancellations of $`z`$ is a symmetric set $`๐S_{n1}\left(Q\right)\times S_{n1}\left(Q\right)`$ with $`(\eta _1,\eta _2)๐\eta _1=\eta _2`$ and $`\eta _1=_k\tau _{i_1},\eta _2=_l\tau _{i_2}`$ for some $`i_1,i_2I,k,l\{0,\mathrm{},n\}`$.
b) If $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$ and $`๐`$ is a set of cancellations for $`z`$, then the associated simplicial set $`\mathrm{{\rm Y}}_{z,๐}`$ is the simplicial set generated<sup>7</sup><sup>7</sup>7That is, the subset of $`S_{}^{sing}\left(Q\right)`$ which contains the $`I`$ n-simplices $`\mathrm{\Delta }_i,iI`$, together with all simplices obtained by iterated applications of face and degeneracy operators, cf. , Example 1.5. by $`\{\mathrm{\Delta }_i:iI\}`$, subject to the identifications $`_k\mathrm{\Delta }_{i_1}=_l\mathrm{\Delta }_{i_2}`$ if and only if $`(_k\tau _{i_1},_l\tau _{i_2})๐`$.
c) Let $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$. Choose a minimal presentation for $`z`$ (i.e. no further cancellation is possible). Let
$$J=J_z:=\left\{\begin{array}{c}(i,a)I\times \{0,\mathrm{},n\}:\\ _a\tau _i\text{ occurs with non-zero coefficient in the chosen presentation of }z\end{array}\right\}.$$
Let $`๐`$ be a set of cancellations for $`z`$. Then the simplicial set $`\mathrm{{\rm Y}}_{z,๐}`$ is defined as the set consisting of $`J`$ n-1-simplices $`\mathrm{\Delta }_{i,a},(i,a)J`$, together with all their iterated faces and degenerations, subject to the identifications $`_a_{a_1}\tau _{i_1}=_a_{a_2}\tau _{i_2}`$ for all $`a=0,\mathrm{},n1`$, whenever $`(_{a_1}\tau _{i_1},_{a_2}\tau _{i_2})๐`$ and $`(i_1,a_1)J`$ .
d) If $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$ is a relative cycle, then a set of cancellations $`๐`$ is called sufficient if the formal sum $`_{iI}_{k=0}^n\left(1\right)^ka_i_k\tau _i`$ can be reduced to a chain in $`C_{n1}^{inf}\left(Q\right)`$ by substracting (possibly infinitely many) multiples of $`\left(_{a_1}\tau _{i_1}_{a_2}\tau _{i_2}\right)`$ with $`(_{a_1}\tau _{i_1},_{a_2}\tau _{i_2})๐`$.
###### Observation 6.
Let $`(Q,_1Q)`$ be a pair of topological spaces.
a) If $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$ is a singular chain, $`๐`$ a set of cancellations, and $`\mathrm{{\rm Y}}:=\mathrm{{\rm Y}}_{z,๐}`$ the associated simplicial set, then the geometric realisation $`\mathrm{{\rm Y}}`$ is obtained from $`I`$ copies of the standard n-simplex $`\mathrm{\Delta }_i,iI`$, with identifications $`_{a_1}\mathrm{\Delta }_{i_1}=_{a_2}\mathrm{\Delta }_{i_2}`$ if and only if $`(_{a_1}\tau _{i_1},_{a_2}\tau _{i_2})๐`$. Moreover, for a minimal presentation of $`z`$ and $`\mathrm{{\rm Y}}:=\mathrm{{\rm Y}}_{z,๐}`$, $`\mathrm{{\rm Y}}`$ is the subspace of $`\mathrm{{\rm Y}}`$ containing all simplices $`_{a_1}\mathrm{\Delta }_{i_1}`$ with $`(i_1,a_1)J`$.
b) There exists an associated continuous map $`\tau :\mathrm{{\rm Y}}Q`$ with $`\tau \mathrm{\Delta }_i=\tau _i`$ (upon the identification $`\mathrm{\Delta }_i=\mathrm{\Delta }^n`$). If $`z`$ is a relative cycle, i.e. $`zC_{n1}^{inf}\left(_1Q\right)`$, then $`\tau `$ maps $`\mathrm{{\rm Y}}`$ to $`_1Q`$.
c) Let $`z_1=_{iI}a_i\tau _i,z_2=_{iI}a_i\sigma _iC_n^{inf}(Q,_1Q)`$ be relative cycles and $`๐_1,๐_2`$ sufficient sets of cancellations of $`z_1`$ resp. $`z_2`$. Assume that $`(_{a_1}\tau _{i_1},_{a_2}\tau _{i_2})๐_1`$ if and only if $`(_{a_1}\sigma _{i_1},_{a_2}\sigma _{i_2})๐_2`$, and that there exist minimal presentations of $`z_1,z_2`$ such that $`J_{z_1}=J_{z_2}`$.
If the associated continuous maps
$`\tau ,\sigma :\mathrm{{\rm Y}}Q`$ are homotopic,
for a homotopy mapping $`\mathrm{{\rm Y}}`$ to $`Q`$, then $`_{iI}a_i\tau _i`$ and $`_{iI}a_i\sigma _iC_{}^{inf}(Q,Q)`$ are relatively homologous.
We emphasize that we do not assume that $`๐`$ is a complete list of cancellations, the simplicial map $`\tau _{}:C_{}^{simp}\left(\mathrm{{\rm Y}}\right)C_{}^{sing}\left(Q\right)`$ need not be injective.
After having set up the necessary notations, we now start with the actual definition of the straightening. We first mention that there is of course an analogue of the classical straightening (, Lemma C.4.3.) in our setting.
###### Observation 7.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I. Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy the conditions i)-viii) from Lemma 11.
Then there exists a โcanonical straighteningโ map
$$str_{can}:C_{}^{simp,inf}\left(K\left(Q\right)\right)C_{}^{simp,inf}\left(K^{str}\left(Q\right)\right),$$
mapping $`C_{}^{simp,inf}\left(K\left(_1Q\right)\right)`$ to $`C_{}^{simp,inf}\left(K^{str}\left(_1Q\right)\right)`$, with the following properties:
i) $`str_{can}`$ is a chain map,
ii) if $`z=_{iI}a_i\tau _iC_{}^{simp,inf}\left(K\left(Q\right)\right)`$ and $`_{iI}a_i\sigma _i:=_{iI}a_istr_{can}\left(\tau _i\right)`$, then the maps
$$\tau ,\sigma :\mathrm{{\rm Y}}Q$$
(defined by Observation 6b) after fixing a set of cancellations $`๐`$ and a minimal presentation of $`z`$) are homotopic.
Moreover, if $`z=_{iI}a_i\tau _i`$ is a relative cycle with $`zC_{}^{simp,inf}\left(K\left(_1Q\right)\right)`$, then the same is true for $`_{iI}a_i\sigma _i`$ and
$$\tau ,\sigma :(\mathrm{{\rm Y}},\mathrm{{\rm Y}})(Q,_1Q)$$
are homotopic as maps of pairs.
In particular, $`_{iI}a_istr_{can}\left(\tau _i\right)`$ is relatively homologous to $`_{iI}a_i\tau _i`$,
###### Proof.
We define $`str_{can}`$, and the homotopy to the identity, by induction on the dimension of simplices. (During the construction we take care that $`str_{can}`$ and the homotopy preserve $`K\left(_1Q\right)`$.)
0-simplices.
If $`C`$ is a path-component of $`_0Q`$ with $`C_1Q=\mathrm{}`$, then we define $`str_{can}\left(v\right)=v`$ for each 0-simplex $`v`$ in $`C`$. The homotopy $`H\left(v\right)`$ is for each $`v`$ given by the constant map.
If $`C`$ is a path-component of $`_0Q`$ with $`C_1Q\mathrm{}`$, then there is at least one path-component $`F`$ of $`_1Q`$ with $`CF\mathrm{}`$. By Construction 1 and condition ii) from Lemma 11, for each such $`F`$, there is a straight 0-simplex $`x_{E_i^F}CF`$. Choose one such straight 0-simplex (among the $`x_{E_i^F}`$โs) for each path-component $`C`$ of $`_0Q`$, denote it $`x_C`$, and for each $`vC`$ we define $`str_{can}\left(v\right):=x_CK_0^{str}\left(Q\right)C`$ and we choose the homotopy $`H(v)`$ to belong to $`C`$.
If $`v_1Q`$, then there is (at least) one straight 0-simplex in the same path-component $`F`$ of $`_1Q`$, we choose $`str_{can}\left(v\right)FK_0^{str}\left(Q\right)`$ and there exists $`H\left(v\right)K_1\left(_1Q\right)`$ with $`H\left(v\right)=vstr_{can}\left(v\right)`$.
If $`vQ`$, then we define $`str_{can}\left(v\right)`$ to be some straight 0-simplex in $`Q`$ and we fix arbitrarily some $`H\left(v\right)K_1\left(Q\right)`$ with $`H\left(v\right)=vstr_{can}\left(v\right)`$.
1-simplices.
For $`eK_1\left(Q\right)`$ we define
$$str_{can}\left(e\right):=\left[\overline{H\left(_1e\right)}eH\left(_0e\right)\right],$$
where, as always, \[.\] denotes the unique 1-simplex in $`K_1^{str}\left(Q\right)`$, which is homotopic rel. boundary to the path in the brackets.
$`e`$ is homotopic to $`str_{can}\left(e\right)`$ by the canonical homotopy which is inverse to the homotopy moving $`\overline{H\left(_1e\right)}`$ resp. $`H\left(_0e\right)`$ into constant maps. In particular, the restriction of this homotopy to $`_1e,_0e`$ gives $`H\left(_1e\right),\overline{H\left(_0e\right)}`$. Thus, for different edges with common vertices, the homotopies are compatible. We thus have constructed a homotopy for the 1-skeleton $`\mathrm{{\rm Y}}_1`$.
We note that, for $`v_1Q`$, the homotopy $`H\left(v\right)`$ is either constant or $`H\left(v\right)K_1\left(_1Q\right)`$, Thus if $`\tau K_1\left(_1Q\right)`$ then $`str_{can}\left(\tau \right)K_1^{str}\left(_1Q\right)`$ and the homotopy between $`\tau `$ and $`str_{can}\left(\tau \right)`$ takes place in $`_1Q`$.
n-simplices.
We assume inductively, that for some $`n1`$, we have defined $`str_{can}`$ on $`K_n\left(Q\right)`$, mapping $`K_n\left(_1Q\right)`$ to $`K_n^{str}\left(_1Q\right)`$, and satisfying i),ii),iii).
Let $`\tau K\left(Q\right)`$ be an n+1-simplex. Then we have by ii) a homotopy between $`\tau `$ and $`str_{can}\left(\tau \right)`$. By Observation 1 this homotopy extends to $`\tau `$. The resulting simplex $`\tau ^{}`$ satisfies $`\tau ^{}K_n^{str}\left(Q\right)`$. Condition v) from Lemma 11 means that $`\tau ^{}`$ is homotopic rel. boundary to a unique simplex $`str_{can}\left(\tau \right)K_{n+1}^{str}\left(Q\right)`$. This proves the inductive step.
If $`\tau K\left(_1Q\right)`$, then we can inductively assume that the homotopy of $`\tau `$ has image in $`_1Q`$. Then condition vii) from Lemma 11 implies $`str_{can}\left(\tau \right)K_{n+1}^{str}\left(_1Q\right)`$. Moreover, since $`_1Q`$ is aspherical, the homotopy of $`\tau `$ can be chosen to have image in $`_1Q`$.
By construction, for any set of cancellations $`๐`$, the induced maps $`\tau `$ and $`\sigma `$ are homotopic. In particular, if we chose a sufficient set of cancellations in the sense of Definition 8d), then Observation 6c) implies that $`_{i=1}^ra_istr_{can}\left(\tau _i\right)`$ is (relatively) homologous to $`_{i=1}^ra_i\tau _i`$.
However, we want to define a more refined straightening, which will be defined only on relative cycles with some kind of additional information.
Before stating the definition of โdistinguished 1-simplicesโ we remark that there is a left and right action of the pseudogroup $`\mathrm{\Gamma }:=\mathrm{\Omega }\left(Q\right)`$ (as defined in Section 3.3) on $`K_1^{str}\left(Q\right)`$: if $`eK_1^{str}\left(Q\right),\gamma _1\pi _1(Q,_1e),\gamma _2\pi _1(Q,_0e)`$, then let $`\gamma _1e\gamma _2`$ be the unique straight 1-simplex homotopic rel. $`\{0,1\}`$ to $`\gamma _1e\gamma _2`$. (The left action agrees with the action defined in Section 3.3.) The cosets $`\mathrm{\Gamma }K_1^{str}\left(Q\right)\mathrm{\Gamma }`$ in Definition 9 are with respect to this action.
For $`x,yK_0^{str}\left(Q\right)`$ we will denote $`K_{1,xy}^{str}:=\{eK_1^{str}\left(Q\right):_1e=x,_0e=y\}`$.
###### Definition 9.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I.
Let $`q:QQ`$ and $`\left\{x_{E_i^F}_0Q_1Q\right\}`$ be given by Construction 1.
Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy conditions i)-viii) from Lemma 11.
A set $`DK_1^{str}\left(Q\right)`$ is called a set of distinguished 1-simplices if
ix) $`_0e,_1eK_0^{str}\left(Q\right)`$ for each $`eD`$,
x) for each
$$(x,y)K_0^{str}\left(Q\right)\times K_0^{str}\left(Q\right)$$
we have that
$$D_{xy}:=\{eD:_1e=x,_0=y\}$$
contains exactly one element in each double coset (w.r.t. $`\mathrm{\Gamma }=\mathrm{\Omega }\left(Q\right)`$)
$$\mathrm{\Gamma }f\mathrm{\Gamma }\mathrm{\Gamma }K_{1,xy}^{str}\left(Q\right)\mathrm{\Gamma },$$
xi) for all $`xK_0^{str}\left(Q\right)`$, the constant loop $`c_x`$ belongs to $`D`$,
xii) if $`eD`$, then $`\overline{e}D`$, where $`\overline{e}`$ denotes the 1-simplex with the opposite orientation,
xiii) if $`F,F^{}`$ are path-components of $`_1Q`$ and $`\left\{x_{E_i^F}_0QF\right\},\left\{x_{E_j^F^{}}_0QF^{}\right\}`$ are given<sup>8</sup><sup>8</sup>8If $`F_0Q=\mathrm{}`$ and/or $`F^{}_0Q=\mathrm{}`$, then there is only one straight 0-simplex $`x_{E_0^F}`$ resp. $`x_{E_0^F^{}}`$ in $`F`$ resp. $`F^{}`$. In particular, if $`F_0Q=\mathrm{}`$ and $`F^{}_0Q=\mathrm{}`$, then condition xiii) is empty. by Construction 1, then $`q\left(D_{x_{E_i^F}x_{E_j^F^{}}}\right)=D_{x_{E_0^F}x_{E_0^F^{}}}`$ for all $`x_{E_i^F},x_{E_j^F^{}}`$,
xiv) if $`x_1,x_2C_1,y_1,y_2C_2`$ for some path-components $`C_1,C_2`$ of $`Q`$, then for each $`e_1D_{x_1y_1}`$ exists some $`e_2D_{x_2y_2}`$ with $`q\left(e_2\right)=gq\left(e_1\right)`$ for some $`gH:=q_{}\left(\mathrm{\Pi }\left(K\left(_0Q\right)\right)\right)`$.
###### Observation 8.
Let the assumptions of Definition 9 be satisfied. Then a set $`D`$ of distinguished 1-simplices exists.
###### Proof.
For each path-component $`C`$ of $`Q`$ we fix some $`x_CK_0^{str}\left(C\right)`$.
For each pair $`\{C_1,C_2\}`$ of path-components we fix one simplex $`e`$ with
$$_1e=x_{C_1},_0e=x_{C_2}$$
in each coset of $`\mathrm{\Gamma }K_{1,x_{C_1}x_{C_2}}^{str}\left(Q\right)\mathrm{\Gamma }`$ to belong to $`D_{x_{C_1}x_{c_2}}`$.
(For all chosen 1-simplices $`eD_{x_{C_1}x_{C_2}}`$, we choose $`\overline{e}`$ to belong to $`D_{x_{C_2}x_{C_1}}`$. If $`C_1=C_2`$, then in particular for the coset of the constant loop we choose the constant loop to belong to $`D_{x_{C_1}x_{C_2}}`$.)
For each path-component $`C`$ of $`Q`$ and each path-component $`F`$ of $`C_1Q`$, we have that $`q\left(x_C\right)`$ and $`q\left(x_{E_0^F}\right)`$ belong to the path-connected set $`q\left(_0QC\right)`$. Therefore we have a sequence of 1-simplices $`\alpha _1,\mathrm{},\alpha _mK_1\left(_0Q\right)`$ with images in distinct path-components of $`_0QC`$, such that
$$_1q\left(\alpha _1\right)=q\left(x_C\right),_0q\left(\alpha _1\right)=_1q\left(\alpha _2\right),\mathrm{},_0q\left(\alpha _{m1}\right)=_1q\left(\alpha _m\right),_0q\left(\alpha _m\right)=q\left(x_{E_0^F}\right).$$
In order to prepare the definition of the $`D_{x,y}`$โs, we first describe, for each $`xCK_0^{str}\left(Q\right)`$ a sequence $`\{\alpha _1,\mathrm{},\alpha _k\}`$ of 1-simplices:
\- if $`C_1Q=\mathrm{}`$, then $`k=1`$ and for each $`xC`$ we choose arbitrarily a 1-simplex $`\alpha _1`$ in $`C`$ with $`_1\alpha _1=x_C,_0\alpha _1=x`$,
\- if $`C_0Q=\mathrm{}`$, then $`CK_0^{str}\left(Q\right)=\left\{x_C\right\}`$ by Lemma 11, condition ii), and we let $`k=0`$,
\- if $`C_0Q_1Q\mathrm{}`$, then by condition ii) from Lemma 11 we have $`x=x_{E_i^F}`$ for some path-component $`F`$ of $`_1Q`$ and some $`i`$, thus we have the above-constructed sequence $`\alpha _1,\mathrm{},\alpha _m`$ with $`_1q\left(\alpha _1\right)=q\left(x_C\right),_0q\left(\alpha _1\right)=_1q\left(\alpha _2\right),\mathrm{},_0q\left(\alpha _{m1}\right)=_1q\left(\alpha _m\right),_0q\left(\alpha _m\right)=q\left(x_{E_i^F}\right)`$, where the last equality holds true because $`q\left(x_{E_i^F}\right)=x_{E_0^F}=q\left(x_{E_0^F}\right)`$.
Let $`x,yK_0^{str}\left(Q\right)`$. Let $`C_1,C_2`$ be the path-components of $`Q`$ with $`xC_1,yC_2`$.
We have constructed sequences of 1-simplices $`\alpha _1,\mathrm{},\alpha _kK_1\left(Q\right)`$ resp. $`\beta _1,\mathrm{},\beta _lK_1\left(Q\right)`$, such that $`_1q\left(\alpha _1\right)=q\left(x_{C_1}\right),_0q\left(\alpha _1\right)=_1q\left(\alpha _2\right),\mathrm{},_0q\left(\alpha _{k1}\right)=_1q\left(\alpha _k\right),_0q\left(\alpha _k\right)=q\left(x\right)`$ resp. $`_1q\left(\beta _1\right)=q\left(x_{C_2}\right),_0q\left(\beta _1\right)=_1q\left(\beta _2\right),\mathrm{},_0q\left(\beta _{k1}\right)=_1q\left(\beta _k\right),_0q\left(\beta _k\right)=q\left(y\right)`$. Note that all $`q\left(\alpha _i\right)`$ and $`q\left(\beta _i\right)`$ are either constant or contained in $`q\left(K_1\left(_0Q\right)\right)`$.
Let $`H:=q_{}\left(\mathrm{\Pi }\left(K\left(_0Q\right)\right)\right)`$. Define
$$g:=\{q\left(\alpha _1\right),q\left(\overline{\alpha }_1\right)\}\mathrm{}\{q\left(\alpha _k\right),q\left(\overline{\alpha }_k\right)\}\{q\left(\beta _l\right),q\left(\overline{\beta }_l\right)\}\mathrm{}\{q\left(\beta _1\right),q\left(\overline{\beta }_1\right)\}H.$$
(If $`k=l=0`$, this means just $`g=1`$.)
We have that $`g=g^1`$ and that
$$geK_{1,q\left(x\right)q\left(y\right)}^{str}\left(Q\right)eK_{1,q\left(x_{C_1}\right)q\left(x_{C_2}\right)}^{str}\left(Q\right).$$
By construction, the $`g`$ associated to $`x_{E_i^F},x_{E_j^F^{}}`$ agrees with the $`g`$ associated to $`x_{E_0^F},x_{E_0^F^{}}`$.
We are given $`D_{x_{C_1}x_{C_2}}`$ and we want to define $`D_{xy}`$ such that condition xiii) is satisfied.
First, if $`C_1_1Q=\mathrm{}`$ or $`C_2_1Q=\mathrm{}`$, then we can fix an arbitrary choice of $`D_{x,y}`$ satisfying conditions x),xi,xii). (Condition xiii) is empty in this case.)
So let us assume $`C_1_1Q\mathrm{}`$ and $`C_2_1Q\mathrm{}`$. We note that
$$q:(Q,Q,_1Q)(Q,Q,_1Q)$$
is homotopic to the identity as a map of triples, by the construction in Section 5.1. This implies that cosets of $`\mathrm{\Gamma }K_{1,xy}^{str}\left(Q\right)\mathrm{\Gamma }`$ are in 1-1-correspondence (by applying $`q`$) to cosets of $`\mathrm{\Gamma }K_{1,q\left(x\right)q\left(y\right)}^{str}\mathrm{\Gamma }`$. It is thus sufficient to describe $`q\left(D_{xy}\right)K_{1,q\left(x\right)q\left(y\right)}^{str}`$.
Let
$$\mathrm{\Gamma }f\mathrm{\Gamma }\mathrm{\Gamma }K_{1,q\left(x\right)q\left(y\right)}^{str}\left(Q\right)\mathrm{\Gamma }$$
be a double coset. Then the double coset
$$\mathrm{\Gamma }\left(gf\right)\mathrm{\Gamma }\mathrm{\Gamma }K_{1,q\left(x_{C_1}\right)q\left(x_{C_2}\right)}^{str}\left(Q\right)\mathrm{\Gamma }$$
is the image under $`q`$ of some double coset
$$\mathrm{\Gamma }e^{}\mathrm{\Gamma }\mathrm{\Gamma }K_{1,x_{C_1}x_{C_2}}^{str}\left(Q\right)\mathrm{\Gamma }$$
Let $`e`$ be the unique distinguished simplex in $`\mathrm{\Gamma }e^{}\mathrm{\Gamma }`$. Then we choose $`gq\left(e\right)`$ to be the distinguished simplex in $`\mathrm{\Gamma }f\mathrm{\Gamma }`$. This is possible because $`gq\left(e\right)`$ belongs to the double coset $`\mathrm{\Gamma }f\mathrm{\Gamma }`$. Indeed
$$q\left(e\right)\mathrm{\Gamma }\left(gf\right)\mathrm{\Gamma }$$
means that $`q\left(e\right)=q_{}\left(\gamma _1\right)gfq_{}\left(\gamma _2\right)`$ for some loops $`\gamma _1`$ and $`\gamma _2`$ based at $`x_{C_1}`$ resp. $`x_{C_2}`$, and this implies $`gq\left(e^{}\right)=q_{}\left(\gamma _1^{}\right)fq_{}\left(\gamma _2^{}\right)`$ with
$$\gamma _1^{}:=\left[\overline{\alpha }_m\mathrm{}\overline{\alpha }_1\gamma _1\alpha _1\mathrm{}\alpha _m\right],\gamma _2^{}:=\left[\overline{\beta }_n\mathrm{}\overline{\beta }_1\gamma _2\beta _1\mathrm{}\beta _n\right].$$
This defines $`D_{xy}`$. By construction, condition xiv) is satisfied if $`e_1D_{x_{C_1}x_{C_2}}`$. In general, if $`e_1D_{x_1y_1}`$, then we get $`eD_{x_{C_1}x_{C_2}}`$ and $`g_1H`$ with $`q\left(e_1\right)=g_1q\left(e\right)`$ and $`e_2D_{x_2y_2},g_2H`$ with $`q\left(e_2\right)=g_2q\left(e\right)`$, thus $`q\left(e_2\right)=g_2g_1^1q\left(e_1\right)`$.
Condition xiii) is implied because $`q\left(x_{E_i^F}\right)=x_{E_0^F},q\left(x_{E_j^F^{}}\right)=x_{E_0^F^{}}`$ and the $`g`$ associated to $`x_{E_i^F},x_{E_j^F^{}}`$ agrees with the $`g`$ associated to $`x_{E_0^F},x_{E_0^F^{}}`$.
One checks easily that xi) and xii) are true for $`D_{xy}`$ since they are true for $`D_{x_{C_1}x_{C_2}}`$. โ
###### Definition 10.
Let $`Q,Q,_0Q,_1Q`$ satisfy Assumption I. Let $`z=_{iI}a_i\tau _iC_n^{inf}\left(Q\right)`$ be a singular chain and $`\mathrm{{\rm Y}}`$ the associated simplicial set (for some set of cancellations $`๐`$).
We say that a labeling of the elements of the 1-skeleton $`\mathrm{{\rm Y}}_1`$ by 0โs and 1โs is admissible, if $`e_1e_2=\mathrm{}`$ for all 1-labeled vertices $`e_1,e_2`$.
###### Lemma 12.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I. Let $`q:QQ`$ be given by Construction 1.
Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy conditions i)-viii) from Lemma 11, and let $`DK_1^{str}\left(Q\right)`$ be a set of distinguished 1-simplices.
Let $`z=_{iI}a_i\tau _iC_{}^{simp,inf}\left(K\left(Q\right)\right)`$ be a relative cycle with $`zC_{}^{simp,inf}\left(K\left(_1Q\right)\right)`$.
Let a set of cancellations $`๐`$ for $`z`$ and a minimal presentation of $`z`$ be given. Let $`\mathrm{{\rm Y}},\mathrm{{\rm Y}}`$ be the associated simplicial sets, $`\tau :(\mathrm{{\rm Y}},\mathrm{{\rm Y}})(Q,_1Q)`$ the associated continuous mapping.
Assume that we have an admissible 0-1-labeling of $`\mathrm{{\rm Y}}_1`$.
Then there exists a relative cycle
$$z^{}=\underset{iI}{}a_i\tau _i^{}C_{}^{simp,inf}(K^{str}\left(Q\right),K^{str}\left(_1Q\right))$$
such that:
i) the associated continuous mappings
$$\tau ,\tau ^{}:(\mathrm{{\rm Y}},\mathrm{{\rm Y}})(Q,_1Q)$$
are homotopic by a homotopy mapping $`\mathrm{{\rm Y}}`$ to $`Q`$,
ii) if an edge of some $`\tau _i`$ is labeled by 1, then the corresponding edge of $`\tau _i^{}`$ belongs to $`D`$,
(Remark: The homotopy in i) does not necessarily map $`\mathrm{{\rm Y}}`$ to $`_1Q`$, but to $`Q`$.)
###### Proof.
First we apply the โcanonical straighteningโ $`str_{can}`$ from Observation 7. The resulting chain $`_{iI}a_istr_{can}\left(\tau _i\right)`$ satisfies i), but not necessarily ii).
$`_{iI}a_istr_{can}\left(\tau _i\right)`$ inherits the admissible labeling from $`_{iI}a_i\tau _i`$. Thus we can w.l.o.g. restrict to the case that all $`\tau _i`$ belong to $`K^{str}\left(Q\right)`$.
Let
$$eK_1^{str}\left(Q\right)$$
be a 1-labeled edge, let $`x=_1eK_0^{str}\left(Q\right),y=_0eK_0^{str}\left(Q\right)`$. By Definition 9, the coset $`\mathrm{\Gamma }e\mathrm{\Gamma }`$ contains a unique distinguished 1-simplex $`str\left(e\right)D_{xy}`$. (We use the notation from Definition 9, in particular $`\mathrm{\Gamma }:=\mathrm{\Omega }\left(Q\right)`$.)
$`str\left(e\right)\mathrm{\Gamma }e\mathrm{\Gamma }`$ means<sup>9</sup><sup>9</sup>9If $`_0e,_1e_1Q`$, then $`str\left(e\right)\mathrm{\Gamma }e\mathrm{\Gamma }`$ means, of course, $`str\left(e\right)=e`$. Similarly, if only one vertex of $`e`$ belongs to $`_1Q`$, then only that vertex is moved during the homotopy. that there are loops $`\gamma _1,\gamma _2Q`$ based at $`x`$ resp. $`y`$ such that $`str\left(e\right)\gamma _1e\gamma _2`$ rel. $`\{0,1\}`$. There is an obvious homotopy between $`e`$ and $`\gamma _1e\gamma _2`$, which moves $`_1e`$ along $`\overline{\gamma }_1`$ and $`_0e`$ along $`\gamma _2`$. (Of course, we change the homotopy class relative boundary, so we can not keep the endpoints fixed during the homotopy.) If $`e`$ and/or $`_0e`$ and/or $`_1e`$ have image in $`_1Q`$, then their images remain in $`Q`$ (and end up in $`_1Q`$) during the homotopy.
Using Observation 1, the so-constructed homotopy between $`e`$ and $`str\left(e\right)`$ can be extended to a homotopy from
$$\tau :(\mathrm{{\rm Y}},\mathrm{{\rm Y}})(Q,_1Q)$$
to some
$$\widehat{\tau }:(\mathrm{{\rm Y}},\mathrm{{\rm Y}})(Q,_1Q),$$
such that $`\widehat{\tau }`$ is a simplicial map from $`\mathrm{{\rm Y}}`$ to $`S_{}\left(Q\right)`$. (If a 0-labeled edge has one or both vertices adjacent to 1-labeled edges, then the 0-labeled edge just follows the homotopy of the vertices. 0-labeled edges that are not adjacent to 1-labeled edges can remain fixed during the homotopy.) The homotopy maps $`\mathrm{{\rm Y}}`$ to $`Q`$.
Next we apply homotopies rel. boundary to the (already homotoped images of) all 0-labeled edges $`fK_1^{str}\left(Q\right)`$, to homotope them to edges in $`K_1^{str}\left(Q\right)`$. If $`f`$ and/or $`_0f`$ and/or $`_1f`$ have image in $`_1Q`$, then their images remain in $`Q`$ (and end up in $`_1Q`$) during the homotopy.
Now we have a simplicial map $`\widehat{\tau }:\mathrm{{\rm Y}}S_{}\left(Q\right)`$, such that all 1-simplices are mapped to $`K_1^{str}\left(Q\right)`$, and such that
$$\widehat{\tau }\left(e\right)DK_1^{str}\left(Q\right)$$
holds for all 1-labeled edges $`e`$. Then we can, as in the proof of Observation 7, by induction on $`n`$, apply homotopies rel. boundary to all n-simplices to homotope them into $`K_n^{str}\left(Q\right)`$. Simplices in $`_1Q`$ remain in $`Q`$ (and end up in $`_1Q`$) during the homotopy.
We obtain a homotopy (of pairs), which keeps the 1-skeleton fixed, to a simplicial map
$$\tau ^{}:\mathrm{{\rm Y}}K^{str}\left(Q\right),$$
mapping $`\mathrm{{\rm Y}}`$ to $`K^{str}\left(_1Q\right)`$ and satisfying i),ii). โ
A somewhat artificial formulation of the conclusion of Lemma 12 is that we have constructed a chain map
$$str:C_{}^{simp,inf}(\mathrm{{\rm Y}},\mathrm{{\rm Y}})C_{}^{simp,inf}(K^{str}\left(Q\right),K^{str}\left(_1Q\right)).$$
Unfortunately, this somewhat artificial formulation can not be simplified because $`str`$ depends on the chain $`_{iI}a_i\tau _i`$. That is, we do not get a chain map $`str:C_{}^{simp,inf}(K\left(Q\right),K\left(_1Q\right))C_{}^{simp,inf}(K^{str}\left(Q\right),K^{str}\left(_1Q\right))`$.
### 5.4 Straightening of crushed cycles
Recall from Section 3.5 that $`._{๐G}๐`$ means the tensor product with the trivial $`๐G`$-module $`๐`$, that is, the quotient under the $`G`$-action. We first state obvious generalizations of Observation 6 to the case of tensor products with a factor with trivial $`G`$-action.
###### Observation 9.
Let $`(Q,_1Q)`$ be a pair of topological spaces. Let $`G`$ be a group acting on a pair $`(K,K)`$ with $`KS_{}\left(Q\right)`$ and $`KS_{}\left(_1Q\right)`$ both closed under face maps.
i) If
$$z=\underset{iI}{}a_i\tau _i1C_{}^{simp,inf}(K,K)_{๐G}๐$$
is a relative cycle, then
$$\widehat{z}=\underset{iI}{}\underset{gG}{}a_i\left(g\tau _i\right)C_{}^{simp,inf}(K,K)$$
is a relative cycle.
If $`๐`$ is a sufficient set of cancellations for $`z`$, then there exists a set of cancellations $`\widehat{๐}`$ for $`\widehat{z}`$ such that $`(\eta _1,\eta _2)\widehat{๐}`$ implies $`(\eta _11,\eta _21)๐`$.
If $`z=_{a,i}c_{ai}_a\tau _i1`$ is a minimal presentation for $`z`$, then $`\widehat{z}=_{gG}_{a,i}c_{ai}_a\left(g\tau _i\right)`$ is a minimal presentation for $`\widehat{z}`$.
ii) Let $`\widehat{\mathrm{{\rm Y}}},\widehat{\mathrm{{\rm Y}}}`$ be the simplicial sets associated to $`\widehat{z}`$, the sufficient set of cancellations $`\widehat{๐}`$ and the minimal presentation of $`\widehat{z}`$. They come with an obvious $`G`$-action. Then we have an associated continuous mapping $`\widehat{\tau }:(\widehat{\mathrm{{\rm Y}}},\mathrm{{\rm Y}})(Q,_1Q)`$.
###### Corollary 3.
Let $`Q,Q,_1Q,_0Q`$ satisfy Assumption I. Let $`q:QQ`$ be given by Construction 1.
Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy conditions i)-viii) from Lemma 11, and let $`DK_1^{str}\left(Q\right)`$ be a set of distinguished 1-simplices.
Let $`G:=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$ with its action on $`K^{str}\left(Q\right)`$ defined in Observation 5, and let $`H:=q_{}\left(G\right)`$ as defined in Section 5.1. Let
$$\underset{iI}{}a_i\tau _i1C_n^{simp,inf}(K\left(Q\right),GK\left(_1Q\right))_{๐G}๐$$
be a relative cycle. Fix a sufficient set of cancellations $`๐`$ and a minimal presentation for $`z`$. Let $`\widehat{\mathrm{{\rm Y}}},\widehat{\mathrm{{\rm Y}}}`$ be defined by Observation 9. Assume that we have a $`G`$-invariant admissible 0-1-labeling of the edges of $`\widehat{\mathrm{{\rm Y}}}`$.
Then there is a well-defined chain map
$$qstr:C_{}^{simp,inf}\left(\widehat{\mathrm{{\rm Y}}}\right)_{๐G}๐C_{}^{simp,inf}\left(HK^{str}\left(Q\right)\right)_{๐H}๐,$$
mapping $`C_{}^{simp,inf}\left(\widehat{\mathrm{{\rm Y}}}\right)_{๐G}๐`$ to $`C_{}^{simp,inf}\left(GK^{str}\left(_1Q\right)\right)_{๐H}๐`$, such that:
i) if $`e\widehat{\mathrm{{\rm Y}}}_1`$ is a 1-labeled edge, $`str\left(e1\right)=f1`$, then $`fD`$.
ii) if $`Q`$ is an orientable manifold with boundary $`Q`$, and if
$$\underset{iI}{}a_i\tau _i1C_{}^{simp,inf}(K\left(Q\right),GK\left(_1Q\right))_{๐G}๐$$
represents<sup>10</sup><sup>10</sup>10Cf. the footnotes in Section 3.4 the image of $`[Q,Q]1`$, then
$$\underset{iI}{}a_iqstr\left(\tau _i1\right)C_{}^{simp,inf}(HK^{str}\left(Q\right),HK^{str}\left(_1Q\right))_{๐H}๐$$
represents the image of $`[Q,Q]1`$ and
$$\underset{iI}{}a_iqstr\left(\tau _i1\right)C_{}^{simp,inf}\left(HK^{str}\left(_1Q\right)\right)_{๐H}๐$$
represents the image of $`\left[Q\right]1`$.
###### Proof.
We can apply Lemma 12 to the infinite chain $`_{iI,gH}a_i\left(g\tau _i\right)`$. Thus Lemma 12 provides us with a chain map $`str:C_{}^{simp,inf}\left(\widehat{\mathrm{{\rm Y}}}\right)C_{}^{simp,inf}\left(K^{str}\left(Q\right)\right)`$, given by
$$str\left(g\tau _i\right):=\left(g\tau _i\right)^{}.$$
$`q:(K^{str}\left(Q\right),K^{str}\left(_1Q\right))(K^{str}\left(Q\right),K^{str}\left(_1Q\right))`$ is defined by Observation 5. (Remark: we actually have $`qstr\left(g\tau _i\right)K^{str}\left(Q\right)`$. We need $`HK^{str}\left(Q\right)`$ in the statement of Corollary 3 just to have the tensor product well-defined.)
We are going to define $`qstr\left(\sigma z\right):=q\left(str\left(\sigma \right)\right)z`$ for each $`\sigma \widehat{\mathrm{{\rm Y}}},z๐`$. For this to be well-defined, we have to check the following claim:
for each $`\sigma K,gG`$, there exists $`hH`$ with $`q\left(str\left(g\sigma \right)\right)=hq\left(str\left(\sigma \right)\right)`$.
By condition viii) from Lemma 11 (asphericity of $`K^{str}\left(Q\right)`$), it suffices to check this for the 1-skeleton.
It is straightforward to check the claim for the 0-skeleton.
If $`\sigma =vS_0\left(_0Q\right)`$ then $`v`$ and $`gv`$ belong to the same path-component $`C`$ of $`_0Q`$, hence $`str\left(v\right)`$ and $`str\left(gv\right)`$ belong to the same path-component $`C`$. Let $`\gamma :[0,1]_0Q`$ be a path with $`\gamma \left(0\right)=str\left(v\right),\gamma \left(1\right)=str\left(gv\right)`$. Let $`\gamma ^{}`$ be the unique 1-simplex in $`K\left(_0Q\right)`$ which is homotopic rel. boundary to $`\gamma `$. Let $`g^{}:=\{\gamma ^{},\overline{\gamma ^{}}\}G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$. Then $`g^{}str\left(v\right)=str\left(gv\right)`$, which implies $`q\left(str\left(gv\right)\right)=hq\left(str\left(v\right)\right)`$ with $`h=q_{}\left(g^{}\right)H`$.
If $`\sigma =v_0Q`$, then $`gv=v`$, hence $`q\left(str\left(gv\right)\right)=q\left(str\left(v\right)\right)`$.
The proof for 1-simplices consists of two steps. In the first step we prove that for $`eK_1\left(Q\right),gG`$ we have $`str_{can}\left(ge\right)=g^{}str_{can}\left(e\right)`$ with $`g^{}G`$. In the second step we show that, if $`eK_1^{str}\left(Q\right)`$ and $`gG`$, then there exists $`hH`$ with $`q\left(str\left(ge\right)\right)=hq\left(str\left(e\right)\right)`$. Hence altogether we will get $`q\left(str\left(ge\right)\right)=q\left(str\left(str_{can}\left(ge\right)\right)\right)=q\left(str\left(g^{}str_{can}\left(e\right)\right)\right)=hq\left(str\left(str_{can}\left(e\right)\right)\right)=hq\left(str\left(e\right)\right)`$.
First step: This is fairly obvious.
First case: If both vertices of $`e`$ do not belong to $`_0Q`$, then also both vertices of $`str_{can}\left(e\right)`$ do not belong to $`_0Q`$, and we have $`ge=e,gstr_{can}\left(e\right)=str\left(e\right)`$, which implies the claim.
Second Case: If both vertices of $`e`$ belong to $`_0Q`$, then $`str_{can}\left(e\right)\alpha _1e\alpha _2,str_{can}\left(ge\right)\beta _1ge\beta _2`$ for some paths $`\alpha _1,\alpha _2,\beta _1,\beta _2`$ in $`_0Q`$. Moreover, by the definition of the action (Section 3.3) we have $`ge\gamma _2e\gamma _1`$ for some $`\gamma _1,\gamma _2K_1\left(_0Q\right)`$. Thus $`str_{can}\left(ge\right)\beta _1\gamma _1\alpha _1^1str_{can}\left(e\right)\alpha _2^1\gamma _2\beta _2`$, in particular $`str_{can}\left(ge\right)=g^{}str_{can}\left(e\right)`$ for some $`g^{}G`$.
Third case: Finally we consider the case that one vertex, say $`_0e`$ belongs to $`_0Q`$, but $`_1e`$ does not belong. Then we are in the situation of the second case with $`\gamma _2=1,\alpha _2=\beta _2`$ (except that $`\alpha _2`$ is not contained in $`_0Q`$). We get $`str_{can}\left(ge\right)\beta _1\gamma _1\alpha _1^1str_{can}\left(e\right)`$. Since $`\beta _1\gamma _1\alpha _1^1`$ is contained in $`_0Q`$, this implies that $`str_{can}\left(ge\right)=g^{}str_{can}\left(e\right)`$ for some $`g^{}G`$.
Second step: Let $`eK_1^{str}\left(Q\right)`$.
If $`e`$ is a 1-labeled edge, with $`x=_1e,y=_0eK_0^{str}\left(Q\right)`$, then we have by condition xiv) from Definition 9 that
$$q\left(str\left(ge\right)\right)=hq\left(e_2\right)$$
for some $`e_2D_{xy}`$ and some $`hH`$. But $`e_2`$ belongs to the same coset in $`\mathrm{\Gamma }K_1^{str}\left(Q\right)\mathrm{\Gamma }`$ as $`e`$, thus $`e_2=str\left(e\right)`$ which proves the claim for $`e`$.
If $`f`$ is adjacent to one 1-labeled edge $`e`$ and $`q\left(str\left(ge\right)\right)=hq\left(str\left(e\right)\right)`$, then $`q\left(str\left(gf\right)\right)=hq\left(str\left(f\right)\right)`$ because the homotopy of $`f`$ resp. $`gf`$ just followed the homotopy of $`e`$ resp. $`ge`$: e.g. if $`_1f=_1e`$ and $`q\left(str\left(ge\right)\right)q_{}\left(\alpha \right)q\left(str\left(e\right)\right)q_{}\left(\beta \right)`$ with $`\alpha ,\beta K_1\left(_0Q\right)`$, then $`q\left(str\left(gf\right)\right)q_{}\left(\alpha \right)q\left(str\left(f\right)\right)`$. Similarly if $`f`$ is adjacent to two 1-labeled edges.
Finally, if a 0-labeled straight 1-simplex $`f`$ is not adjacent to a 1-labeled edge, then $`str\left(f\right)=f`$ and $`str\left(gf\right)=gf`$, which implies $`str\left(gf\right)=gstr\left(f\right)`$ and $`q\left(str\left(gf\right)\right)=q_{}\left(g\right)str\left(f\right)`$.
Thus we have proved $`q\left(str\left(gf\right)\right)=hq\left(str\left(f\right)\right)`$ with some $`hH`$ for any 0-labeled edges $`f`$.
Thus $`qstr`$ is well-defined and satisfies i) by Lemma 12. To prove ii), we first observe that, if $`_{iI}a_i\tau _i`$ represents $`[Q,Q]`$, then, by Observation 6c) and condition i) from Lemma 12 (together with $`qid`$), we have that
$$\underset{iI}{}a_iqstr\left(\tau _i\right)=\underset{i=1}{\overset{r}{}}a_iq\left(\tau _i^{}\right)$$
represents $`[Q,Q]`$ and the claim follows. Thus it suffices to check: if $`_{iI}a_i\tau _i1`$ is (relatively) homologous to $`_{jJ}b_j\kappa _j1`$, then $`qstr\left(_{iI}a_i\tau _i1\right)`$ is (relatively) homologous to $`qstr\left(_{jJ}b_j\kappa _j1\right)`$.
So let
$$\underset{iI}{}a_i\tau _i1\underset{jJ}{}b_j\kappa _j1=\underset{kK}{}c_k\eta _k1modC_{}^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐$$
for some chain $`_{kK}c_k\eta _k1C_{}^{simp,inf}\left(K\left(Q\right)\right)_{๐G}๐`$. In complete analogy with Lemma 12, we may extend $`str`$ to the simplicial set built by the $`g\eta _k`$โs, their faces and degenerations, and obtain a singular chain $`q\left(str\left(_{kK}c_k\eta _k\right)\right)`$ whose boundary is
$$qstr\left(\underset{kK}{}c_k\eta _k\right)=qstr\left(\underset{iI}{}a_i\tau _i1\right)qstr\left(\underset{jJ}{}b_j\kappa _j1\right)modC_{}^{simp,inf}\left(HK^{str}\left(_1Q\right)\right)_{๐H}๐.$$
This gives the first claim of ii). The second claim of ii) follows because $``$ maps $`[Q,Q]`$ to $`\left[Q\right]`$.
### 5.5 Removal of 0-homologous chains
###### Definition 11.
Let $`Q`$ be an n-dimensional compact manifold with boundary $`Q`$. We define $`rmv:S_{}\left(Q\right)S_{}\left(Q\right)`$ by
$$rmv\left(\sigma \right)=0$$
if $`\sigma `$ is weakly degenerate ( Definition 7) and
$$rmv\left(\sigma \right)=\sigma $$
else.
###### Lemma 13.
Assume that $`Q`$ is a $`n`$-dimensional compact manifold with boundary $`Q`$. Let $`K_{}^{str}\left(Q\right)S_{}\left(Q\right)`$ satisfy the conditions i)-viii) from Lemma 11. Then
$$rmv:C_{}^{simp}(K^{str}\left(Q\right),K^{str}\left(_0Q\right)K^{str}\left(_1Q\right))C_{}^{simp}(K^{str}\left(Q\right),K^{str}\left(_0Q\right)K^{str}\left(_1Q\right)),$$
defined by
$$rmv\left(\left[\sigma \right]\right):=\left[rmv\left(\sigma \right)\right],$$
is a well-defined chain map. Moreover, if
$$\underset{j=1}{\overset{r}{}}a_j\tau _jC_{}^{simp}(K^{str}\left(Q\right),K^{str}\left(_0Q\right)K^{str}\left(_1Q\right))C_{}^{sing}(Q,Q)$$
represents $`[Q,Q]`$, then $`_{j=1}^ra_jrmv\left(\tau _j\right)`$ represents $`[Q,Q]`$.
###### Proof.
If $`\sigma K^{str}\left(_0Q\right)K^{str}\left(_1Q\right)`$, then $`rmv\left(\sigma \right)K^{str}\left(_0Q\right)K^{str}\left(_1Q\right)`$, thus $`rmv`$ is well-defined.
In a first step, we prove that $`rmv`$ is a chain map.
Assume that $`rmv\left(\sigma \right)=0`$.
If $`\sigma `$ has image in $`Q`$, then $`rmv\left(\sigma \right)=0`$ and $`rmv\left(\sigma \right)=0`$, thus $`rmv\left(\sigma \right)=rmv\left(\sigma \right)`$.
If some edge $`e`$ of $`\sigma `$, say connecting the $`i`$-th and $`j`$-th vertex, is a constant loop, then all faces of $`\sigma `$ except possibly $`_i\sigma `$ and $`_j\sigma `$ have a constant edge. Thus $`rmv\left(_k\sigma \right)=0`$ if $`k\{i,j\}`$. Moreover, since $`e`$ is constant, corresponding edges of $`_i\sigma `$ and $`_j\sigma `$ are homotopic rel. boundary and thus agree (possibly up to orientation) by condition v) from Lemma 11. By induction on the dimension of subsimplices we get, again using condition v) from Lemma 11, that $`_i\sigma =\left(1\right)^{ij}_j\sigma `$. Altogether we get $`rmv\left(\sigma \right)=0`$, thus $`rmv\left(\sigma \right)=rmv\left(\sigma \right)`$.
Assume that $`rmv\left(\sigma \right)=\sigma `$. Since no edge of $`\sigma `$ is a constant loop, of course also no edge of a face $`_i\sigma `$ is a constant loop. If the image of $`_i\sigma `$ is not contained in $`Q`$, this implies $`rmv\left(_i\sigma \right)=_i\sigma =_irmv\left(\sigma \right)`$. If $`_i\sigma `$ has image in $`Q`$, then of course $`\left[_i\sigma \right]=\left[0\right]=\left[_irmv\left(\sigma \right)\right]`$, which implies $`rmv\left(_i\sigma \right)=_irmv\left(\sigma \right)`$.
Now we prove that $`rmv`$ sends relative fundamental cycles to relative fundamental cycles.
Let $`_{j=1}^ra_j\tau _j`$ be a straight relative cycle, representing the relative homology class $`[Q,Q]`$.
We denote by $`J_1\{1,\mathrm{},r\}`$ the indices of those $`\tau _j`$ which have a constant edge. The sum $`_{jJ_1}a_j\tau _j`$ is a relatively 0-homologous relative cycle. Indeed, each face of $`_i\tau _k`$ not contained in $`Q`$ has to cancel against some face of some $`\tau _l`$, because $`_{j=1}^ra_j\tau _j`$ is a relative cycle. If $`_i\tau _k`$ is degenerate, then necessarily $`lJ_1`$. Moreover, if $`\tau _k`$ is degenerate and $`_i\tau _k`$ is nondegenerate, then we have proved in the first step that $`_i\tau _k`$ cancels against some $`_j\tau _k`$.
Thus $`_{jJ_1}a_j\tau _j`$ represents some relative homology class. The isomorphism $`H_n\left(C_{}^{sing}(Q,Q)\right)๐`$ is given by pairing with the volume form of an arbitrary Riemannian metric. After smoothing the relative cycle, we can apply Sardโs lemma, and conclude that degenerate simplices have volume 0. Thus $`_{jJ_1}a_j\tau _j`$ is 0-homologous.
We denote by $`J_2\{1,\mathrm{},r\}`$ the indices of those $`\tau _j`$ which are contained in $`Q`$. For $`jJ_2`$ we have $`\left[\tau _j\right]=\left[0\right]C_{}^{sing}(Q,Q)`$.
Thus $`_{jJ_1J_2}a_j\tau _j`$ is another representative of the homology class $`[Q,Q]`$. But, by Definition 11, it also represents $`\left(rmv\right)_{}\left([Q,Q]\right)`$. โ
Consider a subgroup $`H\mathrm{\Pi }\left(K\left(A\right)\right)`$ for some $`AQ`$. (E.g. $`A=q\left(_0Q\right)`$ in the setting of Construction 1, and $`H=q_{}\left(\mathrm{\Pi }\left(K\left(_0Q\right)\right)\right)\mathrm{\Pi }\left(K\left(A\right)\right)`$.
A 1-simplex $`e`$ is a constant loop if and only if $`he`$ is a constant loop for all $`hH`$. This implies that a simplex $`\sigma `$ is degenerate if and only if $`h\sigma `$ is degenerate for all $`h\sigma `$. Moreover, $`H`$ maps simplices in $`Q`$ to simplices in $`Q`$. Thus $`rmv\left(\sigma \right)=0`$ if and only if $`rmv\left(h\sigma \right)=0`$ for all $`hH`$, that is, $`rmv`$ is well defined on $`C_{}^{simp,inf}\left(HK^{str}\left(Q\right)\right)_{๐H}๐`$ for each subgroup $`H`$.
###### Lemma 14.
Assume that $`Q`$ is a $`n`$-dimensional compact manifold with boundary $`Q`$. Let the assumptions of Corollary 3 be satisfied. Then we can extend $`rmv`$ to a well-defined chain map
$$rmv:C_{}^{simp,inf}(HK^{str}\left(Q\right),HK^{str}\left(_1Q\right))_{๐H}๐C_{}^{simp,inf}(HK^{str}\left(Q\right),HK^{str}\left(_1Q\right))_{๐H}๐$$
by defining
$$rmv\left(\sigma z\right)=\left\{\begin{array}{cc}0:& rmv\left(\sigma \right)=0\\ \sigma z:& \text{ else }\end{array}\right\}.$$
Moreover, if $`_{jJ}a_j\tau _j1C_{}^{simp,inf}(HK^{str}\left(Q\right),HK^{str}\left(_1Q\right))_{๐H}๐`$ represents the image of $`[Q,Q]1`$, then $`_Ja_jrmv\left(\tau _j1\right)`$ represents the image of $`[Q,Q]1`$.
###### Proof.
Well-definedness of $`rmv`$ follows from the remark before Lemma 14. The same proof as for Lemma 13 shows that $`rmv`$ is a chain map.
If $`_{j=1}^ra_j\tau _j`$ represents $`[Q,Q]`$, then the second claim follows from Lemma 13. If $`_{jJ}a_j\tau _j1`$ is homologous to $`_{i=1}^sb_i\kappa _i1`$ and $`_{i=1}^sb_i\kappa _i`$ represents $`[Q,Q]`$, then (because $`rmv`$ is a chain map) $`rmv\left(_{jJ}a_j\tau _j1\right)`$ is homologous to $`rmv\left(_{i=1}^sb_i\kappa _i1\right)`$, which implies the second claim.โ
The proof of Theorem 1 will pursue the idea of straightening a given cycle such that many simplices either become weakly degenerate or will have an edge in $`_0Q`$. In the first case, they will disappear after application of $`rmv`$. In the second case, they disappear in view of the following observation, which is a variant of an argument used in .
###### Lemma 15.
a) Let Assumption I be satisfied for a manifold $`Q`$ and consider the action of $`G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$ on $`K\left(Q\right)`$. Let $`\sigma K\left(Q\right)`$ be a simplex.
If $`str\left(\sigma \right)`$ has an edge in $`_0Q`$, then
$$str\left(\sigma 1\right)=0C_{}^{simp,inf}\left(K\left(Q\right)\right)_{๐G}๐.$$
b) If $`q:QQ`$ is given by Construction 1, $`H=q_{}\left(G\right)`$, and $`\sigma K\left(Q\right)`$ a simplex such that $`q\left(str\left(\sigma \right)\right)`$ has an edge in $`q\left(_0Q\right)`$, then
$$q\left(str\left(\sigma 1\right)\right)=0C_{}^{simp,inf}\left(K\left(Q\right)\right)_{๐H}๐.$$
###### Proof.
a) Let $`\gamma `$ be the edge of $`str\left(\sigma \right)`$ with image in $`_0Q`$, then $`g=\{\gamma ,\overline{\gamma }\}`$ is an element of $`G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$ and $`gstr\left(\sigma \right)=\overline{str\left(\sigma \right)}`$. In the simplicial chain complex $`C_{}^{simp,inf}\left(K\left(Q\right)\right)`$, one has $`\overline{str\left(\sigma \right)}=str\left(\sigma \right)`$. Thus $`gstr\left(\sigma \right)=str\left(\sigma \right)`$, which implies $`str\left(\sigma 1\right)=str\left(\sigma \right)1=0`$.
b) Let $`\gamma `$ be the edge of $`q\left(str\left(\sigma \right)\right)`$ with image in $`q\left(_0Q\right)`$. Let $`\gamma ^{}`$ be the corresponding edge of $`str\left(\sigma \right)`$. Let $`g=\{\gamma ^{},\overline{\gamma }^{}\}G`$ and $`h=q_{}\left(g\right)=\{\gamma ,\overline{\gamma }\}H`$. The same argument as in a) shows $`hq\left(str\left(\sigma \right)\right)=q\left(str\left(\sigma \right)\right)`$. โ
## 6 Proof of Main Theorem
### 6.1 Motivating examples
Example 1: Let $`M`$ be a connected, orientable, hyperbolic $`n`$-manifold, $`F`$ an orientable, geodesic n-1-submanifold, $`Q=\overline{MF}`$. For simplicity we assume that $`M`$ and $`F`$ are closed, thus $`Q`$ is a hyperbolic manifold with geodesic boundary $`_1Q\mathrm{}`$, and $`_0Q=\mathrm{}`$.
Outline of proof of $`M_F^{norm}\frac{1}{n+1}Q`$: Start with a fundamental cycle $`_{i=1}^ra_i\sigma _i`$ of $`M`$, such that $`\sigma _1,\mathrm{},\sigma _r`$ are normal to $`F`$. Since we want to consider laminations without isolated leaves, we replace $`F`$ by a trivially foliated product neighborhood $``$. We can assume after a suitable homotopy that each component of $`\sigma _i^1\left(Q\right)`$ either contains no vertex of $`\mathrm{\Delta }^n`$ or consists of exactly one vertex, and that each vertex of $`\mathrm{\Delta }^n`$ belongs to $`\sigma _i^1\left(\right)`$, for $`i=1,\mathrm{},r`$.
Each $`\sigma _i^1\left(Q\right)`$ consists of polytopes, which can each be further triangulated (without introducing new vertices) in a coherent way (i.e., such that boundary cancellations between different $`\sigma _i`$โs will remain) into $`\tau _{i1},\mathrm{},\tau _{is\left(i\right)}`$.
$`_{i=1}^ra_i\left(\tau _{i1}+\mathrm{}+\tau _{is\left(i\right)}\right)`$ is a relative fundamental cycle for $`Q`$.
For each $`\sigma _i`$, preimages of the boundary leaves of $``$ cut $`\mathrm{\Delta }^n`$ into regions which we colour with black (components of $`\sigma _i^1\left(\right)`$) and white (components of $`\sigma _i^1\left(Q\right)`$). Moreover, if $`\sigma _i^1\left(Q\right)`$ contains vertices, these vertices are coloured black. This is a canonical colouring (Definition 4).
The edges of the simplices $`\tau _{i,j}`$ fall into two classes: โold edgesโ, i.e. subarcs of edges of $`\sigma _i`$, and โnew edgesโ, which are contained in the interior of some subsimplex of $`\sigma _i`$ of dimension $`2`$.
We define the labeling of the edges of $`\tau _{ij}`$ such that โold edgesโ are labelled 1 and โnew edgesโ are labelled 0. This is an admissible labeling (Definition 10). With this labeling, we apply the straightening procedure<sup>11</sup><sup>11</sup>11Under the assumptions of Example 1, straight simplices can be chosen to be the totally geodesic simplices with vertices in $`S_0^{str}\left(Q\right)`$. Distinguished simplices are chosen according to Observation 8. from Section 5 to get a straight cycle $`_{i=1}^ra_i\left(str\left(\tau _{i1}\right)+\mathrm{}+str\left(\tau _{is\left(i\right)}\right)\right)`$. (Thus โold edgesโ are straightened to distinguished 1-simplices.)
After straightening we remove all weakly degenerate simplices (simplices contained in $`Q`$ or having a constant edge), i.e. we apply the map $`rmv`$ from Section 5.4. By Lemma 13, this does not change the homology class. In particular, the boundary of the relative cycle, $`_{i,j}a_irmv\left(str\left(\tau _{ij}\right)\right)`$ still represents the fundamental class $`\left[Q\right]`$ of $`Q`$.
Claim: for each $`\sigma _i`$, after straightening there remain at most $`n+1`$ faces of nondegenerate simplices $`str\left(\tau _{ij}\right)`$ contributing to $`_{i,j}a_irmv\left(str\left(\tau _{ij}\right)\right)`$.
In view of Lemma 10, it suffices to show the following subclaim: if, for a fixed $`i`$, $`T_1=_{k_1}\tau _{ij_1},T_2=_{k_2}\tau _{ij_2}`$ are faces of some $`\tau _{ij_1}`$ resp. $`\tau _{ij_2}`$ such that $`T_1,T_2`$ have a white-parallel arc (Definition 6), then $`rmv\left(str\left(\tau _{ij_1}\right)\right)=0,rmv\left(str\left(\tau _{ij_2}\right)\right)=0`$. and in particular the corresponding straightened faces $`str\left(T_1\right),str\left(T_2\right)`$ do not occur (with nonzero coefficient) in $`_{i,j}rmv\left(str\left(\tau _{ij}\right)\right)`$. (Notational remark: for a subsimplex $`T`$ of an affine subset $`S\mathrm{\Delta }^n`$ we get a singular simplex $`\sigma _i_T`$ by restricting $`\sigma _i`$ to $`T`$. We denote by $`str\left(T\right)`$ the straightening of $`\sigma _i_T`$.)
To prove the subclaim, let $`W`$ be the white region of $`\mathrm{\Delta }^n`$ containing $`T_1`$ and $`T_2`$ in its boundary. By assumption of the subclaim, there is a white square bounded by two arcs $`e_1T_1,e_2T_2`$ and two arcs $`f_1,f_2`$ which are subarcs of edges of $`\mathrm{\Delta }^n`$. (The square is a formal sum of two triangles, $`U_1+U_2`$, which are 2-dimensional faces of some $`\tau _{ij}`$โs.)
We want to show that all edges of $`str\left(\tau _{ij_1}\right)`$ belong to $`S_1^{str}\left(Q\right)`$. Note that $`T_1,T_2W`$ are mapped to $`Q`$. Let $`x_1S_0^{str}\left(Q\right)`$ resp. $`x_2S_0^{str}\left(Q\right)`$ be the unique elements of $`S_0^{str}\left(Q\right)`$ in the same connected component $`C_1`$ resp. $`C_2`$ of $`Q`$ as $`\sigma _i\left(T_1\right)`$ resp. $`\sigma _i\left(T_2\right)`$. In particular $`_0str\left(e_1\right)=x_1=_1str\left(e_1\right)`$ and $`_0str\left(e_2\right)=x_2=_1str\left(e_2\right)`$. Thus $`e_1`$ and $`e_2`$ are straightened to loops $`str\left(e_1\right)`$ resp. $`str\left(e_2\right)`$ based at $`x_1`$ resp. $`x_2`$. The straightenings of the other two arcs, $`str\left(f_1\right),str\left(f_2\right)`$ connect $`x_1`$ to $`x_2`$, and they are distinguished 1-simplices because they arise as straightenings of โold edgesโ. Thus $`str\left(f_1\right)=str\left(f_2\right)`$, by uniqueness of distinguished 1-simplices in each coset $`\mathrm{\Gamma }K_1^{str}\left(Q\right)\mathrm{\Gamma }`$ of $`\mathrm{\Gamma }=\mathrm{\Omega }\left(Q\right)`$. This is why we have performed the straightening construction in Section 5 such that there should be only one distinguished 1-simplex, in each coset, for any given pair of connected components.
This means that the square is straightened to a cylinder.
But $`(Q,Q)`$ is acylindrical, thus either both $`str\left(e_1\right)`$ and $`str\left(e_2\right)`$ are constant (in which case $`rmv\left(str\left(\tau _{ij_1}\right)\right)=rmv\left(str\left(\tau _{ij_2}\right)\right)=0`$), or the cylinder must be homotopic into $`Q`$. In the latter case, $`str\left(f_1\right)`$ must (be homotopic into and therefore) be contained in $`Q`$. In particular, $`_0str\left(f_1\right)`$ and $`_1str\left(f_1\right)`$ belong to the same component of $`Q`$. This implies $`_0str\left(f_1\right)=_1str\left(f_1\right)`$. Since $`str\left(f_1\right)`$ is a distinguished 1-simplex, this implies that $`str\left(f_1\right)`$ is constant.
Let $`P_1,P_2`$ be the affine planes whose intersections with $`\mathrm{\Delta }^n`$ contain $`T_1`$ resp. $`T_2`$. We have now that there is an arc $`f_1`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_2\mathrm{\Delta }^n`$ such that $`str\left(f_1\right)`$ is contained in $`Q`$. This implies that for each other arc $`f`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_2\mathrm{\Delta }^n`$ its straightening $`str\left(f\right)`$ must (be homotopic into and therefore) be contained in $`Q`$.
If $`P_1`$ and $`P_2`$ are of the same type, then all edges of $`str\left(\tau _{ij_1}\right)`$ connect $`P_1\mathrm{\Delta }^n`$ to $`P_2\mathrm{\Delta }^n`$, hence all edges of $`str\left(\tau _{ij_1}\right)`$ belong to $`S_1^{str}\left(Q\right)`$. If $`P_1`$ and $`P_2`$ are not of the same type, then existence of a parallel arc implies that at least one of them, say $`P_1`$, must be of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`k\{0,n1\}`$. Then, if $`P_3`$ is any other plane bounding $`W`$, it follows from Corollary 2 that $`P_3`$ has a white-parallel arc with $`P_1`$. Thus, repeating the argument in the last paragraph with $`P_1`$ and $`P_3`$ in place of $`P_1`$ and $`P_2`$, we obtain that for each arc $`f`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_3\mathrm{\Delta }^n`$ its straightening $`str\left(f\right)`$ must (be homotopic into and therefore) be contained in $`Q`$. Hence, for each $`\tau _{ij_1}`$ in the chosen triangulation of $`W`$, its 1-skeleton is straightened into $`Q`$.
Since straight simplices $`\sigma `$ (of dimension $`2`$) with $`\sigma `$ in the geodesic boundary $`Q`$, must be in $`Q`$, this implies by induction that the $`k`$-skeleton of $`str\left(\tau _{ij_1}\right)`$ is in $`Q`$, for each $`k`$. In particular, $`str\left(\tau _{ij_1}\right)S_n^{str}\left(Q\right)`$. Hence $`rmv\left(str\left(\tau _{ij_1}\right)\right)=0`$. We have proved the subclaim.
By Lemma 10, the subclaim implies the claim. Since $`_{i=1}^ra_i_jrmv\left(str\left(\tau _{ij}\right)\right)`$ represents the fundamental class $`\left[Q\right]`$, we conclude $`Q\left(n+1\right)_{i=1}^ra_i`$.
The simplifications of Example 1 in comparison to the proof in Section 6.2 are essentially all due to the fact that $`_0Q=\mathrm{}`$. We remark that in Example 2, if $`F`$ is not geodesic, then $`QN`$ and thus $`_0Q\mathrm{}`$ (even though $`M=\mathrm{},F=\mathrm{}`$). Thus the generalization to $`_0Q\mathrm{}`$ would be necessary even if one only wanted to consider closed manfifolds $`M`$ and $`F`$.
Example 2: Let $`M`$ be a connected, closed, hyperbolic 3-manifold, $`FM`$ a closed, incompressible surface, $`N=\overline{MF},Q=Guts\left(N\right)`$.
Outline of proof of $`M_F^{norm}\frac{1}{4}Q`$: Start with a fundamental cycle $`_{i=1}^ra_i\sigma _i`$ of $`M`$, such that $`\sigma _1,\mathrm{},\sigma _r`$ are normal to $`F`$. As in Example 1 we get a relative fundamental cycle $`_{i=1}^ra_i\left(\tau _{i1}+\mathrm{}+\tau _{is\left(i\right)}\right)`$ of $`N`$. We can not apply the argument from Example 1 to $`N`$ because $`N`$ is not acylindrical. Therefore we would like to work with a relative fundamental cycle for the acylindrical manifold $`Q`$.
$`N`$ is aspherical. Using Lemma 2, we can assume that all $`\tau _{ij}`$ belong to $`K\left(N\right)`$. Then we can apply the retraction $`r`$ from Lemma 5. Since $`r`$ is only defined after tensoring $`._{๐G}๐`$, we get $`r\left(\tau _{ij}1\right)=\kappa _{ij}1`$ with $`\kappa _{ij}K\left(Q\right)`$ only determined up to choosing one $`\kappa _{ij}`$ in its $`G`$-orbit.
Since $`Q`$ is aspherical, we have $`K\left(Q\right)=\widehat{K}\left(Q\right)`$, that is, the $`\kappa _{ij}`$ can be considered as simplices in $`Q`$ and we can apply Lemma 6b) to obtain a fundamental cycle for $`Q`$.
The rest of the proof then basically boils down to copying the proof of Example 1 (with $`\tau _{ij}`$ replaced by $`\kappa _{ij}`$), but taking care of the ambiguity in the choice of $`\kappa _{ij}`$. The details can be found in the next section.
### 6.2 Proof
We refer to the introduction for the statement of Theorem 1 and the relevant definitions. In this section we are going to prove Theorem 1.
Proof:
If $`n=1`$, then Theorem 1 is trivially true. Hence we can restrict to the case $`n2`$.
If $`_1Q`$ were empty, then $`Q=_0Q`$ and amenability of $`\pi _1_0Q`$ would imply $`Q=0`$, in particular Theorem 1 would be trivially true. Hence we can restrict to the case $`_1Q\mathrm{}`$. In particular, $`Q`$ satisfies Assumption I from Section 5.
Consider a relative cycle $`_{i=1}^ra_i\sigma _i`$, representing $`[M,M]`$, such that $`\sigma _1,\mathrm{},\sigma _r`$ are normal to $``$. Our aim is to show: $`_{i=1}^ra_i\frac{1}{n+1}Q`$.
Denote
$$N=\overline{M}.$$
Since each $`\sigma _i`$ is normal to $``$, we have for each $`i=1,\mathrm{},r`$ that, after application of a simplicial homeomorphism $`h_i:\mathrm{\Delta }^n\mathrm{\Delta }^n`$, the image of $`\sigma _i^1\left(N\right)`$ consists of polytopes, which can each be further triangulated in a coherent way (i.e., such that boundary cancellations between different $`\sigma _i`$โs will remain) into simplices $`\theta _{ij},j\widehat{J}_i`$. (It is possible that $`\widehat{J}_i=\mathrm{}`$, because $`N`$ may be noncompact.) We choose these triangulations of the $`\sigma _i^1\left(N\right)`$ to be minimal (Definition 6), that is, we do not introduce new vertices. (Indeed, compatible minimal triangulations of the $`\sigma _i^1\left(N\right)`$ do exist: one starts with common minimal triangulations of the common faces and extends them to minimal triangulations of each polytope.)
Because boundary cancellations are preserved, we have that
$$\underset{i=1}{\overset{r}{}}a_i\underset{j\widehat{J}_i}{}\theta _{ij}$$
is a countable (possibly infinite) relative cycle representing the fundamental class $`[N,N]`$ in the sense of section 3.2.
We fix a sufficient set of cancellations $`๐^M`$ for the relative cycle $`_{i=1}^ra_i\sigma _i`$, in the sense of Definition 8. This induces a sufficient set of cancellations $`๐^N`$ for the relative cycle $`_{i=1}^r_{j\widehat{J}_i}a_i\theta _{ij}`$.
If $`M`$ is a leaf of $``$, then all faces of $`z`$ contributing to $`z`$ are contained in $`N`$. We call these faces exterior faces. We can assume that, for each $`i`$,
\- each component of $`\sigma _i^1\left(N\right)`$ either contains no vertex of $`\mathrm{\Delta }^n`$, or consists of exactly one vertex, or consists of an exterior face,
\- and each vertex of $`\mathrm{\Delta }^n`$ belongs to $`\sigma _i^1\left(\right)`$.
Indeed, by a small homotopy of the relative fundamental cycle $`_{i=1}^ra_i\sigma _i`$, preserving normality, we can obtain that no component of $`\sigma _i^1\left(N\right)`$ contains a vertex of $`\mathrm{\Delta }^n`$, except for exterior faces. Afterwards, if some vertices of $`_{i=1}^ra_i\sigma _i`$ do not belong to $``$, we may homotope a small neighborhood of the vertex, until the vertex (and no other point of the neighborhood) meets $`N`$. This, of course, preserves normality to $``$.
Since each $`\sigma _i`$ is normal to $``$, in particular each $`\sigma _i`$ is normal to the union of boundary leaves
$$_1N:=\overline{N\left(MN\right)}.$$
Thus for each $`\sigma _i`$, after application of a simplicial homeomorphism $`h_i:\mathrm{\Delta }^n\mathrm{\Delta }^n`$, the image of $`\sigma _i^1\left(_1N\right)`$ consists of a (possibly infinite) set
$$Q_1,Q_2,\mathrm{}\mathrm{\Delta }^n,$$
such that
$$Q_i=P_i\mathrm{\Delta }^n$$
for some affine hyperplanes $`P_1,P_2,\mathrm{}`$. We define a colouring by declaring that (images under $`h_i`$ of) components of
$$\sigma _i^1\left(int\left(N\right)\right):=\sigma _i^1\left(N_1N\right)$$
are coloured white and (images under $`h_i`$ of) components of $`\sigma _i^1\left(\right)`$ are coloured black. (In particular, all $`Q_i`$ are coloured black.) Since we assume that all vertices of $`\mathrm{\Delta }^n`$ belong to $`\sigma _i^1\left(\right)`$, and since each boundary leaf is adjacent to at least one component of $`\sigma _i^1\left(int\left(N\right)\right)`$, this is a canonical colouring (Definition 4).
By Lemma 2a), we can homotope the relative cycle $`_{i=1}^r_{j\widehat{J}_i}\theta _{ij}C_n^{inf}(N,N)`$ to a relative cycle
$$\underset{i=1}{\overset{r}{}}a_i\underset{j\widehat{J}_i}{}\widehat{\theta }_{ij}$$
such that each $`\widehat{\theta }_{ij}`$ is a simplex of $`\widehat{K}\left(N\right)`$, as defined in Section 3.2, and such that the boundary $`_{i=1}^r_{j\widehat{J}_i}\theta _{ij}`$ is homotoped into $`\widehat{K}\left(N\right)`$. Then consider
$$\underset{i=1}{\overset{r}{}}\underset{j\widehat{J}_i}{}a_i\tau _{ij}:=\underset{i=1}{\overset{r}{}}\underset{j\widehat{J}_i}{}a_ip\left(\widehat{\theta }_{ij}\right)C_n^{simp,inf}\left(K\left(N\right)\right),$$
where $`p:\widehat{K}\left(N\right)K\left(N\right)`$ is the projection defined at the end of Section 3.2, and $`\tau _{ij}:=p\left(\widehat{\theta }_{ij}\right)`$ for all $`i,j`$.
Consider $`QN`$ as in the assumptions of Theorem 1. We denote
$$G:=\mathrm{\Pi }\left(K\left(_0Q\right)\right).$$
We have by assumption that $`N=QR`$ is an essential decomposition (as defined in the introduction), which means exactly that the assumptions of Lemma 5 are satisfied. Thus, according to Lemma 5, there exists a retraction
$$r:C_n^{simp,inf}\left(K\left(N\right)\right)_{๐G}๐C_n^{simp,inf}\left(K\left(Q\right)\right)_{๐G}๐$$
for $`n2`$, mapping $`C_n^{simp,inf}\left(GK\left(N\right)\right)_{๐G}๐`$ to $`C_n^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐`$, such that, for each simplex $`\tau _{ij}K\left(N\right)`$, we either have $`r\left(\tau _{ij}1\right)=0`$ or
$$r\left(\tau _{ij}1\right)=\kappa _{ij}1$$
for some simplex $`\kappa _{ij}K\left(Q\right).`$ (Recall that we assume from the beginning $`n2`$.)
Thus
$$r\left(\underset{i=1}{\overset{r}{}}a_i\underset{j\widehat{J}_i}{}\tau _{ij}1\right)=\underset{i=1}{\overset{r}{}}a_i\underset{jJ_i}{}\kappa _{ij}1$$
with $`J_i\widehat{J}_i`$ for all $`i`$. (It may still be possible that $`J_i=\mathrm{}`$.)
We remark that $`\kappa _{ij}`$ is only determined up to choosing one $`\kappa _{ij}`$ in its G-orbit.
Since $`r`$ is a chain map, we get a sufficient set of cancellations for
$`_{i=1}^ra_i_{jJ_i}\kappa _{ij}1`$ by
$$๐^Q:=\{(_k\kappa _{i_1j_1}1,_l\kappa _{i_2j_2}1):(_k\tau _{i_1j_1},_l\tau _{i_2j_2})๐^N\}.$$
By assumption, $`Q`$ is aspherical. We can therefore apply Lemma 6 and have that
$$\left(\underset{i=1}{\overset{r}{}}a_i\underset{jJ_i}{}\kappa _{ij}1\right)C_{}^{simp,inf}\left(GK\left(_1Q\right)\right)_{๐G}๐$$
represents (the image of) $`\left[Q\right]1`$.
Lemma 4a) gives that $`G`$ is amenable. Together with Lemma 7 this implies
$$Q\underset{i=1}{\overset{r}{}}a_i\left(n+1\right)J_i.$$
In the remainder of the proof, we will use Lemma 14 to improve this inequality and, in particular, get rid of the unspecified (possibly infinite) numbers $`J_i`$.
$`Q,Q,_0Q,_1Q`$ satisfy Assumption I from Section 5. Thus there exists a simplicial set
$$K_{}^{str}\left(Q\right)S_{}\left(Q\right)$$
satisfying conditions i)-viii) from Lemma 11, and a set
$$DK_1^{str}\left(Q\right)$$
of distinguished 1-simplices (Definition 9).
Recall that, for each $`i`$,
$$\underset{j\widehat{J}_i}{}\theta _{i,j}$$
was defined by choosing a triangulation of $`\sigma _i^1\left(N\right)`$. The simplices $`\theta _{i,j}`$ thus have โold edgesโ, i.e. subarcs of edges of $`\sigma _i`$, and โnew edgesโ, whose interior is contained in the interior of some subsimplex of $`\sigma _i`$ of dimension $`2`$.
Associated to $`z=_{i=1}^ra_i_{j\widehat{J}_i}\theta _{ij}`$ and $`๐^N`$ (and an arbitrary minimal presentation of $`z`$) are, by Definition 8, simplicial sets $`\mathrm{{\rm Y}}^N,\mathrm{{\rm Y}}^N`$.
The only possibility that two โold edgesโ have a vertex in $`\mathrm{{\rm Y}}^N`$ in common is that this vertex is a vertex of $`\sigma _i`$.
So the labeling of edges of
$$\underset{i=1}{\overset{r}{}}a_i\underset{j\widehat{J}_i}{}\theta _{ij}$$
by labeling โold edgesโ not containig a vertex of any $`\sigma _i`$ with label 1 and all other edges with label 0 is an admissible labeling (Definition 10).
Associated to
$$w=\underset{i=1}{\overset{r}{}}a_i\underset{jJ_i}{}\kappa _{ij}1$$
and $`๐^Q`$ (and an arbitrary minimal presentation of $`w`$) there are simplicial sets $`\mathrm{{\rm Y}},\mathrm{{\rm Y}}`$. By our definition of $`๐^Q`$, $`\mathrm{{\rm Y}}`$ is isomorphic to a simplicial subset of $`\mathrm{{\rm Y}}^N`$, namely to the subset generated by the set
$$\{\tau \mathrm{{\rm Y}}^N:r\left(\tau 1\right)0\}$$
together with all iterated faces and degenerations. In particular, the admissible 0-1-labeling of $`\mathrm{{\rm Y}}^N`$ induces an admissible 0-1-labeling of $`\mathrm{{\rm Y}}`$.
By Construction 1, there is a map of triples $`q:(Q,Q,_1Q)(Q,Q,_1Q)`$ which is (as a map of triples) homotopic to the identity, and such that $`q\left(_0QC\right)`$ is path-connected for each path-component $`C`$ of $`Q`$.
We denote
$$A:=q\left(_0Q\right),H:=q_{}\left(G\right)=q_{}\left(\mathrm{\Pi }\left(K\left(_0Q\right)\right)\right)\mathrm{\Pi }\left(K\left(A\right)\right).$$
We observe that $`H`$ is a quotient of $`G`$, hence amenable, even though $`\mathrm{\Pi }\left(K\left(A\right)\right)`$ need not be amenable.
Let $`\widehat{\mathrm{{\rm Y}}},\widehat{\mathrm{{\rm Y}}}`$ be defined by Observation 9. By Corollary 3, there is a chain map
$$qstr:C_{}^{simp,inf}\left(\widehat{\mathrm{{\rm Y}}}\right)_{๐G}๐C_{}^{simp,inf}\left(HK^{str}\left(Q\right)\right)_{๐H}๐,$$
mapping $`C_{}^{simp,inf}\left(\widehat{\mathrm{{\rm Y}}}\right)_{๐G}๐`$ to $`C_{}^{simp,inf}\left(HK^{str}\left(_1Q\right)\right)_{๐H}๐`$, such that
$$\underset{i=1}{\overset{r}{}}a_i\underset{jJ_i}{}q\left(str\left(\kappa _{ij}\right)\right)1$$
represents (the image of) $`\left[Q\right]1`$ and such that 1-labeled edges are mapped to distinguished 1-simplices. (We keep in mind that $`\kappa _{ij}`$ is only determined up to G-action, thus $`q\left(str\left(\kappa _{ij}\right)\right)`$ is determined only up to choosing one simplex in its $`H`$-orbit.)
We then apply Lemma 14 to get the cycle
$$\underset{i=1}{\overset{r}{}}a_i\underset{jJ_i}{}rmv\left(q\left(str\left(\kappa _{ij}\right)\right)1\right)C_{}^{simp,inf}\left(HK^{str}\left(_1Q\right)\right)_{๐H}๐$$
representing (the image of) $`\left[Q\right]1`$. We want to show that it is actually a finite chain of $`l^1`$-norm at most $`\left(n+1\right)_{i=1}^ra_i`$.
Claim: For each $`i`$,
$$\underset{jJ_i}{}rmv\left(q\left(str\left(\kappa _{ij}\right)\right)1\right)$$
is the formal sum of at most $`n+1`$ n-1-simplices $`L1`$ with coefficient 1.
The claim will be a consequence of the following subclaim and Lemma 10.
Subclaim: Assume that for some fixed $`iI`$, for the chosen triangulation
$$\sigma _i^1\left(N\right)=\underset{j\widehat{J}_i}{}\theta _{ij},$$
and the associated canonical colouring, there exist $`j_1,j_2\widehat{J}_i,k_1,k_2\{0,\mathrm{},n\}`$ such that the faces
$$T_1=_{k_1}\theta _{ij_1}S_{n1}\left(N\right),T_2=_{k_2}\theta _{ij_2}S_{n1}\left(N\right)$$
have a white-parallel arc (Definition 6). Then
$$rmv\left(q\left(str\left(\kappa _{ij_1}\right)\right)1\right)=0,rmv\left(q\left(str\left(\kappa _{ij_2}\right)\right)1\right)=0.$$
We are going to prove the subclaim.
$$_{k_l}\theta _{ij_l}S_{n1}\left(N\right)$$
implies (by Lemma 5 and Construction 1)
$$_{k_l}q\left(str\left(\kappa _{ij_l}\right)\right)HK_{}^{str}\left(_1Q\right)$$
for $`l=1,2`$. Argueing by contradiction, we assume that
$$rmv\left(q\left(str\left(\kappa _{ij_1}\right)\right)1\right)0.$$
By assumption of the subclaim, there are white-parallel arcs $`e_1,e_2`$ of $`T_1`$ resp. $`T_2`$. This means that there are arcs $`e_1,e_2`$ in a 2-dimensional subsimplex $`\tau ^2\mathrm{\Delta }^n`$ of the standard simplex, and that there are arcs $`f_1,f_2`$, which are subarcs of some edge of $`\tau ^2`$, such that
$$_0e_1=_1f_2,_0f_2=_0e_2,_1e_2=_0f_1,_1f_1=_1e_1$$
and such that
$$e_1,f_2,e_2,f_1$$
bound a square in the boundary of a white component. (Cf. the picture in section 6.1. We will use the same letter for an affine subset of $`\mathrm{\Delta }^n`$ and for the singular simplex obtained by restricting $`\sigma _i`$ to this subset.) The square is of the form $`U_1+U_2`$, where $`U_1,U_2`$ are n-2-fold iterated faces of some $`\theta _{ij}`$โs. Hence
$$U_1=e_1+f_2+_2U_1$$
and
$$U_2=e_2f_1_2U_1,$$
i.e.
$$\left(U_1+U_2\right)=e_1+f_2e_2f_1$$
and
$$_2U_1=_2U_2.$$
We emphasize that we assume $`e_1`$ resp. $`e_2`$ to be edges of $`\theta _{ij_1}`$ resp. $`\theta _{ij_2}`$ but that $`f_1,f_2`$ need not be edges of $`\theta _{ij_1}`$ or $`\theta _{ij_2}`$.
Notational convention: for each iterated face $`f=_{k_1}\mathrm{}_{k_l}\theta _{ij}`$ with $`iI,jJ_i`$, we will denote $`f^{}`$ the $`nl`$-simplex with
$$f^{}1=_{k_1}\mathrm{}_{k_l}\kappa _{ij}1=r\left(_{k_1}\mathrm{}_{k_l}\tau _{ij}1\right)=r\left(_{k_1}\mathrm{}_{k_l}p\left(\widehat{\theta }_{ij}\right)1\right).$$
(The last two equations are true because $`r,p`$ and the homotopy from $`_{i,j}a_i\theta _{ij}`$ to $`_{i,j}\widehat{\theta }_{ij}`$ are chain maps.) In other words, if $`f`$ is an iterated face of some $`\tau _{ij}`$, then $`f^{}`$ is, up to the ambiguity by the $`H`$-action, the corresponding iterated face of $`\kappa _{ij}`$.
By Lemma 5 we have $`e_1^{},e_2^{}GK\left(_1Q\right)`$. Thus we can (and will) choose $`\kappa _{ij_1},\kappa _{ij_2}`$ in their $`G`$-orbits such that we have $`e_1^{},e_2^{}K\left(_1Q\right)`$, hence $`str\left(e_1^{}\right),str\left(e_2^{}\right)K^{str}\left(_1Q\right)`$.
Since $`r,p`$ and the homotopy are chain maps, we have
$$_2U_1^{}1=_2U_2^{}1.$$
That is,
$$_2U_1^{}=g\overline{_2U_2^{}}$$
for some $`gG`$.
Since $`U_1^{}`$ and $`U_2^{}`$ belong to different $`\kappa _{ij}`$โs, say $`\kappa _{ij_1}`$ and $`\kappa _{ij_2}`$, we can, upon replacing $`\kappa _{ij_2}`$ by $`g\kappa _{ij_2}`$, assume that $`_2U_1^{}=\overline{_2U_2^{}}`$, that is, $`U_1^{}+U_2^{}`$ is a square. (Since $`g`$ maps $`e_2^{}`$ to $`e_1^{}`$, this second choice of $`\kappa _{ij_2}`$ in its $`G`$-orbit preserves the condition that $`e_2^{}K^{str}\left(_1Q\right)`$.)
Let $`F`$ resp. $`F^{}`$ be the path-components of $`_1Q`$ with $`e_1^{}F`$ resp. $`e_2^{}F^{}`$. Then we have $`_1str\left(f_1^{}\right),_0str\left(f_2^{}\right)F,_0str\left(f_1^{}\right),_1str\left(f_2^{}\right)F^{}`$.
We note that $`f_1^{}`$ and $`f_2^{}`$ are edges with label 1. By condition (i) of Corollary 3, this implies that $`str\left(f_1^{}\right)`$ and $`str\left(f_2^{}\right)`$ are distinguished 1-simplices.
By Condition ix) and Condition xiii) of Definition 9 we have that
$$_1q\left(str\left(f_1^{}\right)\right)=x_{E_0^F}=_0q\left(str\left(f_2^{}\right)\right),_0q\left(str\left(f_1^{}\right)\right)=x_{E_0^F^{}}=_1q\left(str\left(f_2^{}\right)\right).$$
That is, $`q(str\left(e_1^{}\right)))`$ and $`q\left(str\left(e_2^{}\right)\right)`$ are loops in $`_1Q`$, based at $`x_{E_0^F}`$ resp. $`x_{E_0^F^{}}`$.
Since the square $`q\left(str\left(U_1^{}+U_2^{}\right)\right)`$ realizes a homotopy between $`q\left(str\left(f_1^{}\right)\right)`$ and $`q\left(str\left(f_2^{}\right)\right)`$, we have that
$$q\left(str\left(f_1^{}\right)\right)=\gamma _1q\left(str\left(f_2^{}\right)\right)\gamma _2$$
with
$$\gamma _1=q(str\left(e_1^{}\right))),\gamma _2=q(str\left(e_2^{}\right)))\mathrm{\Omega }(_1Q)\mathrm{\Gamma }=\mathrm{\Omega }(Q).$$
By condition x) from Definition 9 this implies
$$q\left(str\left(f_1^{}\right)\right)=q\left(str\left(f_2^{}\right)\right).$$
This means that $`q\left(str\left(U_1^{}\right)\right)+q\left(str\left(U_2^{}\right)\right)`$ is a cylinder with the boundary circles $`q\left(str\left(e_1^{}\right)\right)`$ and $`q\left(str\left(e_2^{}\right)\right)`$ in $`_1Q`$.
(This is why we have performed the straightening construction in Section 5 such that there should be only one distinguished 1-simplex in each coset.)
The assumption $`rmv\left(qstr\left(\kappa _{ij_1}\right)1\right)0`$ implies that the loops $`q\left(str\left(e_1^{}\right)\right)`$ and $`q\left(str\left(e_2^{}\right)\right)`$ are not 0-homotopic.
Indeed, if one of them, say $`q\left(str\left(e_1^{}\right)\right)`$, were a 0-homotopic (thus constant) loop, then also $`q\left(str\left(e_2^{}\right)\right)`$ would be a 0-homotopic (thus constant) loop, because they are homotopic through the cylinder. But $`q\left(str\left(e_1^{}\right)\right),q\left(str\left(e_2^{}\right)\right)`$ are edges of $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ resp. $`q\left(str\left(\kappa _{ij_2}\right)\right)`$. In particular, $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ would have a constant loop as an edge. By Lemma 14 and Definition 7, this would prove the wanted equalities $`rmv\left(qstr\left(\kappa _{ij_1}\right)1\right)=0,rmv\left(qstr\left(\kappa _{ij_2}\right)1\right)=0`$.
Thus we can assume that $`q\left(str\left(e_1^{}\right)\right)`$ and $`q\left(str\left(e_2^{}\right)\right)`$ are not 0-homotopic, that is, the cylinder
$$q\left(str\left(U_1^{}\right)\right)+q\left(str\left(U_2^{}\right)\right)$$
is $`\pi _1`$-injective as a map of pairs. Since $`(Q,_1Q)`$ is a pared acylindrical manifold, the cylinder must then be homotopic into $`Q`$, as a map of pairs
$$(๐^1\times [0,1],๐^1\times \{0,1\})(Q,_1Q).$$
Since $`_1Q`$ is acylindrical, the cylinder must then either degenerate ($`๐^1\times [0,1]Q`$ homotopes to a map that factors over the projection $`๐^1\times [0,1]๐^1`$, in particular $`q\left(str\left(e_1^{}\right)\right)=q\left(str\left(e_2^{}\right)\right)`$) or be homotopic into $`_0Q`$ (and hence into $`q\left(_0Q\right)`$, since $`qid`$). In the second case the vertices $`x_{E_0^F},x_{E_0^F^{}}`$ must belong to $`_0Q`$ and we get by condition vii) from Lemma 11 that $`q\left(str\left(e_1^{}\right)\right),q\left(str\left(e_2^{}\right)\right)K_1^{str}\left(_0Q\right)`$. By Lemma 15 this implies that $`q\left(str\left(\kappa _{ij_1}\right)\right)1=0,q\left(str\left(\kappa _{ij_2}\right)\right)1=0`$.
Thus we can assume that the cylinder degenerates. In particular $`q\left(str\left(f_1^{}\right)\right),q\left(str\left(f_2^{}\right)\right)K_1^{str}\left(_1Q\right)`$.
Let $`P_1,P_2`$ be the affine planes whose intersections with $`\mathrm{\Delta }^n`$ contain $`T_1`$ resp. $`T_2`$. Let $`W`$ be the white component whose boundary contains the white-parallel arcs of $`T_1,T_2`$. We have seen that there is are arcs $`f_1,f_2`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_2\mathrm{\Delta }^n`$ such that
$$q\left(str\left(f_1^{}\right)\right),q\left(str\left(f_2^{}\right)\right)K_1^{str}\left(_1Q\right).$$
This implies that for each other arc $`f`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_2\mathrm{\Delta }^n`$ the straightening $`q\left(str\left(f^{}\right)\right)`$ must be (homotopic into and therefore by condition vii) from Lemma 11) contained in $`_1Q`$.
If $`P_1`$ and $`P_2`$ are of the same type (Definition 2), then this shows that for all arcs $`fW`$:
$$q\left(str\left(f^{}\right)\right)K_1^{str}\left(_1Q\right)$$
If $`P_1`$ and $`P_2`$ are not of the same type, then the existence of a parallel arc implies that at least one of them, say $`P_1`$, must be of type $`\left\{0a_1\mathrm{}a_k\right\}`$ with $`k\{0,n1\}`$. Then, for each plane $`P_3P_1`$ with $`P_3\mathrm{\Delta }^nW`$, it follows from Corollary 2 that $`P_3\mathrm{\Delta }^n`$ has a white-parallel arc with $`P_1\mathrm{\Delta }^n`$. Thus, repeating the argument with $`P_1`$ and $`P_3`$ in place of $`P_1`$ and $`P_2`$, we prove that there are arcs in $`_1Q`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_3\mathrm{\Delta }^n`$, and consequently for each arc $`fW`$ connecting $`P_1\mathrm{\Delta }^n`$ to $`P_3\mathrm{\Delta }^n`$, the straightening $`str\left(f^{}\right)`$ must be (homotopic into and therefore) contained in $`_1Q`$.
Consequently, also for arcs connecting $`P_2\mathrm{\Delta }^n`$ to $`P_3\mathrm{\Delta }^n`$, we have that $`q\left(str\left(f^{}\right)\right)`$ must be (homotopic into and therefore) contained in $`_1Q`$. This finally shows that the 1-skeleta of $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ belong to $`K_1^{str}\left(_1Q\right)`$. By $`\pi _1`$-injectivity of $`_1QQ`$, asphericity of $`K\left(_1Q\right)`$, and condition vii) from Lemma 11, this implies that the 2-skeleta of $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ belong to $`K_1^{str}\left(_1Q\right)`$. Inductively, if the $`k`$-skeleta of $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ belong to $`K_k^{str}\left(_1Q\right)`$, then by asphericity of $`K\left(Q\right)`$, asphericity of $`K\left(_1Q\right)`$, and condition vii) from Lemma 11 we obtain that the $`k+1`$-skeleta of $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ belong to $`K_{k+1}^{str}\left(_1Q\right)`$. This provides the inductive step and thus our inductive proof shows that $`q\left(str\left(\kappa _{ij_1}\right)\right)`$ and $`q\left(str\left(\kappa _{ij_2}\right)\right)`$ belong to $`K^{str}\left(_1Q\right)`$.
By Definition 7, Definition 11 and Lemma 14 this implies
$$rmv\left(q\left(str\left(\kappa _{ij_1}\right)\right)1\right)=0,rmv\left(q\left(str\left(\kappa _{ij_2}\right)\right)1\right)=0.$$
So we have shown the subclaim: if $`T_1=_{k_1}\theta _{ij_1},T_2=_{k_2}\theta _{ij_2}`$ have a white-parallel arc, then $`rmv\left(q\left(str\left(\kappa _{ij_1}\right)\right)1\right)=0,rmv\left(q\left(str\left(\kappa _{ij_2}\right)\right)1\right)=0`$. In particular,
$$q\left(str\left(T_1^{}\right)\right),q\left(str\left(T_2^{}\right)\right)$$
do not occur (with non-zero coefficient) in
$$\underset{jJ_i}{}rmv\left(q\left(str\left(\kappa _{ij}\right)\right)1\right).$$
By Lemma 10, for a canonical colouring associated to a set of affine planes $`P_1,P_2,\mathrm{}`$, and a fixed triangulation of each $`Q_i=P_i\mathrm{\Delta }^n`$, we have at most $`n+1`$ n-1-simplices whose 1-skeleton does not contain a white-parallel arc. Therefore the subclaim implies the claim.
Thus we have presented $`\left[Q\right]1`$ as a finite chain of $`l^1`$-norm at most
$`\left(n+1\right)_{i=1}^ra_i`$. By Lemma 4a) we know that $`G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$ is amenable. Hence $`H=q_{}\left(G\right)`$ is amenable. Thus Lemma 7, applied to $`X=Q`$ and $`K=HK^{str}\left(_1Q\right)`$ with its $`H`$-action, implies
$$Q\left(n+1\right)\underset{i=1}{\overset{r}{}}a_i.$$
QED
We remark that Theorem 1 is not true without assuming amenability of $`\pi _1_0Q`$. Counterexamples can be found, for example, using or , Theorem 6.3.
In , Theorem 1 has been proven for incompressible surfaces in hyperbolic 3-manifolds. We compare the steps of the proof in with the arguments in our paper:
Step 1 in is the normalization procedure, which we have restated in Lemma 1.
Step 2 in consists in choosing compatible triangulations of the polytopes $`\sigma _i^1\left(N\right)`$.
Step 3 in boils down to the statement that, for each component $`Q_i`$ of $`Q`$, there exists a retraction $`r:\widehat{N}p^1\left(Q_i\right)`$, for the covering $`p:\widehat{N}N`$ corresponding to $`\pi _1Q_i`$. Such a statement can not be correct because it would (together with step 7 from ) imply $`NQ`$ whenever $`Q`$ is a $`\pi _1`$-injective submanifold of $`N`$. This inequality is true for submanifolds with amenable boundary, but not in general. In fact, one only has the more complicated retraction $`r:C_{}(K\left(N\right),K\left(N^{}\right))_{๐G}๐C_{}(K\left(Q\right),K\left(Q\right))_{๐G}๐`$, with $`G=\mathrm{\Pi }\left(K\left(_0Q\right)\right)`$. This more complicated retraction is the reason that much of the latter arguments become notationally awkward, although conceptually not much is changing. Moreover, the action of the group $`G`$ is basically the reason that Theorem 1 is true only for amenable $`G`$.
Basically, the reason why the retraction $`r:\widehat{N}Q`$ does not exist, is as follows. Let $`R_j`$ be the connected components of $`\widehat{N}p^1\left(Q_i\right)`$. Then $`R_j`$ is homotopy equivalent to each connected component of $`R_j`$. If $`R_j`$ were connected for each $`j`$, this homotopy equivalence could be extended to a homotopy equivalence $`r:\widehat{N}p^1\left(Q_i\right)`$. However, in most cases $`R_j`$ will be disconnected, and then such an $`r`$ can not exist.
We note that also the weaker construction of cutting off simplices does not work. A simplex may intersect $`Q_i`$ in many components and it is not clear which component to choose.
Step 5 in is the straightening procedure, it corresponds to sections 5.2-5.4 in this paper. We remark that the straightening procedure must be slightly more complicated than in because it is not possible, as suggested in , to homotope all edges between boundary components of $`Q`$ into shortest geodesics. This is the reason why we can only straighten chains with an admissible 0-1-labeling of their edges (and why our straightening homomorphism in Section 5.3 is only defined on $`C_{}^{simp}\left(\mathrm{{\rm Y}}\right)`$ and not on all of $`C_{}^{sing}\left(Q\right)`$).
Step 6 in consists in removing degenerate simplices. This corresponds to Section 5.5 in this paper.
Step 7 in proves that each triangle in $`\sigma _i^1\left(N\right)`$ contributes only once to the constructed fundamental cycle of $`Q`$. Since, in our argument, we do not work with the covering $`p:\widehat{N}N`$, we have no need for this justification.
Step 8 in counts the remaining triangles per simplex (after removing degenerate simplices). It seems to have used the combinatorial arguments which we work out for arbitrary dimensions in Section 4.
We mention that the arguments of Section 4 are the only part of the proof which gets easier if one restricts to 3-manifolds rather than arbitrary dimensions. Moreover, the proof for laminations is the same as for hypersurfaces except for Lemma 1. Thus, upon these two points it seems that even in the case of incompressible surfaces in 3-manifolds the proof of Theorem 1 can not be further simplified.
## 7 Specialization to 3-manifolds
Guts of essential laminations. We start with recalling the guts-terminology. Let $`M`$ be a compact 3-manifold with (possibly empty) boundary consisting of incompressible tori, and $``$ an essential lamination transverse or tangential to the boundary. $`N=\overline{M}`$ is a, possibly noncompact, irreducible 3-manifold with incompressible, aspherical boundary $`N`$. We denote $`_0N=NM`$ and $`_1N=\overline{N_0N}`$. (Thus $`_1N`$ is the union of boundary leaves of the lamination.) By the proof of , Lemma 1.3., the noncompact ends of $`N`$ are essential $`I`$-bundles over noncompact subsurfaces of $`_1N`$. After cutting off each of these ends along an essential, properly fibered annulus, one obtains a compact 3-manifold to which one can apply the JSJ-decomposition of , . Hence we have a decomposition of $`N`$ into the characteristic submanifold $`Char\left(N\right)`$ (which consists of $`I`$-bundles and Seifert fibered solid tori, where the fibrations have to respect boundary patterns as defined in , p.83) and the guts of $`N`$, $`Guts\left(N\right)`$. The $`I`$-fibered ends of $`N`$ will be added to the characteristic submanifold, which thus may become noncompact, while $`Guts\left(N\right)`$ is compact. (We mention that there are different notions of guts in the literature. Our notion is compatible with , , but differs from the definition in or by taking the Seifert fibered solid tori into the characteristic submanifold and not into the guts. Thus, solid torus guts in the paper of Calegari-Dunfield is the same as empty guts in our setting.) If $`_0NQ\mathrm{}`$ consists of annuli $`A_1,\mathrm{},A_k`$, then, to be consistent with the setting of Theorem 1, we add components $`A_i\times [0,1]`$ to $`Char\left(N\right)`$ (without changing the homeomorphism type of $`N`$), which implies $`_0NQ=\mathrm{}`$.
For $`Q=Guts\left(N\right)`$ we denote $`_1Q=_1NQ=NQ=QN`$ and $`_0Q=\overline{Q_1Q}`$. For $`R=Char\left(N\right)`$ we denote $`_1R=NR`$ and $`_0R=\overline{R_1R}`$. $`_0NQ=\mathrm{}`$ implies then $`_0Q=QR`$.
$`_0Q`$ consists of essential tori and annuli, in particular $`\pi _1_0Q`$ is amenable. The guts of $`N`$ has the following properties: the pair $`(Q,_1Q)`$ is a pared acylindrical manifold as defined in Definition 3, $`Q,_1Q,_1R`$ are aspherical, and the inclusions $`_0QQ,_1QQ,QN`$, $`_0RR,_1RR,RN`$ are $`\pi _1`$-injective (see ,). It follows from Thurstonโs hyperbolization theorem for Haken manifolds that $`Q`$ admits a hyperbolic metric with geodesic boundary $`_1Q`$ and cusps corresponding to $`_0Q`$. (In particular, $`\chi \left(Q\right)0`$, thus $`Q`$ is aspherical, and $`_1Q`$ is a hyperbolic surface, thus acylindrical.)
Theorem 2 : Let $`M`$ be a compact 3-manifold with (possibly empty) boundary consisting of incompressible tori, and let $``$ be an essential lamination of $`M`$. Then
$$M,M_{}^{norm}\chi \left(Guts\left(\right)\right).$$
More generally, if $`P`$ is a polyhedron with $`f`$ faces, then
$$M,M_{,P}^{norm}\frac{2}{f2}\chi \left(Guts\left(\right)\right).$$
###### Proof.
Let $`N=\overline{M}`$. Since $``$ is essential, $`N`$ is irreducible (hence aspherical, since $`N\mathrm{}`$) and has incompressible, aspherical boundary. Let $`R=Char\left(N\right)`$ be the characteristic submanifold and $`Q=Guts\left(N\right)`$ be the complement of the characteristic submanifold of $`N`$. The discussion before Theorem 2 shows that the decomposition $`N=QR`$ satisfies the assumptions of Theorem 1.
From the computation of the simplicial volume for surfaces (, section 0.2.) and $`\chi \left(Q\right)=\frac{1}{2}\chi \left(Q\right)`$ (which is a consequence of Poincare duality for the closed 3-manifold $`Q_QQ`$), it follows that
$$\chi \left(Guts\left(\right)\right)=\frac{1}{2}\chi \left(Guts\left(\right)\right)=\frac{1}{4}Guts\left(\right).$$
Thus, the first claim is obtained as application of Theorem 1 to $`Q=Guts\left(\right)`$.
The second claim, that is the generalisation to arbitrary polyhedra, is obtained as in . Namely, one uses the same straightening as above, and asks again how many nondegenerate 2-simplices may, after straightening, occur in the intersection of $`Q`$ with some polyhedron $`P_i`$. In , p. 11, it is shown that this number is at most $`2f4`$, where $`f`$ is the number of faces of $`P_i`$. The same argument as above shows then $`_{i=1}^ra_i\frac{1}{2f4}Guts\left(\right)`$, giving the wanted inequality. โ
The following corollary applies, for example, to all hyperbolic manifolds obtained by Dehn-filling the complement of the figure-eight knot in $`๐^3`$. (Note that Hatcher has proved in that each hyperbolic manifold obtained by Dehn-filling the complement of the figure-eight knot in $`๐^3`$ carries essential laminations.)
###### Corollary 4.
If $`M`$ is a finite-volume hyperbolic manifold with $`Vol\left(M\right)<2V_3=2.02\mathrm{}`$, then $`M`$ carries no essential lamination $``$ with $`M,M_{,P}^{norm}=M,M_P`$ for all polyhedra P, and nonempty guts. In particular, there is no tight essential lamination with nonempty guts.
###### Proof.
The derivation of Corollary 4 from Theorem 2 is exactly the same as in for the usual (non-laminated) Gromovnorm. Namely, by (or , end of Section 6) there exists a sequence $`P_n`$ of straight polyhedra in $`๐^3`$ with $`lim_n\mathrm{}\frac{Vol\left(P_n\right)}{f_n2}=V_3`$, with $`f_n`$ denoting the number of faces of $`P_n`$. Assuming that $`M`$ carries a lamination $``$ with $`M,M_{,P_n}^{norm}=M,M_{P_n}`$ for all $`n`$, one gets
$$\chi \left(Guts\left(\right)\right)\frac{f_n2}{2}M,M_{,P_n}=\frac{f_n2}{2}M,M_{P_n}$$
$$\frac{f_n2}{2}\frac{Vol\left(M\right)}{Vol\left(P_n\right)}\frac{Vol\left(M\right)}{2V_3}<1.$$
On the other hand, if $`Guts\left(\right)`$ is not empty, then it is a hyperbolic manifold with nonempty geodesic boundary, hence
$$\chi \left(Guts\left(\right)\right)1,$$
giving a contradiction.โ
###### Definition 12.
The Weeks manifold is the closed 3-manifold obtained by $`(\frac{5}{1},\frac{5}{2})`$-surgery at the Whitehead link (, p.68).
It is known that the Weeks manifold is hyperbolic and that its hyperbolic volume is approximately $`0.94..`$. (It is actually the hyperbolic 3-manifold of smallest volume.)
###### Corollary 5.
(, Conjecture 9.7.): The Weeks manifold admits no tight lamination $``$.
###### Proof.
According to , the Weeks manifold can not carry a tight lamination with empty guts. Since tight laminations satisfy $`M_{,P}^{norm}=M`$ for each polyhedron (see Lemma 1), and since the Weeks manifold has volume smaller than $`2V_3`$, it follows from Corollary 4 that it can not carry a tight lamination with nonempty guts neither.โ
The same argument shows that a hyperbolic 3-manifold $`M`$ with
\- $`Vol\left(M\right)<2V_3`$, and
\- no injective homomorphism $`\pi _1MHomeo^+\left(๐^1\right)`$
can not carry a tight lamination, because it was shown by Calegari-Dunfield in that the existence of a tight lamination with empty guts implies the existence of an injective homomorphism $`\pi _1MHomeo^+\left(๐^1\right)`$. Some methods for excluding the existence of injective homomorphisms $`\pi _1MHomeo^+\left(๐^1\right)`$ have been developed in (which yielded in particular the nonexistence of such homomorphisms for the Weeks manifold, used in the corollary above), but in general it is still hard to apply this criterion to other hyperbolic 3-manifolds of volume $`<2V_3`$.
As indicated in , an approach to a generalization of some of the above arguments to essential, non-tight laminations, yielding possibly a proof for nonexistence of essential laminations on the Weeks manifold, could consist in trying to define a straightening of cycles (as in the proof of Lemma 1) upon possibly changing the essential lamination.
As a consequence of a recent paper of Tao Li, one can at least exclude the existence of transversely orientable essential laminations on the Weeks manifold.
###### Corollary 6.
The Weeks manifold admits no transversely orientable essential lamination $``$.
###### Proof.
According to , Theorem 1.1, the following statement is true: if a closed, orientable, atoroidal 3-manifold $`M`$ contains a transversely orientable essential lamination, then it contains a transversely orientable tight essential lamination. Hence Corollary 6 is a direct consequence of Corollary 5.โ
## 8 Higher dimensions
We want to finish this paper with showing that Theorem 1 is interesting also in higher dimensions. While in dimension 3 the assumptions of Theorem 1 hold for each essential lamination, it is likely that this will not be the case for many laminations in higher dimensions. However, the most straightforward, but already interesting application of the inequality is Corollary 7 which means that, for a given negatively curved manifold $`M`$, we can give an explicit bound on the topological complexity of geodesic hypersurfaces. Such a bound seems to be new except, of course, in the 3-dimensional case where it is due to Agol () and (with nonexplicit constants) to Hass ().
###### Corollary 7.
Let $`M`$ be a compact Riemannian $`n`$-manifold of negative sectional curvature and finite volume. Let $`FM`$ be a geodesic $`n1`$-dimensional hypersurface of finite volume. Then $`F\frac{n+1}{2}M`$.
###### Proof.
Consider $`N=\overline{MF}`$. $`(N,N)`$ is acylindrical. This is well-known and can be seen as follows: assume that $`N`$ contained an essential cylinder, then the double $`DN=N_{_1N}N`$ would contain an essential 2-torus. But, since $`N`$ is a negatively curved manifold with geodesic boundary, we can glue the Riemannian metrics to get a complete negatively curved Riemannian metric on $`DN`$. In particular, $`DN`$ contains no essential 2-torus, giving a contradiction.
Moreover, the geodesic boundary $`N`$ is $`\pi _1`$-injective and negatively curved, thus aspherical. Therefore we can choose $`Q=N`$, in which case the other assumptions of Theorem 1 are trivially satisfied. From Theorem 1 we conclude
$`M_F^{norm}\frac{1}{n+1}N`$. The boundary of $`N`$ consists of two copies of $`F`$, hence $`N=2F`$. The leaf space of $`\stackrel{~}{F}\stackrel{~}{M}`$ is a Hausdorff tree, thus Lemma 1b) implies $`M_F^{norm}=M`$. The claim follows.โ
This statement should be read as follows: for a given manifold $`M`$ (with given volume) one has an upper bound on the topological complexity of compact geodesic hypersurfaces.
For hyperbolic manifolds one can use the proportionality principle and the Chern-Gauร-Bonnet Theorem to reformulate Corollary 7 as follows: If $`M`$ is a closed hyperbolic n-manifold and $`F`$ a closed n-1-dimensional geodesic hypersurface, then $`Vol\left(M\right)C_n\chi \left(F\right)`$ for a constant $`C_n`$ depending only on $`n`$.
Thilo Kuessner
Mathematisches Institut, Universitรคt Mรผnster
Einsteinstraรe 62
D-48149 Mรผnster
Germany
e-mail: kuessner@math.uni-muenster.de
|
warning/0506/cond-mat0506712.html
|
ar5iv
|
text
|
# Unrestricted renormalized mean field theory of strongly correlated electron systems
## Abstract
We generalized systematically the renormalized mean field theory in the case of uniform states to the unrestricted case of general inhomogeneous states with competing spin-, charge- and superconducting orders. Applying the theory to high-$`T_c`$ superconductors, we discuss the issues of electronic inhomogeneity, the superfluid density, and in particular the local electron density of states. The results account for many intriguing aspects of the phenomenology.
As a prototype of strongly correlated electron systems, high-$`T_c`$ superconductors reveal rich phenomenologies. Apart from the apparent competing anti-ferromagnetic and superconducting orders, recent scanning tunnelling microscopy (STM) measurements have suggested checkerboard charge-density-wave order in under-doped NaCaCuOCl and BiSrCaCuO superconductors checkerboard . On the other hand, earlier STM measurements already revealed nano-scale spatial inhomogeneity of the quasi-particle energy gap gapdisorder , but yet the local density of states (LDOS) at very low energy are remarkably homogeneous, seemingly ignorant of the gap inhomogeneity robust . The physics behind the robust low energy quasi-particle states are intriguing. In short, strongly correlated electrons may support complex phases and their response to extrinsic impurities may be highly nontrivial. This motivates us to develop a reliable effective theory that takes proper care of the strong correlations and allows competing orders and inhomogeneous electronic states from the starting point.
A widely adopted model for high-$`T_c`$ superconductors is the $`t`$-$`J`$ model,
$`H_{tJ}=t{\displaystyle \underset{ij\sigma }{}}(C_{i\sigma }^{}C_{j\sigma }+\mathrm{h}.\mathrm{c}.)+J{\displaystyle \underset{ij}{}}S_iS_j,`$ (1)
where the constraint of no double occupation at any site is implied. In order to fit the band structure, hopping terms connecting next-nearest-neighbor sites and further neighboring sites may be necessary but they do not affect the following development of the theory. Recently the $`tUJ`$ model has also been used to study the Gossamer superconductivity tUJ , in which the constraint of no-double-occupation is relaxed but a Hubbard $`U`$-term is added,
$`H=H_{tJ}+U{\displaystyle \underset{i}{}}n_in_i.`$ (2)
The following development is formulated within this general model, Eq.(2), with obvious specific limits to the $`tJ`$ and Hubbard models at our disposal.
A very powerful way to take proper care of the strong correlation effects is to Gutzwiller project a trial wave-function, $`|\psi =P|\psi _0`$, where $`|\psi _0`$ denotes a free-particle many-body wave-function, with possible order parameters, and $`P=\mathrm{\Pi }_i(1\alpha D_i)`$ is a projection operator. Here $`D_i=n_in_i`$, and $`\alpha `$ is a variational parameter between $`0`$ and $`1`$. The calculation can be done by Monte Carlo methods vmc . However, it is limited by the lattice size and the parameter space. Alternatively the effect of Gutzwiller projection is taken into account approximately in a re-normalized mean field theory (RMFT) rmft , which agrees qualitatively with the Monte Carlo result. The Gutzwiller projected $`d`$-wave BCS states appear to capture the basic features of high-$`T_c`$ phase diagram vmc ; rmft ; plainvanilla . In this paper, we develop a systematic extension of the RMFT to the case of unrestricted spin/charge densities. We then discuss the physical quantities based on the unrestricted RMFT and explain the robust low energy states observed in the STM.
Let $`|\psi _0`$ be a Hatree-Fock state which carries the bare variational unrestricted order parameters. In the usual mean field theory this wave function is used to calculate the expectation value of the Hamiltonian, and the optimization with respect to the order parameters leads to the mean field self-consistent equations. For strongly-correlated electrons, however, this usual type of approach misses the most important physics, namely, the possible emergence of Mott insulating phases. This physics is conveniently picked up by Gutzwiller projecting the Hatree-Fock state, namely, the trial state is given by $`|\psi =P|\psi _0`$. However, we are no longer justified to use a projector $`P`$ with identical projection parameter $`\alpha `$ at each site, since the effect of strong correlations depends on the electron density. This motivates us to use a generalized projection operator $`๐ซ=\mathrm{\Pi }_i๐ซ_i`$ so that $`|\psi =๐ซ|\psi _0`$, with
$`๐ซ_i=y_i^{n_i}(1\alpha _iD_i)=E_i+y_iQ_i+\eta _iy_i^2D_i,`$ (3)
where $`E_i=(1n_i)(1n_i)`$ and $`Q_i=_\sigma n_{i\sigma }(1n_{i\overline{\sigma }})`$ are the empty and single occupation operators, respectively. These are standard projection operators, satisfying $`E_i^2=E_i`$, $`Q_i^2=Q_i`$ and $`D_i^2=D_i`$. Therefore
$`๐ซ_i^2=E_i+y_i^2Q_i+\eta _i^2y_i^4D_i.`$ (4)
Here $`\eta _i=1\alpha _i`$, and the fugacity $`y_i`$ is introduced so that the local charge density is not changed by projection laughlin ; anderson ; gros , or $`f_i=\psi _0|n_i|\psi _0=\psi |n_i|\psi /\psi |\psi `$, the advantage of which is that the variation of $`\eta _i`$ does not influence the local charge density. (In the $`tJ`$ model $`\eta _i0`$.) We note that we are working in the grand-canonical ensemble, which proves more convenient than the usual canonical ensemble approach usually used in the case of uniform state RMFT.
We shall use the shorthand notations $`_0\psi _0||\psi _0`$ and $`Z^1\psi _0|๐ซ๐ซ|\psi _0`$ with the normalization factor $`Z=\psi _0|๐ซ^2|\psi _0=๐ซ^2_0`$. Since $`|\psi _0`$ describes free particles, one can use Wickโs theorem to obtain all the contractions in $`Z`$. Under the Gutzwiller approximation, inter-site correlations are ignored when expectation values of projection operators are evaluated. One thus obtains $`Z=\mathrm{\Pi }_iz_i`$, with
$`z_i=๐ซ_i^2_0=e_{0i}+y_i^2q_{0i}+\eta _i^2y_i^4d_{0i},`$ (5)
where $`e_{0i}=E_i_0=1f_i+d_{0i}`$, $`q_{0i}=Q_i_0=f_i2d_{0i}`$ and $`d_{0i}=D_i_0=r_ir_i`$. In the above derivation, we have assumed the absence of on-site pairing and spin flipping order parameters in the wave-function $`|\psi _0`$. Similarly, we have $`e_i=E_i=e_{0i}/z_i`$, $`q_i=Q_i=q_{0i}y_i^2/z_i`$, and $`d_i=D_i=d_{i0}\eta _i^2y_i^4/z_i`$. The fugacity $`y_i`$ is determined by enforcing $`f_i=q_i+2d_i`$, and is a function of $`\eta _i`$, $`f_i`$ and $`d_{0i}`$ (or equivalently $`r_{i\sigma }`$). The calculation of $`y_i`$ is even unnecessary. Eliminating $`y_i`$ we find $`d_ie_i/q_i^2=\eta _i^2d_{0i}e_{0i}/q_{0i}^2`$, which determines $`d_i`$ uniquely.
We now discuss the effect of projection on the local spin moment directed in the $`z`$-direction. Since $`S_i`$ is already projective, we have $`๐ซ_iS_i๐ซ_i=y_i^2S_i`$, and therefore
$`m_i=S_i^z=S_i^z_0y_i^2/z_i=g_s(i)m_{0i},`$ (6)
where $`g_s(i)=q_i/q_{0i}`$ is the re-normalization factor for the variational spin moment $`m_{0i}=S_i^z_0=(r_ir_i)/2`$. Similar consideration applies to the spin-spin exchange,
$`S_iS_j=g_s(i)g_s(j)S_iS_j_0.`$ (7)
Using the identities
$`๐ซ_iC_{i\sigma }๐ซ_i=[y_i(1n_{i\overline{\sigma }})+\eta _iy_i^3n_{i\overline{\sigma }}]C_{i\sigma },`$ (8)
$`๐ซ_iC_{i\sigma }^{}๐ซ_i=[y_i(1n_{i\overline{\sigma }})+\eta _iy_i^3n_{i\overline{\sigma }}]C_{i\sigma }^{},`$ (9)
we obtain
$`C_{i\sigma }^{}C_{j\sigma }=g_{t\sigma }(i)g_{t\sigma }(j)C_{i\sigma }^{}C_{j\sigma }_0,`$ (10)
with the re-normalization factors
$`g_{t\sigma }(i)=(1r_{i\overline{\sigma }})\sqrt{{\displaystyle \frac{e_iq_i}{e_{0i}q_{0i}}}}+r_{i\overline{\sigma }}\sqrt{{\displaystyle \frac{q_id_i}{q_{0i}d_{0i}}}}.`$ (11)
In Eq.(11) the first term arises from the hopping process that does not encounter $`\overline{\sigma }`$-electrons at site $`i`$, and the second is from that involving a double occupation. We note that the re-normalization factors for spin-spin exchange and the hopping reduce in disguise to the known results in uniform charge density and uniform magnetically ordered states rmft . The result in the non-magnetic case for the $`tJ`$ model was used for granted in Ref. liyq .
The total internal energy $`E=H`$ can then be written as,
$`E=t{\displaystyle \underset{ij\sigma }{}}g_{t\sigma }(i)g_{t\sigma }(j)(\chi _{ij\sigma }^{}+\chi _{ij\sigma })`$
$`{\displaystyle \frac{3J}{8}}{\displaystyle \underset{ij}{}}g_s(i)g_s(j)(\chi _{ij}^{}\chi _{ij}+\mathrm{\Delta }_{ij}^{}\mathrm{\Delta }_{ij})`$
$`+J{\displaystyle \underset{ij}{}}g_s(i)g_s(j)m_{0i}m_{0j}+{\displaystyle \underset{i}{}}Ud_i,`$
where we have defined $`\chi _{ij\sigma }=C_{i\sigma }^{}C_{j\sigma }_0`$ and $`\chi _{ij}=_\sigma \chi _{ij\sigma }`$, $`\mathrm{\Delta }_{ij}=B_{ij}_0`$ with $`B_{ij}=C_iC_jC_iC_j`$, and used the standard decomposition
$`S_iS_j_0=m_{0i}m_{0j}{\displaystyle \frac{3}{8}}(\chi _{ij}^{}\chi _{ij}+\mathrm{\Delta }_{ij}^{}\mathrm{\Delta }_{ij}).`$ (12)
It is understood that $`r_{i\sigma }`$, $`\chi _{ij\sigma }`$ and $`\mathrm{\Delta }_{ij}`$ are determined by $`\psi _0`$, $`d_i`$ is determined by $`r_{i\sigma }`$ and $`\eta _i`$, and therefore $`E`$ is eventually a functional of $`\psi _0`$ and $`\{\eta _i\}`$.
The optimization of $`E`$ requires $`\delta E/\delta \psi _0|=0`$ and $`E/\eta _i=0`$. The first condition leads to a Schรถrdinger equation $`H_{MF}|\psi _0=\lambda |\psi _0`$, with the mean field Hamiltonian
$`H_{MF}={\displaystyle \underset{ij\sigma }{}}({\displaystyle \frac{E}{\chi _{ij\sigma }}}C_{i\sigma }^{}C_{j\sigma }+\mathrm{h}.\mathrm{c}.)`$
$`+{\displaystyle \underset{ij}{}}({\displaystyle \frac{E}{\mathrm{\Delta }_{ij}}}B_{ij}+\mathrm{h}.\mathrm{c}.)+{\displaystyle \underset{i\sigma }{}}({\displaystyle \frac{E}{r_{i\sigma }}}\mu )n_{i\sigma },`$ (13)
where $`\mu `$ is the chemical potential. This is a self-consistent unrestricted RMFT in the sense that 1) $`|\psi _0`$ is determined by $`H_{MF}`$, and so are the variational order parameters; 2) There is no restriction on the spatial variation of the order parameters; 3) The effect of strong correlation is reflected in the re-normalization coefficients $`g_{t\sigma }`$ and $`g_s`$; 4) The re-normalization coefficients depend on $`\eta _i`$, and the global minimum of the energy $`E`$ is reached after the self-consistency of the mean field theory and the optimization of $`\eta _i`$ are simultaneously achieved. In the case of $`tJ`$ model one simply sets $`\eta _i=0`$ ogata-extension . The theory has the advantage that the Mott physics is built in. In the literature gdft , the unrestricted Gutzwiller projection has been applied to the multi-band Hubbard models in dealing with para- and ferromagnetic states in transition metals. Further encoded in the present theory are the intriguing unrestricted competing anti-ferromagnetic and superconducting orders, which are most interesting in layered cuprate superconductors.
The mean field internal energy $`H_{MF}+\mu N_0`$ is different from $`E`$, but a self-consistent $`|\psi _0`$ minimizes $`E`$ so that $`\delta E=_nฯต_n\delta f_n+_{n,n^{}}V_{n,n^{}}\delta f_n\delta f_n^{}+\mathrm{}`$ where $`\{ฯต_n\}`$ is the single-particle spectrum of $`H_{MF}`$, $`f_n`$ is the Fermi-Dirac occupancy of the $`n`$-th single-particle orbital, and $`V_{n,n^{}}`$ is the residual two-body interaction kernel. The mean-field single-particles are not necessarily the electrons themselves, but are the bare Landau quasi-particles. The interpretation of these states, and in particular, the excitation energy spectrum, would be obscured if the charge density in $`|\psi _0`$ were different from that in $`|\psi `$.
In what follows we shall apply the theory to the $`tJ`$ model, relevant to the cuprates. First, we discuss the effect of the projection on the observables. Since the theory is projective, the order parameters in the RMFT are different from the measured ones. The re-normalization in the spin moment $`m_i=g_s(i)m_{0i}`$ has been derived previously. For the superconducting pairing order parameters, it can be shown easily using the method described above that $`\stackrel{~}{\mathrm{\Delta }}_{ij}=B_{ij}=g_\mathrm{\Delta }(i,j)\mathrm{\Delta }_{ij}`$, with $`g_\mathrm{\Delta }(i,j)=\frac{1}{2}_\sigma g_{t\sigma }(i)g_{t\overline{\sigma }}(j).`$ In the charge-uniform nonmagnetic states, $`g_\mathrm{\Delta }=2x/(1+x)<1`$ where $`x`$ (with $`e_i=x`$) is the hole-doping level away from half-filling. This result is well-known in the literature, and is argued to be related to the existence of pseudo-gap rmft ; plainvanilla . The pair-pair correlation function on disconnected bonds is re-normalized in a similar fashion, $`B_{ij}^{}B_{kl}=g_\mathrm{\Delta }^2B_{ij}^{}B_{kl}_0`$. This justifies the assignment of $`\stackrel{~}{\mathrm{\Delta }}`$ as representing the long-range off-diagonal order. However, the pair-pair correlation on the same bond is re-normalized quite differently, since $`B_{ij}^{}B_{ij}`$ projects out a singlet spin pair on the bond $`ij`$, $`B_{ij}^{}B_{ij}=2S_iS_j\frac{1}{4}n_in_j=2g_s(i)g_s(j)S_iS_j_0+\frac{1}{2}f_if_j`$. (In the $`tJ`$ model $`Q_i=n_i`$ so that $`q_i=f_i`$.) Therefore we expect significant incoherent part in the pair-pair susceptibility due to short-range spin correlations.
The superfluid density $`\rho _s`$ is related to the second order response of free energy to a vector potential. At zero temperature, it is related to the kinetic energy and is therefore re-normalized in the same way as the hopping parameters are, so that $`\rho _s=g_{t\sigma }^2t\chi =2xt\chi /(1+x)`$ in a uniform state, where $`\chi `$ is the variational hopping order parameter in the RMFT. (The expression is slightly changed in the presence of longer-range hopping, but the re-normalization factor is the same.) One realizes that this relation is consistent with the result of elaborate gauge theory gaugetheory . Indeed integrating over gauge fluctuations in the gauge theory effectively restores, to some extent, the strong correlation effects. The appealing feature of the projected density functional theory is that no slave degrees of freedom, and therefore no gauge fields are involved.
The factor $`g_{t\sigma }(i)`$ measures the overlap between the bare quasi-particle state in the RMFT and a corresponding real electron wave function at the given site. Therefore, the electron Greenโs function may be written as $`G(i\sigma ,j\sigma )=g_{t\sigma }(i)g_{t\sigma }(j)G_0(i\sigma ,j\sigma )+G_{inc}(i\sigma ,j\sigma )`$ anderson , where the first contribution is the coherent part from RMFT. We suppressed the energy arguments for brevity. Consider the on-site Greenโs function. A corresponding decomposition applies for the spectral function, $`A_\sigma (i)=g_{t\sigma }^2(i)A_{0\sigma }(i)+A_{inc,\sigma }(i)`$. Even though RMFT does not provide $`A_{inc,\sigma }`$ directly, some properties of it can be argued, as discussed previously in the literature randeria . We extend the discussion here for generally inhomogeneous cases. Since $`_{\mathrm{}}^0_\sigma A_\sigma (i)=_{\mathrm{}}^0_\sigma A_{0\sigma }(i)=f_i`$ but $`g_{t\sigma }(i)<1`$, one concludes that there is incoherent spectral weight in the occupied side. While the general expression is quite complicated in general, it has a simple form for a nonmagnetic site, which is given by $`_{\mathrm{}}^0_\sigma A_{inc,\sigma }(i)=(1e_i)^2/(1+e_i)`$. On the other hand, the number of unoccupied states at site $`i`$ is given by $`_\sigma ๐\omega C_{i\sigma }\delta (\omega H)C_{i\sigma }^{}=2e_i`$, where we dropped a contribution $`q_i`$ from the upper Hubbard band , which is out of the Hilbert space of the $`tJ`$ model. To this total unoccupied spectral weight, the contribution from the coherent part of $`G`$, which does describe excitations below the Mott gap, is given by
$`{\displaystyle \underset{\sigma }{}}g_{t\sigma }(i)^2(1r_{i\sigma })={\displaystyle \frac{2e_i}{1+4m_{0i}^2/(1e_i^2)}}.`$
Of particular interest is the nonmagnetic case ($`m_{0i}=0`$), in which the above discussion predicts that the unoccupied spectral weight is exhausted by the coherent part, in agreement with a recent argument randeria ; note . By continuity, one would expect that the incoherent part in the occupied side vanishes while approaching the Fermi level. The fact that incoherent excitations are abundant in the occupied side preferably at higher energies implies that 1) the nodal excitations at the Fermi energy are quite robust against strong correlations and 2) the antinodal excitations are largely incoherent due to strong correlations. This seems to be consistent with the the energy-dependence of the spectral function widely observed in angle-resolved photo-emission measurements arpesbackground .
We now calculate the LDOS in the $`tJ`$ model with weak random on-site potential, relevant to STM data. The inset of Fig.1(a) shows the scalar impurity potential profiles. The main panel of Fig.1(a) shows the coherent part of the LDOS calculated from the RMFT at an average doping level $`x=0.1`$. The system carries $`d`$-wave pairing order without spin moments. While the DOS at and above the gap energy scale disperse from site to site, the low energy DOS seems to be independent of position and ignorant of the random impurities, in qualitative agreement with experiments gapdisorder . In order to see whether this is a universal feature of doped Mott insulators, we deliberately erased the pairing order to find that the system evolves to a staggered-flux (or $`d`$-density-wave) state, with circulating currents whose chirality switches from plaquette to plaquette ddw . The LDOS in this state is shown in Fig.1(b). Now the low energy LDOS at different sites no longer merge together, but those at an energy above the Fermi level does. Since the DOS at the band-gap minimum vanishes linearly with energy in a charge-uniform staggered-flux state in the same way as at the Fermi level in a uniform $`d`$-wave paired state, we conclude that the homogeneity in the low energy DOS in a $`d`$-wave superconductor is merely a result of vanishing DOS in the parent uniform phase. In other words, the strong correlation effect stabilizes the $`d`$-wave pairing or density-wave states, and then some quasi-particles are protected by the small phase space available to elastic scattering.
A few remarks are in order. First, we are exaggerating the density of impurities in Fig.1. The phase-space protection of the quasi-particles is limited to weak disorder and low impurity concentrations. At stronger disorder and denser impurities, the inhomogeneities of the LDOS prevail at all energies. Second, in comparison with the U(1) slave boson mean field theory u1mft , our results are in better agreement with experiments. This is because the U(1) mean field theory underestimates the kinetic energy so that the holes are more susceptible to impurity localization. In addition, as compared to calculations without the projection and the self-consistency, the LDOS in our case are more uniform against potential impurities in space. The physics is best seen in the Mott insulator limit, where the low-lying excitations are spinons, which are completely ignorant to potential impurities. This means that strong correlation effect reduces the charge susceptibility. Third, the peak-to-peak gap in our case varies less severely than in experiments gapdisorder ; robust . We note that in order to mimic the experimental data, gap-impurities were introduced phenomenologically in a recent model hirschfeld . It would be highly interesting to find a mechanism that could lead to such gap impurities.
###### Acknowledgements.
We thank Dung-Hai Lee, P. W. Anderson, T. M. Rice, Dieter Vollhardt, V. N. Muthukumar, and T. K. Ng for stimulating communications. The work at Nanjing University was supported by NSFC 10325416, 10429401 and 10021001, the Fok Ying Tung Education Foundation No.91009, and the Ministry of Science and Technology of China (973 project No: 2006CB601002), and the work at the University of Hong Kong was supported by the RGC grants of Hong Kong.
|
warning/0506/quant-ph0506107.html
|
ar5iv
|
text
|
# Limits and restrictions of private quantum channel
## 1 Introduction
Quantum cryptography (for a popular review see ) is a rapidly developing branch of quantum information processing. The results of quantum cryptography include quantum key distribution , quantum secret sharing , quantum oblivious transfer and other cryptographic protocols . Quantum cryptography has two main goals: solutions to classical cryptographic primitives, and quantum cryptographic primitives.
The first goal is to design solutions of cryptographic primitives, which achieve a higher (provable) degree of security than their classical counterparts. The degree of security should be better than the security of any known classical solution, or it should be of the degree that is even not achievable by using classical information theory at all. Another alternative is to design a solution which is more efficient<sup>1</sup><sup>1</sup>1According to time, space or communication complexity. than any classical solution of comparable security.
The second class of cryptosystems is motivated by the evolution of applications of quantum information processing, regardless whether their purpose is cryptographic, communication complexity based or algorithmic. These cryptosystems are designed to manipulate quantum information. As applications of quantum information processing start to challenge a number of their classical counterparts, the need to secure quantum communications in general is getting more urgent. Therefore, there is a large class of quantum primitives which should secure quantum communication in the same way as classical communication is secured. These primitives include encryption of quantum information using both classical and quantum key , authentication of quantum information , secret sharing of quantum information , quantum data hiding and even commitment to a quantum bit , oblivious transfer of quantum information and others.
In this paper we concentrate on the encryption of quantum information with classical key described and explained in Section 2. At the end of Section 2 we introduce notation and one theorem we will be using through the remaining sections. To begin our analysis, in Section 3 we investigate for a given set $`๐`$ the set of all possible states $`\rho ^{(0)}`$ such that there exists a private quantum channel $`\widehat{๐}`$ with the property $`\rho ๐:\widehat{๐}(\rho )=\rho ^{(0)}`$ and we determine that it forms a ball within the Bloch sphere centered in $`\frac{1}{2}๐`$. In the Sections 47 we derive all possible private quantum channels for a given set $`๐`$ and state $`\rho ^{(0)}`$ and analyze necessary and sufficient entropy of the key of such PQC. We also explicitly construct PQCs achieving this bound. Another interesting result contained in Section 7 is that any PQC encrypting given set $`๐`$ of two linearly independent states encrypts also any two-dimensional set $`๐^{}`$ parallel to $`๐`$ and lying in the plane spanned by $`๐`$ and $`\frac{1}{2}๐`$.
We conclude our paper in Section 8 by few comments on possible generalizations of the described techniques to systems of higher dimension.
## 2 Private quantum channel
The private quantum channel is a general framework designed to perfectly encrypt an arbitrary quantum system using a classical key.
###### Definition 2.1.
Let $`๐๐ฎ(_{1,\mathrm{},n})`$ be a set of $`n`$-qubit states<sup>2</sup><sup>2</sup>2To make the definition easier we work with qubits. To obtain an equivalent definition for arbitrary quantum systems $`A`$ it suffices to replace $`_{1,\mathrm{},n}`$ by $`_A`$ and $`_{n+1,\mathrm{},m}`$ by $`_{anc}`$., $`\widehat{๐}=\{(p_i,U_i)\}_i`$ be a superoperator, where each $`U_i`$ is a unitary operator on $`_{1,\mathrm{},m},nm`$, $`p_i1`$ and $`_ip_i=1`$. Let $`\rho _{anc}`$ be an $`(mn)`$ qubit density matrix and $`\rho ^{(0)}`$ be an $`m`$-qubit density matrix. Then $`[๐,\widehat{๐},\rho _{anc},\rho ^{(0)}]`$ is a private quantum channel (PQC) if and only if for all $`\rho ๐`$ it holds that
$$\widehat{๐}(\rho \rho _{anc})=\underset{i}{}p_iU_i(\rho \rho _{anc})U_i^{}=\rho ^{(0)}.$$
(1)
The definition of the private quantum channel establishes the following cryptosystem. Alice wants to establish a communication (quantum) channel with Bob with the property that any state $`\rho ๐`$ will be transmitted securely. The security in this case means that Eve gets no advantage (information) by intercepting the transmitted message.
The encryption of the plaintext is done in the way that one operator, chosen randomly out of the operators $`\{U_i\}_i`$, is applied to the plaintext system. The operator $`U_i`$ is chosen with probability $`p_i`$. The classical key specifies which of the unitary operators was applied. The unitary operators $`U_i`$ are acting on $`_{1,\mathrm{},m}`$, while the state $`\rho `$ is only $`n`$-qubit state. The encryption operation $`U_i`$ is performed on the Hilbert space $`_{1,\mathrm{},m}`$, the plaintext space $`_{1,\mathrm{},n}`$ is a subspace of $`_{1,\mathrm{},m}`$. The encryption operation is defined on the (possibly) larger space than the plaintext to allow optional encryption of the plaintext together with an ancillary system. The encryption operators, therefore, act on a tensor product of the plaintext Hilbert space $`_{1,\mathrm{},n}`$ and the ancillary Hilbert space $`_{n+1,\mathrm{},m}`$, which is originally factorized (decoupled) from the plaintext. The ancillary Hilbert space is initially in the state $`\rho _{anc}`$ (see Figure 1).
The security of the scheme can be explained in the following way: Without knowledge of the key (i.e. without specific knowledge about which of the operators was used) any initial state $`\rho ๐`$ together with the ancilla appears to be in the state $`\rho ^{(0)}`$ after the encryption. The state $`\rho ^{(0)}`$ is the same for all $`\rho ๐`$, it is independent of the input state. It means that all states from the set $`๐`$ are physically indistinguishable after the encryption.
A dual point of view of the security is also possible. Let us denote by $`๐=\widehat{๐}[๐]`$ the set of all ciphertexts, i.e. $`\rho _i^{(c)}=U_i\rho U_i^{}๐`$ for each encryption operation $`U_i`$ and plaintext state $`\rho ๐`$. The encryption key (represented by the sequence $`i_1,\mathrm{},i_n`$) is used also for the decryption. The only difference in the case of the decryption is that the inverse operations $`U_i^{}`$ are applied, i.e. $`\rho _i^{(c)}U_i^{}\rho _i^{(c)}U_i=\rho ๐`$. Formally the decryption procedure induces a transformation $`\widehat{๐}[\rho ]=_ip_iU_i^{}\rho U_i`$. It describes the result of a decryption of a particular ciphertext without knowledge which key was used to encrypt it. The probabilities $`\{p_i\}_i`$ are the same as in the case of Eq. (1), because the probability that the key $`U_i`$ was used is $`p_i`$. In general, each encryption operation $`U_i`$ defines a different set of ciphertexts $`๐_i`$. The linear span $`\overline{๐}_i`$ is just rotated set of plaintexts $`\overline{๐}`$.
From Eq. (1) it follows that the information about the ciphertext contained in the plaintext is $`I(\rho _P:\rho _C)=0`$. However, from the symmetry of the mutual information it follows that the information about plaintext contained in the ciphertext is also $`0`$, therefore, the dual equation also holds
$$\widehat{๐}(\rho )=\underset{i}{}p_iU_{}^{}{}_{i}{}^{}(\rho )U_i=\rho ^{(1)},$$
(2)
where $`\rho ^{(1)}`$ is fixed for all ciphertext states $`\rho `$. The superoperator $`\widehat{๐}=\{(p_i,U_{}^{}{}_{i}{}^{})\}_i`$ is not an inverse of the superoperator $`\widehat{๐}`$ in the standard meaning. It describes the result of a decryption of a particular ciphertext without knowledge which key was used to decrypt it.
From the mathematical point of view the encryption transformation $`\widehat{๐}`$ is defined as a convex combination of unitary maps. Using the ancilliary system the encryption can be still defined only in terms of the system under consideration. Tracing out the ancilla we obtain a map $`\widehat{๐}_s`$ with the action defined by $`\widehat{๐}_s[\rho ]=\mathrm{Tr}_{anc}\widehat{๐}[\rho \rho _{anc}]`$. The usage of ancilla results in most general form of the quantum channel, i.e. $`\widehat{๐}_s[\rho ]=_ip_i\widehat{๐}_i[\rho ]`$ with $`\widehat{๐}_i[\rho ]=\mathrm{Tr}_{anc}[U_i\rho \rho _{anc}U_{}^{}{}_{i}{}^{}]`$. However, for the decryption the ancilliary system is necessary. In this paper we will analyze PQC without additional ancillas, so the encryption is formally a convex combination of unitary transformations. It means the encryption is described by a unital completely positive map, i.e. it preserves the total mixture $`\frac{1}{2}๐`$.
Finally, we will introduce one definition and one theorem, which will be used through this paper.
###### Definition 2.2.
Let $`๐=\{\rho _i|i๐\}`$, where $`๐`$ is an index set. We define the set $`\overline{๐}`$ as
$$\overline{๐}=\left\{\rho =\underset{i๐}{}\lambda _i\rho _i\right|\rho _i๐,\lambda _i,\underset{i๐}{}\lambda _i=1\}.$$
(3)
Especially in the case when $`๐=\{\rho _1,\rho _2\}`$ the set $`\overline{๐}`$ contains all operators of the form $`\lambda \rho _1+(1\lambda )\rho _2`$.
From now on we will denote the maximally mixed state<sup>3</sup><sup>3</sup>3I.e. the state nearest to the maximally mixed state $`\frac{1}{2}๐`$ according to the trace distance. It is also the nearest state in the Bloch ball. in $`\overline{๐}`$ as $`\overline{\rho }`$.
###### Theorem 2.3.
Let $`[๐,\widehat{๐},\rho ^{(0)}]`$ be a PQC. Then $`\widehat{๐}(\rho )=\rho ^{(0)}`$ for any operator $`\rho \overline{๐}`$. Note that we are interested only in operators $`\rho `$ with nonnegative eigenvalues, since the operators with negative eigenvalue(s) are not valid quantum states.
###### Proof.
The proof follows from the linearity of $`\widehat{๐}`$. โ
## 3 Achievable states $`\rho ^{(0)}`$
In this section we will derive several results on PQC on a single qubit. Our first question is โWhat are the possible states $`\rho ^{(0)}`$ given a specific set $`๐`$?โ. In it was proved that the only possible candidate for the state $`\rho ^{(0)}`$ is $`\frac{1}{2}๐`$, whenever $`\frac{1}{2}๐`$ can be expressed as a convex combination of states from $`๐`$. We will generalize this result for any set $`๐`$ to calculate the minimal entropy of the key necessary and sufficient to encrypt a specific set $`๐`$.
All information about the action of $`\widehat{๐}`$ we have is its behaviour on the set of plaintexts $`๐`$. Each element of this set is transformed into the fixed state $`\rho ^{(0)}`$. This determines the channel $`\widehat{๐}`$ completely, or incompletely depending on the set $`๐`$. However, in the case of incomplete specifications the choice of the channel $`\widehat{๐}`$ has no impact on the security. Under the action of the channel $`\widehat{๐}`$ each operator $`\rho \overline{๐}`$ is transformed into $`\rho ^{(0)}`$. This follows directly from the linearity of the transformation $`\widehat{๐}`$.
Let us assume a PQC given by operators $`\{(p_i,U_i)\}_i`$ with some $`\rho ^{(0)}`$. By applying the unitary transformations $`U_i^{}=VU_i`$ ($`V`$ is unitary) we obtain that $`_ip_iU_i^{}\rho U_i^{}=_ip_iVU_i\rho U_i^{}V^{}=V\rho ^{(0)}V^{}=\rho ^{(0)}`$ is fixed for all plaintext states. It follows that the unitarily transformed PQC is again a PQC with unitarily transformed average output state. Moreover, a convex combination of two PQC channels $`\widehat{๐}_1,\widehat{๐}_2`$ for the given set $`๐`$ is again a PQC channel for $`๐`$. In particular, $`\widehat{๐}=\pi _1\widehat{๐}_1+\pi _2\widehat{๐}_2`$ is PQC with $`\rho ^{(0)}=\pi _1\rho _1^{(0)}+\pi _2\rho _2^{(0)}`$, i.e. $`\widehat{๐}[\rho ]=\rho ^{(0)}`$ for all $`\rho ๐`$. Thus, for a given set of plaintexts $`๐`$ the set of all possible private quantum channels is convex. The set of achievable states is convex as well. It is formed by orbits of states under the action of the whole unitary group.
For any TCP map $`\widehat{๐}`$ the following inequality holds
$$D(\rho ,\sigma )D(\widehat{๐}(\rho ),\widehat{๐}(\sigma ))$$
(4)
for the distance measure $`D(\rho ,\sigma )=\mathrm{Tr}|\rho \sigma |`$ on mixed states, i.e. two quantum states $`\rho ,\sigma `$ cannot become more distinguishable after applying a TCP transformation. We have already mentioned that we consider that the encryption superoperator $`\widehat{๐}`$ is unital (we do not consider the ancilla here) and therefore from Eq. (4) we have
$$D(\rho ,\frac{1}{2}๐)D(\rho ^{(0)},\frac{1}{2}๐),$$
(5)
where $`\rho \overline{๐}`$ is any state in the set $`\overline{๐}`$ and $`\rho ^{(0)}=\widehat{๐}(\rho )`$ is fixed for all states $`\rho \overline{๐}`$. We use the fact that for unital maps $`\widehat{๐}[\frac{1}{2}๐]=\frac{1}{2}๐`$. Especially this equation holds for the state $`\overline{\rho }`$, which is the most mixed density operator in $`\overline{๐}`$ (the nearest point to $`\frac{1}{2}๐`$ in the Bloch ball).
Therefore the condition is that given the set of plaintext states $`\overline{๐}`$ any achievable state $`\rho ^{(0)}`$ fulfills the condition
$$D(\overline{\rho },\frac{1}{2}๐)D(\rho ^{(0)},\frac{1}{2}๐).$$
(6)
As a consequence, we obtain the result of that the state $`\rho ^{(0)}=\frac{1}{2}๐`$ whenever $`\frac{1}{2}๐`$ is contained in the convex span of the set $`๐`$.
Provided that the most mixed state $`\overline{\rho }`$ in $`\overline{๐}`$ is not $`\frac{1}{2}๐`$, the state $`\rho ^{(0)}`$ must have the same or a smaller distance from $`\frac{1}{2}๐`$ than $`\overline{\rho }`$. This is the necessary condition each candidate to the state $`\rho ^{(0)}`$ must obey. In this sense the set of potential candidates $`\rho ^{(0)}`$ forms a ball within the Bloch ball, with the center in $`\frac{1}{2}๐`$ and the radius given by the distance of $`\overline{\rho }`$ and $`\frac{1}{2}๐`$. Let us denote this ball (set of allowed states) by $`b`$. This condition is necessary, it remains to verify whether it is also sufficient, i.e. whether for a given $`๐`$ and $`\rho ^{(0)}b`$ there exists a suitable TCP superoperator $`\widehat{๐}_{\rho ^{(0)}}`$, or equivalently whether the set of all achievable states coincides with those allowed by inequality (6).
In what follows we will analyze the achievability of $`\rho ^{(0)}`$. Let us first consider two trivial cases. When $`๐`$ has only a single member, then there is nothing to encrypt. If the set $`๐`$ contains at least four linearly independent members, then $`\overline{๐}`$ already spans the whole Bloch ball and $`\frac{1}{2}๐\overline{๐}`$. It follows that the PQC maps all states to $`\frac{1}{2}๐`$, so the set of achievable states contains only single element. In subsequent sections we will analyze the remaining two cases: i) set $`\overline{๐}`$ is two-dimensional, and ii) set $`\overline{๐}`$ is three-dimensional.
In Sections 46 we will adopt analytical approach to prove that for all states $`\rho ^{(0)}b`$ given a set of plaintext states $`๐`$ there exists a PQC sending the set $`๐`$ to $`\rho ^{(0)}`$. In Section 7 we will analyze concrete PQC realizations as well as the minimal entropy of the key for a given set $`๐`$ and $`\rho ^{(0)}`$. We will show an easily understandable geometrical method how to construct PQCs for encryption of the given set of plaintexts.
## 4 General remarks
In our analysis we will exploit the geometric picture of the Bloch ball (see Appendix). In both cases we will define specific representatives of the set of plaintexts $`\overline{๐}`$. We will choose the basis of the state space as four operators $`\xi _j`$ represented by mutually orthogonal Bloch vectors $`\stackrel{}{v}_j`$. In particular, we will rotate the coordinate system (this rotation is just unitary change of the basis operators) to work with not necessarily positive, but trace-one operators
$$\begin{array}{ccccc}\hfill \xi _x& =& \frac{1}{2}(๐+\alpha S_x)\hfill & & \stackrel{}{v}_x=(\alpha ,0,0)\hfill \\ \hfill \xi _y& =& \frac{1}{2}(๐+\beta S_y)\hfill & & \stackrel{}{v}_y=(0,\beta ,0)\hfill \\ \hfill \xi _z& =& \frac{1}{2}(๐+S_z)\hfill & & \stackrel{}{v}_z=(0,0,1)\hfill \\ \hfill \xi _0& =& \frac{1}{2}๐\hfill & & \stackrel{}{v}_0=(0,0,0)\hfill \end{array}$$
(7)
The S-basis is just a suitably rotated $`\sigma `$-basis (basis consisting of Pauli operators), i.e. $`S_j=U\sigma _jU^{}`$ for some unitary $`U`$.
This new operator basis shares all the properties of the original Pauli basis. In fact, the operators $`S_x,S_y,S_z`$ specify only a rotated Cartesian coordinate system. Each private quantum channel $`\widehat{๐}`$ induces a contraction of the given set $`\overline{๐}`$ into the state $`\rho ^{(0)}`$. Our first aim is to explicitly specify the maximally mixed state in $`\overline{\rho }\overline{๐}`$. The second goal will be to show the achievability of this state, i.e. the construction of the PQC that transforms the whole set of plaintexts states into the state $`\rho ^{(0)}`$ having the same mixedness (i.e. distance from $`\frac{1}{2}๐`$) as $`\overline{\rho }`$. In particular, $`\rho ^{(0)}=V\overline{\rho }V^{}`$ ($`V`$ is unitary). Let us denote by $`s`$ the mixedness of $`\overline{\rho }`$. Then the PQC $`\widehat{๐}_{\rho ^{(0)}}`$ acts (in a suitably chosen basis) as follows: $`\widehat{๐}_{\rho ^{(0)}}[\xi _j]=\rho ^{(0)}=\frac{1}{2}(๐+sS_z)`$ for all $`\xi _j\overline{๐}`$. Its action on the linear complement of $`\overline{๐}`$ must be defined in a way that the whole transformation is TCP. The existence of such PQC will be proved in subsequent sections.
## 5 Two states
Given two linearly independent states $`๐=\{\rho _1,\rho _2\}`$ the set $`\overline{๐}`$ defines a line crossing the Bloch sphere in two pure states, e.g. $`|\psi _1,|\psi _2`$. The mixedness of $`\rho _\lambda =\lambda \rho _1+(1\lambda )\rho _2`$ (i.e. the distance from the total mixture) is characterized by the length of the corresponding Bloch vector $`\stackrel{}{r}_\lambda `$. In particular, $`|\stackrel{}{r}_\lambda |^2=\lambda ^2|\stackrel{}{r}_1|^2+(1\lambda )^2|\stackrel{}{r}_2|^2+2\lambda (1\lambda )\stackrel{}{r}_1\stackrel{}{r}_2`$ can be easily minimized (with respect to $`\lambda `$) providing that we use two pure states $`|\psi _j\stackrel{}{r}_j`$. In this case $`|\stackrel{}{r}_1|=|\stackrel{}{r}_2|=1`$ and $`\stackrel{}{r}_1\stackrel{}{r}_2=|\stackrel{}{r}_1||\stackrel{}{r_2}|\mathrm{cos}\theta `$ with $`\theta [0,\pi ]`$ being an angle between the vectors. The minimum we obtain by calculating the equation
$$\frac{d}{d\lambda }[\lambda ^2+(1\lambda )^22\lambda (1\lambda )\mathrm{cos}\theta ]=2(2\lambda 1)(1\mathrm{cos}\theta )=0$$
(8)
The minimum is achieved for $`\lambda =1/2`$, i.e. for the equal mixture of two pure states from $`\overline{๐}`$ and reads $`|\stackrel{}{r}_{\mathrm{min}}|=\sqrt{\frac{1}{2}(1+\mathrm{cos}\theta )}`$. For general (nonpure) states $`\rho _1\stackrel{}{r}_1`$ and $`\rho _2\stackrel{}{r}_2`$ the state $`\rho _\lambda =\lambda \rho _1+(1\lambda )\rho _2`$ is maximally mixed for the value $`\lambda =(|\stackrel{}{r}_2|^2\stackrel{}{r}_1\stackrel{}{r}_2)/|\stackrel{}{r}_1\stackrel{}{r}_2|^2`$.
Let us assume that the state $`|\psi _1`$ corresponds to the North Pole of the Bloch ball and the $`y`$ coordinate of $`|\psi _2`$ vanishes, i.e. we choose the operator basis $`S_x,S_y,S_z`$ such that $`|\psi _1\psi _1|=\frac{1}{2}(๐+S_z)`$ and $`|\psi _2\psi _2|=\frac{1}{2}(๐+\mathrm{sin}\theta S_x+\mathrm{cos}\theta S_z)`$. In other words, the state $`|\psi _2`$ is represented by the vector $`\stackrel{}{r}_2=(\mathrm{sin}\theta ,0,\mathrm{cos}\theta )`$. The norm of the vector $`\stackrel{}{r}_\lambda `$ is minimal for $`\lambda _{\mathrm{min}}=1/2`$, i.e. $`\stackrel{}{r}_{\mathrm{min}}=(\frac{1}{2}\mathrm{sin}\theta ,0,\frac{1}{2}(1+\mathrm{cos}\theta ))`$ with norm $`|\stackrel{}{r}_{\mathrm{min}}|=\sqrt{\frac{1}{2}(1+\mathrm{cos}\theta )}`$.
The possible quantum private channels form a set
$$\widehat{๐}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& \frac{1}{2}(1\mathrm{cos}\theta )& a& \frac{1}{2}\mathrm{sin}\theta \\ 0& 0& b& 0\\ 0& \frac{1}{2}\mathrm{sin}\theta & c& \frac{1}{2}(1+\mathrm{cos}\theta )\end{array}\right)$$
(9)
and the complete positivity with respect to parameters $`a`$, $`b`$ and $`c`$ must be verified. Our aim is to find at least one valid TCP transformation. Therefore, let us consider that only the parameter $`b`$ is nonvanishing<sup>4</sup><sup>4</sup>4We will show that even in this particular case there exists a completely positive superoperator. (i.e. we set $`a=c=0`$). In this case the matrix is symmetric, so the singular values coincide with the eigenvalues that read
$$\{\lambda _1,\lambda _2,\lambda _3\}=\{1,b,0\}.$$
(10)
The complete positivity constraint requires the validity of the following inequalities
$`1+\lambda _1\lambda _2\lambda _30`$ $``$ $`b2`$ (11)
$`1\lambda _1+\lambda _2\lambda _30`$ $``$ $`b0`$ (12)
$`1\lambda _1\lambda _2+\lambda _30`$ $``$ $`b0`$ (13)
$`1+\lambda _1+\lambda _2+\lambda _30`$ $``$ $`b2.`$ (14)
It turns out that the only possibility to satisfy these conditions is that the value of $`b`$ must set to zero, i.e. $`b=0`$. As a result, we get that the channel $`\widehat{๐}`$ with $`b=0`$ is for sure completely positive for all values of $`\mathrm{cos}\theta `$. Consequently, for two linearly independent states the derived bound on the choice of the state $`\rho ^{(0)}`$ is achievable. The achievability of the states inside the ball $`b`$ we obtain from the fact that the set of PQCs encrypting given set $`๐`$ is convex as well as the set of all achievable states, see Section 3. Later we will specify the unitary transformations forming the private quantum channel explicitly.
## 6 Three linearly independent states
In case the set $`๐=\{\rho _1,\rho _2,\rho _3\}`$ contains precisely three linearly independent states, the set $`\overline{๐}`$ forms a plane and valid quantum states from this plane (the intersection with the Bloch ball) form a circle $`c`$. Since all points of the ball $`b`$, containing all possible candidates for the state $`\rho ^{(0)}`$, have the distance from $`\frac{1}{2}๐`$ the same or smaller than the most mixed state from $`c`$, it follows that the circle $`c`$ touches the ball $`b`$ precisely in the middle of the circle $`c`$. Moreover, this point $`\overline{\rho }=\rho ^{(0)}`$ is the most mixed state from $`c`$.
To solve the general case explicitely, we will exploit the tools of analytic geometry. A plane determined by three points $`A=\stackrel{}{r}_1`$, $`B=\stackrel{}{r}_2`$, $`C=\stackrel{}{r}_3`$ reads $`ax+by+cz+d=0`$, where
$`d`$ $`=`$ $`det(\stackrel{}{r}_1\stackrel{}{r}_2\stackrel{}{r}_3)`$ (15)
$`a`$ $`=`$ $`det(\stackrel{}{1}\stackrel{}{r}_2\stackrel{}{r}_3)`$ (16)
$`b`$ $`=`$ $`det(\stackrel{}{r}_1\stackrel{}{1}\stackrel{}{r}_3)`$ (17)
$`c`$ $`=`$ $`det(\stackrel{}{r}_1\stackrel{}{r}_2\stackrel{}{1}).`$ (18)
The symbol $`\stackrel{}{1}=(1,1,1)^T`$ denotes a column vector. The distance from the origin of the coordinate system (the total mixture) equals to
$$s=\frac{|d|}{\sqrt{a^2+b^2+c^2}}.$$
(19)
This number coincides (if $`s1`$) with the distance between the maximally mixed state $`\overline{\rho }\overline{๐}`$ and the total mixture. It follows that we can use directly the given set of plaintext states $`\{\rho _1,\rho _2,\rho _3\}`$ as the basis. However, for our purposes it will be useful to choose operators $`\xi _1,\xi _2,\xi _3\overline{๐}`$ of the form given in Eq. (7). Let us assume that the set $`\overline{๐}`$ does not contain the total mixture. In this case
$$a=\beta b=\alpha c=d=\alpha \beta $$
(20)
and
$$s=|\stackrel{}{s}|=\frac{|\alpha \beta |}{\sqrt{\alpha ^2\beta ^2+\beta ^2+\alpha ^2}}.$$
(21)
The question is, whether the transformations $`\xi _j\rho ^{(0)}=\frac{1}{2}(๐+sS_z)=\widehat{๐}[\xi _j]`$ is completely positive, or not. Due to the unitality of the PQC channels, the transformation $`\widehat{๐}`$ is completely specified as
$$\widehat{๐}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& s/\alpha & s/\beta & s\end{array}\right).$$
(22)
Our task is only to verify the condition of complete positivity. The singular values of $`\widehat{๐}`$ reads $`\{\lambda _1,\lambda _2,\lambda _3\}=\{0,0,s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}}\}`$. It follows that the map is completely positive if and only if $`01s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}}`$. Inserting the derived result for the value of $`s`$ into this complete positivity constraint, we find that it is always satisfied, because
$$01s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}}=1\frac{|\beta \alpha |}{\sqrt{\beta ^2+\alpha ^2+\beta ^2\alpha ^2}}\sqrt{\frac{\beta ^2\alpha ^2+\alpha ^2+\beta ^2}{\beta ^2\alpha ^2}}=11=0.$$
(23)
As a result we obtain that it is always possible to define private quantum channel so that the norm bound is saturated and the state $`\rho ^{(0)}=\overline{\rho }`$ is achievable. Let us note that the value of $`s`$ is always less than 1, which is in agreement with the fact that $`\rho ^{(0)}`$ is a quantum state (it belongs to the Bloch ball). The fact that for a given set of states $`๐`$ it is always possible to find a unital channel $`\widehat{๐}`$ such that $`\rho ๐:\widehat{๐}(\rho )=\rho ^{(0)}`$, where $`\rho ^{(0)}`$ is the closest state to the total mixture belonging to the linear span of the set $`๐`$, is interesting per se.
The case when $`\rho ^{(0)}`$ lies inside (not on the surface) the ball $`b`$ we obtain again from the convexity of the set of private quantum channels as discussed in Section 3.
## 7 Realizations of PQC and entropy of the key
So far, we studied the existence of PQC with states $`\rho ^{(0)}`$ for a given set of plaintexts $`\overline{๐}`$. Next we will analyze the optimal realizations of these private quantum channels, i.e. we will ask the question: how many classical bits one needs to design the PQC. These classical bits represent the key that must be shared between sender and receiver to perfectly encrypt/decrypt the quantum states from plaintext. The efficiency of PQC is quantified by the entropy ($`H(p)=_jp_j\mathrm{log}_2p_j`$) of the probability distribution of unitary transformations that specify the length of the shared classical key. It is known that the ideal realization of PQC for general qubit states requires two bits. It is realized by arbitrary collection of four unitary transformations $`\{U_k\}_k`$ satisfying the following orthogonality condition $`\mathrm{Tr}(U_j^{}U_k)=2\delta _{jk}`$. Each of these transformations is applied with the same probability $`p=1/4`$, i.e. $`H(p)=2`$.
Each private quantum channel $`\widehat{๐}`$ is a convex combination of unitary transformations. Our task is to find a representation for arbitrary PQC, which is optimal. The action channel $`\widehat{๐}`$ can be written in the form $`\widehat{๐}[\rho ]=U\mathrm{\Phi }_{\widehat{๐}}[V\rho V^{}]U^{}`$, where $`U,V`$ are unitary transformations. For qubit unital channels the induced transformation $`\mathrm{\Phi }_{\widehat{๐}}`$ is diagonal, i.e. $`\mathrm{\Phi }_{\widehat{๐}}=\mathrm{diag}\{1,\lambda _1,\lambda _2,\lambda _3\}`$. It turns out that these transformations are of a simple form and can be written as convex combination of four Pauli transformations
$$\mathrm{\Phi }_{\widehat{๐}}[\rho ]=p_0\rho +p_x\sigma _x\rho \sigma _x+p_y\sigma _y\rho \sigma _y+p_z\sigma _z\rho \sigma _z$$
(24)
Consequently, the original transformation $`\widehat{๐}`$ is realized by four unitary transformations $`W_j=U\sigma _jV`$, i.e. $`\widehat{๐}[\rho ]=_jp_jW_j\rho W_{}^{}{}_{j}{}^{}`$. Since the probabilities do not change, the PQCs $`\widehat{๐}`$ and $`\mathrm{\Phi }_{\widehat{๐}}`$ can be realized with the same entropy. In fact, this holds in general: two unitarily equivalent PQCs can be always realized with the same efficiency, i.e. with the classical keys of the same entropy. Thus, it is sufficient to analyze the optimality of the realization of Pauli channels $`\mathrm{\Phi }_{\widehat{๐}}`$. Finding the singular values corresponding to $`\widehat{๐}`$ we obtain the diagonal elements $`\lambda _1,\lambda _2,\lambda _3`$ of $`\mathrm{\Phi }_{\widehat{๐}}`$. The probabilities $`p_j`$ are related to these values $`\lambda _k`$ via the following equations
$$\begin{array}{ccc}\hfill p_x& =& \frac{1}{4}(1+\lambda _1\lambda _2\lambda _3)\hfill \\ \hfill p_y& =& \frac{1}{4}(1\lambda _1+\lambda _2\lambda _3)\hfill \\ \hfill p_z& =& \frac{1}{4}(1\lambda _1\lambda _2+\lambda _3)\hfill \\ \hfill p_0& =& 1p_xp_yp_z\hfill \end{array}$$
(25)
The entropy rate of the given PQC $`H(\widehat{๐})=H(p)`$, $`p=\{p_j\}_j`$, is given by the entropy of the distribution $`p`$. Let $`\rho =_jp_j|\psi _j\psi _j|`$, where $`\{|\psi _j\}_j`$ is a set of not necessarily orthogonal quantum states. It follows that $`S(\rho )=S(_jp_j|\psi _j\psi _j|)H(p)`$ and the inequality is saturated if and only if $`\{|\psi _j\}_j`$ are mutually orthogonal. Let us consider any pure plaintext state $`|\psi `$ and let $`|\psi _j=U_j|\psi `$. It is clear that
$$S(\widehat{๐}(|\psi \psi |))=S(\rho )=S\left(\rho ^{(0)}\right)H(p).$$
(26)
Therefore the entropy of the encryption operation can always be bounded from below by the entropy of $`\rho ^{(0)}`$ as long as $`\overline{๐}`$ contains at least one pure state. This always holds in the case of qubit, however, not in general for systems of larger dimension. In example in the case of two qubits we can define the set $`๐=\{1/4๐,1/2(|0000|+|1111|)\}`$. It is clear that $`\overline{๐}`$ contains no pure state, it is encrypted by the superoperator
$$\{(1/2,๐),(1/2,๐\sigma _x)\}$$
(27)
and $`\rho ^{(0)}=\frac{1}{4}๐`$.
The limit (26) is saturated only if the encoding operators $`U_j`$ generate mutually orthogonal (noncommuting) states (ciphertexts) for each given plaintext. In particular, if we consider a PQC for all possible states of a qubit, this limit can be achieved only (up to unitary equivalence) by encoding with the identity and the universal NOT operation. However, this map is not completely positive, and therefore unphysical .
Next we shall study the realization of PQC when the set of plaintexts is two-dimensional and three-dimensional, respectively, and the state $`\rho ^{(0)}`$ is the maximally mixed one from the set $`\overline{๐}`$. For two-dimensional set of plaintexts the induced transformation is $`\mathrm{\Phi }_{\widehat{๐}}=\mathrm{diag}\{1,0,0,1\}`$, i.e. $`\lambda _1=\lambda _2=0`$ and $`\lambda _3=1`$. It follows that
$$\mathrm{\Phi }_{\widehat{๐}}[\rho ]=\frac{1}{2}\rho +\frac{1}{2}\sigma _z\rho \sigma _z$$
(28)
and one bit is sufficient for encoding.
One can specify the precise form of unitary transformations, but the explicit calculation is quite lengthy. Instead, we will use the geometric picture of Bloch sphere to guess the unitaries.
### Two states
Let us suppose that the set $`๐=\{\rho _1,\rho _2\}`$ has only two (linearly independent) states. By Theorem 2.3 it also encrypts any state on the line segment $`l\overline{๐}`$ defined by the points corresponding to the states $`\rho _1`$ and $`\rho _2`$ in the Bloch ball. Let us choose the state $`\rho ^{(0)}`$ as the point where the line segment $`l`$ touches the ball<sup>5</sup><sup>5</sup>5Of possible candidates to the state $`\rho ^{(0)}`$. $`b`$ (see fig. 2), i.e. $`\rho ^{(0)}=\overline{\rho }`$. It is clear that this point is in the center of the line segment $`l`$, since the extremal points of this line segment are on the surface of the Bloch ball.
We will design a specific superoperator $`\widehat{๐}`$, which encrypts this line segment to the state $`\rho ^{(0)}`$. This superoperator can be realized using two unitary operators with uniform distribution,
$$\widehat{๐}(\rho )=\frac{1}{2}๐\rho ๐+\frac{1}{2}U\rho U^{},$$
(29)
where $`U`$ is the unitary operation, which realizes the rotation of the Bloch ball by $`180`$ degrees around the axis intersecting points $`\rho ^{(0)}`$ and $`\frac{1}{2}๐`$. It is easy to see that such a superoperator takes any state $`\rho `$ from $`l`$ to a convex combination of the original state $`\rho `$ and the state which lies on the line $`l`$ in the same distance from $`\rho ^{(0)}`$, but on the opposite half line (starting in the point $`\rho ^{(0)}`$). The consequence is that the convex combination
$$\frac{1}{2}\rho +\frac{1}{2}U\rho U^{}=\rho ^{(0)}.$$
(30)
The way to achieve any other point lying on the surface of the ball $`b`$ is straightforward. For any such point $`\rho ^{(0)}`$ there exists a two dimensional rotation $`R_{\rho ^{(0)},\rho ^{(0)}}`$, which rotates the point $`\rho ^{(0)}`$ to the point $`\rho ^{(0)}`$. This rotation is realized by some unitary operation $`U_{\rho ^{(0)},\rho ^{(0)}}`$ on the density operators. Therefore, the superoperator encrypting the whole line segment $`l`$ into the point $`\rho ^{(0)}`$ is
$$\widehat{๐}(\rho )=\frac{1}{2}U_{\rho ^{(0)},\rho ^{(0)}}๐\rho ๐U_{}^{}{}_{\rho ^{(0)},\rho ^{(0)}}{}^{}+\frac{1}{2}U_{\rho ^{(0)},\rho ^{(0)}}U\rho U^{}U_{}^{}{}_{\rho ^{(0)},\rho ^{(0)}}{}^{}.$$
(31)
This superoperator has again Kraus decomposition with only two unitary operators, and therefore only a single bit of key is needed. The method how to achieve any $`\rho ^{(0)}b`$ (not only on the surface) is based on the method of encryption of three linearly independent states and will be discussed later in this section.
In this way we have demonstrated that one bit of key is sufficient to encrypt set $`๐`$ containing two linearly independent states. It remains to verify whether one bit is also necessary. There might e.g. exist some encryption operation $`\{(p_1,U_1),(p_2,U_2)\},p_1+p_2=1,p_1p_2`$ encrypting the set $`๐`$. Clearly the entropy of the key of such an operation is smaller than one. We will prove that one bit is necessary by showing that any encryption superoperator $`\widehat{๐}`$ encrypting a line $`l`$ encrypts also the line $`l^{}`$, which is parallel to $`l`$ and intersects $`\frac{1}{2}๐`$. Then the derived inequality (26) implies that one bit is indeed necessary, since $`S(\frac{1}{2}๐)=1`$. Also, $`\widehat{๐}`$ can be used to encrypt a classical bit ($`\{|0,|1\}`$) and this result is in accordance with .
Let us denote $`\rho _1`$ and $`\rho _2`$ the extremal points of the line segment $`l`$ (lying on the surface of the Bloch ball, see figure 3) and $`\rho _1^{}`$ and $`\rho _2^{}`$ the extremal points of the line segment $`l^{}`$. Let us express each of the points $`\rho _1^{}`$ and $`\rho _2^{}`$ as a linear combination of the points $`\rho _1`$, $`\rho _2`$ and $`\frac{1}{2}๐`$. It is easy to see that the coordinates of the points satisfy
$$\rho _1^{}=x\rho _1+y\rho _2+z\frac{1}{2}๐\rho _2^{}=y\rho _1+x\rho _2+z\frac{1}{2}๐.$$
(32)
This relation holds for extremal points of any line segment parallel to $`l`$ and lying in the plane spanned by $`l`$ and $`\frac{1}{2}๐`$. Let $`\widehat{๐}`$ encrypts $`l`$ to some state $`\rho ^{(0)}`$. Then from linearity and unitality of $`\widehat{๐}`$ we obtain
$$\widehat{๐}(\rho _1^{})=x\widehat{๐}(\rho _1)+y\widehat{๐}(\rho _2)+z\frac{1}{2}๐=y\widehat{๐}(\rho _1)+x\widehat{๐}(\rho _2)+z\frac{1}{2}๐=\widehat{๐}(\rho _2^{})$$
(33)
since $`\widehat{๐}(\rho _1)=\widehat{๐}(\rho _2)=\rho ^{(0)}`$ from assumption that $`\widehat{๐}`$ encrypts $`l`$. It follows that $`\widehat{๐}`$ encrypts any line segment parallel to $`l`$ lying in the plane spanned by $`l`$ and $`\frac{1}{2}๐`$.
From Eq. (33) we can also easily determine the state $`\rho ^{(0)}=\widehat{๐}(\rho _1^{})`$, i.e. the state where $`\widehat{๐}`$ sends the line segment $`l^{}`$. It is the state $`\rho ^{(0)}`$ shifted towards $`\frac{1}{2}๐`$, from the equation
$$\widehat{๐}(\rho _1^{})=x\widehat{๐}(\rho _1)+y\widehat{๐}(\rho _2)+z\frac{1}{2}๐=(x+y)\rho ^{(0)}+z\frac{1}{2}๐.$$
(34)
The ratio between $`(x+y)`$ and $`z`$ determines the distance from $`\frac{1}{2}๐`$.
We can also derive an analogical result for PQC encrypting a circle in the Bloch sphere. In this case the fact that it also encrypts all parallel circles implies that this PQC establishes an approximative encryption of the whole Bloch sphere, as defined in . Also, in general, PQC encrypting any set $`\overline{๐}`$ in any Hilbert space encrypts also all spaces parallel in the superplane spanned by $`\overline{๐}`$ and $`\frac{1}{d}๐`$. This will be discussed in detail in a separate paper.
### Three states
In case the set $`๐`$ contains precisely three linearly independent states $`\rho _1,\rho _2,\rho _3`$, their linear span is a plane and valid quantum states from this plane (the intersection with the Bloch ball) form a circle $`c`$ (see figure 4). Since all points of the ball $`b`$, containing all possible candidates for the state $`\rho ^{(0)}`$, have the distance from $`\frac{1}{2}๐`$ the same or smaller than the most mixed state from $`c`$, it follows that the circle $`c`$ touches the ball $`b`$ precisely in the middle of the circle $`c`$. Moreover, this point $`\overline{\rho }=\rho ^{(0)}`$ is the most mixed state from $`c`$.
Following analogical argumentation as in the case of the two states, we construct the TCP superoperator, which encrypts the whole circle and sends it to $`\rho ^{(0)}`$. This superoperator is the same as in the case of two states, it is the superoperator (29). The operator $`U`$ is the rotation around the axis of the circle $`c`$. The saturation of any other point on the surface of the ball $`b`$ is the same as in the case of two states, see Eq. (31).
Using this result we may also design a PQC, which encrypts a set $`๐`$ of two linearly independent states into the arbitrary state $`\rho ^{(0)}`$ inside the ball $`b`$ by using just single bit of key. The ball $`b`$ is now specified by the given line $`l`$ associated with the set $`\overline{๐}`$. The state $`\rho ^{(0)}`$ specifies uniquelly a sphere $`g`$ of the radius $`r=D(\rho ^{(0)},\frac{1}{2}๐)`$, centered in total mixture and containing this state on its surface. There exists a tangent plane $`\kappa `$ to this sphere determined by the original line $`l`$. This plane is generated by three linearly independent states and following the reasoning of this subsection, it can be encrypted into its maximally mixed state by PQC with $`H(\widehat{๐})=1`$. However, the maximally mixed state equals to the only point in the intersection of the plane with the sphere (see figure 5). This point is unitarily equivalent to $`\rho ^{(0)}`$. It means that the original set $`\overline{๐}`$ given by two linearly independent states (forming the line in the plane, $`l\kappa `$) is encrypted by the same PQC (up to unitary transformation) into the state $`\rho ^{(0)}`$.
Let us now proceed with the analytic approach to see how large key is required to encrypt the set $`\overline{๐}`$ into arbitrary state $`\rho ^{(0)}b`$. For three-dimensional set of plaintexts we have a unique PQC that transforms $`\overline{๐}`$ into the state $`\rho ^{(0)}`$. The singular values of the corresponding mapping $`\widehat{๐}`$ reads $`\{\lambda _1,\lambda _2,\lambda _3\}=\{0,0,s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}}\}`$. It follows that
$$\begin{array}{ccc}p_0=p_z\hfill & =& \hfill (1+s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}})/4\\ p_x=p_y\hfill & =& \hfill (1s\sqrt{1+\frac{1}{\alpha ^2}+\frac{1}{\beta ^2}})/4\end{array}$$
(35)
The parameter $`s`$ corresponds to the distance between the state $`\rho ^{(0)}`$ and the total mixture. It is bounded by the inequality $`s\frac{|\alpha \beta |}{\sqrt{\alpha ^2\beta ^2+\alpha ^2+\beta ^2}}`$. Except the case when this inequality is saturated we need four unitary transformations to realize the PQC. For maximal value of $`s`$ two unitary transformations are sufficient. Moreover, they are used with equal probabilities. It means that the limiting case ($`\rho ^{(0)}=\overline{\rho }`$) for two-dimensional and three-dimensional set of plaintexts has the same entropy rates.
If you examine the dependence of the probabilities $`\{p_i\}_i`$ on the parameter $`s`$, then you realize that to encrypt three states a single bit of key is sufficient if $`\rho ^{(0)}`$ is on the surface of the ball $`b`$ (see the beginning of this section). However, as the state $`\rho ^{(0)}`$ is getting closer to $`\frac{1}{2}๐`$, the entropy of the key grows up to two bits.
The key question is whether the derived bound on entropy of PQC encrypting the given set of three linearly independent states is also necessary, i.e. whether it is possible for given $`๐`$ and $`\rho ^{(0)}`$ design a PQC with lower entropy of key than in the derived example. Let us recall the proof that one bit of key is necessary to encrypt the set of two linearly independent plaintexts, see Section 5. It immediately follows that one bit is also necessary to encrypt plaintext containing at least three linearly independent states.
The final step is to prove that also the entropy of the private quantum channel encrypting three dimensional set $`๐`$, with $`D(\rho ^{(0)},\frac{1}{2}๐)<D(\overline{\rho },\frac{1}{2}๐)`$, realized using the Pauli channel is minimal. Let us introduce the entropy exchange $`S_{ex}(\rho _A,\widehat{๐})`$ as the quantity measuring the part of quantum information which is lost into the environment under the action of the channel $`\widehat{๐}`$ providing that the system is initially prepared in the state $`\rho _A`$. Provided that the channel $`\widehat{๐}`$ on the system $`A`$ is realized using unitary operation $`G`$ on a larger system $`AE`$, where $`E`$ is the environment, the entropy of exchange is defined as the von Neumann entropy of the reduced density matrix of the environment after applying the operation $`G`$. It turns out that this entropy is independent of concrete realization of the superoperator.
In particular, any channel $`\widehat{๐}`$ has unitary representation $`\widehat{๐}[\rho _A]=\mathrm{Tr}_E[G(\rho _A|00|)G^{}]=_jA_j\rho _AA_{}^{}{}_{j}{}^{}`$ with $`G=_jA_j|j0|`$. It is known that the entropy of the environment state $`\omega _E=\mathrm{Tr}_A[G(\rho _A|00|)G^{}]=_{jk}\mathrm{Tr}[A_j\rho _AA_{}^{}{}_{k}{}^{}]|jk|`$ does not depend on the particular Kraus representation.
In our case this function is the lower bound of the entropy of the key $`H(\{p_k\}_k)`$, i.e. $`H(\{p_k\}_k)\mathrm{max}_{\rho _A}S_{ex}(\rho ,\widehat{๐})`$. This inequality follows from the fact that $`S(\omega _E)S(\mathrm{diag}_{}[\omega _E])`$<sup>6</sup><sup>6</sup>6The $`\mathrm{diag}_{}[\omega _E]`$ is the all-zero matrix except for the diagonal elements, which are equal to diagonal elements of $`\omega _E`$ in the basis $``$. (definition of von Neumann entropy as minimum of Shannon entropy ovel all measurements) and for PQC channels we have $`\mathrm{diag}_{}[\omega _E]=_kp_k\mathrm{Tr}[U_k\rho _AU_{}^{}{}_{k}{}^{}]|kk|`$. Using the the trace properties and normalization of $`\rho _A`$ we obtain $`\mathrm{diag}_{}[\omega _E]=\{p_k\}_k`$, i.e. $`S(\mathrm{diag}_{}[\omega _E])=H(\{p_k\}_k)`$.
In what follows we will show that for qubit the inequality is saturated for decomposition into orthogonal unitaries, i.e. for Pauli channels. From the previous paragraph it is clear that it is sufficient to show that for some $`\rho `$ the induced enviroment state $`\omega _E`$ is diagonal. Hence, we have to verify the conditions under which the identity $`\mathrm{Tr}[U_j\rho U_{}^{}{}_{k}{}^{}]=0`$ holds for $`jk`$. In such case the inequality is saturated. It is easy to see that by choosing $`\rho =\frac{1}{2}๐`$ this is the condition for orthogonality of transformations $`U_j`$ and this justifies our statement.
We have shown that for orthogonal decomposition of the channel $`\widehat{๐}`$ the entropy of the inequality is saturated, i.e. entropy of the key equals to entropy exchange and this is indeed the maximal value of entropy exchange. Fortunately, the entropy exchange does not depend on the particular decomposition and therefore the entropy of the key cannot be lower for another decompositions. It turns out that for qubits any unital channel can be written as a convex combination of orthogonal unitaries. However, for larger systems this is not the case in general. Consequently, the qubit PQC channel with minimal entropy of the key is the one with orthogonal encoding operations, i.e. the corresponding Pauli channel $`\mathrm{\Phi }_{\widehat{๐}}`$.
The necessary and sufficient entropy of the key is $`1`$ when $`\rho ^{(0)}=\overline{\rho }`$ and it grows up to $`2`$ bits as the state $`\rho ^{(0)}`$ approaches $`\frac{1}{2}๐`$. Therefore, it is natural to express the entropy as a function of the parameter
$$r=\frac{D(\rho ^{(0)},1/2๐)}{D(\overline{\rho },1/2๐)},$$
(36)
where the radius of the Bloch ball is $`1`$.
Let us use the parametrization of states $`\rho _1,\rho _2,\rho _3๐`$ introduced in the Eq. (7) and put $`s=D(\rho ^{(0)},1/2๐)`$ and $`p=D(\overline{\rho },1/2๐)=|\alpha \beta |/\sqrt{1+\alpha ^2+\beta ^2}`$. Comparing it with the Eq. (35) we obtain that the probabilites reads $`p_0=p_z=\frac{1}{4}(1+r)`$ and $`p_x=p_y=\frac{1}{4}(1r)`$, where we used the relation $`r=s/p`$. The evaluation of the entropy for this realization of PQC channel leads us to formula
$$\begin{array}{cc}\hfill H(\{p_j\}_j)& =\underset{j}{}p_j\mathrm{log}_2p_j=2\frac{1}{2}\left[(1+r)\mathrm{log}_2(1+r)+(1r)\mathrm{log}_2(1r)\right],\hfill \end{array}$$
(37)
where $`0r1`$. It is easy to see that $`1H(\{p_j\}_j)2`$. The graph of the function $`H(\{p_j\}_j)`$ depending on the variable $`r`$ is on the Figure 6. Unfortunately, as we see from the graph, the entropy grows very fast as $`r`$ goes to $`0`$. In example for $`r=1/2`$ the entropy is already $`H(\{p_j\}_j)1.81128`$.
## 8 Conclusion
### All single-qubit private quantum channels
In this paragraph we will answer the following question: which unital maps constitute a PQC? We have shown that it is sufficient to consider only Pauli channels, i.e. the maps $`\mathrm{\Phi }_{\widehat{๐}}=\mathrm{diag}\{1,\lambda _1,\lambda _2,\lambda _3\}`$. Nontrivial private quantum channels are characterized by the property, that at least two pure states $`|\psi _1,|\psi _2`$ are mapped into the same state $`\rho ^{(0)}`$. In the Bloch sphere parametrization ($`\stackrel{}{r}=(r_x,r_y,r_z)`$) this means that $`\stackrel{}{r}_1\stackrel{}{r}_1^{}=\stackrel{}{s}`$ and $`\stackrel{}{r}_2\stackrel{}{r}_2^{}=\stackrel{}{s}`$. Using these relations and explicit form of the Pauli channel we come to the following โPQCโ conditions $`0=\stackrel{}{r}_1^{}\stackrel{}{r}_2^{}`$, i.e. $`\lambda _j(r_{1j}r_{2j})=0`$ for all components $`j=x,y,z`$. This equality is satisfied only if $`\lambda _j=0`$ for some $`j`$, or $`r_{1j}=r_{2j}`$. Consider the case when none of the $`\lambda `$s vanishes, i.e. $`\lambda _1\lambda _2\lambda _30`$. It follows that in order to fulfill the PQC conditions $`r_{1j}=r_{2j}`$ for every $`j`$. But it means that the states are the same. Therefore, at least one of the parameters $`\lambda _j`$ must vanish. Otherwise the transformation does not correspond to private quantum channel. The complete positivity condition restricts the possible values of $`\lambda _1,\lambda _2`$ (we put $`\lambda _3=0`$) so that the inequality $`|\lambda _1\pm \lambda _2|1`$ characterize all the possible qubit private quantum channels.
### Multi-qubit generalization
This result can be generalized for a specific class of multi-qubit states. In each of the $`n`$ qubits we choose a set of plaintexts $`\overline{๐}_k`$ in the corresponding Bloch ball. Each of the qubits can be encoded by PQC $`\widehat{๐}_k:\overline{๐}_k\rho _k^{(0)}`$. Following the single qubit results, we design a PQC on each of the qubits, which encrypts any state of the form
$$\underset{i}{}\mu _i\rho _1^{(i)}\mathrm{}\rho _n^{(i)},$$
(38)
where $`i:\mu _i`$, $`_i\mu _i=1`$ and $`i,k:\rho _k^{(i)}\overline{๐}_k`$. Let us denote the set of such states by $`\overline{๐}`$. Note note that this set contains entangled states as well, because not only convex combinations of factorized states are allowed. The values of $`\mu _i`$ are arbitrary. Consider for instance two qubits. Using PQC encryption for $`๐=๐ฎ()`$ on each qubit enables us to encrypt each two-qubit (even entangled) quantum state.
### Other implications
In this paper we derived the restriction that the state $`\rho ^{(0)}`$ of the private quantum channel can be any state which has the distance from the maximally mixed state $`\frac{1}{2}๐`$ the same or smaller than the state $`\overline{\rho }`$, where $`\overline{\rho }`$ denotes the most mixed state in the linear span of $`๐`$. We showed that any of these states can be achieved in the case of the qubit and therefore this condition is also sufficient.
Further, we demonstrated that it is enough to use a single bit of key to encrypt the set $`\overline{๐}`$, which is spanned by two linearly independent states, and that any state of the previously described candidates to the state $`\rho ^{(0)}`$ can be achieved. We derived the same result for the set $`\overline{๐}`$ containing three linearly independent states, but with the restriction that a single bit of the key suffices provided that the state $`\rho ^{(0)}`$ has the same distance from $`1/2๐`$ as the state $`\overline{\rho }`$. As the distance of the state $`\rho ^{(0)}`$ to $`\frac{1}{2}๐`$ approaches $`0`$, the necessary and sufficient entropy of the key approaches $`2`$.
As a special consequence of our derivation we obtain the result of that the state $`\rho ^{(0)}=\frac{1}{2}๐`$ when $`\frac{1}{2}๐`$ is in the convex span of $`๐`$ and two bits of the key are needed to encrypt a qubit. Another special consequence of the above derivations is the result of that to encrypt real combinations of two orthogonal basis states it is necessary and sufficient to use a single bit of key. These real combinations form a circle on the surface of the Bloch ball with center coinciding with the center of the Bloch ball.
Moreover, from the discussion in Section 6 it follows that the impossibility of universal not operation on qubit can be derived from the fact that one bit of the key is not sufficient to encrypt a qubit.
## Acknowledgements
Support of the project GAฤR GA201/01/0413 is acknowledged. M.Z. acknowledges the support of the Slovak Academy of Sciences via the project CE-PI and of project INTAS (04-77-7289).
## Appendix A Bloch sphere and qubit channels
Qubit (two-dimensional quantum system) provides us a very simple and illustrative picture of the state space. Any state can be expressed as a linear combination of the operators $`\{๐,\sigma _x,\sigma _y,\sigma _z\}`$. In particular, each operator $`\rho =\frac{1}{2}(๐+\stackrel{}{r}\stackrel{}{\sigma })`$ has a unit trace and if $`|\stackrel{}{r}|1`$, then it is also positive. Consequently, the state space forms a ball with the unit radius. The equivalence $`\rho \stackrel{}{r}`$ is called Bloch sphere representation (see for instance Refs. ). From the orthogonality relation $`\mathrm{Tr}\sigma _k\sigma _l=2\delta _{kl}`$ the parameters of state are given by a simple formula $`\stackrel{}{r}=\mathrm{Tr}\rho \stackrel{}{\sigma }`$, i.e. as the mean values of the hermitian operators (measurements) $`\sigma _x,\sigma _y,\sigma _z`$.
Let us describe the relation between the density operators $`๐ฎ()`$ (three-parametric subset) embedded in four-dimensional space of Hermitian operators and the Bloch sphere contained in three-dimensional space. Let us denote by $`\rho _j`$ ($`j=1,2,3,4`$) the basis of this space corresponding to four density operators. The vectors $`\stackrel{}{r}_j`$ represents the associated points in the Bloch sphere (in three dimensional real vector space). Only trace-preserving linear combinations, i.e. $`\rho =_ja_j\rho _j`$ with $`_ja_j=1`$ for real $`a_j`$, can be understand as linear combinations of the vectors within the Bloch sphere picture, i.e. $`\stackrel{}{r}=_ja_j\stackrel{}{r}_j`$. In fact, Bloch sphere is situated in the three-dimensional space of Hermitian operators with unit trace, but only special linear combinations ($`_ja_j=1`$) of Bloch vectors has its counterparts in the original space of Hermitian operators.
The structure of qubit channels is known mainly due to work of Ruskai et al. . Let us now briefly present a corresponding geometrical picture. From the mathematical point of view the channels are described by linear trace-preserving completely positive maps $`\widehat{๐}`$ defined on the set of operators. The complete positivity is guaranteed if the operator $`\mathrm{\Omega }_{\widehat{๐}}=(\widehat{๐}๐)P_+`$ is a valid quantum state<sup>7</sup><sup>7</sup>7$`P_+`$ is a projection onto maximally entangled state $`|\psi _+=\frac{1}{\sqrt{2}}(|00+|11)`$.. Any qubit channel $`\widehat{๐}`$ can be illustrated as an affine transformation of the Bloch vector $`\stackrel{}{r}`$, i.e. $`\stackrel{}{r}\stackrel{}{r}^{}=T\stackrel{}{r}+\stackrel{}{t}`$, where $`T`$ is a real 3x3 matrix and $`\stackrel{}{t}`$ is a translation. This form guarantees that the transformation $`\widehat{๐}`$ is hermitian and trace preserving, but the complete positivity conditions define (nontrivial) constraints on possible values of parameters. In fact, the set of all completely positive trace-preserving maps forms a specific convex subset of all affine transformations.
Any matrix $`T`$ can be written in the so-called singular value decomposition, i.e. $`T=R_UDR_V`$ with $`R_U,R_V`$ corresponding to orthogonal rotations and $`D=\mathrm{diag}\{\lambda _1,\lambda _2,\lambda _3\}`$ being diagonal with $`\lambda _k`$ the singular values of $`T`$. This means that any map $`\widehat{๐}`$ is a member of less-parametric family of maps of the โdiagonal formโ $`\mathrm{\Phi }_{\widehat{๐}}`$. In particular $`\widehat{๐}[\rho ]=U\mathrm{\Phi }_{\widehat{๐}}[V\rho V^{}]U^{}`$ with $`U,V`$ unitary operators. The reduction of parameters is very helpful, and most of the properties (also complete positivity) of $`\widehat{๐}`$ is reflected by the properties of $`\mathrm{\Phi }_{\widehat{๐}}`$. The map $`\widehat{๐}`$ is completely positive only if $`\mathrm{\Phi }_{\widehat{๐}}`$ is. Let us note that $`\mathrm{\Phi }_{\widehat{๐}}`$ is determined not only by the matrix $`D`$, but also by a new translation vector $`\stackrel{}{\tau }=R_U\stackrel{}{t}`$, i.e. under the action of the map $`\mathrm{\Phi }_{\widehat{๐}}`$ the Bloch sphere transforms as follows $`r_jr_j^{}=\lambda _jr_j+\tau _j`$.
A special type of completely positive maps are the unital ones, i.e. those for which the total mixture (center of the Bloch sphere) is preserved. For these channels the translation term vanishes, $`\stackrel{}{t}=\stackrel{}{\tau }=\stackrel{}{0}`$, and the Bloch sphere is โshrinkedโ without shifting its center. In this case the analysis of all possible channels is quite simple, because the induced map $`\mathrm{\Phi }_{\widehat{๐}}`$ is uniquely specified only by three real parameters. Positivity of the transformation $`\mathrm{\Phi }_{\widehat{๐}}`$ corresponds to conditions $`|\lambda _k|1`$, i.e. all points lying inside a cube. The conditions of complete positivity demands the validity of the following four inequalities $`|\lambda _1\pm \lambda _2||1\pm \lambda _3|`$. This specifies the tetrahedron lying inside a cube of all positive unital maps with extremal points being four unitary transformations $`๐,\sigma _x,\sigma _y,\sigma _z`$.
It follows that each unital map is unitarily equivalent to the map of the form $`\mathrm{\Phi }_{\widehat{๐}}=\mathrm{diag}\{1,\lambda _1,\lambda _2,\lambda _3\}`$. The set of all unital channels is convex. Obviously the unitary channels are extremal points of this set. Let us consider a Pauli channel $`\widehat{๐}[\rho ]=_kp_k\sigma _k\rho \sigma _k`$, i.e. a general convex combination of four Pauli unitary transformations. Rewriting this action in the Bloch sphere parameters we obtain the transformation
$$\widehat{๐}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 12(p_y+p_z)& 0& 0\\ 0& 0& 12(p_x+p_z)& 0\\ 0& 0& 0& 12(p_x+p_y)\end{array}\right).$$
(39)
As a result we get that unital channels are unitarily equivalent to Pauli channel. Consequently, each unital channel $`\widehat{๐}`$ can be written as a convex combination of (at least) four unitary channels. The probabilities are determined by the parameters of the induced map $`\mathrm{\Phi }_{\widehat{๐}}`$
$$\begin{array}{ccc}\hfill p_x& =& \frac{1}{4}(1+\lambda _1\lambda _2\lambda _3)\hfill \\ \hfill p_y& =& \frac{1}{4}(1\lambda _1+\lambda _2\lambda _3)\hfill \\ \hfill p_z& =& \frac{1}{4}(1\lambda _1\lambda _2+\lambda _3)\hfill \\ \hfill p_0& =& 1p_xp_yp_z.\hfill \end{array}$$
(40)
Unitary channels are rotations of the Bloch sphere. Unital channels are rotations combined with the deformation so that the output states form an ellipsoid centered in the total mixture. The values $`\lambda _j`$ define the size of the ellipsoid along three main axes.
|
warning/0506/cond-mat0506644.html
|
ar5iv
|
text
|
# Critical dynamics and effective exponents of magnets with extended impurities
## I Introduction
Critical properties of structurally disordered magnets remain a problem of great interest in condensed matter physics, since real magnetic crystals are usually non ideal. A simple and natural case of disorder is implemented via the point-like uncorrelated quenched non-magnetic impurities and is experimentally realized as substitutional disorder in uniaxial Folk03 as well as in Heisenberg Pelissetto02 ; Holovatch02 magnets. Examples are given by substitute alloys $`\mathrm{Mn}_\mathrm{x}\mathrm{Zn}_{1\mathrm{x}}\mathrm{F}_2`$, $`\mathrm{Fe}_\mathrm{x}\mathrm{Zn}_{1\mathrm{x}}\mathrm{F}_2`$ for the uniaxial (Ising) magnets Belanger , and by amorphous magnets $`\mathrm{Fe}_{90+\mathrm{x}}\mathrm{Zr}_{10\mathrm{x}}`$, $`\mathrm{Fe}_{90\mathrm{y}}\mathrm{M}_\mathrm{y}\mathrm{Zr}_{10}`$ ( $`\mathrm{M}=\mathrm{Co},\mathrm{Mn},\mathrm{Ni}`$) ammagn ; Perumal01 , transition-metal based magnetic glasses magglass ; Kellner86 as well as disordered crystalline materials $`\mathrm{Fe}_{100\mathrm{x}}\mathrm{Pt}_\mathrm{x}`$ Boxberg94 , $`\mathrm{Fe}_{70}\mathrm{Ni}_{30}`$ Kellner86 and Eu-chalcogenide solid solutions Westerholt for the Heisenberg magnets. The question of a great interest arising here is: does the disorder change critical properties of the systems? The answer is given by the famous Harris criterion Harris74 . It states, that disorder changes the critical exponents, if the critical exponent $`\alpha _p`$ of the pure system is positive: $`\alpha _p=2d\nu _p>0`$. Here, $`\nu _p`$ is the critical exponent governing the divergence of correlation length of the corresponding pure system and $`d`$ is the space dimension.
On the base of this inequality, one can estimate the marginal value $`m_c`$ for the spin $`m`$-vector model of magnetic systems, such that for $`m>m_c`$ the critical exponents remain unchanged by point-like defects whereas for $`m<m_c`$ they cross over to new values. The present estimates for $`m_c`$ in three dimensions definitely imply $`m_c<2`$: $`m_c=1.942\pm 0.026`$ Bervillier86 and $`m_c=1.912\pm 0.004`$ Dudka01 and thus only the pure Ising model ($`m=1`$) is affected by weak point-like uncorrelated disorder at criticality.
But in real magnets one encounters non-idealities of structure, which can not be modelled by simple point-like uncorrelated impurities. Indeed, magnetic crystals often contain defects of a more complex structure: linear dislocations, disclinations, complexes of non-magnetic impurities, embedded in the matrix of the original crystal defectbook . Theoretical studies of critical behavior of magnets containing such โextendedโ (macroscopic) defects have attracted considerable interest Dorogovtsev80 ; Boyanovsky82 ; Prudnikov83 ; Lawrie84 ; Yamazaki86 ; Blavatska02 ; Blavatska03 ; Fedorenko04 ; Weinrib83 ; lr ; mcext ; Lee92 , however, a systematic experimental analysis still remains to be performed.
One of the possible ways to treat defects that extend through the system being randomly distributed in space but oriented in the same direction is to consider them as quenched $`\epsilon _d`$-dimensional nonmagnetic impurities of parallel orientation. This was proposed in the work of Dorogovtsev Dorogovtsev80 . The case $`\epsilon _d=0`$ is associated with point-like defects, and extended parallel linear (planar) defects are described by $`\epsilon _d`$ = 1(2). To give an interpretation to the non-integer values of $`\epsilon _d`$, one may consider patterns of extended defects like aggregation clusters, and treat $`\epsilon _d`$ as the fractal dimension of these clusters Yamazaki86 . However, relation of analytically continued non-integer Euclidean dimension to the fractal dimension is not straightforward dimfract .
It was shown Boyanovsky82 , that presence of extended impurities leads to a generalized Harris criterion, namely, in this case disorder alters the critical behavior of the pure system, if $`\epsilon _d>d2/\nu _p`$. If one considers the point-like disorder with $`\epsilon _d=0`$, the generalized Harris criterion turns to ordinary one cited above. Again, this inequality defines for each value of $`\epsilon _d`$ the critical value $`m_c`$, below which the extended disorder alters the universality class. But now the disorder induced critical behavior of $`d=3`$ systems holds not only for the Ising magnets ($`m=1`$) but for $`m>1`$ as well, as shown in Fig. 1 (region denoted as โDilutedโ in the figure). Therefore, the class of magnets where the new critical behavior may be found is not restricted to the uniaxial ones ($`m=1`$), but include easy-plane ($`m=2`$) and Heisenberg ($`m=3`$) systems. Thus predicted phenomena of a new universal behavior may be experimentally checked for a wider class of magnets.
Another interesting feature of systems with parallel extended defects is that due to the spatial anisotropy they are described by two correlation lengths, one perpendicular, $`\xi _{}`$, and one parallel, $`\xi _{||}`$, to the extended impurities direction Dorogovtsev80 . As the critical temperature $`T_c`$ is approached, their divergences are characterized by corresponding critical exponents $`\nu _{}`$, $`\nu _{||}`$:
$$\xi _{}|t|^\nu _{},\xi _{||}|t|^{\nu _{||}},$$
(1)
where $`t`$ is the reduced distance to the critical temperature $`t=(TT_c)/T_c`$. Anisotropic scaling holds also for the spin-spin pair correlation function at $`T_c`$ and is governed by exponents $`\eta _{}`$ and $`\eta _{||}`$. Whereas the magnetic susceptibility is isotropic, as far as all order parameter components interact with defects in a similar way. However, the dynamic critical behavior is modified, two times of relaxation in directions perpendicular and parallel to the extended impurities $`\tau _{}`$ and $`\tau _{||}`$ behave correspondingly as:
$$\tau _{}\xi _{}^z_{},\tau _{||}\xi _{||}^{z_{||}},$$
(2)
with dynamical exponents $`z_{}`$ and $`z_{||}`$.
Therefore, magnets with quenched extended defects of parallel orientation constitute a large class of systems with a bulk of unusual phenomena worth to be analyzed. Their asymptotic critical behavior was a subject of the field-theoretical renormalization group (RG) analyses Dorogovtsev80 ; Boyanovsky82 ; Prudnikov83 ; Blavatska02 ; Blavatska03 . In particular, a double expansion in both $`\epsilon =4d`$, $`\epsilon _d`$ was suggested and RG functions were calculated Dorogovtsev80 to order $`\epsilon `$, $`\epsilon _d`$; qualitatively, the crossover to a new universality class in the presence of extended defects was supported. These calculations were extended to the second order in Refs. Boyanovsky82 ; Lawrie84 . However, these divergent RG expansion did not give a reliable numerical estimate for the critical exponents - the goal highly desirable both for experimental and simulational purposes. Numerical estimates for the exponents describing static critical behavior of $`d=3`$ magnets with parallel extended impurities were obtained only recently Blavatska02 ; Blavatska03 by applying special resummation technique to the two-loop RG functions of Refs. Boyanovsky82 ; Lawrie84 . Currently, no estimates of similar accuracy exist for the dynamic exponents describing critical slowing down in the magnets under consideration. The theoretical investigation of critical dynamics of system with parallel extended defects was performed in the one-loop approximation Prudnikov83 for the simple models with non-conserved order parameter (model A) and conserved order parameter (model B). Then the RG study of model A critical dynamics was extended to the two-loop order Lawrie84 . But again, the convergence properties of the series obtained did not allow numerical evaluations. Recently, the short-time critical dynamics of the model A was considered in Ref. Fedorenko04 .
The goal of this paper is to apply the state-of-the-art analysis of the divergent RG perturbation theory series to get numerical estimates of the exponents describing model A dynamics of $`d=3`$ magnets with extended impurities for a wide region of impurity dimension $`\epsilon _d`$ and for different values of $`m`$. Complementing existing estimates of the static exponents Blavatska02 ; Blavatska03 , our results will give comprehensive values for the critical exponents of systems with extended impurities and, in this way, will facilitate their experimental analysis. Moreover, since both in experiments and in Monte Carlo (MC) simulations often the asymptotic region is not reached and effective exponents are observed effective , we will calculate these as well. By these estimates we will predict the possible scenarios of approaching the critical point. These are important to perform corresponding experimental checks, as it was shown recently by theoretical Dudka03 and experimental expereff studies of the $`d=3`$ disordered Heisenberg magnets with point-like defects.
The setup of the paper is as follows. In the next section we present the model; in section III the renormalization procedure is discussed. In section IV we apply the resummation techniques to analyze the renormalization group functions in two-loop approximation and present the quantitative estimates for the asymptotic dynamical critical exponents. Section V gives description of possible scenarios for the effective critical behavior and section VI concludes our study.
## II The model
The starting point is the effective Hamiltonian of the model of an $`m`$-vector magnet with $`\epsilon _d`$-dimensional defects, extending throughout the system along the coordinate directions symbolized as $`x_{||}`$ and randomly distributed in perpendicular directions $`x_{}`$ Boyanovsky82 ; Lawrie84 :
$``$ $`=`$ $`{\displaystyle }d^dx[{\displaystyle \frac{1}{2}}(\mu _0^2+V(x))\stackrel{}{\varphi }^2(x)+(_{}\stackrel{}{\varphi }(x))^2`$ (3)
$`+a_0(_{||}\stackrel{}{\varphi }(x))^2+{\displaystyle \frac{u_0}{4!}}(\stackrel{}{\varphi }^2(x))^2].`$
Here, $`\stackrel{}{\varphi }`$ is an $`m`$-component vector field: $`\stackrel{}{\varphi }=\{\varphi ^1\mathrm{}\varphi ^m\}`$, $`\mu _0`$ and $`u_0`$ are the bare mass and the coupling, $`a_0`$ is the bare anisotropy constant, $`_{||}`$ and $`_{}`$ denote differentiation in the coordinates $`x_{||}`$ and $`x_{}`$ and the impurity potential $`V(x)`$ is introduced. The probability distribution of defects has zero mean and variance given by:
$`V(x)V(y)`$ $`=`$ $`v_0\delta ^{d\epsilon _d}(x_{}y_{}).`$ (4)
Here, $`\mathrm{}`$ denotes the averaging over the distribution of defects, (-$`v_0`$) is a positive coupling constant, which is proportional to the concentration of impurities. The constant $`a_0`$ parameterizes the anisotropy, arising in the system due to the presence of extended defects.
We consider critical dynamics of the model (9) for the case of a non-conserved order parameter. For this case the dynamics can be expressed in the Langevin equation form Hohenberg77 :
$$\frac{\varphi ^i(x,t)}{t}=\lambda _0\frac{}{\varphi ^i(x,t)}+\eta ^i(x,t),i=1\mathrm{}m,$$
(5)
where $`\lambda _0`$ is the Onsager kinetic coefficient and $`\eta ^i(x,t)`$ is the Gaussian random-noise source with zero mean and correlation:
$$\eta ^i(x,t)\eta ^j(x^{},t^{})=2\lambda _0\delta (xx^{})\delta (tt^{})\delta _{ij}.$$
(6)
The brackets $`\mathrm{}`$ mean an average with respect to the thermal noise.
Within the field theory approach it is convenient to use the Bausch-Janssen-Wagner formulation Bauch76 which is given by the Lagrangian:
$$[\stackrel{~}{\varphi },\varphi ]=d^dx๐t\underset{i}{}\stackrel{~}{\varphi }^i\left[\frac{\varphi ^i}{t}+\lambda _0\frac{\delta }{\delta \varphi ^i}\lambda _0\stackrel{~}{\varphi }^i\right].$$
(7)
Here, $`\stackrel{~}{\varphi }^i`$ are components of an auxiliary response field introduced to average over the thermal noise. Then correlation and response functions are computed with the help of a weight function $`We^{[\stackrel{~}{\varphi },\varphi ]}`$.
It is known that studying dynamical properties of disordered systems averaging over random impurities can be applied directly to dynamical weight function $`W`$. As it is established in Ref. DeDominicis78 , the configurational averaging can be performed avoiding replica trick Emery75 . However it leads to the same perturbative expansions for the RG functions as those obtained with replica formalism. Therefore both approaches giving equivalent results are possible for dynamics. However the above alternative does not exist when the static critical behavior is analyzed (see e.g. Folk03 ; Pelissetto02 ). As far as both static and dynamic criticality is addressed in this paper, we use the replica trick, representing the logarithm of a weight function in the following form:
$$\mathrm{ln}W=\underset{n0}{lim}\frac{W^n1}{n}.$$
(8)
Finally it leads to study of properties of replicated Lagrangian $`[\stackrel{~}{\varphi },\varphi ]`$ note :
$`[\stackrel{~}{\varphi },\varphi ]`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}\{{\displaystyle }d^dxdt{\displaystyle \underset{i}{}}\stackrel{~}{\varphi }_\alpha ^i[{\displaystyle \frac{\varphi _\alpha ^i}{t}}+\lambda _0(\mu _0^2_{}^2a_0_{}^2)\varphi _\alpha ^i\lambda _0\stackrel{~}{\varphi }_\alpha ^i+{\displaystyle \underset{j}{}}\lambda _0{\displaystyle \frac{u_0}{3!}}\varphi _\alpha ^i\varphi _\alpha ^j\varphi _\alpha ^j]`$ (9)
$`+{\displaystyle \underset{i,j,\beta }{}}\lambda _0^2{\displaystyle \frac{v_0}{2}}{\displaystyle }d^dxd^dydtdt^{}\delta (x_{}y_{})\varphi _\alpha ^i(x,t)\varphi _\alpha ^i(x,t)\varphi _\beta ^j(y,t^{})\varphi _\beta ^j(y,t^{})\},`$
here, the summation over Greek indices spans from 1 to $`n`$ denoting the different replicas and the Latin indices go from 1 to $`m`$ denoting the components of the order parameter. To study behavior of this model (9) in the vicinity of the critical point we apply the minimal subtraction scheme within field-theoretical RG. The description of this approach is given in the next section.
## III RG study
The description of the long-distance properties of the model (9) near the second order phase transition point is performed using the field-theoretical RG method rgbooks . Let us present the renormalization algorithm, developed for Lagrangian field theory. It is well known, that such a theory encounters with ultraviolet divergences, the removal of which is achieved within an appropriate renormalization procedure by a controlled rearangement of the perturbation theory series. The change of bare couplings $`u_0,v_0`$ and of the anisotropy constant $`a_0`$ under renormalization is described by the RG functions:
$`\beta _u(u,v)`$ $`=`$ $`{\displaystyle \frac{u}{\mathrm{ln}\kappa }}|_0,\beta _v(u,v)={\displaystyle \frac{v}{\mathrm{ln}\kappa }}|_0,`$ (10)
$`\zeta _a(u,v)`$ $`=`$ $`{\displaystyle \frac{\mathrm{ln}a}{\mathrm{ln}\kappa }}|_0.`$ (11)
Here, $`u,v,a`$ are the renormalized couplings and anisotropy constant, respectively, $`\kappa `$ is the rescaling parameter, the notation $`|_0`$ indicates differentiation at fixed bare parameters. The bare fields $`\varphi `$, $`\stackrel{~}{\varphi }`$, the mass $`\mu _0`$ and the Onsager kinetic coefficient $`\lambda _0`$ are related to the renormalized ones $`\phi `$, $`\stackrel{~}{\phi }`$ $`\mu `$ and $`\lambda `$ by:
$`\varphi `$ $`=`$ $`Z_\phi ^{1/2}\phi ,\stackrel{~}{\varphi }=Z_{\stackrel{~}{\phi }}^{1/2}\stackrel{~}{\phi },`$
$`\mu _0^2`$ $`=`$ $`Z_{\mu ^2}\mu ^2,\lambda _0^1=Z_\lambda \lambda ^1,`$
where $`Z`$-factors are dimensionless functions of renormalized parameters $`m,u,v,a`$. Their flows are defined by corresponding RG functions:
$`\gamma _\varphi (u,v)`$ $`=`$ $`{\displaystyle \frac{\mathrm{ln}Z_\phi }{\mathrm{ln}\kappa }}|_0,`$ (12)
$`\gamma _{\stackrel{~}{\varphi }}(u,v)`$ $`=`$ $`{\displaystyle \frac{\mathrm{ln}Z_{\stackrel{~}{\phi }}}{\mathrm{ln}\kappa }}|_0,`$ (13)
$`\overline{\gamma }_{\varphi ^2}(u,v)`$ $`=`$ $`{\displaystyle \frac{\mathrm{ln}Z_{\mu ^2}^1}{\mathrm{ln}\kappa }}|_0\gamma _\varphi ,`$ (14)
$`\zeta (u,v)`$ $`=`$ $`{\displaystyle \frac{\mathrm{ln}Z_\lambda }{\mathrm{ln}\kappa }}|_0.`$ (15)
Functions $`\zeta `$ and $`\gamma _{\stackrel{~}{\varphi }}`$ are connected by the relation: $`\zeta =(\gamma _\varphi \gamma _{\stackrel{~}{\varphi }})/2`$.
The fixed points (FPs) $`u^{},v^{}`$ of the RG transformation are defined as common zeroes of the $`\beta `$-functions:
$$\beta _u(u^{},v^{})=0,\beta _v(u^{},v^{})=0.$$
(16)
A FP is stable, if the eigenvalues of the stability matrix, defined as: $`B_{ij}=\beta _{u_i}(u^{},v^{})/\beta _{u_j}(u^{},v^{}),u_i=\{u,v\}`$ have positive real parts. In the case when the stable FP is physically accessible, i.e. can be reached starting from the initial values of the renormalized couplings (in our case, $`u>0`$, $`v<0`$), corresponds to the critical point of the system. The RG functions (12)-(15) taken at this point give the critical exponents of magnetic susceptibility, correlation length and relaxation time:
$`\gamma ^1`$ $`=`$ $`1{\displaystyle \frac{\overline{\gamma }_{\varphi ^2}(u^{},v^{})}{2\gamma _\varphi (u^{},v^{})}},`$
$`\nu _{}^1`$ $`=`$ $`2\overline{\gamma }_{\varphi ^2}(u^{},v^{})\gamma _\varphi (u^{},v^{}),`$ (17)
$`z_{}`$ $`=`$ $`2+\zeta (u^{},v^{}).`$ (18)
Note, that the following relations between the exponents describing the parallel and perpendicular correlation length and relaxation times hold Dorogovtsev80 :
$`\nu _{||}`$ $`=`$ $`\nu _{}(1{\displaystyle \frac{\zeta _a}{2}}),`$
$`z_{||}`$ $`=`$ $`z_{}/(1{\displaystyle \frac{\zeta _a}{2}}).`$ (19)
To obtain the quantitative characteristics of the dynamical critical behavior of magnetic systems with extended impurities, we turn our attention to the RG functions derived in Ref. Lawrie84 in two-loop approximation:
$`\beta _u/u`$ $`=`$ $`\epsilon +{\displaystyle \frac{(m+8)}{6}}u+2v{\displaystyle \frac{(3m+14)}{12}}u^2`$ (20)
$``$ $`{\displaystyle \frac{1}{12}}uv\left[{\displaystyle \frac{2}{3}}(11m+58)+(m4){\displaystyle \frac{\epsilon _d}{3(\epsilon +\epsilon _d)}}\right]`$
$``$ $`v^2{\displaystyle \frac{1}{144}}\left[328+32{\displaystyle \frac{\epsilon _d}{\epsilon +\epsilon _d}}\right],`$
$`\beta _v/v`$ $`=`$ $`\epsilon \epsilon _d+{\displaystyle \frac{4}{3}}v+{\displaystyle \frac{m+2}{3}}u{\displaystyle \frac{7}{6}}v^2`$ (21)
$``$ $`vu{\displaystyle \frac{m+2}{18}}\left[11{\displaystyle \frac{\epsilon _d}{\epsilon +\epsilon _d}}\right]{\displaystyle \frac{5}{12}}{\displaystyle \frac{m+2}{3}}u^2,`$
$`\gamma _\varphi `$ $`=`$ $`{\displaystyle \frac{1}{36}}v^2+{\displaystyle \frac{m+2}{36}}vu+{\displaystyle \frac{m+2}{72}}u^2,`$ (22)
$`\overline{\gamma }_{\varphi ^2}`$ $`=`$ $`u{\displaystyle \frac{m+2}{6}}+{\displaystyle \frac{1}{3}}v{\displaystyle \frac{m+2}{6}}u^224v^2`$ (23)
$``$ $`{\displaystyle \frac{m+2}{24}}vu\left[6{\displaystyle \frac{\epsilon _d}{\epsilon +\epsilon _d}}\right],`$
$`\zeta _a`$ $`=`$ $`{\displaystyle \frac{1}{3}}v{\displaystyle \frac{5}{36}}v^2{\displaystyle \frac{(m+2)}{36}}vu,`$ (24)
$`\zeta `$ $`=`$ $`{\displaystyle \frac{1}{3}}v+{\displaystyle \frac{(m+2)(6\mathrm{ln}\frac{4}{3}1)}{72}}u^2+`$ (25)
$`{\displaystyle \frac{(m+2)}{36}}vu+{\displaystyle \frac{5}{36}}v^2.`$
Here, $`\epsilon =4d`$ and the replica limit $`n=0`$ has been assumed. Putting in Eqs. (20)-(23), (25) $`\epsilon _d=0`$ one regains the RG functions of the $`m`$-vector magnet with point-like defects Janssen95 ; Kleinert95 ; Folk00 .
The description of the critical behavior of magnets with extended impurities carried out in Refs. Dorogovtsev80 ; Boyanovsky82 ; Lawrie84 ; Yamazaki86 was based on the double expansion in two parameters $`\epsilon `$, $`\epsilon _d`$, considering both to be of the same order of magnitude. Being quite clear technically, such a statement causes however certain cautions. Indeed, taking the $`\epsilon `$ as a small parameter is due to the fact that it deviates from the upper critical dimension, where the $`\varphi ^4`$ theory becomes asymptotically free rgbooks , whereas $`\epsilon _d`$ is the dimension of defect itself and obviously has a different physical origin. Therefore, it is desirable to search for an alternative way to analyze the RG functions (20)-(25). Fortunately, such an alternative exists and it is exploited in the field-theoretical approach to critical phenomena Schloms . Namely, it consists in the analysis of minimal subtraction RG functions directly at dimension of space of interest, fixing $`\epsilon `$ and solving the FP equations numerically Schloms (the so-called fixed dimension scheme). It was proposed Blavatska02 ; Blavatska03 to extend the approach of direct evaluation to the RG functions of the present model, i.e. to treat them directly at $`d=3`$ $`(\epsilon =1)`$ for different fixed values of the (non-integer) defect dimensionality $`\epsilon _d`$. In our calculations (see Section IV), we will make use of both possibilities, exploiting $`\epsilon `$, $`\epsilon _d`$ expansion as well as working within the fixed dimension scheme.
In the field-theoretical RG approach the series expansions of the RG functions in powers of the coupling appear to be divergent; moreover, they are characterized by a factorial growth of the coefficients implying a zero radius of convergence rgbooks . To take into account the higher order contributions, the application of special tools of resummation is required Hardy48 . In our previous paper Blavatska03 , analyzing static critical behavior we applied the Chisholm-Borel resummation technique. Here, the Borel image of the initial function is extrapolated by a rational Chisholm Chisholm73 approximant $`[K/L](x,y)`$. One constructs this ratio of two polynomials of order $`K`$ and $`L`$ such that its truncated Taylor expansion is equal to that of the Borel image of the initial function. The resummed function is then calculated by an inverse Borel transform of this approximant. The details can be found in IJMP .
## IV The results
The critical behavior of the model (9) is influenced by the presence of the following FPs: the Gaussian FP G ($`u^{}=v^{}=0`$), the $`O(m)`$-symmetric FP of a pure magnet P ($`u^{}0,v^{}=0`$) and the random FP R ($`u^{}0,v^{}0`$) that governs disorder-induced critical behavior. A polymer FP with $`u^{}=0,v^{}>0`$ is not reachable from the initial values of couplings and therefore is out of interest for our study. Depending on the values of global parameters $`d,\epsilon _d,m`$ one of the above FPs is stable and in the asymptotic limit governs the criticality. However, as we will show in the section V an approach to the asymptotic limit and hence the effective critical behavior is influenced by all the FPs.
For the $`d=3`$ magnets considered here, it is the crossover between FPs P and R that corresponds to the change of the universal properties of an $`m`$-vector magnet upon dilution by $`\epsilon _d`$-dimensional defects. According to the generalized Harris criterion, this crossover occurs at certain marginal value $`m_c(\epsilon _d)`$. At $`m>m_c`$, the FP P is stable, indicating no change in critical behavior, whereas for $`m<m_c`$ one finds the stability of the FP R, displaying the fact of relevance of disorder. From the generalized Harris criterion one can obtain the marginal value $`\epsilon _d^{\mathrm{marg}}`$ as function of $`m`$. Using the six-loop results Guida98 for the correlation length critical exponent $`\nu _p(m)`$ of the pure $`m`$-vector magnet, one obtains: $`\epsilon _d^{\mathrm{marg}}(m=1)=0.173`$; $`\epsilon _d^{\mathrm{marg}}(m=2)=0.016`$; $`\epsilon _d^{\mathrm{marg}}(m=3)=0.172`$; $`\epsilon _d^{\mathrm{marg}}(m=4)=0.300`$. These estimates are shown in Fig. 1 by filled squares connected by a solid line. The figure may serve as a phase diagram of an $`m`$-vector magnet with extended impurities: the new critical behavior is expected in the region of $`\epsilon _dm`$ plane denoted as โDilutedโ. In the region denoted โPureโ the asymptotic critical behavior does not indicate presence of defects, however the effective critical behavior does. Below, we will perform an analysis of the critical dynamics for both regions of the phase diagram.
### IV.1 Fixed $`d=3`$ scheme
We start from the RG $`\beta `$-functions (20), (21): fixing the value $`\epsilon =1`$ (i.e. $`d=3`$) and treating $`\epsilon _d`$ as a varying parameter, we look for the common zeros of the resummed functions $`\beta _u`$ and $`\beta _v`$. The numerical values of the stable FP coordinates obtained for the three-dimensional $`m`$-component magnets with $`m=1,2,3,4`$ can be found in our previous paper Blavatska03 .
To calculate the values of the critical exponents $`z_{}`$ we substitute Eq. (25) for $`\zeta `$ in Eq. (18), apply the resummation procedure for the resulting series, and, finally, estimate them at the stable FP. Note, that because the function $`\zeta `$ is not symmetric in variables $`u`$ and $`v`$ (namely, it does not contain term linear in $`u`$) the Chisholm approximant chosen for its analytic continuation differs from those chosen for the $`\beta `$-functions. The value of critical exponent $`z_{||}`$ is obtained using the relation (19).
In Table 1 we give the obtained results for critical exponents $`z_{||}`$, $`z_{}`$ of three-dimensional $`m`$-component magnets. The case $`\epsilon _d=0`$ corresponds to point-like defects. As it was already noted in the Introduction, for this case only the Ising model ($`m=1`$) is influenced by disorder, and thus for systems with $`m>1`$ the values of the critical exponents are not altered by disorder. Here, one can compare our result with the dynamic exponent of the pure and diluted Ising models, as shown in Tables 2, 3. When $`\epsilon _d`$ increases, for $`m=2,3,4`$ the critical exponents remain constant and equal to the corresponding exponents of the pure model, until $`m`$ becomes $`m_c`$ for given $`\epsilon _d`$ and for $`m>m_c`$ the values of exponents start to increase because they take their new values belonging to the disordered universality class.
An interesting feature of the data shown in Table 1 is that for each fixed $`\epsilon _d`$ the relation $`z_{||}<z_{}`$ holds. One can give the following physical interpretation to this fact. As it has been noted for isotropic systems Hohenberg77 , the dynamical critical exponent $`z`$ is proportional to the pair correlation function critical exponents $`\eta `$:
$$z=2+c\eta .$$
(26)
For the systems we consider here the critical exponents $`\eta _{}`$ and $`\eta _{||}`$, that characterize the behavior of the pair spin-spin correlation function in the directions, perpendicular and parallel to the extended defects, are distinguished Dorogovtsev80 . As far as the extended defects cut interacting bonds of spins perpendicular to the extended-defect direction, in the parallel direction the fluctuations are stronger and therefore $`\eta _{||}<\eta _{}`$ and thus, by Eq. (26) with $`z_{||}`$, $`z_{}`$ we obtain the confirmation to our results. However, in the anisotropic case the coefficient $`c`$ in Eq. (26) is also direction dependent. Thus the above argumentation is rather of qualitative nature to give a hint to physical interpretation of the observed relation $`z_{||}<z_{}`$.
In our previous papers Blavatska02 ; Blavatska03 we have also touched the question of existence of an upper marginal value for the defect dimensionality $`\epsilon _d`$. In our analysis, we observe the disappearance of a stable reachable FP for $`\epsilon _d`$ slightly above $`1`$. The following physical interpretation was proposed: extended defects of large dimension (e.g. parallel planar defects with $`\epsilon _d=2`$), may divide the system into non-interacting regions and thus inhibit ferromagnetic order. Recently in Refs. Fendler05 the Ising magnet with spin interaction bonds, correlated in 2 dimensions and randomly distributed in the perpendicular direction, was studied. Although such a system differs from those considered here, both models possesses a number of common features. In particular, the smearing of a phase transition due to the presence of planar defects was predicted and explained by the existence - although rare - of infinite spatial regions, which are free of defects and therefore may be locally in the ordered phase.
To confirm the obtained results, we have also tried to apply the Padรฉ-Borel resummation to estimate the values for $`z_{}`$, $`z_{||}`$. In this procedure, one rewrites the two-variable series (20), (21) in terms of a resolvent series Watson74 of one variable and then applies a Padรฉ approximant for its analytic continuation. We do not present the results obtained, they give rather close numbers to those in Table 3 and reproduce the same behavior.
### IV.2 $`\epsilon ,\epsilon _d`$-expansion
Another possibility to obtain estimates for dynamic critical exponents of the $`m`$-vector magnet with extended defects is to resum the double $`\epsilon ,\epsilon _d`$ expansions obtained in the two-loop approximation in Ref. Lawrie84 . For $`m<m_c`$, the random FP R is stable and the expression for $`z_{}`$ and anisotropy function $`\zeta _\alpha `$ for $`m1`$ read Lawrie84 :
$`z_{}`$ $`=`$ $`2{\displaystyle \frac{m+2}{4(m1)}}\epsilon +{\displaystyle \frac{(m+8)}{8(m1)}}\stackrel{~}{\epsilon }+\{4(m+2)[5m^2+42m+112192(m1)\mathrm{ln}(4/3)]\epsilon ^2`$ (27)
$``$ $`4(m+2)[27m^2264m240+576(m1)\mathrm{ln}(4/3)]\epsilon \stackrel{~}{\epsilon }`$
$`+`$ $`[59m^3528m^22928m896+1728(m+2)(m1)\mathrm{ln}(4/3)]\stackrel{~}{\epsilon }^2\}(1024(m1)^3)^1;`$
$`\zeta _\alpha `$ $`=`$ $`{\displaystyle \frac{(m+2)}{4(m1)}}\epsilon +{\displaystyle \frac{m+8}{8(m1)}}\stackrel{~}{\epsilon }`$ (28)
$`+`$ $`[4(m+2)(5m^2+10m+144)\epsilon ^24(m+2)(27m^2168m336)\epsilon \stackrel{~}{\epsilon }`$
$`+`$ $`(59m^3240m^22640m1472)\stackrel{~}{\epsilon }^2][1024(m1)^3]^1,\stackrel{~}{\epsilon }=\epsilon +\epsilon _d.`$
Whereas for $`m>m_c`$, the pure FP P is stable. Therefore the critical behavior is isotropic and the dynamic exponent $`z`$ reads:
$$z=2+0.363\frac{\left(m+2\right)\epsilon ^2}{\left(m+8\right)^2}.$$
(29)
As it is known sqrt , due to degeneracy of the $`\beta `$-functions of the weakly diluted Ising model with point-like defects, the usual $`\epsilon `$-expansion for critical exponents turns into the $`\sqrt{\epsilon }`$-expansion. The same holds for the extended defects: indeed, the case $`m=1`$ is to be analyzed separately as far as expressions (27), (28) contain poles at $`m=1`$, and one arrives Lawrie84 at the $`\sqrt{\epsilon }`$-expansion for the critical exponents. In particular, the corresponding expressions for $`z`$ can be given only to the lowest non-trivial order. Moreover, the $`\sqrt{\epsilon }`$-expansion does not allow for a reliable numerical estimate Folk00 . Therefore, for the disordered Ising model, the fixed dimension scheme considered in the subsection IV.1 remains the only way to get numerical estimates.
To get the numerical estimate for $`m_c`$ in the frames of $`\epsilon ,\epsilon _d`$-expansion we substitute into the modified Harris criterion the five-loop $`\epsilon `$ \- expansion for the correlation length critical exponent $`\nu _p`$ for pure $`m`$-component model, given in Kleinert91 and obtain the following expansion:
$`m_c`$ $`=`$ $`(4+8{\displaystyle \frac{\stackrel{~}{\epsilon }}{\epsilon }}2.50000{\displaystyle \frac{\stackrel{~}{\epsilon }^2}{\epsilon }}1.500000\stackrel{~}{\epsilon }2.448919{\displaystyle \frac{\stackrel{~}{\epsilon }^3}{\epsilon }}+3.014557\stackrel{~}{\epsilon }^2+4.141561\epsilon \stackrel{~}{\epsilon }1.682130{\displaystyle \frac{\stackrel{~}{\epsilon }^4}{\epsilon }}`$ (30)
$``$ $`14.12940\epsilon ^2\stackrel{~}{\epsilon }0.5736055\epsilon \stackrel{~}{\epsilon }^2+7.657623\stackrel{~}{\epsilon }^3+55.57104\epsilon ^3\stackrel{~}{\epsilon }16.25104\epsilon ^2\stackrel{~}{\epsilon }^237.62878\epsilon \stackrel{~}{\epsilon }^3+`$
$`+`$ $`22.53257\stackrel{~}{\epsilon }^43.345417{\displaystyle \frac{\stackrel{~}{\epsilon }^5}{\epsilon }}\left)\right(2{\displaystyle \frac{\stackrel{~}{\epsilon }}{\epsilon }})^1,\stackrel{~}{\epsilon }=\epsilon +\epsilon _d.`$
Putting here $`\stackrel{~}{\epsilon }=\epsilon `$ (i.e $`\epsilon _d=0`$) one recovers the $`\epsilon `$-expansion for $`m_c`$ of the model with point-like uncorrelated defects Dudka01 .
To estimate $`m_c`$ numerically for different fixed values of $`d,\epsilon _d`$, one should apply a resummation. As it is known Dudka01 ; Dudka04 already simple Padรฉ-analysis gives convergent results for marginal dimensions. In the Padรฉ analysis Baker81 , it is known that the best convergence is achieved along the main diagonal of a Padรฉ table, therefore we make use of the diagonal \[2/2\] Padรฉ-approximant. In the three-dimensional case ($`\epsilon =1`$) it gives: $`m_c(\epsilon _d=0)=1.92`$ (which is in a good agreement with the known six-loop results for the point-like disorder $`m_c=1.942\pm 0.026`$ Bervillier86 and $`m_c=1.912\pm 0.004`$ Dudka01 ), $`m_c(\epsilon _d=0.1)=2.48`$, $`m_c(\epsilon _d=0.2)=3.10`$, $`m_c(\epsilon _d=0.3)=3.77`$, $`m_c(\epsilon _d=0.4)=4.54`$. The results obtained are plotted by the dotted line in Fig. 1.
To estimate dynamical critical exponents in the FP R we applied the Chisholm-Borel resummation technique to the functions (27), (28) treating them as series in two variables $`\epsilon ,\epsilon _d`$ and using the Chisholm approximant. In the corresponding two-loop approximation the dynamical critical exponents in the pure FP P is given by Eq. (29) and the expression is too short to be resummed. Therefore it has been estimated at $`\epsilon =1`$ and different $`m`$ by a direct substitution. The results are presented in Table 4. They qualitatively confirm our conclusions made on the basis of fixed dimension technique in subsection IV.1 (see Table 3). The discrepancy between the data of the two techniques may serve also as an estimate for accuracy of the perturbation schemes applied.
Let us note, that currently the three-loop series for the exponent $`z`$ of the pure $`m`$-vector magnet model A dynamics is available. The expression found in Antonov84 reads:
$$z=2+0.726(1\epsilon 0.1885)\eta .$$
(31)
Substituting into (31) the $`\epsilon `$-expansion for $`\eta `$ Kleinert91 , one gets:
$$z=2+0.363\frac{\left(m+2\right)\epsilon ^2}{\left(m+8\right)^2}+\left(0.09075\frac{\left(m+2\right)\left(m^2+56m+272\right)}{\left(m+8\right)^4}0.363\frac{0.1885m+0.3770}{\left(m+8\right)^2}\right)\epsilon ^3.$$
(32)
Numerical estimates of this series based on the Padรฉ-Borel resummation are given in the Table 1 for $`m=1`$ and compared with values found by other approaches.
## V Effective critical behavior
The previous discussion concerns the asymptotic critical dynamics of the model (9). Here we discuss the effective critical behavior which is observed in approaching the critical point $`T_c`$ effective . This behavior is characterized by effective critical exponents governing scaling laws when $`T_c`$ still is not reached. The calculation of effective critical exponents for models with extended impurities was not considered so far in dynamics as well as in the statics. However these exponents are mainly observed in numerical and real experiments, which may be outside the asymptotic region.
The effective critical exponents are defined as logarithmic derivatives of the appropriate quantities of interest with respect to the reduced temperature $`t=|TT_c|/T_c`$ effective . For instance, effective critical exponent for perpendicular correlation length is defined as:
$$\nu _{}^{\mathrm{eff}}(t)=\frac{\mathrm{d}\mathrm{ln}\xi _{}(t)}{\mathrm{d}\mathrm{ln}t}.$$
(33)
In the limit $`TT_c`$ the effective exponent coincides with the asymptotic one $`\nu _{}^{\mathrm{eff}}=\nu _{}`$. Within the field-theoretical RG approach effective exponents are calculated using the values of couplings (solution of flow equation) depending on the flow parameter $`\mathrm{}`$. For instance, effective exponents for correlation length and relaxation time perpendicular to the extended impurities direction are defined as:
$`1/\nu _{}^{\mathrm{eff}}(\mathrm{})=2\overline{\gamma }_{\varphi ^2}(u(\mathrm{}),v(\mathrm{}))`$
$`\gamma _\varphi (u(\mathrm{}),v(\mathrm{}))+\mathrm{};`$ (34)
$`z_{}^{\mathrm{eff}}(\mathrm{})=2+\zeta (u(\mathrm{}),v(\mathrm{}))+\mathrm{}.`$ (35)
The flow parameter $`\mathrm{}`$ is related to the temperature distance to the critical point via the inverse correlation length. Therefore the dependence of the effective exponents on the flow parameter corresponds to a dependence on the reduced temperature remark . In (V) and (35) the parts denoted by dots come from the change of the amplitude part of the perpendicular correlation length and critical relaxation time correspondingly. In the subsequent calculations we will neglect this part, since the contribution of the amplitude function to the crossover seems to be small note2 .
Solving the flow equation for different initial conditions we can obtain different flows in the space of couplings. We choose these condition near the Gaussian FP (G), expecting that couplings are small in the background region. In flow equations (10) we use the $`\beta `$-function (20), (21) at fixed $`d`$ resummed by the Chisholm-Borel method. This allows us to investigate also the case of Ising systems $`(m=1)`$, for which the $`\beta `$-function are degenerated on the one-loop level and the $`\sqrt{\epsilon }`$-expansion does not allow for numerical estimates Folk00 .
The RG flows at $`m=1`$ for the two most interesting cases $`\epsilon _d=0`$ and $`\epsilon _d=1`$ are shown in the Fig. 2 (dashed lines for $`\epsilon _d=0`$ and solid lines for $`\epsilon _d=1`$). The first case corresponds to the point defects, whereas the second one corresponds to the lines of defects. We consider the same initial condition for both cases with a small (flows 1 and $`1^{}`$) and with a large (flows 2 and $`2^{}`$) value of the ratio $`v/u`$. The ratio $`v/u`$ defines the degree of disorder, thus we can observe difference in the behavior of systems with low and high dilution.
Using the flows given in the Fig. 2 we can obtain static and dynamic effective exponents. Below, we choose to display the exponents $`\nu _{}^{\mathrm{eff}}`$ and $`z_{}^{\mathrm{eff}}`$. Due to Eq. (19) the parallel effective exponents $`\nu _{||}^{\mathrm{eff}}`$ and $`z_{||}^{\mathrm{eff}}`$ have qualitatively similar behavior. The deviations of effective exponents from the mean-field values $`z_{}^{\mathrm{eff}}2`$ and $`\nu _{}^{\mathrm{eff}}1/2`$ corresponding to flows of Fig. 2 are given in the Fig. 3 and Fig. 4 respectively. First we compare the behavior of the effective exponents of the uniaxial $`m=1`$ magnets at $`\epsilon _d=0`$ and $`\epsilon _d=1`$. As it can be seen from the Figs. 3 and 4, even a weak dilution by lines of defects leads to a faster crossover to the new universality class (curves 1 in Figs. 3 and 4) and the effective exponents are not influenced by the pure FP P. Such a behavior qualitatively differs from that at dilution by point-like impurities, where effective exponents corresponding to the pure FP P are observed in a relatively wide region (curves 1โ in both figures). Note that the dynamical critical exponent $`z_{}^{\mathrm{eff}}`$ at $`\epsilon _d=1`$ approaches its asymptotics from above. Such a nonmonotonic behavior is typical in statics for system with point-like defects Folk03 ; Folk00 . However we do not observe nonmonotonic behavior for $`\nu _{}^{\mathrm{eff}}`$. It may be a feature of static effective critical exponents for system with extended impurities or consequence of our approximation.
We also consider static and dynamic effective behavior for different values of order parameter dimension. Fig. 5 and Fig. 6 present the $`z_{}^{\mathrm{eff}}2`$ and $`\nu _{}^{\mathrm{eff}}1/2`$ correspondingly obtained for the same initial conditions and different values $`m=1,\mathrm{\hspace{0.17em}2},\mathrm{\hspace{0.17em}3},\mathrm{\hspace{0.17em}4}`$. The nonmonotonic character of dependence of exponents on the logarithm of flow parameter is observed only for $`z_{}^{\mathrm{eff}}`$ at $`m=1`$.
## VI Conclusions
The present study is dedicated to providing the reliable numerical estimates for the dynamical critical exponents of magnetic systems with $`\epsilon _d`$-dimensional non-magnetic impurities of parallel orientation. The presence of such impurities leads to anisotropy in the systems; thus, two correlation lengths, parallel and perpendicular to the extended defects exist, which diverge in the vicinity of critical temperature with different exponents. The dynamic critical behavior is modified as well, so that two times of critical relaxation in directions parallel and transverse to extended impurities appear. The former results obtained for such systems are based on the double expansion in parameters $`\epsilon ,\epsilon _d`$ and thus are rather of a qualitative character.
An example of related systems is given by magnets with extended randomly oriented defects: these can be described by the model of Weinrib and Halperin Weinrib83 . There, the distribution of defects is characterized by the pair correlation function, falling off with distance for large separations according to a power law. The influence of such correlated defects on criticality was a subject of several theoretical studies Weinrib83 ; lr and predicted new universal critical behavior found recently its confirmation in the MC simulations mcext .
We applied the resummation to the RG functions of the model, obtained in the minimal subtraction scheme Boyanovsky82 ; Lawrie84 , treating them directly for fixed $`d=3`$ and fixed parameter $`\epsilon _d`$. The case $`\epsilon _d=0`$ describes point-like quenched disorder and reproduces well-known results. For $`\epsilon _d>0`$ it was found, that the relation $`z_{}>z_{||}`$ holds for every $`\epsilon _d`$ and $`m`$, which is connected to the fact, that the extended defects cut interacting paths of spins perpendicular to the extended-defect direction, so in the parallel direction the fluctuations are stronger.
The data of Tables 1, 4 give numerical values of the dynamical critical exponents for the model A dynamics of the $`m`$-vector magnets with extended $`\epsilon _d`$-dimensional defects. Together with possible scenarios of the effective critical behavior discussed in the Section V it should facilitate experimental studies of the influence of extended defects on criticality. As we already noted, the point defects change the universality class at $`d=3`$ only for the Ising magnets ($`m=1`$). Values of the critical exponent $`z`$ of the pure and the diluted (by point-like impurities) Ising model are given in Tables 2, 3, respectively. Comparing the data in these tables one sees an increase of the exponent for the diluted model with respect to the pure one. This corresponds to a stronger divergency of the relaxation time (increase of the critical slowing-down effects). Further, comparing Tables 3 and 1 one sees that extended impurities make this effect more pronounced, leading to further increase of both exponents $`z_{},z_{||}`$. However this increase (of the order of 8 % if one compares $`z`$-exponents for point and line defects) is not dramatic to make MC simulations impossible due to enhanced critical slowing down. On the other hand, the same order of difference in critical exponents values holds also for the static exponents Blavatska02 ; Blavatska03 . Therefore, we expect that the predicted change in the critical behavior of three dimensional magnets with extended impurities is within current experimental accuracy. Although it is more difficult to extract dynamical critical exponents from simulations to the same accuracy, it would be worthwhile to look for them (i) in the cases of different defect dimensionality and (ii) in the non-asymptotic regime.
## Acknowledgements
We thank the Austrian Fonds zur Fรถrderung der wissenschaftlichen Forschung, project No.16574-PHY which supported in part this research. V.B. thanks the W. Macke Stiftung for enabling her research stay in Linz.
|
warning/0506/nucl-th0506006.html
|
ar5iv
|
text
|
# A dynamical model for pion electroproduction on the nucleon
## I Introduction
The electron scattering on the nucleon is the main source of information about the nucleon structure and the nature of strong interaction. As such it has been a subject of comprehensive experimental and theoretical studies for the past several decades. Most recently, electron and photon beams have been used at several facilities such as JLab Frolov ; Joo , LEGS LEGS , MAMI MAMI , and MIT-Bates Bates , to investigate the electroexcitation of nucleon resonances with an unprecedented precision. In these works the parameters of the $`\mathrm{\Delta }`$(1232)-resonance electroexcitation โ the $`\gamma ^{}N\mathrm{\Delta }`$ transition form factors โ are extracted from the observables of pion electroproduction on the proton ($`epe^{}p\pi `$).
Such determinations of resonance properties from experimental data require theoretical input, which, at present, is provided by the partial-wave analyses SAID SAID and MAID MAID1 , as well as by various dynamical models, e.g., Sato2 ; DMT ; Azn:2004 . We recently developed a new dynamical model of pion electroproduction on the nucleon electroCaia . The model, henceforth called the OHIO model, has the advantage of building in the correct electromagnetic form factors for the nucleon and pion-exchange (Born) contributions. Therefore we avoid the problem of preserving electromagnetic gauge-invariance by using the same electromagnetic form factors, as is done in the other descriptions, see, e.g., in MAID1 ; Sato2 ; DMT .
Our model is based on solving a quasipotentially-reduced Bethe-Salpeter equation for pion-nucleon system where the photon is then subsequently attached to describe the photopion reaction. We use the equal-time quasipotential reduction which amounts to neglecting the relative-energy dependence of the potential of the Bethe-Salpeter equation thus leading to the Salpeter equation. In this approximation the pion, nucleon, and resonance exchanges which take place in the one-hadron-exchange potential appear instantaneously โ the retardation effects are neglected.
While in the case of pion-nucleon scattering and pion-photoproduction this quasipotential reduction can be implemented rather straightforwardly, inclusion of the virtual photons technically more difficult because of appearance of new singularities. These singularities are associated with the production channels (cuts) that involve particles exchanged in the driving force. We do not include these production channels in a unitary fashion and therefore would like to evade corresponding singularities. In the Salpeter formalism this can be achieved by fixing the relative-energy variable to a $`Q^2`$-dependent value, where $`Q`$ is the photon virtuality. Another viable choice, adopted by us previously electroCaia ; CaiaThesis , is the โspectatorโ approximation applied to the electroproduction potential. In this paper we elaborate on the details of our model with an emphasis on how we deal with the problem of singularities of the Salpeter equation. We also compare the results of our model with the recent polarization data from JLab.
The paper is organized as follows. In Sec. II, we describe the model and outline the problem of the exchange singularities. In Sec. III we present the two choices of quasipotential reduction which evade these singularities. In Sec. IV we study the model results for both choices, in comparison with recent experimental data from CLAS at JLab.
## II The model and particle-exchange singularities
The OHIO model is based on the unitarity dynamics of the $`\pi N`$ scattering model presented in piNscatt , where it is shown to be possible to approach the electromagnetic induced reactions in a way which satisfies the unitarity in the photo-pion channel space, and hence obeys exactly the Watson theorem. The model is based on a $`\pi N`$-$`\gamma N`$ coupled-channel equation which, when solved to first order in the electromagnetic coupling $`e`$, leads to the electroproduction amplitude, $`T_{\pi \gamma ^{}}=V_{\pi \gamma ^{}}+T_{\pi \pi }G_\pi V_{\pi \gamma ^{}}`$, where $`V_{\pi \gamma ^{}}`$ is the basic electroproduction potential, $`G_\pi `$ is the pion-nucleon propagator and $`T_{\pi \pi }`$ is the full $`\pi N`$ amplitude. Thus, pion rescattering effects are included as the final state interaction.
The advantage of this approximation is that the scattering equation has to be solved iteratively only for the $`\pi N`$ scattering amplitude and then one can evaluate the electromagnetic amplitudes in a one loop calculation.
Our model for the pion production potential (i.e., the driving term $`V_{\pi \gamma ^{}}`$), is given in VladThesis ; CaiaThesis and includes the following tree-level contributions: $`N`$ direct and crossed terms, $`t`$\- channel $`\pi `$, $`\omega `$ and $`\rho `$ exchanges, the Kroll-Ruderman (contact) term, and the direct and crossed $`\mathrm{\Delta }`$ terms (see Fig.1). The model for the $`\pi N`$ driving term (i.e. $`V_{\pi \pi }`$) has been described in detail in piNscatt ; VladThesis .
The possibility of a large negative mass squared of the virtual photon poses a serious problem in the case when one calculates the off-shell elements of the matrix $`V_{\pi \gamma ^{}}`$ (i.e. $`Q^20`$). Namely, singularities are encountered in the integration path when the one-loop integration is performed. In the following we will show how this happens, and give a prescription for avoiding this problem. The singularities only arise in the $`t`$\- and $`u`$-channel exchange terms.
Consider for instance the denominator of the nucleon propagator in the case of the $`u`$channel exchange:
$`S_N(pk)={\displaystyle \frac{p/k/+m_N}{u(Q^2,W,|\stackrel{}{q}^{\prime \prime }|,\mathrm{cos}(\theta ))m_N^2}}`$ (1)
where
$`u(Q^2,W,\stackrel{}{q}^{\prime \prime },\mathrm{cos}\theta ))m_N^2`$
$`=\left(E_Nk_0\right)^2\left(\stackrel{}{p}\stackrel{}{k}\right)^2m_N^2`$
$`=\left(E_N\beta \sqrt{s}+q_0^{\prime \prime }\right)^2`$
$`\left(\sqrt{(\stackrel{}{p}+\beta \stackrel{}{P}\stackrel{}{q}^{\prime \prime })^2+m_N^2}\right)^2,`$ (2)
where $`m_N`$, $`m_\pi `$ are the masses of the nucleon and pion, respectively, $`W=\sqrt{s}=P_\mu P^\mu `$ is the total CM energy of the system, and the relative intermediate energy and momentum are designated by $`q_o^{\prime \prime }`$ and $`\stackrel{}{q^{\prime \prime }}`$ respectively. In Eq. (II) we have made use of the fact that the incoming nucleon has 4-momentum defined by $`p^\mu =(E_N,\stackrel{}{p})`$ (it is fully on-shell), and the outgoing $`\pi `$ has 4-momentum $`k^\mu =(k_0,\stackrel{}{k})`$, which written in terms of the relative 4-momentum of the outgoing channel has the energy $`k_0=\beta \sqrt{s}q_0^{\prime \prime }`$ and the 3-momentum $`\stackrel{}{k}=\beta \stackrel{}{P}\stackrel{}{q}^{\prime \prime }`$, where the Lorentz scalars $`\alpha `$ and $`\beta `$ are: $`\alpha =pP/s=(s+m_N^2m_\pi ^2)/2s`$ and $`\beta =kP/s=(sm_N^2+m_\pi ^2)/2s`$. We have also assumed that all the kinematics are in the CM frame (i.e., $`P^\mu =(P^0,0)=(\sqrt{s},0)`$). In this frame, and from kinematics considerations the asymptotic energies of the incoming nucleon and outgoing pion are: $`\omega _\pi =\left(W^2m_N^2+m_\pi ^2\right)/2W,`$ and $`E_N=\left(W^2+m_N^2+Q^2\right)/2W,`$ respectively. The on-shell 3-momentum of the incoming nucleon is $`|\stackrel{}{p}|=\sqrt{E_N^2m_N^2}`$.
The problem of singularities arises when the denominator of the propagator described by the function
$`f`$ $`=`$ $`\left[E_N\omega _\pi +q_0^{\prime \prime }\sqrt{\left(\stackrel{}{p}+\stackrel{}{q}^{\prime \prime }\right)^2+m_N^2}\right]`$
$`\times `$ $`\left[E_N\omega _\pi +q_0^{\prime \prime }+\sqrt{\left(\stackrel{}{p}+\stackrel{}{q}^{\prime \prime }\right)^2+m_N^2}\right],`$
vanishes. Note that in the equal time approximation which we use for the real photon case, $`q_0^{\prime \prime }=0`$. The relative asymptotic energy of the nucleon-pion system is given by:
$`E_N\omega _\pi ={\displaystyle \frac{2m_N^2m_\pi ^2+Q^2}{2W}},`$ (4)
while the square root term ranges from $`m_N`$ to $`\mathrm{}`$ as the integration variable $`|\stackrel{}{q}^{\prime \prime }|`$ ranges over all possible values. Since the minimum value of $`W`$ is $`m_N+m_\pi `$, the function $`f`$ cannot vanish for $`Q^2=0`$. (Note that it is the first term in $`f`$ which could possibly have a zero, but for $`Q^2=0`$ this term is always negative.) However, as $`Q^2`$ increases and $`E_N\omega _\pi `$ becomes larger (by the addition of $`Q^2/2W`$ as compared to the $`Q^2=0`$ case), the first term in $`f`$ can vanish. It vanishes at the smallest value of $`Q^2/2W`$ when $`\stackrel{}{p}`$ and $`\stackrel{}{q}^{\prime \prime }`$ are antiparallel and equal in magnitude.
### II.1 $`Q^2/2W`$ choice
The singularity in the $`u`$-channel nucleon exchange can be evaded by fixing the relative energy $`q_0^{\prime \prime }=Q^2/2W`$, instead of to zero as in the equal-time reduction. This choice has the advantage that the propagators in the $`u`$-channel exchanges are not modified at the photon point. Similar considerations also apply for the case of the $`u`$channel $`\mathrm{\Delta }`$ exchange, $`t`$channel $`\pi `$, $`\rho `$, $`\omega `$ exchanges, as well as for the case of the hadronic form factors, such as the pion (monopole), and the rho and omega (one boson) form factors. The same choice for $`q_0^{\prime \prime }`$ works for the u-channel $`\mathrm{\Delta }`$ exchange while for the $`t`$-channel terms and hadronic form factors, one must choose $`q_0^{\prime \prime }=+Q^2/2W`$. We shall refer to these modifications of the propagators and form factors as the $`Q^2/2W`$ approximation.
The main objection to this choice would be that it violates current conservation even at the on-shell values of the potential matrix ($`V_{\pi \gamma ^{}}`$). This problem could be fixed by a global restoration of the current such as:
$$J^\mu J_{}^{}{}_{}{}^{\mu }(Q^2)=J^\mu n^\mu \frac{qJ}{qn},$$
(5)
where $`J^\mu `$ is the $`4`$component electromagnetic current and $`n^\mu `$ is an arbitrary $`4`$vector. However, since one of the main goals of our work was to construct a gauge invariant current, at least for the on-shell tree level, such a restoration of current conservation is unsatisfactory.
Instead, we choose to restore the electromagnetic gauge-invariance in the following way. The isospin decomposition of the pion electroproduction amplitude is done as follows:
$`m_{(T)}^\mu `$ $`=`$ $`{\displaystyle \frac{1}{3}}\tau _a\tau _3m_{(1/2)}^\mu +\tau _am_{(0)}^\mu `$ (6)
$`+`$ $`\left(\delta _{a3}{\displaystyle \frac{1}{3}}\tau _a\tau _3\right)m_{(3/2)}^\mu `$
where $`\tau `$ are the usual Pauli matrices, index a stands for the pion isospin states ($`\pi ^\pm `$, $`\pi ^0`$) and the lower index, in brackets, refers to the total isospin. Using Eq. (6) one can determine the contribution of each exchange to the isospin decomposed amplitudes.
$`m_{(T=\frac{1}{2})}^\mu `$ $`=`$ $`3m_{(s,N)}^\mu m_{(u,N)}^\mu +4m_{(t,\pi )}^\mu `$ (7a)
$`+`$ $`4m_{(KR)}^\mu +m_{(t,\omega )}^\mu +2m_{(u,\mathrm{\Delta })}^\mu `$
$`m_{(T=\frac{3}{2})}^\mu `$ $`=`$ $`2m_{(u,N)}^\mu 2m_{(t,\pi )}^\mu 2m_{(KR)}^\mu `$ (7b)
$`+`$ $`m_{(t,\omega )}^\mu +{\displaystyle \frac{3}{2}}m_{(s,\mathrm{\Delta })}^\mu +{\displaystyle \frac{1}{2}}m_{(s,\mathrm{\Delta })}^\mu `$
$`m_{(T=0)}^\mu `$ $`=`$ $`m_{(s,N)}^\mu +m_{(u,N)}^\mu +m_{(t,\rho )}^\mu .`$ (7c)
The gauge invariance of the electromagnetic interactions can be imposed by the current conservation condition, $`q_\mu m_T^\mu =0`$, for all values of the isospin: $`T=0,`$ $`\frac{1}{2}`$, $`\frac{3}{2}`$. Due to the violation of current conservation introduced by our approximation in the denominators of the on shell potential matrix $`m_T^\mu `$, Eqs. (7) have to be modified accordingly. In the following we introduce the correction terms necessary for each isospin channel at tree level. The following notation will be used for this derivation: the unmodified denominators are denoted by a unprimed symbol, while the modified denominators, which include the $`{}_{}{}^{\prime \prime }x_{}^{\mu }=(Q^2/2W,\stackrel{}{0})^{\prime \prime }`$ approximation, are denoted by a primed symbol. The $`u`$ and $`t`$channel denominators and the hadronic form factor of the exchanged pion can be written:
$`d_u`$ $`=`$ $`p_up_um_N^2`$
$`d_u^{}`$ $`=`$ $`(p_ux)(p_ux)m_N^2,`$ (8a)
$`d_t`$ $`=`$ $`p_tp_tm_\pi ^2`$
$`d_t^{}`$ $`=`$ $`(p_t+x)(p_t+x)m_\pi ^2,`$ (8b)
$`d_{f_\pi }`$ $`=`$ $`p_tp_t\mathrm{\Lambda }_\pi ^2`$
$`d_{f_\pi ^{}}`$ $`=`$ $`(p_t+x)(p_t+x)\mathrm{\Lambda }_\pi ^2,`$ (8c)
where $`p_u=(pk)^\mu `$ and $`p_t=(qk)^\mu `$. From Eqs. (8) the following relationships between denominators result:
$`d_u^{}d_u`$ $`=`$ $`q_\mu q^\mu \mathrm{\Phi }`$ (9a)
$`d_t^{}d_t`$ $`=`$ $`q_\mu q^\mu \mathrm{\Psi }`$ (9b)
$`d_{f_\pi ^{}}d_{f_\pi }`$ $`=`$ $`q_\mu q^\mu \mathrm{\Psi }`$ (9c)
where
$`\mathrm{\Phi }`$ $`=`$ $`\left(Q^2+4m_N^22m_\pi ^2\right)/\left(4W^2\right),`$ (10)
$`\mathrm{\Psi }`$ $`=`$ $`\left(Q^2+2m_\pi ^2\right)/\left(4W^2\right).`$ (11)
Using Eq. (9) the u and t-channels propagators, and the pion form factor in our approximation write:
$`S_N^{}(p_u)`$ $`=`$ $`S_N(p_u)\left(d_u/d_u^{}\right)`$ (12a)
$`S_\pi ^{}(p_t)`$ $`=`$ $`S_\pi (p_t)\left(d_t/d_t^{}\right)`$ (12b)
$`f_\pi ^{}(p_t)`$ $`=`$ $`f_\pi (p_t)\left(d_{f_\pi }/d_{f_\pi ^{}}\right)^2.`$ (12c)
Current conservation has to be satisfied by the isospin projected on shell amplitudes (Eqs. (7)) separately. Since the resonant contributions as well as rho and omega exchanges satisfy the condition by their Lorentz structure, then the problem reduces just to the Born terms (i.e. nucleon direct and crossed terms, pion and Kroll-Rudermann terms). Using the modifications introduced in Eqs. (7) and Eqs. (9), and constructing the invariant $`q_\mu m_T^\mu `$ for each isospin separately the following violating terms result:
$`q_\mu \mathrm{\Delta }m_{\frac{1}{2}}^\mu `$ $`=`$ $`[3+{\displaystyle \frac{d_u}{d_u^{}}}4{\displaystyle \frac{d_t}{d_t^{}}}\left({\displaystyle \frac{d_{f_\pi }}{d_{f_\pi ^{}}}}\right)^2]\gamma _5k/`$ (13a)
$`+`$ $`4[{\displaystyle \frac{d_t}{d_t^{}}}\left({\displaystyle \frac{d_{f_\pi }}{d_{f_\pi ^{}}}}\right)^21]\gamma _5q/`$
$`q_\mu \mathrm{\Delta }m_{\frac{3}{2}}^\mu `$ $`=`$ $`2[{\displaystyle \frac{d_u}{d_u^{}}}{\displaystyle \frac{d_t}{d_t^{}}}\left({\displaystyle \frac{d_{f_\pi }}{d_{f_\pi ^{}}}}\right)^2]\gamma _5k/`$ (13b)
$``$ $`2[{\displaystyle \frac{d_t}{d_t^{}}}\left({\displaystyle \frac{d_{f_\pi }}{d_{f_\pi ^{}}}}\right)^21]\gamma _5q/`$
$`q_\mu \mathrm{\Delta }m_0^\mu `$ $`=`$ $`(1{\displaystyle \frac{d_u}{d_u^{}}})\gamma _5k/`$ (13c)
After some straightforward algebra in Eqs. (13) the analytical terms necessary for correcting the violation of the current conservation, introduced by our approximation, for the on-shell amplitude are:
$`\mathrm{\Delta }m_{\frac{1}{2}}^\mu `$ $`=`$ $`{\displaystyle \frac{\gamma _5k/\mathrm{\Phi }q^\mu }{d_u^{}}}+{\displaystyle \frac{4\mathrm{\Psi }q^\mu \gamma _5(k/q/)}{d_t^{}}}\mathrm{\Omega }`$ (14a)
$`\mathrm{\Delta }m_{\frac{3}{2}}^\mu `$ $`=`$ $`{\displaystyle \frac{2\gamma _5k/\mathrm{\Phi }q^\mu }{d_u^{}}}+{\displaystyle \frac{2\mathrm{\Psi }q^\mu \gamma _5(k/q/)}{d_t^{}}}\mathrm{\Omega }`$ (14b)
$`\mathrm{\Delta }m_0^\mu `$ $`=`$ $`{\displaystyle \frac{\gamma _5k/\mathrm{\Phi }q^\mu }{d_u^{}}}`$ (14c)
where
$`\mathrm{\Omega }=1{\displaystyle \frac{(d_t^{}Q^2\mathrm{\Psi })(2d_{f_\pi ^{}}+Q^2\mathrm{\Psi })}{d_{f_\pi ^{}}^2}}`$ (15)
Comparing the isospin factors from Eqs. (14) with those from Eqs. (7) one can exactly identify where each of the contributions from Eqs. (14) has to be incorporated into the calculations. Notice that, as expected, they do not have an effect on the transverse components of the current, hence only the longitude multipoles would be affected by this correction.
While these additional terms clearly restore current conservation at tree level, when we carry out the full dynamical calculation using this procedure or the spectator approximation, we also impose current conservation numerically in the off-shell contributions. We confirmed that the use of a numerical restoration of current conservation at the tree level reproduced the same results as given by the analytical results given above.
### II.2 Spectator choice
Previously electroCaia ; CaiaThesis we have used a different method of avoiding the particle-exchange singularities namely the spectator approximation Gross made in the electroproduction potential. As pointed out in Refs. GrS93 ; PaT99 the choice of the spectator particle depends on whether the potential is of the $`t`$-channel or $`u`$-channel exchange. For the $`u`$channel exchange the outgoing pion is the spectator and therefore should be placed on the mass shell. Then the outgoing nucleon has the energy:
$$E_N^{}=W\sqrt{\stackrel{}{q}^{\prime \prime 2}+m_\pi ^2}.$$
(16)
Under the spectator approximation the first term in Eq. (II) becomes equal to $`W2\sqrt{\stackrel{}{q}^{\prime \prime 2}+m_\pi ^2}\sqrt{(\stackrel{}{p}+\stackrel{}{q}^{\prime \prime })^2+m_N^2}`$, which remains negative for all values of $`|\stackrel{}{q}^{\prime \prime }|`$.
If the potential is of the form of $`t`$channel exchange the outgoing nucleon is set on its mass shell. Thus the outgoing pion energy is:
$$\omega _\pi =W\sqrt{\stackrel{}{q}^{\prime \prime 2}+m_N^2},$$
(17)
which also avoids the singularities in the $`t`$-channel terms. An advantage of the spectator approximation is that the on-shell potential matrix ($`V_{\pi \gamma ^{}}`$) does not violate current conservation, since this approximation, at the tree level, corresponds to the asymptotic kinematics.
The main disadvantage of the spectator approximation is that it modifies the $`\mathrm{๐๐}\mathrm{๐ โ๐๐๐}`$ behavior even at the photon point, where no anomalous singularities arise in the u-channel and t-channel terms. In order to describe the $`Q^2=0`$ data in the spectator approximation, it is necessary to refit some of the electromagnetic couplings (the most dramatic change arose in the magnetic coupling of $`\mathrm{\Delta }`$, $`G_M`$). We have done this refitting and all the results presented in electroCaia ; CaiaThesis were calculated using this approximation.
## III Results and Discussion
Using the $`Q^2/2W`$ approximation with the restoration of current conservation as described above, we have fitted the major electroproduction multipoles that have been extracted from experiment using the coupling constants determined by pion photoproduction.
The parametrization of the $`\gamma N\mathrm{\Delta }`$ form factors we universally use the form as in electroCaia :
$$G_I(Q^2)=g_I\frac{1+(Q^2/A_I)e^{Q^2/B_I}}{(1+Q^2/\mathrm{\Lambda }_I)^2},I=M,E,C.$$
(18)
Here we have built in a constraint from perturbative QCD (pQCD) such that these form factors fall as $`Q^4`$ (modulo logs) for asymptotically large $`Q`$, see e.g. Carlson98 . In Table 1 are shown the parameters we used in the $`N\mathrm{\Delta }`$ form factors using the spectator approximation ($`II`$), while ($`I`$) denotes the same parameters using our $`Q^2/2W`$modification.
In Fig. 2 we plot the $`Q^2`$ dependence of the determined electromagnetic form factors in comparison with the standard dipole form factor $`F_D(Q^2)`$ using $`\mathrm{\Lambda }`$=$`0.71`$ $`GeV^2`$. The solid lines are the form factors determined in the $`Q^2/2W`$approximation, while the dashed lines are the same form factors determined using the spectator approximation. Note that all the form factors are normalized to $`1`$ at the photon point and that the coupling constants for the two fits are different as given in Table I. One sees that the magnetic form factor $`G_M(Q^2)`$, in both parameterizations, is overall harder than the dipole at low $`Q^2`$, but it tends toward the dipole values at higher $`Q^2`$, and the two different methods of regularizing the scattering equation lead to moderate effects in the $`Q^2`$ dependence of the form factor. The electric and Coulomb form factors $`G_E(Q^2)`$ and $`G_C(Q^2)`$ start much softer than the dipole; but $`G_E`$ at higher $`Q^2`$ tends to return to $`F_D`$, while $`G_C(Q^2)`$ remains a fraction of $`F_D`$ at high $`Q^2`$. Unlike the magnetic case, both the electric and Coulomb form factors $`G_E`$ and $`G_C`$ have almost the same $`Q^2`$ dependence in both fits. Note that in all cases the overall trend of the $`N\mathrm{\Delta }`$ transition form factors as a function of $`Q^2`$ is similar. We can consider the differences as a measure of the uncertainty of our model.
We have been trying to determine quantitatively the effect of the two approaches used in avoiding the singularities in the u- and t-channel terms. Why do the coupling constants have to be modified from one approach to the other? The method of avoiding the singularities affects only the u- and t-channel contributions. Hence, in Fig. 3 we plot our calculations of the resonant multipoles with the full rescattering model, but in the driving term $`V_{\pi \gamma ^{}}`$ we include just the pion pole (right panel) contribution (t-channel pion exchange) or just the nucleon u-channel exchange (left panel). The t-channel contribution to the $`M_{1+}`$ multipole is weakly affected by this modification, while the u-channel changes significantly. This shift results in $`g_M`$ changing from 2.67 to 3.10. The contribution to the $`E_{1+}`$ multipole at low $`Q^2`$ between model ($`I`$) and model ($`II`$) is quite large, resulting in $`g_E`$ changing from 0.112 to 0.048. The differences in the two models at higher $`Q^2`$ are not so large since the effects in the t- and u-channel terms partially cancel. The largest effect is in the $`S_{1+}`$ multipole. The u-channel contribution varies significantly at low $`Q^2`$ between models ($`I`$) and ($`II`$), while at larger $`Q^2`$ it is the t-channel contribution which varies. Using the $`Q^2=0`$ results we had to change $`g_C`$ from -0.38 to -0.18.
Depending on the size of the u- and t-channel contributions to the various multipoles, the values at $`Q^2=0`$ can change significantly and the extracted $`Q^2`$ dependence for $`G_M`$, $`G_E`$ and $`G_C`$ may change from model ($`I`$) to model ($`II`$). Fortunately, the overall $`Q^2`$ dependence as shown in Fig. 2 is not changed very much. The nucleon exchange term has a larger effect in the $`M_{1+}`$ than the pion pole, in fact, the u-channel change leads to a modification of the shape of the fitted $`G_M(Q^2)`$. In the case of the $`E_{1+}`$ quadrupole, the $`Q^2`$ dependence is not affected in any of the instances, hence we did not have to refit the $`Q^2`$ dependence of $`G_E(Q^2)`$. The $`Q^2`$ dependence of $`G_C(Q^2)`$ had to be slightly modified since we see an overall strengthening of both the pion pole and nucleon crossed terms (above $`Q^21`$ $`(GeV/c)^2`$) from model ($`II`$) to model ($`I`$), hence we had to make $`G_C(Q^2)`$ slightly harder from model ($`II`$) to model ($`I`$). We did not plot the effects of these two models in the $`\mathrm{\Delta }`$ crossed term but we analyzed them, and as expected, they did not play an important role in the resonant multipoles (most of the $`\mathrm{\Delta }`$ contribution to the resonant multipoles comes from the direct and the Born terms).
In Fig. 4 and Fig.5 we compare our calculations with some recent results from CLAS alt ; siglt . This asymmetry measurement is important in determining the $`t`$channel pion pole and contact Born terms contributions in $`\pi ^+n`$, which otherwise are weak in the $`\pi ^0p`$ channel. Measurements of beam asymmetry $`A_{LT^{}}`$ where made only for the neutral channel, but since in this reaction the non-resonant amplitude strongly interferes with the imaginary part of the dominant $`\mathrm{\Delta }(1232)`$ $`M_{1+}^{(3/2)}`$ it is difficult to extract the information about the non-resonant contribution to the observables. The resulting amplitude for $`\pi ^0p`$ is strongly dependent on the rescattering correction and largely dependent on the way the model generates the width of the resonance.
The calculated angular distribution of $`\sigma _{LT^{}}`$ for the $`\pi ^+n`$ channel show a strong forward peaking for energies around the $`\mathrm{\Delta }`$, in contrast to $`\pi ^0p`$ channel which shows backward peaking. In Fig. 4 and Fig. 5 we plot the asymmetry $`A_{LT^{}}`$ and the longitudinal-transverse polarized structure function $`\sigma _{LT^{}}`$, respectively. The relationship between these two quantities, following the reference MAID\_polarization conventions, is as follows:
$`A_{LT^{}}=\sqrt{2ฯต_L(1ฯต)}R_{LT^{}}\mathrm{sin}\varphi _\pi /\sigma _0,`$ (19)
where
$`\sigma _0`$ $`=`$ $`R_T+ฯต_LR_L+\sqrt{2ฯต_L(1+ฯต)}R_{TL}\mathrm{cos}\varphi _\pi `$ (20)
$`+`$ $`ฯตR_{TT}\mathrm{cos}2\varphi _\pi ,`$
where $`R_i`$ ($`i=T,L,LT,TT`$) are the response functions, $`ฯต=\left(1+2|๐ช|^2/Q^2\mathrm{tan}^2\left(\theta _e/2\right)\right)^1`$ is the degree of transverse polarization and $`ฯต_L=\left(Q^2/\omega ^2\right)ฯต`$ is the degree of longitudinal polarization of the virtual photon. Therefore the longitudinal-transverse polarized structure function shown in Fig. 4 is: $`\sigma _{LT^{}}=R_{LT^{}}/\mathrm{sin}\theta _\pi `$, which is written in terms of the electromagnetic multipoles,
$`R_{LT^{}}=`$ $``$ $`\mathrm{sin}\theta _\pi Im\{(F_2^{}+F_3^{}+\mathrm{cos}\theta _\pi F_4^{})F_5`$ (21)
$`+`$ $`(F_1^{}+F_4^{}+\mathrm{cos}\theta _\pi F_3^{})F_6\},`$
where $`F_i`$ with $`i=1,\mathrm{},6`$, are the CGLN amplitudes CGLN which are defined in the appendix of MAID\_polarization .
The two approximations (solid lines is the $`Q^2/2W`$-approximation and dashed line is the spectator approximation) are used in our calculations and a fairly good agreement is obtained in both cases. The largest difference can be noticed in the case of the asymmetry $`A_{LT^{}}`$ for the neutral channel for c.m. total energy equal to the $`\mathrm{\Delta }`$ mass, while for the other kinematics there is no practical difference. Since these observables are directly proportional to the longitudinal component of the electromagnetic current, then the overall difference can be a direct measure of the uncertainty of our model related to the theoretical error introduced by either of the approximations.
In view of the upcoming data from JLab for higher values of the $`Q^2`$, in Figs. 6 and 7 we show the prediction of our model for differential cross sections although we only fit our $`\mathrm{\Delta }`$ electromagnetic form-factors up to $`Q^2=4(GeV/c)^2`$ (see electroCaia ). For the differential cross section both approximations give very similar results, proving once again that the cross section is not sensitive enough to possible theoretical errors in the model.
## IV Conclusions
A dynamical model for pion electro-production was first introduced in electroCaia where it was used for determining the $`\mathrm{\Delta }`$ electromagnetic form-factors. The major theoretical uncertainty in our model is the treatment of the u- and t-channel terms in the potential $`V_{\pi \gamma ^{}}`$. For large $`Q^2`$ (i.e., large negative mass squared of the virtual photon), non-physical singularities occur in these terms when they go off shell in solving the scattering integral equation. We have investigated two approximate ways of solving this problem. The first was the spectator approximation for these diagrams and the second was to modify the relative energy in these diagrams by $`\pm Q^2/2W`$. We have shown that the singularity problem can be partially solved without affecting the $`Q^2=0`$ results by proposing an ad-hoc solution which avoids these poles. Within the theoretical uncertainty of how to regularize the u- and t-channel terms for pion electro-production, this model provides an overall good prediction of the available data up through the first resonance region. Comparison with recent single polarization data also shows fairly good agreement. Furthermore, we predict cross sections at higher $`Q^2`$ that are being analyzed at Jlab. In our view, the $`\pm Q^2/2W`$ approximation with restoration of current conservation provides a good dynamic model for investigating pion electroproduction at large $`Q^2`$. The parameters needed also describe pion photoproduction quite well and allows the two processes to be studied in the same model.
## Acknowledgements
This work was performed in part under the auspices of the U. S. Department of Energy, under the contracts DE-FG02-93ER40756, DE-FG02-04ER41302, DE-FG05-88ER40435, and the National Science Foundation under grant NSF-SGER-0094668. The work of VP is also supported in part by DOE contract DE-AC05-84ER-40150 under which the Southeastern Universities Research Association (SURA) operates the Thomas Jefferson National Accelerator Facility.
|
warning/0506/math-ph0506072.html
|
ar5iv
|
text
|
# New applications of pseudoanalytic function theory to the Dirac equation
## 1 Introduction
The Dirac equation with a fixed energy and the Vekua equation describing pseudoanalytic (=generalized analytic) functions both are first order elliptic systems, and it would be quite natural to expect a deep interrelation between their theories especially in the case when all potentials and wave functions in the Dirac equation depend on two space variables only. Nevertheless there is not much work done in this direction<sup>1</sup><sup>1</sup>1We refer to the work where the theory of pseudoanalytic functions was used in a way completely different from ours for studying the two-dimensional Dirac equation with a scalar or a pseudoscalar potential. due to the fact that any of the traditional matrix representations of the Dirac operator does not allow to visualize a relation between the Dirac equation in the two-dimensional case and the Vekua equation. The Dirac equation is a system of four complex equations which does not decouple in a two-dimensional situation but decouples in the one-dimensional case only.
In the present work we establish a simple relation between the Dirac equation with a scalar and an electromagnetic potentials in a two-dimensional case from one side and a pair of decoupled Vekua equations from the other. As a first step we use the matrix transformation proposed in (see also and ) which allows us to rewrite the Dirac equation in a covariant form as a biquaternionic equation. This is not our aim to discuss here the advantages of our biquaternionic reformulation of the Dirac equation compared with other its representations (the interested reader can find some of the arguments in ). We point out only that our transformation is $``$-linear as well as the resulting Dirac operator, which is not the case for a better known biquaternionic reformulation of the Dirac operator introduced by C. Lanczos in (see and for more references). Moreover, in the time-dependent case, with a vanishing electromagnetic potential our Dirac operator is real quaternionic.
Here we exploit another attractive facet of our biquaternionic Dirac equation. In the two-dimensional case it decouples into two separate Vekua equations. In general these Vekua equations are bicomplex. However we show that the whole theory of pseudoanalytic functions without modifications can be applied to these equations under a certain not restrictive condition. As an example we formulate the similarity principle which is the central reason why a pseudoanalytic function and as a consequence a spinor field depending on two space variables share many of the properties of analytic functions; e.g., they are either identically zero or have isolated zeros. In this way more results of the theory developed in and in posterior works (see, e.g., and ) can be applied to the two-dimensional Dirac equation with a scalar and an electromagnetic potentials. Nevertheless in the present work we concentrate on another non-trivial and surprising consequence of the established relation with pseudoanalytic functions. Consider the Dirac equation with a scalar potential depending on one variable with fixed energy and mass. In general this equation cannot be solved explicitly even if one looks for wave functions of one variable. None the less the result of this work is an algorithmically simple procedure for obtaining in explicit form a complete system of exact solutions depending on two variables for such Dirac equation. This system of solutions is a generalization of the system of powers $`1,z,z^2,\mathrm{}`$ in complex analysis and as such they are not appropriate for studying the Dirac equation on the whole plane. However the very fact that it is always possible to obtain explicitly a complete system of exact solutions of the Dirac equation with scalar potential of one variable as well as the hope to be able to obtain explicitly not only the generalizations of positive powers but also those of the negative ones makes in our opinion this approach attractive and promising. The system of exact solutions for the Dirac equation with a one-dimensional scalar potential is obtained due to the proposed reduction of the Dirac equation to Vekua equations and due to L. Bersโ theory of Taylor series in formal powers.
In Section 2 we introduce notations. In Section 3 we give the biquaternionic reformulation of the Dirac equation. Let us emphasize that our biquaternionic Dirac equation is completely equivalent to the โtraditionalโ Dirac equation written in $`\gamma `$-matrices, we have a simple matrix transformation giving us a relation between their solutions. In Section 4 we show that in a two-dimensional situation the Dirac equation with a scalar and an electromagnetic potentials decouples into a pair of bicomplex Vekua equations. We establish that if one of the coefficients in such Vekua equation has not zeros and does not turn into a zero divisor at any point of the domain of interest, the solutions will not be zero divisors either, and the whole theory of generalized analytic functions without modifications is applicable to the bicomplex Vekua equation.
In Section 5 we adapt some definitions and results from L. Bersโ theory to bicomplex pseudoanalytic functions. Section 6 is dedicated to a special class of Vekua equations which have been studied recently (see and ) due to their close relation to stationary Schrรถdinger equations. In Section 7 we show that the Dirac equation with a scalar potential depending on one space variable can be represented as a Vekua equation from the special class mentioned above. Here we should notice that the case of the scalar potential is only an example. The same is true, for example, for the electric potential. To the Vekua equation we apply L. Bersโ procedure for constructing corresponding formal powers which as was mentioned above are exact solutions of the Vekua equation and generalize the system of analytic functions $`1,z,z^2,\mathrm{}`$. With their aid any regular solution of the Vekua equation can be represented by its Taylor series in formal powers.
## 2 Preliminaries
We denote by $`()`$ the algebra of complex quaternions (= biquaternions). The elements of $`()`$ have the form $`q=_{k=0}^3q_ke_k`$ where $`\left\{q_k\right\},`$ $`e_0`$ is the unit and $`\left\{e_k\right|k=1,2,3\}`$ are the standard quaternionic imaginary units.
We denote the imaginary unit in $``$ by $`i`$ as usual. By definition $`i`$ commutes with $`e_k,`$ $`k=\overline{0,3}.`$We will use also the vector representation of $`q():`$ $`q=Sc(q)+Vec(q)`$, where $`Sc(q)=q_0`$ and $`Vec(q)=\stackrel{}{q}=_{k=1}^3q_ke_k.`$ The quaternionic conjugation is defined as follows $`\overline{q}=q_0\stackrel{}{q}`$.
By $`M^p`$ we denote the operator of multiplication by $`p`$ from the right hand side
$$M^pq=qp.$$
More information on complex quaternions the interested reader can find, e.g., in or .
Let $`q`$ be a complex quaternion valued differentiable function of $`๐ฑ=(x_1,x_2,x_3)`$. Denote
$$Dq=\underset{k=1}{\overset{3}{}}e_k\frac{}{x_k}q.$$
This operator is called sometimes the Moisil-Theodorescu operator or the Dirac operator but the true is that it was introduced already by W.R. Hamilton himself and studied in a great number of works (see, e.g., , , , , ).
## 3 Quaternionic reformulation of the Dirac equation
Consider the Dirac operator with scalar and electromagnetic potentials
$$๐ป=\gamma _0_t+\underset{k=1}{\overset{3}{}}\gamma _k_k+i\left(m+p_{el}\gamma _0+\underset{k=1}{\overset{3}{}}A_k\gamma _k+p_{sc}\right)$$
where $`\gamma _j,`$ $`j=0,1,2,3`$ are usual $`\gamma `$-matrices (see, e.g., , ), $`m`$, $`p_{el}`$, $`A_k`$ and $`p_{sc}`$ are real valued functions.
In a simple matrix transformation was obtained which allows us to rewrite the classical Dirac equation in quaternionic terms.
Let us introduce an auxiliary notation $`\stackrel{~}{f}:=f(t,x_1,x_2,x_3)`$. The domain $`\stackrel{~}{G}`$ is assumed to be obtained from the domain $`G^4`$ by the reflection $`x_3x_3`$. The transformation announced above we denote as $`๐`$ and define it in the following way. A function $`\mathrm{\Phi }:G^4^4`$ is transformed into a function $`F:\stackrel{~}{G}^4()`$ by the rule
$$F=๐[\mathrm{\Phi }]:=\frac{1}{2}\left((\stackrel{~}{\mathrm{\Phi }}_1\stackrel{~}{\mathrm{\Phi }}_2)e_0+i(\stackrel{~}{\mathrm{\Phi }}_0\stackrel{~}{\mathrm{\Phi }}_3)e_1(\stackrel{~}{\mathrm{\Phi }}_0+\stackrel{~}{\mathrm{\Phi }}_3)e_2+i(\stackrel{~}{\mathrm{\Phi }}_1+\stackrel{~}{\mathrm{\Phi }}_2)e_3\right).$$
The inverse transformation $`๐^1`$ is defined as follows
$$\mathrm{\Phi }=๐^1[F]=(i\stackrel{~}{F}_1\stackrel{~}{F}_2,\stackrel{~}{F}_0i\stackrel{~}{F}_3,\stackrel{~}{F}_0i\stackrel{~}{F}_3,i\stackrel{~}{F}_1\stackrel{~}{F}_2)^T.$$
Let us present the introduced transformations in a more explicit matrix form which relates the components of a $`^4`$-valued function $`\mathrm{\Phi }`$ with the components of an $`()`$-valued function $`F`$:
$$F=๐[\mathrm{\Phi }]=\frac{1}{2}\left(\begin{array}{cccc}\hfill 0& \hfill 1& \hfill 1& \hfill 0\\ \hfill i& \hfill 0& \hfill 0& \hfill i\\ \hfill 1& \hfill 0& \hfill 0& \hfill 1\\ \hfill 0& \hfill i& \hfill i& \hfill 0\end{array}\right)\left(\begin{array}{c}\stackrel{~}{\mathrm{\Phi }}_0\\ \stackrel{~}{\mathrm{\Phi }}_1\\ \stackrel{~}{\mathrm{\Phi }}_2\\ \stackrel{~}{\mathrm{\Phi }}_3\end{array}\right)$$
and
$$\mathrm{\Phi }=๐^1[F]=\left(\begin{array}{cccc}\hfill 0& \hfill i& \hfill 1& \hfill 0\\ \hfill 1& \hfill 0& \hfill 0& \hfill i\\ \hfill 1& \hfill 0& \hfill 0& \hfill i\\ \hfill 0& \hfill i& \hfill 1& \hfill 0\end{array}\right)\left(\begin{array}{c}\stackrel{~}{F}_0\\ \stackrel{~}{F}_1\\ \stackrel{~}{F}_2\\ \stackrel{~}{F}_3\end{array}\right).$$
Denote
$$R=D_tM^{e_1}+๐+M^{i(\stackrel{~}{p}_{el}e_1i(\stackrel{~}{p}_{sc}+m)e_2)}$$
where $`๐=i(\stackrel{~}{A}_1e_1+\stackrel{~}{A}_2e_2\stackrel{~}{A}_3e_3)`$. The following equality holds
$$R=๐\gamma _1\gamma _2\gamma _3๐ป๐^1.$$
That is, a $`^4`$-valued function $`\mathrm{\Phi }`$ is a solution of the equation
$$๐ป\mathrm{\Phi }=0\text{in }G$$
iff the complex quaternionic function $`F=๐\mathrm{\Phi }`$ is a solution of the quaternionic equation
$$RF=0\text{in }\stackrel{~}{G}.$$
Note that in the absence of the electromagnetic potential the operator $`R`$ becomes real quaternionic which is an important property (see ).
In what follows we assume that potentials are time-independent and consider solutions with fixed energy: $`\mathrm{\Phi }(t,๐ฑ)=\mathrm{\Phi }_\omega (๐ฑ)e^{i\omega t}`$. The equation for $`\mathrm{\Phi }_\omega `$ has the form
$$๐ป_\omega \mathrm{\Phi }_\omega =0\text{in }\widehat{G}$$
(1)
where $`\widehat{G}`$ is a domain in $`^3`$,
$$๐ป_\omega =i\omega \gamma _0+\underset{k=1}{\overset{3}{}}\gamma _k_k+i\left(m+p_{el}\gamma _0+\underset{k=1}{\overset{3}{}}A_k\gamma _k+p_{sc}\right).$$
We have
$$R_\omega =๐\gamma _1\gamma _2\gamma _3๐ป_\omega ๐^1,$$
where
$$R_\omega =D+๐+M^๐$$
with $`๐=i((\stackrel{~}{p}_{el}+\omega )e_1i(\stackrel{~}{p}_{sc}+m)e_2)`$. Thus, equation (1) turns into the complex quaternionic equation
$$R_\omega q=0$$
(2)
where $`q`$ is a complex quaternion valued function.
## 4 The Dirac equation in a two-dimensional case as a bicomplex Vekua equation
Let us introduce the following notation. For any complex quaternion $`q`$ we denote by $`Q_1`$ and $`Q_2`$ its bicomplex components:
$$Q_1=q_0+q_3e_3\text{and }Q_2=q_2q_1e_3.$$
Then $`q`$ can be represented as follows $`q=Q_1+Q_2e_2`$. For the operator $`D`$ we have $`D=D_1+D_2e_2`$ with $`D_1=e_3_3`$ and $`D_2=_2_1e_3`$. Notice that $`๐=Be_2`$ with $`B=(\stackrel{~}{p}_{sc}+m)+i(\stackrel{~}{p}_{el}+\omega )e_3`$, $`๐=A_1+A_2e_2`$ with $`A_1=a_3e_3`$ and $`A_2=a_2a_1e_3`$.
We obtain that equation (2) is equivalent to the system
$$D_1Q_1D_2\overline{Q}_2+A_1Q_1A_2\overline{Q}_2\overline{B}Q_2=0,$$
(3)
$$D_2\overline{Q}_1+D_1Q_2+A_2\overline{Q}_1+A_1Q_2+BQ_1=0,$$
(4)
where $`Q_1`$ and $`Q_2`$ are bicomplex components of $`q`$. We stress that the system (3), (4) is equivalent to the Dirac equation in $`\gamma `$-matrices (1).
Let us suppose all fields in our model to be independent of $`x_3`$, and $`A_1=a_3e_30`$. Then the system (3), (4) decouples, and we obtain two separate bicomplex equations
$$\overline{D}_2Q_2=\overline{A}_2Q_2B\overline{Q}_2$$
and
$$\overline{D}_2Q_1=\overline{A}_2Q_1\overline{B}\overline{Q}_1.$$
Denote $`\overline{}=\overline{D}_2`$, $`a=\overline{A}_2`$, $`b=B`$, $`w=Q_2`$, $`W=Q_1`$, $`z=x+y๐ค`$, where $`x=x_2`$, $`y=x_1`$ and for convenience we denote $`๐ค=e_3`$. Then we reduce the Dirac equation with electromagnetic and scalar potentials independent of $`x_3`$ to a pair of Vekua-type equations
$$\overline{}w=aw+b\overline{w}$$
(5)
and
$$\overline{}W=aW+\overline{bW}.$$
(6)
The difference between the bicomplex equations (5), (6) and the usual complex Vekua equations is revealed if only $`w`$ or $`W`$ can take values equal to bicomplex zero divisors (otherwise equations (5), (6) can be analyzed following Bers-Vekua theory , ). Let us study this possibility with the aid of the following pair of projection operators
$$P^+=\frac{1}{2}(1+i๐ค)\text{and }P^{}=\frac{1}{2}(1i๐ค).$$
The set of bicomplex zero divisors, that is of nonzero elements $`q=q_0+q_1๐ค`$, $`\{q_0,q_1\}`$ such that
$$q\overline{q}=\left(q_0+q_1๐ค\right)\left(q_0q_1๐ค\right)=0$$
(7)
we denote by $`๐`$.
###### Lemma 1
Let $`q`$ be a bicomplex number of the form $`q=q_0+q_1๐ค`$, $`\{q_0,q_1\}`$. If $`q๐`$ then $`q=2P^+q_0`$ or $`q=2P^{}q_0`$.
Proof. From (7) it follows that $`q_0^2+q_1^2=0`$ which gives us that $`q_1=\pm iq_0`$. That is $`q=q_0(1+i๐ค)`$ or $`q=q_0(1i๐ค)`$.
For other results on bicomplex numbers we refer to .
Let $`\mathrm{\Omega }`$ denote a bounded, simply connected domain in the plane of the variable $`z`$.
###### Theorem 2
Let $`b(z)๐\left\{0\right\}`$, $`z\mathrm{\Omega }`$ and $`w`$, $`W`$ be solutions of (5) and (6) respectively. Then $`w(z)๐`$ and $`W(z)๐`$, $`z\mathrm{\Omega }`$.
Proof. Assume that $`w(z)๐`$ for some $`z\mathrm{\Omega }`$. For definiteness let $`w(z)=2P^+w_0(z)`$. Then from (5) we have
$$\overline{}P^+w_0=aP^+w_0+bP^{}w_0.$$
Applying $`P^{}`$ to this equality we find that $`P^{}b=0`$ which is a contradiction.
Thus, if the coefficient $`b`$ does not have zeros and does not turn into a zero divisor at any point of the domain of interest, the solutions of (5) and (6) will not be zero divisors either and the whole theory of pseudoanalytic functions is applicable without changes to the bicomplex equations (5) and (6). As an example let us formulate one of the main results of the theory, the similarity principle which is the basic tool for studying the distribution of zeros and of singularities of pseudoanalytic functions as well as boundary value problems .
###### Theorem 3
Let $`w`$ be a regular solution of (5) in a domain $`\mathrm{\Omega }`$ and let $`b(z)๐\left\{0\right\}`$, $`z\mathrm{\Omega }`$. Then the bicomplex function $`\mathrm{\Phi }=we^h`$, where
$$h(z)=\frac{1}{2\pi }_\mathrm{\Omega }\frac{g(\tau )d\tau }{\tau z},$$
$$g(z)=\{\begin{array}{c}a(z)+b(z)\frac{\overline{w}(z)}{w(z)}\text{ if }w(z)0,z\mathrm{\Omega },\hfill \\ a(z)+b(z)\text{ if }w(z)=0,z\mathrm{\Omega }\hfill \end{array}$$
is a solution of the equation $`\overline{}\mathrm{\Phi }=0`$ in $`\mathrm{\Omega }`$.
The proof of this theorem is completely analogous to that given in . It would be interesting to extend this result to the case of $`b`$ being a zero divisor in the whole domain $`\mathrm{\Omega }`$ or in some points.
This theorem opens the way to generalize many classical results from theory of analytic functions to the case of solutions of equations (5) and (6) by analogy with . Nevertheless in the present work we prefer to explore another possibility. Namely, we show how the application of Bersโ theory of pseudoanalytic functions allows us to obtain explicitly a complete system of solutions of the Dirac equation with a scalar potential depending on one variable.
## 5 Some definitions and results from Bersโ theory for bicomplex pseudoanalytic functions
### 5.1 Generating pair, derivative and antiderivative
Following we introduce the notion of a bicomplex generating pair.
###### Definition 4
A pair of bicomplex functions $`F=F_0+F_1๐ค`$ and $`G=G_0+G_1๐ค`$, possessing in $`\mathrm{\Omega }`$ partial derivatives with respect to the real variables $`x`$ and $`y`$ is said to be a generating pair if it satisfies the inequality
$$\mathrm{Vec}(\overline{F}G)0\text{in }\mathrm{\Omega }.$$
The following expressions are called characteristic coefficients of the pair $`(F,G)`$
$$a_{(F,G)}=\frac{\overline{F}G_{\overline{z}}F_{\overline{z}}\overline{G}}{F\overline{G}\overline{F}G},b_{(F,G)}=\frac{FG_{\overline{z}}F_{\overline{z}}G}{F\overline{G}\overline{F}G},$$
$$A_{(F,G)}=\frac{\overline{F}G_zF_z\overline{G}}{F\overline{G}\overline{F}G},B_{(F,G)}=\frac{FG_zF_zG}{F\overline{G}\overline{F}G},$$
where the subindex $`\overline{z}`$ or $`z`$ means the application of $`\overline{}`$ or $``$ respectively.
Every bicomplex function $`W`$ defined in a subdomain of $`\mathrm{\Omega }`$ admits the unique representation $`W=\varphi F+\psi G`$ where the functions $`\varphi `$ and $`\psi `$ are complex valued.
The $`(F,G)`$-derivative $`\stackrel{}{W}=\frac{d_{(F,G)}W}{dz}`$ of a function $`W`$ exists and has the form
$$\stackrel{}{W}=\varphi _zF+\psi _zG=W_zA_{(F,G)}WB_{(F,G)}\overline{W}$$
(8)
if and only if
$$\varphi _{\overline{z}}F+\psi _{\overline{z}}G=0.$$
(9)
This last equation can be rewritten in the following form
$$W_{\overline{z}}=a_{(F,G)}W+b_{(F,G)}\overline{W}$$
which we call the bicomplex Vekua equation. Solutions of this equation are called $`(F,G)`$-pseudoanalytic functions.
###### Remark 5
The functions $`F`$ and $`G`$ are $`(F,G)`$-pseudoanalytic, and $`\stackrel{}{F}\stackrel{}{G}0`$.
###### Definition 6
Let $`(F,G)`$ and $`(F_1,G_1)`$ \- be two generating pairs in $`\mathrm{\Omega }`$. $`(F_1,G_1)`$ is called successor of $`(F,G)`$ and $`(F,G)`$ is called predecessor of $`(F_1,G_1)`$ if
$$a_{(F_1,G_1)}=a_{(F,G)}\text{and}b_{(F_1,G_1)}=B_{(F,G)}\text{.}$$
The importance of this definition becomes obvious from the following statement.
###### Theorem 7
Let $`W`$ be an $`(F,G)`$-pseudoanalytic function and let $`(F_1,G_1)`$ be a successor of $`(F,G)`$. Then $`\stackrel{}{W}`$ is an $`(F_1,G_1)`$-pseudoanalytic function.
###### Definition 8
Let $`(F,G)`$ be a generating pair. Its adjoint generating pair $`(F,G)^{}=(F^{},G^{})`$ is defined by the formulas
$$F^{}=\frac{2\overline{F}}{F\overline{G}\overline{F}G},G^{}=\frac{2\overline{G}}{F\overline{G}\overline{F}G}.$$
The $`(F,G)`$-integral is defined as follows
$$_\mathrm{\Gamma }Wd_{(F,G)}z=\frac{1}{2}\left(F(z_1)\mathrm{Sc}_\mathrm{\Gamma }G^{}W๐z+G(z_1)\mathrm{Sc}_\mathrm{\Gamma }F^{}W๐z\right)$$
where $`\mathrm{\Gamma }`$ is a rectifiable curve leading from $`z_0`$ to $`z_1`$.
If $`W=\varphi F+\psi G`$ is an $`(F,G)`$-pseudoanalytic function where $`\varphi `$ and $`\psi `$ are complex valued functions then
$$_{z_0}^z\stackrel{}{W}d_{(F,G)}z=W(z)\varphi (z_0)F(z)\psi (z_0)G(z),$$
(10)
and as $`\stackrel{}{F}=\stackrel{}{\stackrel{}{G}=}0`$, this integral is path-independent and represents the $`(F,G)`$-antiderivative of $`\stackrel{}{W}`$.
### 5.2 Generating sequences and Taylor series in formal powers
###### Definition 9
A sequence of generating pairs $`\left\{(F_m,G_m)\right\}`$, $`m=0,\pm 1,\pm 2,\mathrm{}`$ , is called a generating sequence if $`(F_{m+1},G_{m+1})`$ is a successor of $`(F_m,G_m)`$. If $`(F_0,G_0)=(F,G)`$, we say that $`(F,G)`$ is embedded in $`\left\{(F_m,G_m)\right\}`$.
###### Theorem 10
Let $`(F,G)`$ be a generating pair in $`\mathrm{\Omega }`$. Let $`\mathrm{\Omega }_1`$ be a bounded domain, $`\overline{\mathrm{\Omega }}_1\mathrm{\Omega }`$. Then $`(F,G)`$ can be embedded in a generating sequence in $`\mathrm{\Omega }_1`$.
###### Definition 11
A generating sequence $`\left\{(F_m,G_m)\right\}`$ is said to have period $`\mu >0`$ if $`(F_{m+\mu },G_{m+\mu })`$ is equivalent to $`(F_m,G_m)`$ that is their characteristic coefficients coincide.
Let $`W`$ be an $`(F,G)`$-pseudoanalytic function. Using a generating sequence in which $`(F,G)`$ is embedded we can define the higher derivatives of $`W`$ by the recursion formula
$$W^{[0]}=W;W^{[m+1]}=\frac{d_{(F_m,G_m)}W^{[m]}}{dz},m=1,2,\mathrm{}\text{.}$$
###### Definition 12
The formal power $`Z_m^{(0)}(a,z_0;z)`$ with center at $`z_0\mathrm{\Omega }`$, coefficient $`a`$ and exponent $`0`$ is defined as the linear combination of the generators $`F_m`$, $`G_m`$ with complex constant coefficients $`\lambda `$, $`\mu `$ chosen so that $`\lambda F_m(z_0)+\mu G_m(z_0)=a`$. The formal powers with exponents $`n=1,2,\mathrm{}`$ are defined by the recursion formula
$$Z_m^{(n+1)}(a,z_0;z)=(n+1)_{z_0}^zZ_{m+1}^{(n)}(a,z_0;\zeta )d_{(F_m,G_m)}\zeta .$$
(11)
This definition implies the following properties.
1. $`Z_m^{(n)}(a,z_0;z)`$ is an $`(F_m,G_m)`$-pseudoanalytic function of $`z`$.
2. If $`a^{}`$ and $`a^{\prime \prime }`$ are complex constants, then
$$Z_m^{(n)}(a^{}+๐คa^{\prime \prime },z_0;z)=a^{}Z_m^{(n)}(1,z_0;z)+a^{\prime \prime }Z_m^{(n)}(๐ค,z_0;z).$$
3. The formal powers satisfy the differential relations
$$\frac{d_{(F_m,G_m)}Z_m^{(n)}(a,z_0;z)}{dz}=nZ_{m+1}^{(n1)}(a,z_0;z).$$
4. The asymptotic formulas
$$Z_m^{(n)}(a,z_0;z)a(zz_0)^n,zz_0$$
hold.
Assume now that
$$W(z)=\underset{n=0}{\overset{\mathrm{}}{}}Z^{(n)}(a,z_0;z)$$
(12)
where the absence of the subindex $`m`$ means that all the formal powers correspond to the same generating pair $`(F,G),`$ and the series converges uniformly in some neighborhood of $`z_0`$. It can be shown that the uniform limit of pseudoanalytic functions is pseudoanalytic, and that a uniformly convergent series of $`(F,G)`$-pseudoanalytic functions can be $`(F,G)`$-differentiated term by term. Hence the function $`W`$ in (12) is $`(F,G)`$-pseudoanalytic and its $`r`$th derivative admits the expansion
$$W^{[r]}(z)=\underset{n=r}{\overset{\mathrm{}}{}}n(n1)\mathrm{}(nr+1)Z_r^{(nr)}(a_n,z_0;z).$$
From this the Taylor formulas for the coefficients are obtained
$$a_n=\frac{W^{[n]}(z_0)}{n!}.$$
(13)
###### Definition 13
Let $`W(z)`$ be a given $`(F,G)`$-pseudoanalytic function defined for small values of $`\left|zz_0\right|`$. The series
$$\underset{n=0}{\overset{\mathrm{}}{}}Z^{(n)}(a,z_0;z)$$
(14)
with the coefficients given by (13) is called the Taylor series of $`W`$ at $`z_0`$, formed with formal powers.
The Taylor series always represents the function asymptotically:
$$W(z)\underset{n=0}{\overset{N}{}}Z^{(n)}(a,z_0;z)=O\left(\left|zz_0\right|^{N+1}\right),zz_0,$$
(15)
for all $`N`$. This implies (since a pseudoanalytic function can not have a zero of arbitrarily high order without vanishing identically) that the sequence of derivatives $`\left\{W^{[n]}(z_0)\right\}`$ determines the function $`W`$ uniquely.
If the series (14) converges uniformly in a neighborhood of $`z_0`$, it converges to the function $`W`$.
###### Theorem 14
The formal Taylor expansion (14) of a pseudoanalytic function in formal powers defined by a periodic generating sequence converges in some neighborhood of the center.
## 6 Special class of Vekua equations
The following important class of Vekua equations was considered in . Let $`f_0`$ be a complex valued (with respect to $`i`$), twice differentiable nonvanishing function defined on $`\mathrm{\Omega }`$. Consider the equation
$$\overline{}W=\frac{\overline{}f_0}{f_0}\overline{W}\text{in }\mathrm{\Omega }.$$
(16)
Denote $`\nu _1=\mathrm{\Delta }f_0/f_0`$.
###### Theorem 15
If $`W=W_1+W_2๐ค`$ is a solution of (16) then $`W_1=\mathrm{Sc}W`$ is a solution of the stationary Schrรถdinger equation
$$\mathrm{\Delta }W_1+\nu _1W_1=0\text{in }\mathrm{\Omega }$$
(17)
and $`W_2=\mathrm{Vec}W`$ is a solution of the associated Schrรถdinger equation
$$\mathrm{\Delta }W_2+\nu _2W_2=0\text{in }\mathrm{\Omega }$$
(18)
where $`\nu _2=2(\overline{}f_0f_0)/f_0^2\nu _1`$.
Moreover, in a simple formula was obtained which allows us for any given solution $`W_1`$ of (17) to construct such a solution $`W_2`$ of (18) that $`W=W_1+W_2๐ค`$ will be a solution of (16) generalizing in this way the well known procedure for constructing conjugate harmonic functions in complex analysis.
## 7 Dirac equation with a scalar potential
Let us show that the Dirac equation with a scalar potential depending on one real variable reduces to a bicomplex Vekua equation of the form (16).
Let $`p_{sc}=p(x)`$ and $`p_{el}0,`$ $`A_k0,`$ $`k=1,2,3`$. Then according to Section 4 the Dirac equation is equivalent to the pair of bicomplex Vekua equations
$$\overline{}w=b\overline{w}$$
(19)
and
$$\overline{}W=\overline{bW}$$
(20)
with $`b=p(x)+mi\omega ๐ค`$.
Let $`f_0=e^{P(x)+mx+i\omega y}`$, where $`P`$ is an antiderivative of $`p`$. Then we have
$$\overline{b}=\overline{}f_0/f_0.$$
Note that due to theorem 15 if the bicomplex function $`W`$ is a solution of (20) then the complex function $`W_1=\mathrm{Sc}W`$ is a solution of the stationary Schrรถdinger equation (17) where
$$\nu _1(x)=p^{}(x)+(p(x)+m)^2\omega ^2,$$
(21)
and the function $`W_2=\mathrm{Vec}W`$ is a solution of equation (18) where
$$\nu _2(x)=p^{}(x)+(p(x)+m)^2\omega ^2.$$
(22)
Let us notice that both Schrรถdinger equations (17) and (18) in this case admit separation of variables. Nevertheless this does not imply they can be solved explicitly. In general this is not the case. However we will show how using our approach and Bersโ theory for both of them one can construct in explicit form a locally complete system of exact solutions.
Consider equation (20). It is easy to see that the pair of functions
$$F=f_0\text{and}G=\frac{๐ค}{f_0}$$
(23)
represents a generating pair for (20). Note that $`F=e^\sigma `$ and $`G=e^\sigma ๐ค`$, where $`\sigma =\alpha (x)+\beta (y)`$ and $`\alpha (x)=P(x)+mx`$, $`\beta (y)=i\omega y`$. For a generating pair of such special kind it is easy to construct a successor . Let $`\tau =\alpha (x)+\beta (y)`$. Then the pair $`F_1=e^\tau `$ and $`G_1=e^\tau ๐ค`$ is a successor of $`(F,G)`$. Moreover, $`(F,G)`$ is a successor of $`(F_1,G_1)`$. Thus, for $`(F,G)`$ we obtain a complete periodic generating sequence of a period $`2`$ in explicit form (for explicitly constructed, in general non-periodic generating sequences in a far more general situation we refer to ).
The fact that we have a generating sequence in explicit form implies that we are able to construct the corresponding formal powers of any order explicitly and therefore to obtain a locally complete system of exact solutions of the Dirac equation with a scalar potential depending on one variable as well as of the stationary Schrรถdinger equations (17) and (18) with potentials (21) and (22) respectively.
As a first step we construct the adjoint generating pair (see definition 8):
$$F^{}=f_0๐ค\text{and}G^{}=\frac{1}{f_0}.$$
Next, we write down the expression for the $`(F,G)`$-integral:
$$_\mathrm{\Gamma }Wd_{(F,G)}z=\frac{1}{2}\left(f_0(z_1)\mathrm{Sc}_\mathrm{\Gamma }\frac{W(z)}{f_0(z)}๐z\frac{๐ค}{f_0(z_1)}\mathrm{Sc}_\mathrm{\Gamma }f_0(z)W(z)๐ค๐z\right).$$
By definition, the formal power $`Z^{(0)}(a,z_0;z)`$ for equation (20) has the form
$$Z^{(0)}(a,z_0;z)=\lambda F(z)+\mu G(z),$$
where the complex constants $`\lambda `$ and $`\mu `$ are chosen so that $`\lambda F(z_0)+\mu G(z_0)=a`$. That is,
$$Z^{(0)}(a,z_0;z)=\lambda e^{P(x)+mx+i\omega y}+\mu e^{(P(x)+mx+i\omega y)}๐ค.$$
In order to obtain $`Z^{(1)}(a,z_0;z)`$ we should take the $`(F,G)`$-integral of $`Z_1^{(0)}(a,z_0;z)`$, where
$$Z_1^{(0)}(a,z_0;z)=\lambda _1F_1(z)+\mu _1G_1(z),$$
with $`\lambda _1F_1(z_0)+\mu _1G_1(z_0)=a`$. Thus,
$$Z^{(1)}(a,z_0;z)=_{z_0}^z(\lambda _1F_1(\zeta )+\mu _1G_1(\zeta ))d_{(F,G)}\zeta $$
$`={\displaystyle \frac{1}{2}}\{e^{P(x)+mx+i\omega y}\mathrm{Sc}{\displaystyle _{z_0}^z}e^{P(x^{})mx^{}i\omega y^{}}(\lambda _1e^{P(x^{})mx^{}+i\omega y^{}}+\mu _1e^{P(x^{})+mx^{}i\omega y^{}}๐ค)d\zeta `$
$`e^{P(x)mxi\omega y}๐ค\mathrm{Sc}{\displaystyle _{z_0}^z}e^{P(x^{})+mx^{}+i\omega y^{}}๐ค(\lambda _1e^{P(x^{})mx^{}+i\omega y^{}}+\mu _1e^{P(x^{})+mx^{}i\omega y^{}}๐ค)d\zeta \}`$
$`={\displaystyle \frac{1}{2}}\{e^{P(x)+mx+i\omega y}\mathrm{Sc}{\displaystyle _{z_0}^z}(\lambda _1e^{2(P(x^{})+mx^{})}+\mu _1e^{2i\omega y^{}}๐ค)d\zeta `$
$`e^{P(x)mxi\omega y}๐ค\mathrm{Sc}{\displaystyle _{z_0}^z}(\lambda _1e^{2i\omega y^{}}๐ค\mu _1e^{2(P(x^{})+mx^{})})d\zeta \}`$
where $`\zeta =x^{}+y^{}๐ค`$.
For $`Z^{(2)}(a,z_0;z)`$ by definition 12 we have
$$Z^{(2)}(a,z_0;z)=2_{z_0}^zZ_1^{(1)}(a,z_0;\zeta )d_{(F,G)}\zeta ,$$
(24)
where $`Z_1^{(1)}(a,z_0;\zeta )`$ in its turn can be found from the equality
$$Z_1^{(1)}(a,z_0;z)=_{z_0}^zZ_2^{(0)}(a,z_0;\zeta )d_{(F_1,G_1)}\zeta .$$
(25)
We note that due to periodicity of the generating sequence containing the generating pair (23),
$$Z_2^{(0)}(a,z_0;\zeta )=Z^{(0)}(a,z_0;\zeta ).$$
The adjoint pair for $`(F_1,G_1)`$ necessary for the $`(F_1,G_1)`$-integral in (25) has the form
$$F_1^{}=e^\tau ๐ค\text{and}G_1^{}=e^\tau .$$
Thus,
$$Z_1^{(1)}(a,z_0;z)=$$
$`={\displaystyle \frac{1}{2}}\{e^{P(x)mx+i\omega y}\mathrm{Sc}{\displaystyle _{z_0}^z}e^{P(x^{})+mx^{}i\omega y^{}}(\lambda e^{P(x^{})+mx^{}+i\omega y^{}}+\mu e^{P(x^{})mx^{}i\omega y^{}}๐ค)d\zeta `$
$`e^{P(x)+mxi\omega y}๐ค\mathrm{Sc}{\displaystyle _{z_0}^z}e^{P(x^{})mx^{}+i\omega y^{}}๐ค(\lambda e^{P(x^{})+mx^{}+i\omega y^{}}+\mu e^{P(x^{})mx^{}i\omega y^{}}๐ค)d\zeta \}`$
$`={\displaystyle \frac{1}{2}}\{e^{P(x)mx+i\omega y}\mathrm{Sc}{\displaystyle _{z_0}^z}(\lambda e^{2(P(x^{})+mx^{})}+\mu e^{2i\omega y^{}}๐ค)d\zeta `$
$`e^{P(x)+mxi\omega y}๐ค\mathrm{Sc}{\displaystyle _{z_0}^z}(\lambda e^{2i\omega y^{}}๐ค\mu e^{2(P(x^{})+mx^{})})d\zeta \}.`$
Substitution of this expression into (24) gives us the formal power $`Z^{(2)}(a,z_0;z)`$, and this algorithmically simple procedure can be continued indefinitely. As a result we obtain an infinite system of formal powers which at least locally gives us a complete system of solutions of (20) in the sense that any regular solution of (20) can be approximated arbitrarily closely by a finite linear combination of formal powers (formula (15)). Moreover, as the corresponding generating sequence is periodic, theorem 14 is valid, and therefore we can guarantee the convergence of a Taylor expansion in the formal powers to a corresponding solution of (20) in some neighborhood of $`z_0`$.
A similar procedure works also for equation (19). Note that the pair of functions $`F_1๐ค=e^\tau ๐ค`$ and $`G_1๐ค=e^\tau `$ is a generating pair corresponding to (19).
As any solution of the Schrรถdinger equation (17) with the potential $`\nu _1`$ defined by (21) is the scalar part of some solution of (20), and any solution of (18) with the potential (22) is the vector part of some solution of (20), the scalar and the vector parts of the constructed system of formal powers give us locally complete systems of solutions of (17) and (18) respectively.
This last result can also be interpreted in the following way. Consider the equation
$$\mathrm{\Delta }f+\nu f=\omega ^2f\text{in }\mathrm{\Omega }$$
(26)
where $`f`$ is a complex twice continuously differentiable function of two real variables $`x`$ and $`y`$, and $`\nu `$ is a complex valued function of one real variable $`x`$, $`\omega `$ is a complex constant. Suppose we are given a particular solution $`f_0=f_0(x)`$ of the ordinary differential equation
$$\frac{d^2f_0}{dx^2}+\nu f_0=0.$$
(27)
This implies that we are able to represent $`\nu `$ in the form $`\nu =p^{}+p^2`$ where $`p=f_0^{}/f_0`$. Then we observe that (26) is precisely equation (17) with $`m=0`$ in (21). Thus our result means that if we are able to solve the ordinary differential equation (27) then we can construct explicitly a locally complete system of exact solutions to (26) for any $`\omega `$. For this one should consider the bicomplex Vekua equation (20) and follow the procedure described above for constructing the corresponding system of formal powers. Then the scalar part of the system gives us a locally complete system of exact solutions to (26).
|
warning/0506/cond-mat0506723.html
|
ar5iv
|
text
|
# Images of a Bose-Einstein condensate at finite temperature Invited talk given at the conference โQuantum Optics VIโ, June 13-18 2005, Krynica, Poland
## I Introduction
Quantum measurements on Bose-condensed systems can give quite unexpected results. For example, in the classic paper by Javanainen and Yoo a density measurement on a Fock state $`|N/2,N/2`$ with $`N`$ particles equally divided between two counter-propagating plane waves $`e^{\pm ix}`$ reveals an interference pattern $`\rho (x|\phi )\mathrm{cos}^2(x\phi )`$ with a phase $`\phi `$ chosen randomly in every realization of the experiment. The Fock state has a uniform single particle density distribution, but its measurement unexpectedly reveals interference between the two counter-propagating condensates. The Fock state is a quantum superposition over $`N`$-particle condensates with different relative phases $`\phi `$ in their wave functions , $`|N/2,N/2๐\phi |N:e^{+i(x\phi )}+e^{i(x\phi )}`$, but every single realization of the experiment reveals such a density distribution as if the state before the density measurement were one of the condensates $`|N:e^{+i(x\phi )}+e^{i(x\phi )}`$ with a randomly chosen phase $`\phi `$. This effect is best explained when the density measurement, which is a destructive measurement of all particle positions at the same time, is replaced by an equivalent sequential measurement of one position after another. With an increasing number $`n`$ of measured positions a quantum state of the remaining $`Nn`$ particles gradually โcollapsesโ from the initial uniform superposition over all phases to a state with a more and more localized phase $`\phi `$. For a large $`N`$ a measurement of only a small fraction $`\frac{n}{N}1`$ of all particles practically collapses the state of remaining $`Nn`$ particles to a condensate with definite phase $`\phi `$.
A lesson from this instructive example is that quantum measurement on an $`N`$-particle state with highly occupied single particle modes โcollapsesโ the state to a definite condensate with a definite condensate wave function $`\varphi (x)`$. The question is: what is the probability distribution for different measurement outcomes $`\varphi (x)`$? As the set of condensates is not an orthonormal basis this is not a trivial question.
In Ref. we derived this probability distribution in the framework of the time-dependent Bogoliubov theory at zero temperature. At zero temperature a condensate initially prepared in its $`N`$-particle ground state evolves under external time-dependent perturbation into a time-dependent excited state. The excited state is a time-dependent Bogoliubov vacuum i.e. at any time $`t`$ there exists a complete set of quasiparticle annihilation operators for which the excited state is a vacuum. In Ref. it was shown that the time-dependent vacuum has a simple diagonal structure which directly leads to a compact gaussian probability distribution for different condensate wave functions $`\varphi (x)`$. As the case of zero temperature is covered in Ref., in these notes I describe the general case of finite temperature when the initial state is a condensate in equilibrium with a thermal cloud of atoms. I derive gaussian probability distribution for $`\varphi (x)`$ at any time $`t`$ when the external perturbation drives the initial thermal state into an excited mixed state.
## II N-conserving Bogoliubov theory
Number conserving Bogoliubov theory is a quadratic approximation to the second quantized Hamiltonian (in trap units)
$$\widehat{H}=๐x\left[\frac{1}{2}_x\widehat{\mathrm{\Psi }}^{}_x\widehat{\mathrm{\Psi }}+\frac{1}{2}x^2\widehat{\mathrm{\Psi }}^{}\widehat{\mathrm{\Psi }}+V(t,x)\widehat{\mathrm{\Psi }}^{}\widehat{\mathrm{\Psi }}+\frac{1}{2}g\widehat{\mathrm{\Psi }}^{}\widehat{\mathrm{\Psi }}^{}\widehat{\mathrm{\Psi }}\widehat{\mathrm{\Psi }}\right].$$
(1)
Here $`\widehat{\mathrm{\Psi }}(x)`$ is the bosonic annihilation operator, $`V(t,x)`$ is the external perturbation potential, and $`g`$ is strength of contact interaction between atoms. Here and in the following I will use one-dimensional notation but all equations can be generalized by the simple replacement $`x\stackrel{}{x}`$. The annihilation operator is split into condensate and non-condensate part
$$\widehat{\mathrm{\Psi }}(x)=\widehat{a}_0\varphi _0(x)+\delta \widehat{\psi }(x).$$
(2)
It is assumed that most atoms occupy the condensate mode $`\varphi _0(x)`$. Equation (2) is substituted to the Hamiltonian (1) and then the Hamiltonian is expanded in powers of the fluctuation operator $`\delta \widehat{\psi }`$, see Ref..
Many experiments on dilute atomic condensates can be clearly divided in two steps: as a first step a condensate is prepared in its ground state and then in the second step an external potential $`V(t,x)`$ is applied to manipulate with the condensate wave function. Generic examples are phase imprinting of dark solitons , atomic interferometry , or generation of shock waves in Bose-Einstein condensates . At finite temperature the initial state before the manipulation is a thermal state including thermal excitations above the $`N`$-particle ground state. The initial condensate wave function $`\varphi _0`$ solves the stationary Gross-Pitaevskii equation
$$\mu \varphi _0=\frac{1}{2}_x^2\varphi _0+\frac{1}{2}x^2\varphi _0+g|\varphi _0|^2\varphi _0.$$
(3)
In Bogoliubov approximation the ground state is Bogoliubov vacuum $`|0_b`$ which can be written as a gaussian superposition over condensates
$$|0_b=d^2be^{\frac{1}{2}_{m=1}^Mb_m^{}b_m}|N:\varphi _0(x)+\frac{1}{\sqrt{N}}\underset{m=1}{\overset{M}{}}b_mu_m(x)+b_m^{}v_m^{}(x).$$
(4)
Here the state $`|N:\varphi `$ is a condensate of $`N`$ atoms in the normalized condensate wave function $`\frac{\varphi }{\sqrt{\varphi |\varphi }}`$. The Bogoliubov modes $`u_m`$ and $`v_m`$ are eigenmodes of the stationary Bogoliubov-de Gennes equations
$`\omega _mu_m`$ $`=`$ $`{\displaystyle \frac{1}{2}}_x^2u_m+{\displaystyle \frac{1}{2}}x^2u_m+2g|\varphi _0|^2u_m+g\varphi _0^2v_m,`$ (5)
$`\omega _mv_m`$ $`=`$ $`{\displaystyle \frac{1}{2}}_x^2v_m+{\displaystyle \frac{1}{2}}x^2v_m+2g|\varphi _0|^2v_m+g\left(\varphi _0^{}\right)^2u_m.`$ (6)
Numerical solution of these equations gives a finite number of modes $`M`$. At finite temperature the initial state is a thermal state $`\widehat{\rho }(0)`$ with thermal quasiparticle excitations. The thermal state is also a gaussian state
$`\widehat{\rho }(0)`$ $`=`$ $`{\displaystyle d^2b_Ld^2b_Re^{\frac{1}{2}b_L^{}b_L\frac{1}{2}b_R^{}b_R+b_L^{}e^{\beta \omega }b_R}}`$ (8)
$`|N:\varphi _0(x)+{\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{m=1}{\overset{M}{}}}b_{L,m}u_m(x)+b_{L,m}^{}v_m^{}(x)N:\varphi _0(x)+{\displaystyle \frac{1}{\sqrt{N}}}{\displaystyle \underset{m=1}{\overset{M}{}}}b_{R,m}u_m(x)+b_{R,m}^{}v_m^{}(x)|.`$
Here $`b_L^{}b_L=_{m=1}^Mb_{L,m}^{}b_{L,m}`$ and $`b_L^{}e^{\beta \omega }b_R=_{m=1}^Mb_{L,m}^{}e^{\beta \omega _m}b_{R,m}`$.
In Bogoliubov theory the initial thermal state evolves under external perturbation $`V(t,x)`$ into an excited state $`\widehat{\rho }(t)`$ which has the same form as the initial $`\widehat{\rho }(0)`$ in Eq.(8) but with time-dependent Bogoliubov modes $`u_m(t,x)`$ and $`v_m(t,x)`$ which solve time-dependent Bogoliubov-de Gennes equations
$`i_tu_m`$ $`=`$ $`{\displaystyle \frac{1}{2}}_x^2u_m+{\displaystyle \frac{1}{2}}x^2u_m+2g|\varphi _0|^2u_m+g\varphi _0^2v_m,`$ (9)
$`i_tv_m`$ $`=`$ $`{\displaystyle \frac{1}{2}}_x^2v_m+{\displaystyle \frac{1}{2}}x^2v_m+2g|\varphi _0|^2v_m+g\left(\varphi _0^{}\right)^2u_m`$ (10)
with initial conditions being the eigenmodes of the stationary BdG equations (6). The time-dependent condensate wave function $`\varphi _0(t,x)`$ solves the time-dependent Gross-Pitaevskii equation
$$i_t\varphi _0=\frac{1}{2}_x^2\varphi _0+\frac{1}{2}x^2\varphi _0+g|\varphi _0|^2\varphi _0.$$
(11)
## III Probability distribution for outcomes
As mentioned before, density measurement is โcollapsingโ $`N`$-particle state to a Bose-Einstein condensate. The aim of the measurement theory is to provide probability distribution for different condensate wave functions $`\varphi `$. In the present context of Bogoliubov theory it is convenient to split possible condensate wave functions into the condensate part and the non-condensate part: $`\varphi =\varphi _0+\frac{1}{\sqrt{N}}\delta \varphi `$. The aim is to find gaussian probability distribution for $`\delta \varphi `$ in the gaussian state $`\widehat{\rho }(t)`$. Ideally the gaussian distribution would be fully determined by the following equalities between second order correlators of the gaussian $`\delta \varphi (x)`$ and second order correlators of the field operators:
$`\overline{\delta \varphi ^{}(x)\delta \varphi (y)}`$ $`\stackrel{\mathrm{?}}{=}\delta \widehat{\psi }^{}(x)\delta \widehat{\psi }(y)=`$ $`{\displaystyle \underset{m}{}}n_mu_m^{}(x)u_m(y)+(1+n_m)v_m(x)v_m^{}(y),`$ (12)
$`\overline{\delta \varphi (x)\delta \varphi ^{}(y)}`$ $`\stackrel{\mathrm{?}}{=}\delta \widehat{\psi }(x)\delta \widehat{\psi }^{}(y)=`$ $`{\displaystyle \underset{m}{}}n_mv_m^{}(x)v_m(y)+(1+n_m)u_m(x)u_m^{}(y),`$ (13)
$`\overline{\delta \varphi (x)\delta \varphi (y)}`$ $`=\delta \widehat{\psi }(x)\delta \widehat{\psi }(y)=`$ $`{\displaystyle \underset{m}{}}n_mv_m^{}(x)u_m(y)+(1+n_m)u_m(x)v_m^{}(y),`$ (14)
$`\overline{\delta \varphi ^{}(x)\delta \varphi ^{}(y)}`$ $`=\delta \widehat{\psi }^{}(x)\delta \widehat{\psi }^{}(y)=`$ $`{\displaystyle \underset{m}{}}n_mu_m^{}(x)v_m(y)+(1+n_m)v_m(x)u_m^{}(y).`$ (15)
Here the most right hand sides follow from the Bogoliubov theory . $`n_m=(e^{\beta \omega _m}1)^1`$ is average number of thermally excited Bogoliubov quasiparticles in the initial state. Unfortunately, because of the non-zero commutator $`[\delta \widehat{\psi }(x),\delta \widehat{\psi }^{}(y)]=\delta (xy)\varphi _0^{}(x)\varphi _0(y)`$, the first two conditions cannot be satisfied simultaneously. I replace them with the condition
$`\overline{\delta \varphi ^{}(x)\delta \varphi (y)}=\left(\overline{\delta \varphi (x)\delta \varphi ^{}(y)}\right)^{}={\displaystyle \frac{1}{2}}\delta \widehat{\psi }^{}(x)\delta \widehat{\psi }(y)+\delta \widehat{\psi }(y)\delta \widehat{\psi }^{}(x).`$ (16)
It is convenient to expand the fluctuation as
$$\delta \varphi (x|z)=\underset{\alpha =1}{\overset{\mathrm{}}{}}z_\alpha \varphi _\alpha (x)$$
(17)
in the ortonormal basis of the eigenmodes $`\varphi _\alpha `$ of the reduced single particle density matrix
$$\delta \widehat{\psi }^{}(x)\delta \widehat{\psi }(y)=\underset{\alpha =1}{\overset{M}{}}\delta N_\alpha \varphi _\alpha ^{}(x)\varphi _\alpha (y).$$
(18)
Here the left hand side is given by Eq.(12). The right hand side is obtained after diagonalization of the hermitean operator on the left. The real eigenvalues $`\delta N_\alpha `$ are average occupation numbers of the corresponding non-condensate modes $`\varphi _\alpha `$. The correlators (15) after the semiclassical approximation (16) determine the matrix of correlators of the complex gaussian random variables $`z_\alpha `$:
$$\left(\begin{array}{cc}\overline{z_\alpha ^{}z_\beta }& \overline{z_\alpha ^{}z_\beta ^{}}\\ \overline{z_\alpha z_\beta }& \overline{z_\alpha z_\beta ^{}}\end{array}\right)=\left(\begin{array}{cc}D_{\alpha \beta }& C_{\alpha \beta }\\ C_{\alpha \beta }^{}& D_{\alpha \beta }\end{array}\right).$$
(19)
Here the $`M\times M`$ matrices on the right hand side are
$`D_{\alpha \beta }`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(U_{\alpha m}^{}n_mU_{\beta m}+V_{\alpha m}^{}(1+n_m)V_{\beta m}+U_{\alpha m}(1+n_m)U_{\beta m}^{}+V_{\alpha m}n_mV_{\beta m}^{}\right),`$ (20)
$`C_{\alpha \beta }`$ $`=`$ $`U_{\alpha m}^{}n_mV_{\beta m}^{}+V_{\alpha m}^{}(1+n_m)U_{\beta m}^{},`$ (21)
with the matrix elements $`U_{\alpha m}=\varphi _\alpha |u_m`$ and $`V_{\alpha m}=\varphi _\alpha |v_m^{}`$. Replacing $`z_\alpha `$โs with real coordinates, $`z_\alpha =x_\alpha +iy_\alpha `$, we get a real symmetric matrix of correlators
$$\left(\begin{array}{cc}\overline{x_\alpha x_\beta }& \overline{x_\alpha y_\beta }\\ \overline{y_\alpha x_\beta }& \overline{y_\alpha y_\beta }\end{array}\right)=\frac{1}{2}\left(\begin{array}{cc}\mathrm{Re}D_{\alpha \beta }+\mathrm{Re}C_{\alpha \beta }& \mathrm{Im}C_{\alpha \beta }\\ \mathrm{Im}C_{\alpha \beta }& \mathrm{Re}D_{\alpha \beta }\mathrm{Re}C_{\alpha \beta }\end{array}\right)$$
(22)
and a condition that $`\mathrm{Im}D_{\alpha \beta }=0`$ \- a good test of the correctness of the calculations. Diagonalization of the correlation matrix (22) gives eigenvalues $`\lambda _s0`$ with $`s=1,..,2M`$. Corresponding eigenvectors are columns of an orthogonal matrix $`O`$. The eigenvectors define convenient parametrization of the gaussian fluctuation as
$$\delta \varphi (x)=\underset{\alpha =1}{\overset{M}{}}z_\alpha \varphi _\alpha (x)=\underset{\alpha =1}{\overset{M}{}}\varphi _\alpha (x)\underset{s=1}{\overset{2M}{}}\left(O_{\alpha ,s}+iO_{M+\alpha ,s}\right)q_s\underset{s=1}{\overset{2M}{}}\mathrm{\Phi }_s(x)q_s$$
(23)
with independent real gaussian random variables $`q_s`$ of zero mean and variances $`\overline{q_s^2}=\lambda _s`$. However, this is not the end of the story yet.
As a result of the semiclassical approximation in Eq.(16) averages like e.g. average density of depletion $`\overline{\delta \varphi ^{}(x)\delta \varphi (x)}`$ are divergent because there is infinite number of unoccupied modes $`\varphi _\alpha (x)`$, every one of them contributing to this depletion density a term $`\frac{1}{2}\varphi _\alpha ^{}(x)\varphi _\alpha (x)`$. In stochastic averages like $`\overline{\delta \varphi ^{}(x)\delta \varphi (x)}`$ average occupation numbers of modes $`\varphi _\alpha (x)`$ seem to be $`\delta N_\alpha +\frac{1}{2}`$ instead of the correct $`\delta N_\alpha `$. This artifact of the semiclassical approximation can be corrected by introducing to Eq.(23) of regularization factors:
$$\delta \varphi (x)|_{\mathrm{reg}}=\underset{\alpha =1}{\overset{M}{}}z_\alpha \left(\frac{\delta N_\alpha }{\delta N_\alpha +\frac{1}{2}}\right)^{1/2}\varphi _\alpha (x)=\underset{\alpha =1}{\overset{M}{}}\left(\frac{\delta N_\alpha }{\delta N_\alpha +\frac{1}{2}}\right)^{1/2}\varphi _\alpha (x)\underset{s=1}{\overset{2M}{}}\left(O_{\alpha ,s}+iO_{M+\alpha ,s}\right)q_s\underset{s=1}{\overset{2M}{}}\mathrm{\Phi }_s(x)q_s$$
(24)
As expected in semiclassical approximation, the regularizing factors $`\left(\frac{\delta N}{\delta N+\frac{1}{2}}\right)^{1/2}`$ are approximately $`1`$ for the highly occupied modes with $`\delta N_\alpha 1`$ which dominate in the density distribution, but at the same time they remove the divergence coming from the infinity of unoccupied modes. The wave functions $`\mathrm{\Phi }_s(x)`$ are in general neither normalized nor orthogonal, except in the quantum limit of zero temperature, see the proof in Ref..
## IV Conclusion
In conclusion, a recipe to simulate density measurement on the time-dependent excited thermal state has the following steps:
* Solve stationary Gross-Pitaevskii and Bogoliubov-de Gennes equations (3,6) to provide initial conditions for $`\varphi _0(t,x)`$, $`u_m(t,x)`$ and $`v_m(t,x)`$, and the initial quasiparticle frequencies $`\omega _m`$.
* Solve time-dependent Gross-Pitaevskii and Bogoliubov-de Gennes equations (11,10) with respect to $`\varphi _0(t,x)`$, $`u_m(t,x)`$, and $`v_m(t,x)`$.
* Diagonalize the reduced single particle matrix (18) to get its non-condensate eigenmodes $`\varphi _\alpha `$ with their average occupation numbers $`\delta N_\alpha `$.
* Build the matrices $`D_{\alpha \beta }`$ and $`C_{\alpha \beta }`$ in Eqs.(20,21), and then the real symmetric correlation matrix in Eq.(22).
* Diagonalize the correlation matrix in Eq.(22) to get its real eigenvalues $`\lambda _s`$ and correposnding eigenvectors $`O_{\alpha s}`$.
* Build the regularized modes $`\mathrm{\Phi }_s`$ according to their definition implicit in Eq.(24):
$$\mathrm{\Phi }_s(x)=\underset{\alpha =1}{\overset{M}{}}\left(\frac{\delta N_\alpha }{\delta N_\alpha +\frac{1}{2}}\right)^{1/2}\varphi _\alpha (x)\left(O_{\alpha ,s}+iO_{M+\alpha ,s}\right).$$
(25)
* Choose independent real random variables $`q_s`$โs from their gaussian distributions of zero mean and variance $`\overline{q_s^2}=\lambda _s`$, and then combine the chosen $`q`$โs into condensate density
$$\rho (x|q)=\left|\sqrt{N}\varphi _0(x)+\underset{s=1}{\overset{2M}{}}q_s\mathrm{\Phi }_s(x)\right|^2.$$
(26)
The $`\rho (x|q)`$ defines a family of all possible density measurement outcomes with a gaussian probability distribution for different $`q`$โs.
In the limit of zero temperature this general recipe coincides with the recipe derived by different methods in Ref.. At zero temperature the wave functions become $`\mathrm{\Phi }_\alpha (x)\varphi _\alpha (x)`$ for $`\alpha =1,\mathrm{},M`$ and zero otherwise (here the $``$ means equality up to a phase factor). Corresponding variances are $`\overline{q_\alpha ^2}=\delta N_\alpha `$.
## Acknowledgements
I would like to thank Zbyszek Karkuszewski and Krzysztof Sacha for stimulating discussions. This work was supported in part by Polish government scientific funds (2005-2008) as a research project.
|
warning/0506/math0506289.html
|
ar5iv
|
text
|
# Stability of FDโTD schemes for MaxwellโDebye and MaxwellโLorentz equations.
## 1 Introduction
To describe the propagation of an electromagnetic wave through a dispersive medium some extensions to Maxwell equations are used. They involve time differential equations which accounts for the constitutive laws of the material that link the displacement $`๐`$ to the electric field $`๐`$ or equivalently the polarization $`๐`$ to $`๐`$. We focus on two of these models (Debye and Lorentz models) which are addressed in in view of specific applications to the interaction of an electromagnetic wave with a human body. In contrast we treat any medium which is described by these models. We only consider the stability analysis of numerical schemes whereas also treated phase error issues.
### 1.1 MaxwellโDebye and MaxwellโLorentz models
In our context (no magnetization) the Maxwell equations read
$$\begin{array}{cccc}\text{(Faraday)}\hfill & \hfill _t๐(t,๐ฑ)& =& \mathrm{curl}๐(t,๐ฑ),\hfill \\ \text{(Ampรจre)}\hfill & \hfill _t๐(t,๐ฑ)& =& \frac{1}{\mu _0}\mathrm{curl}๐(t,๐ฑ),\hfill \end{array}$$
(1)
where $`๐ฑ^N`$ together with a linear constitutive law
$$๐(t,๐ฑ)=\epsilon _0\epsilon _{\mathrm{}}๐(t,๐ฑ)+\epsilon _0_{\mathrm{}}^t๐(t\tau ,๐ฑ)\chi (\tau )๐\tau ,$$
(2)
where $`\epsilon _{\mathrm{}}`$ is the relative infinite frequency permittivity and $`\chi `$ is the linear susceptibility. The discretization of the integral expression (2) leads to recursive schemes (see e.g. , ). However, differentiating Eq. (2) leads to a time differential equation for $`๐`$ which depends on the specific form of $`\chi `$. For a Debye medium
$$t_\mathrm{r}_t๐+๐=t_\mathrm{r}\epsilon _0\epsilon _{\mathrm{}}_t๐+\epsilon _0\epsilon _\mathrm{s}๐,$$
(3)
where $`t_\mathrm{r}>0`$ is the relaxation time and $`\epsilon _\mathrm{s}\epsilon _{\mathrm{}}`$ is the relative static permittivity. Defining the polarization by $`๐(t,๐ฑ)=๐(t,๐ฑ)\epsilon _0\epsilon _{\mathrm{}}๐(t,๐ฑ)`$, an equivalent form is
$$t_\mathrm{r}_t๐+๐=\epsilon _0(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})๐.$$
(4)
For a Lorentz medium with one resonant frequency $`\omega _1`$, we likewise have
$$_t^2๐+\nu _t๐+\omega _1^2๐=\epsilon _0\epsilon _{\mathrm{}}_t^2๐+\epsilon _0\epsilon _{\mathrm{}}\nu _t๐+\epsilon _0\epsilon _\mathrm{s}\omega _1^2๐,$$
(5)
where $`\nu 0`$ is a damping coefficient, and
$$_t^2๐+\nu _t๐+\omega _1^2๐=\epsilon _0(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})\omega _1^2๐.$$
(6)
If we denote by $`๐`$ the time derivative of $`๐`$, system (1) can be cast as
$$\begin{array}{ccc}\hfill _t๐(t,๐ฑ)& =& \mathrm{curl}๐(t,๐ฑ),\hfill \\ \hfill \epsilon _0\epsilon _{\mathrm{}}_t๐(t,๐ฑ)& =& \frac{1}{\mu _0}\mathrm{curl}๐(t,๐ฑ)๐(t,๐ฑ).\hfill \end{array}$$
(7)
### 1.2 Numerical schemes
A classical and very efficient way to compute the Maxwell equations is the Yee scheme . We restrict our study to existing Yee based schemes. Other methods may be found in the literature in the context of Maxwell-Debye and Maxwell-Lorentz equations: see e.g. for pseudo-spectral schemes or for finite elementโtime domain (FEโTD) schemes.
The Yee scheme consists in discretizing $`๐`$ and $`๐`$ on staggered grids in space and time. This allows to use only centered discrete differential operators. We denote by $`h`$ the space step (supposed here to be the same in all directions in the case of multi-dimensional equations) and by $`k`$ the time step. In space dimension 1, we only consider the dependence in the space variable $`z`$ and classically two polarizations for the field may be decoupled. For example, the transverse electric polarization only involves $`EE_x`$ and $`BB_y`$. The discretized variables are $`E_j^nE(nk,jh)`$ (and similar notations for $`DD_x`$) and $`B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B((n+\frac{1}{2})k,(j+\frac{1}{2})h)`$, and the Yee scheme for system (1) reads
$$\begin{array}{ccc}\hfill \frac{1}{k}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j+\frac{1}{2}}^{n\frac{1}{2}})& =& \frac{1}{h}(E_{j+1}^nE_j^n),\hfill \\ \hfill \frac{1}{k}(D_j^{n+1}D_j^n)& =& \frac{1}{\mu _0h}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j\frac{1}{2}}^{n+\frac{1}{2}}).\hfill \end{array}$$
(8)
Similarly the Yee scheme for system (7) reads
$$\begin{array}{ccc}\hfill \frac{1}{k}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j+\frac{1}{2}}^{n\frac{1}{2}})& =& \frac{1}{h}(E_{j+1}^nE_j^n),\hfill \\ \hfill \frac{\epsilon _0\epsilon _{\mathrm{}}}{k}(E_j^{n+1}E_j^n)& =& \frac{1}{\mu _0h}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j\frac{1}{2}}^{n+\frac{1}{2}})J_j^{n+\frac{1}{2}}.\hfill \end{array}$$
(9)
Usual Maxwell equations consist in taking $`J_j^{n+\frac{1}{2}}0`$ in Eq. (9) or equivalently $`D_j^n=\epsilon _0\epsilon _{\mathrm{}}E_j^n`$ in Eq. (8) and leads to a stable second order scheme under a CourantโFriedrichsโLewy (CFL) stability condition. Namely, if $`c_{\mathrm{}}=1/\sqrt{\epsilon _0\epsilon _{\mathrm{}}\mu _0}`$ denotes the infinite frequency light speed, the CFL condition reads $`c_{\mathrm{}}kh`$ if the space dimension is $`N=1`$ and $`c_{\mathrm{}}kh/\sqrt{2}`$ for $`N=2`$ or 3.
In contrast to the recursive schemes, we are interested in direct integration schemes which are based on the finite differenceโtime domain (FDโTD) discretization of Eqs (3) to (6) (see , , ).
### 1.3 Outline
The von Neumann stability analysis is recalled in Sect. 2. We also describe the sketch of our proofs which is common for all the schemes. In Section 3 two one dimensional direct integration schemes for Debye media are presented and analyzed, pointing carefully out the physical properties needed to ensure stability and the specific cases which have to be handled separately. Numerical applications to physical media are also given. The same point of view is carried out for Lorentz media in Section 4. Two-dimensional results are given in Section 5.
## 2 Principles of the von Neumann analysis
The von Neumann analysis allows to localize roots of certain classes of polynomials, which proves to be crucial here. We recall the main principles of this technique. Details and proofs of theorems may be found in .
### 2.1 Schur and von Neumann polynomials
We define two families of polynomials: Schur polynomials and simple von Neumann polynomials.
###### Definition 1
A polynomial is a Schur polynomial if all its roots, $`r`$, satisfy $`|r|<1`$.
###### Definition 2
A polynomial is a simple von Neumann polynomial if all its roots, $`r`$, lie on the unit disk ($`|r|1`$) and its roots on the unit circle are simple roots.
If a polynomial is of high degree or has sophisticated coefficients, it may be difficult to locate its roots. However, there is a way to split this difficult problem into many simpler ones. For this aim, we construct a sequence of polynomials of decreasing degree. Let $`\varphi `$ be written as
$$\varphi (z)=c_0+c_1z+\mathrm{}+c_pz^p,$$
where $`c_0`$, $`c_1`$ โฆ, $`c_p`$ and $`c_p0`$. We define its conjugate polynomial $`\varphi ^{}`$ by
$$\varphi ^{}(z)=c_p^{}+c_{p1}^{}z+\mathrm{}+c_0^{}z^p.$$
Given a polynomial $`\varphi _0`$, we may define a sequence of polynomials
$$\varphi _{m+1}(z)=\frac{\varphi _m^{}(0)\varphi _m(z)\varphi _m(0)\varphi _m^{}(z)}{z}.$$
It is clear that $`\text{deg}\varphi _{m+1}<\text{deg}\varphi _m`$, if $`\varphi _m0`$. Besides, we have the two following theorems.
###### Theorem 1
A polynomial $`\varphi _m`$ is a Schur polynomial of exact degree $`d`$ if and only if $`\varphi _{m+1}`$ is a Schur polynomial of exact degree $`d1`$ and $`|\varphi _m(0)||\varphi _m^{}(0)|`$.
###### Theorem 2
A polynomial $`\varphi _m`$ is a simple von Neumann polynomial if and only if
$``$ $`\varphi _{m+1}`$ is a simple von Neumann polynomial and $`|\varphi _m(0)||\varphi _m^{}(0)|`$,
or
$``$ $`\varphi _{m+1}`$ is identically zero and $`\varphi _m^{}`$ is a Schur polynomial.
The main ingredient in the proof of both theorems is the Rouchรฉ theorem (see ). To analyze $`\varphi _0`$, at each step $`m`$, conditions should be checked (leading coefficient is non-zero, $`|\varphi _m(0)||\varphi _m^{}(0)|`$, โฆ) until a definitive negative answer arises or the degree is 1.
### 2.2 Stability analysis
The models we deal with are linear models. They may therefore be analyzed in the frequency domain. Thus we assume that the scheme handles a variable $`U_๐ฃ^n`$ with spatial dependence
$$U_๐ฃ^n=U^n\mathrm{exp}(i๐๐ฃ),$$
where $`๐`$ and $`๐ฃ^N`$, $`N=1,2,3`$. The amplification matrix $`G`$ is the matrix such that $`U^{n+1}=GU^n`$. We assume that $`G`$ does not depend on time or on $`h`$ and $`k`$ separately but only on the ratio $`h/k`$. Let $`\varphi _0`$ be the characteristic polynomial of $`G`$, then we have a sufficient stability condition.
###### Theorem 3
A sufficient stability condition is that $`\varphi _0`$ be a simple von Neumann polynomial.
This condition is not necessary. A scheme is stable if and only if the sequence $`(U^n)_n`$ is bounded. Since we assume that $`G`$ does not depend on time, $`U^n=G^nU^0`$ and stability is also the boundedness of $`(G^n)_n`$. If the eigenvalues of $`G`$, i.e. the roots $`r`$ of $`\varphi _0`$, lie inside the unit circle ($`|r|<1`$), then $`lim_n\mathrm{}G^n=0`$ and the sequence is bounded. If any root lies outside the unit circle then $`G^n`$ grows exponentially and the scheme is unstable. The intermediate case when some roots may be on the unit circle (and the others inside) may lead to different situations. The good case is for example given when $`G`$ is the identity. Then $`U^n=U^0`$ and the scheme is clearly stable. However there are other examples of matrices with multiple roots on the unit circle that lead either to bounded or unbounded sequences $`(G^n)_n`$. We will call this property $`G^n`$-boundedness in the sequel. It is clearly a property of the amplification matrix and not of its characteristic polynomial. If the minimal stable subspaces associated to the multiple root are one-dimensional then $`G^n`$ is bounded (identity example). If the minimal stable subspaces are multidimensional then $`G^n`$ grows linearly. Such cases (which occur for our schemes) should therefore be handled specifically.
### 2.3 Sketch of proofs
In the next sections, we will not give the proofs, but only list in a table the arguments used for each situation. We describe here the general plan and give names to specific final arguments used. The detailed proofs may be found in for space dimensions 1 and 2. The three dimensional case is much more tedious and is work in progress.
Usually the system is given in a implicit form. The first step consists in writing it in an explicit form. This yields the amplification matrix $`G`$. Then we compute its characteristic polynomial $`\varphi _0`$. In order to perform a von Neumann analysis, we compute the series $`(\varphi _m)`$. In the general case, under the assumption that the stability condition cannot be better than Maxwellโs, we can apply either Theorem 1 (Theorem 1 argument) or Theorem 2 (Theorem 2 argument), check estimates at each level until $`\varphi _m`$ is a one degree polynomial. Special cases arise when $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$, $`\mathrm{sin}(\xi /2)=0`$ or $`\pm 1`$, and sometimes for limit values of physical coefficients. In these cases, different points of view have to be considered:
* Theorem 2 has to be used instead of Theorem 1,
* Some eigenvalues lie on the unit circle (mostly $`\pm 1`$ or $`\pm i`$) and are simple, it is then sufficient to study only the other eigenvalues (sub-polynomial argument) and we conclude to a simple von Neumann polynomial and stability,
* Some eigenvalues lie on the unit circle and are not simple, and besides the study of the other eigenvalues (to prove that the polynomial is a von Neumann one), we have to find out if the associated minimal stable subspaces are one- (stable case) or multidimensional (unstable case). This may be checked directly on the form of matrix $`G`$ ($`G`$ form argument), or necessitates the computation of eigenvectors (eigenvectors argument). If only one eigendirection is found for a multiple eigenvalue, the minimal subspace is necessarily multidimensional.
## 3 Debye media
We address two discretizations of MaxwellโDebye equations. The first one uses a $`(๐,๐,๐)`$ setting for the equations and the second a $`(๐,๐,๐,๐)`$ formulation.
### 3.1 DebyeโJoseph et al. model
In , Joseph et al. close System (8) by a discretization for Eq. (3), namely
$$\begin{array}{c}\epsilon _0\epsilon _{\mathrm{}}t_\mathrm{r}\frac{E_j^{n+1}E_j^n}{k}+\epsilon _0\epsilon _\mathrm{s}\frac{E_j^{n+1}+E_j^n}{2}=t_\mathrm{r}\frac{D_j^{n+1}D_j^n}{k}+\frac{D_j^{n+1}+D_j^n}{2}.\hfill \end{array}$$
(10)
System (8)โ(10) may be cast in an explicit form which handles the variable
$$U_j^n={}_{}{}^{t}(c_{\mathrm{}}B_{j+\frac{1}{2}}^{n\frac{1}{2}},E_j^n,D_j^n/\epsilon _0\epsilon _{\mathrm{}})$$
and the amplification matrix $`G`$ reads
$$\left(\begin{array}{ccc}1& \lambda (e^{i\xi }1)& 0\\ \frac{(1+\delta )\lambda (1e^{i\xi })}{1+\delta \epsilon _\mathrm{s}^{}}& \frac{(1\delta \epsilon _\mathrm{s}^{})+(1+\delta )\lambda ^2(e^{i\xi }2+e^{i\xi })}{1+\delta \epsilon _\mathrm{s}^{}}& \frac{2\delta }{1+\delta \epsilon _\mathrm{s}^{}}\\ \lambda (1e^{i\xi })& \lambda ^2(e^{i\xi }2+e^{i\xi })& 1\end{array}\right)$$
where $`\lambda =c_{\mathrm{}}k/h`$ is the CFL constant, $`\delta =k/2t_\mathrm{r}>0`$ is the normalized time step and $`\epsilon _\mathrm{s}^{}=\epsilon _\mathrm{s}/\epsilon _{\mathrm{}}1`$ denotes the normalized static permittivity. Moreover we define
$$q=\lambda ^2(e^{i\xi }2+e^{i\xi })=4\lambda ^2\mathrm{sin}^2(\xi /2).$$
The characteristic polynomial is proportional to
$`\varphi _0(Z)`$ $`=`$ $`[1+\delta \epsilon _\mathrm{s}^{})]Z^3[3+\delta \epsilon _\mathrm{s}^{}(1+\delta )q]Z^2`$
$`+[3\delta \epsilon _\mathrm{s}^{}(1\delta )q]Z[1\delta \epsilon _\mathrm{s}^{}].`$
The proofs are summed up in Table 1 and we deduce that the stability condition is $`q4`$ if $`\epsilon _\mathrm{s}>\epsilon _{\mathrm{}}`$ and $`q<4`$ if $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$.
### 3.2 DebyeโYoung model
In , Young closes System (9) by two discretizations for Eq. (4), namely
$$t_\mathrm{r}\frac{P_j^{n+\frac{1}{2}}P_j^{n\frac{1}{2}}}{k}=\frac{P_j^{n+\frac{1}{2}}+P_j^{n\frac{1}{2}}}{2}+\epsilon _0(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})E_j^n,$$
(11)
$$t_\mathrm{r}J_j^{n+\frac{1}{2}}=P_j^{n+\frac{1}{2}}+\epsilon _0(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})\frac{E_j^{n+1}+E_j^n}{2}.$$
(12)
Although $`J_j^{n+\frac{1}{2}}`$ is used for the computations, this not a genuine variable for System (9)โ(11)โ(12) which handles the variable
$$U_j^n={}_{}{}^{t}(c_{\mathrm{}}B_{j+\frac{1}{2}}^{n\frac{1}{2}},E_j^n,P_j^{n\frac{1}{2}}/\epsilon _0\epsilon _{\mathrm{}})$$
and the amplification matrix $`G`$ reads
$$\left(\begin{array}{ccc}1& \lambda (e^{i\xi }1)& 0\\ \frac{\lambda (1e^{i\xi })}{1+\delta \alpha }& \frac{1+\delta \delta \alpha +3\delta ^2\alpha (1+\delta )q}{(1+\delta )(1+\delta \alpha )}& \frac{1\delta }{1+\delta }\frac{2\delta }{1+\delta \alpha }\\ 0& \frac{2\delta \alpha }{1+\delta }& \frac{1\delta }{1+\delta }\end{array}\right)$$
with the same notation as above and $`\alpha =\epsilon _\mathrm{s}^{}10`$.
The characteristic polynomial is proportional to
$`\varphi _0(Z)`$ $`=`$ $`[(1+\delta \alpha )(1+\delta )]Z^3[3+\delta +\delta \alpha +3\delta ^2\alpha (1+\delta )q]Z^2`$
$`+[3\delta \delta \alpha +3\delta ^2\alpha (1\delta )q]Z[(1\delta \alpha )(1\delta )].`$
Again, the proofs are summed up in Table 2.
The stability condition is therefore $`q4`$ and $`\delta 1`$ if $`\epsilon _\mathrm{s}>\epsilon _{\mathrm{}}`$ and $`q<4`$ if $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$.
### 3.3 Conclusion for one-dimensional Debye schemes
If $`\epsilon _\mathrm{s}>\epsilon _{\mathrm{}}`$, the pure CFL condition $`q4`$ is the same for both models. It is exactly the condition for Maxwell equations. However Young model necessitates another condition, $`\delta 1`$, which corresponds to a sufficient discretization of Debye equation (4). Even if we are interested here in stability properties, such conditions are to be taken to ensure equations to be correctly taken into account. Results are given in physical variables in Table 3.
To compare conditions on $`q`$ and $`\delta `$, let us consider a simple physical case. We assume that a matter with $`\epsilon _{\mathrm{}}=1`$ (and thus $`c_{\mathrm{}}\mathrm{3\hspace{0.17em}10}^8\text{m\hspace{0.17em}s}^1`$) is lighted by an optical wave of say wavelength 1 $`\mu `$m. The space step $`h`$ has to be smaller than this wavelength, and therefore $`q<4`$ reads at least $`k<\frac{1}{3}\mathrm{\hspace{0.17em}10}^{14}\text{s}`$. In a Debye medium, relaxation times $`t_\mathrm{r}`$ are of the order of a picosecond (or even a nanosecond) which is many decades larger than the previous bound. The estimate $`q<4`$ is thus predominant and both models present the same advantages. Only the value of $`\epsilon _{\mathrm{}}`$ yields the CFL condition. A typical example is water for which $`\epsilon _{\mathrm{}}=1.8`$, $`\epsilon _\mathrm{s}=81.0`$ and $`t_\mathrm{r}=\mathrm{9.4\hspace{0.17em}10}^{12}\text{s}`$ . Condition $`k2t_\mathrm{r}`$ comes to $`k\mathrm{1.88\hspace{0.17em}10}^{11}\text{s}`$. Condition $`q4`$ yields a similar condition if $`h=\mathrm{4.2\hspace{0.17em}10}^3\text{m}`$. This is of course much larger than any reasonable space step for Maxwell equations and optical waves. The stability condition for water is $`q<4`$ for both schemes. A quite different material is for example the 0.25-dB loaded foam given in for which $`\epsilon _{\mathrm{}}=1.01`$, $`\epsilon _\mathrm{s}=1.16`$ and $`t_\mathrm{r}=\mathrm{6.497\hspace{0.17em}10}^{10}\text{s}`$. Condition $`k2t_\mathrm{r}`$ comes to $`k\mathrm{1.3\hspace{0.17em}10}^9\text{s}`$ and $`q4`$ yields a similar condition if $`h=\mathrm{3.9\hspace{0.17em}10}^1\text{m}`$. Once more, the stability condition for water is $`q<4`$ for both schemes.
In conclusion for current material the stability condition is the same for MaxwellโDebye equations as for the usual Yee scheme. The result announced in was $`q4`$ for Joseph et al. scheme and for water, which is consistent with our result.
## 4 Lorentz media
Three discretizations of MaxwellโLorentz equations are now addressed. The first one uses a $`(๐,๐,๐)`$ setting and the two others a $`(๐,๐,๐,๐)`$ formulation, but differ from the time-discretization of $`๐`$.
Each of these models reads the same in the harmonic ($`\nu =0`$) or an-harmonic ($`\nu >0`$) cases. However the analysis will differ greatly since $`\varphi _10`$ for all the schemes in the harmonic cases.
### 4.1 LorentzโJoseph et al. model
In , system (8) is closed by a discretization for Eq. (3), namely
$$\begin{array}{c}\epsilon _0\epsilon _{\mathrm{}}\frac{E_j^{n+1}2E_j^n+E_j^{n1}}{k^2}+\nu \epsilon _0\epsilon _{\mathrm{}}\frac{E_j^{n+1}E_j^{n1}}{2k}+\epsilon _0\epsilon _\mathrm{s}\omega _1^2\frac{E_j^{n+1}+E_j^{n1}}{2}\hfill \\ =\frac{D_j^{n+1}2D_j^n+D_j^{n1}}{k^2}+\nu \frac{D_j^{n+1}D_j^{n1}}{2k}+\omega _1^2\frac{D_j^{n+1}+D_j^{n1}}{2}\hfill \end{array}$$
(13)
The explicit version of system (8)โ(13) does not use explicitly the value of $`D_j^{n1}`$ and therefore this system handles the variable
$$U_j^n={}_{}{}^{t}(c_{\mathrm{}}B_{j+\frac{1}{2}}^{n\frac{1}{2}},E_j^n,E_j^{n1},D_j^n/\epsilon _0\epsilon _{\mathrm{}}).$$
The amplification matrix $`G`$ reads
$$\left(\begin{array}{cccc}1& \lambda (e^{i\xi }1)& 0& 0\\ \frac{2\delta \lambda (1e^{i\xi })}{1+\delta +\omega \epsilon _\mathrm{s}^{}}& \frac{2q(1+\delta +\omega )}{1+\delta +\omega \epsilon _\mathrm{s}^{}}& \frac{1\delta +\omega \epsilon _\mathrm{s}^{}}{1+\delta +\omega \epsilon _\mathrm{s}^{}}& \frac{2\omega }{1+\delta +\omega \epsilon _\mathrm{s}^{}}\\ 0& 1& 0& 0\\ \lambda (1e^{i\xi })& q& 0& 1\end{array}\right)$$
where $`\delta =\nu k/20`$ is the new normalized time step, and $`\omega =\omega _1^2k^2/2>0`$ denotes the normalized squared frequency. The other notations used for the Debye model remain valid.
The characteristic polynomial is proportional to
$`\varphi _0(Z)`$ $`=`$ $`[1+\delta +\omega \epsilon _\mathrm{s}^{}]Z^4[4+2\delta +2\omega \epsilon _\mathrm{s}^{}(1+\delta +\omega )q]Z^3`$
$`+[6+2\omega \epsilon _\mathrm{s}^{}2q]Z^2[42\delta +2\omega \epsilon _\mathrm{s}^{}(1\delta +\omega )q]Z`$
$`+[1\delta +\omega \epsilon _\mathrm{s}^{}].`$
The proofs are summed up in Table 4 for the an-harmonic and the harmonic case.
In the an-harmonic case the stability condition is $`q2`$ whatever $`\epsilon _\mathrm{s}\epsilon _{\mathrm{}}`$ is. The $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$ harmonic case, needs some explanation. For $`q]0,2]`$, $`\varphi _0`$ may be cast as the product of two second order polynomials. The roots are two couples of conjugate complex roots of modulus 1. For the specific value $`q=2\omega /(1+\omega )`$, which always lies in the interval $`]0,2]`$, the two couples degenerate in one double couple, and the associated minimal stable sub-spaces are two-dimensional. To avoid this instability one may think to bound $`q`$ and say that the scheme is stable provided $`q[0,2\omega /(1+\omega )[`$. But if we come back to the original variables, we see that this is not an upper bound on $`k`$ but rather a lower bound on $`h`$, which we surely do not want. It is therefore better to avoid using Joseph et al. scheme in this very specific case, $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$ and $`\nu =0`$, and we hope to find a better scheme for this case in the following examples.
### 4.2 LorentzโKashiwa et al. model
In , Kashiwa et al. close a modified version of System (9), which consists of the three first equations in System (14), by a discretization for Eq. (6), namely
$$\begin{array}{ccc}\hfill \frac{1}{k}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j+\frac{1}{2}}^{n\frac{1}{2}})& =& \frac{1}{h}(E_{j+1}^nE_j^n),\hfill \\ \hfill \frac{\epsilon _0\epsilon _{\mathrm{}}}{k}(E_j^{n+1}E_j^n)& =& \frac{1}{\mu _0h}(B_{j+\frac{1}{2}}^{n+\frac{1}{2}}B_{j\frac{1}{2}}^{n+\frac{1}{2}})\frac{1}{k}(P_j^{n+1}P_j^n),\hfill \\ \hfill \frac{1}{k}(P_j^{n+1}P_j^n)& =& \frac{1}{2}(J_j^{n+1}+J_j^n),\hfill \\ \hfill \frac{1}{k}(J_j^{n+1}J_j^n)& =& \frac{\nu }{2}(J_j^{n+1}+J_j^n)+\frac{\omega _1^2(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})\epsilon _0}{2}(E_j^{n+1}+E_j^n)\hfill \\ & & \frac{\omega _1^2}{2}(P_j^{n+1}+P_j^n).\hfill \end{array}$$
(14)
The explicit version of system (14) handles the variable
$$U_j^n={}_{}{}^{t}(c_{\mathrm{}}B_{j+\frac{1}{2}}^{n\frac{1}{2}},E_j^n,P_j^n/\epsilon _0\epsilon _{\mathrm{}},kJ_j^n/\epsilon _0\epsilon _{\mathrm{}})$$
and the amplification matrix $`G`$ reads
$$\left(\begin{array}{cccc}1& \lambda (e^{i\xi }1)& 0& 0\\ \frac{\lambda (1e^{i\xi })(\mathrm{\Delta }\frac{1}{2}\omega \alpha )}{\mathrm{\Delta }}& \frac{\mathrm{\Delta }q\mathrm{\Delta }(2q)\frac{1}{2}\omega \alpha }{\mathrm{\Delta }}& \frac{\omega }{\mathrm{\Delta }}& \frac{1}{\mathrm{\Delta }}\\ \frac{\lambda (1e^{i\xi })\frac{1}{2}\omega \alpha }{\mathrm{\Delta }}& \frac{(2q)\frac{1}{2}\omega \alpha }{\mathrm{\Delta }}& \frac{\mathrm{\Delta }\omega }{\mathrm{\Delta }}& \frac{1}{\mathrm{\Delta }}\\ \frac{\lambda (1e^{i\xi })\omega \alpha }{\mathrm{\Delta }}& \frac{(2q)\omega \alpha }{\mathrm{\Delta }}& \frac{2\omega }{\mathrm{\Delta }}& \frac{2\mathrm{\Delta }}{\mathrm{\Delta }}\end{array}\right)$$
where together with the previously defined notations, $`\mathrm{\Delta }=1+\delta +\omega \epsilon _\mathrm{s}^{}/2`$.
The characteristic polynomial is proportional to
$`\varphi _0(Z)`$ $`=`$ $`[1+\delta +{\displaystyle \frac{1}{2}}\omega \epsilon _\mathrm{s}^{}]Z^4[4+2\delta (1+\delta +{\displaystyle \frac{1}{2}}\omega )q]Z^3`$
$`+[6\omega \epsilon _\mathrm{s}^{}+(\omega 2)q]Z^2[42\delta (1\delta +{\displaystyle \frac{1}{2}}\omega )q]Z`$
$`+[1\delta +{\displaystyle \frac{1}{2}}\omega \epsilon _\mathrm{s}^{}].`$
The proofs are summed up in Table 5. Both in the an-harmonic and harmonic cases, the stability condition is $`q<4`$ which is much better than the previous scheme since we gain a factor 2 on $`k`$ and we have no problem when $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$ and $`\nu =0`$ as for the previous model.
### 4.3 LorentzโYoung model
In , System (9) is closed by a discretization for Eq. (6), namely
$$\begin{array}{c}\frac{1}{k}(P_j^{n+1}P_j^n)=J^{n+\frac{1}{2}},\hfill \\ \frac{1}{k}(J_j^{n+\frac{1}{2}}J_j^{n\frac{1}{2}})=\frac{\nu }{2}(J_j^{n+\frac{1}{2}}+J_j^{n\frac{1}{2}})\hfill \\ +\omega _1^2(\epsilon _\mathrm{s}\epsilon _{\mathrm{}})\epsilon _0E_j^n\omega _1^2P_j^n.\hfill \end{array}$$
(15)
The explicit version of System (9)โ(15) handles once more the variable
$$U_j^n={}_{}{}^{t}(c_{\mathrm{}}B_{j+\frac{1}{2}}^{n\frac{1}{2}},E_j^n,P_j^n/\epsilon _0\epsilon _{\mathrm{}},kJ_j^n/\epsilon _0\epsilon _{\mathrm{}})$$
and the amplification matrix $`G`$ reads
$$\left(\begin{array}{cccc}1& \lambda (e^{i\xi }1)& 0& 0\\ \lambda (1e^{i\xi })& \frac{(1q)(1+\delta )2\omega \alpha }{1+\delta }& \frac{2\omega }{1+\delta }& \frac{1\delta }{1+\delta }\\ 0& \frac{2\omega \alpha }{1+\delta }& \frac{1+\delta 2\omega }{1+\delta }& \frac{1\delta }{1+\delta }\\ 0& \frac{2\omega \alpha }{1+\delta }& \frac{2\omega }{1+\delta }& \frac{1\delta }{1+\delta }\end{array}\right)$$
The characteristic polynomial is proportional to
$`\varphi _0(Z)`$ $`=`$ $`[1+\delta ]Z^4[4+2\delta 2\omega \epsilon _\mathrm{s}^{}(1+\delta )q]Z^3`$
$`+2[32\omega \epsilon _\mathrm{s}^{}+(\omega 1)q]Z^2`$
$`[42\delta 2\omega \epsilon _\mathrm{s}^{}(1\delta )q]Z+[1\delta ].`$
The proofs are summed up in Table 6. This scheme combines three drawbacks we have already encountered. First as for the Debye model, there is an extra condition on the time step: $`\omega <2/(2\epsilon _\mathrm{s}^{}1)`$. This will have to be compared to the condition on $`q`$ for physical examples. Second, as for the LorentzโJoseph et al. scheme we need a twice smaller $`k`$ than for raw Maxwell equations: $`q2`$ instead of $`q4`$. Last, and also as for the Lorentz-Joseph et al. model, the $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$ and $`\nu =0`$ leads to an instability. This is exactly the same story. This time $`q=2\omega `$ leads to double couples of conjugate complex roots of modulus 1, with two-dimensional minimal stable sub-spaces. If $`\omega <1`$ this value of $`q`$ is however never reached, but $`\omega <1`$ is a stronger assumption than $`\omega <2/(2\epsilon _\mathrm{s}^{}1)`$. We will see what this amounts to in numerical applications.
### 4.4 Conclusion for one-dimensional Lorentz schemes
We can summarize all our results for Lorentz schemes in Table 7. We chose not to translate the result for the Young scheme for $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$ as a condition on $`h`$ ($`q<2\omega `$) but as a condition on $`k`$ ($`\omega <1`$, and therefore $`q=2\omega `$ is not reached).
For the harmonic Young scheme if $`\epsilon _\mathrm{s}>\epsilon _{\mathrm{}}`$ the condition is slightly better since $`q=2`$ and $`\omega <2/(2\epsilon _\mathrm{s}^{}1)`$, or $`q<2`$ and $`\omega =2/(2\epsilon _\mathrm{s}^{}1)`$ also yield stable schemes.
Contrarily to Debye materials, for which Joseph et al. model and Young model compete, the Kashiwa et al. model seems to overcome others for Lorentz material. First, there is a gain in CFL condition $`q<4`$ is twice better as $`q2`$, second, there are no instabilities for limiting values of the physical coefficients and last there are no extra condition on the time step. In practice, an extra condition is however needed to account for the dynamics of the Lorentz equation, but not for stability reasons.
However we can compare the relative strength of the different conditions on $`k`$ for Joseph et al. and Young models. The values used in are $`\epsilon _{\mathrm{}}=1`$, $`\epsilon _\mathrm{s}=2.25`$, $`\omega _1=\mathrm{4\hspace{0.17em}10}^{16}\text{rad\hspace{0.17em}s}^1`$ and $`\nu =\mathrm{0.56\hspace{0.17em}10}^{16}\text{rad\hspace{0.17em}s}^1`$. Condition $`\omega 2/\sqrt{2\epsilon _\mathrm{s}^{}1}`$ comes to $`k\mathrm{2.7\hspace{0.17em}10}^{17}\text{s}`$ which is very small and corresponds to $`h=\mathrm{1.13\hspace{0.17em}1}^8\text{m}`$ in the $`q<2`$ condition. This space step is more than sufficient to discretize optical waves. For such a material the extra condition imposed by the Joseph et al. scheme is stronger than the basic CFL condition. The Kashiwa et al. model is then more advisable.
In there is a totally different material for which $`\epsilon _{\mathrm{}}=1.5`$, $`\epsilon _\mathrm{s}=3`$, $`\omega _1=2\pi \mathrm{\hspace{0.17em}5\hspace{0.17em}10}^{10}\text{rad\hspace{0.17em}s}^1`$ and $`\nu =10^{10}\text{rad\hspace{0.17em}s}^1`$ (these round values certainly refer to a model material). In this case $`\omega 2/\sqrt{2\epsilon _\mathrm{s}^{}1}`$ comes to $`k\mathrm{3.6\hspace{0.17em}10}^{12}\text{s}`$ which corresponds to $`h=\mathrm{1.9\hspace{0.17em}1}^3\text{m}`$ in the $`q<2`$ condition. For this material condition $`q<2`$ is the strongest for optical waves. The Kashiwa et al. model is however more advisable, since it allows $`q<4`$ instead of $`q2`$.
The results obtained in where obtained for our first cited material and for Joseph et al. and Kashiwa et al. models. He observed instabilities for $`\xi >\frac{\pi }{2}`$. We note that if $`\xi \frac{\pi }{2}`$ then $`\mathrm{sin}(\xi /2)1/\sqrt{2}`$ and $`q2`$ instead of $`q4`$. This is exactly our result. He found also the Kashiwa et al. scheme to stable for $`q4`$.
## 5 Two-dimensional results
In a two-dimensional context where unknowns depend only on space variables $`x`$ and $`y`$, Maxwell system may be split in two decoupled systems corresponding to the transverse electric (TE) ($`B_x`$, $`B_y`$, $`E_z`$) and the transverse magnetic (TM) ($`B_z`$, $`E_x`$, $`E_y`$) polarizations. In the one-dimensional case, MaxwellโDebye equations were represented by three equations and MaxwellโLorentz by four equations. In the TE polarization, one more Faraday equation is added and we have four equations for MaxwellโDebye and five equations for MaxwellโLorentz. In the TM polarization for the MaxwellโDebye model, one Ampรจre equation and one Debye equation have to be added, leading to five equations systems. For the MaxwellโLorentz model, there are one Ampรจre equation and two Lorentz equations more, and the system consists of seven equations.
The principle of the stability analysis is exactly the same, but we now have larger polynomials to study. A small miracle however happens: one-dimensional polynomials are a factor in two-dimensional polynomials. More precisely we now denote by $`h_x`$ and $`h_y`$ the space steps in the $`x`$\- and $`y`$-directions respectively and by $`q`$ the quantity
$$q=q_x+q_y=4c_{\mathrm{}}^2\left(\frac{k^2}{h_x^2}\mathrm{sin}^2(\xi _x/2)+\frac{k^2}{h_y^2}\mathrm{sin}^2(\xi _y/2)\right)$$
(recall $`q=4c_{\mathrm{}}^2\frac{k^2}{h_x^2}\mathrm{sin}^2(\xi _x/2)`$ in 1D). Then in the two-dimensional TE polarization
$$\varphi _0^{2D,TE}(Z)=[Z1]\varphi _0^{1D}(Z),$$
for all the MaxwellโDebye and Maxwell-Lorentz schemes we study here. This could be a problem, if 1 is already a root of $`\varphi _0^{1D}(Z)`$, i.e. when $`q=0`$, but it happens that it is never a problem: minimal stable sub-spaces are always one-dimensional. In the TM polarization, the same factorization occurs but the remaining polynomial is slightly more complicated, namely
$$\varphi _0^{2D,TM}(Z)=[Z1]\psi _0(Z)\varphi _0^{1D}(Z),$$
where $`\psi _0(Z)`$ is equal to:
โ DebyeโJoseph et al. model
$$[(1+\delta \epsilon _\mathrm{s}^{})Z(1\delta \epsilon _\mathrm{s}^{})].$$
โ DebyeโYoung model
$$[(1+\alpha )(1+\delta \alpha )Z(1\alpha )(1\delta \alpha )].$$
โ LorentzโJoseph et al. model
$$[(1+\delta +\omega \epsilon _\mathrm{s}^{})Z^22Z+(1\delta +\omega \epsilon _\mathrm{s}^{})].$$
\- LorentzโKashiwa et al. model
$$[(1+\delta +\frac{1}{2}\omega \epsilon _\mathrm{s}^{})Z^2(2\omega \epsilon _\mathrm{s}^{})Z+(1\delta +\frac{1}{2}\omega \epsilon _\mathrm{s}^{})].$$
โ LorentzโYoung model
$$[(1+\delta )Z^22(1\omega \epsilon _\mathrm{s}^{})Z+(1\delta )].$$
As for the TE polarization the extra eigenvalue 1 is never a source of instability. The other extra eigenvalues always lie inside or on the unit circle (conjugate complex roots). The only problem is when modulus 1 eigenvalues are also eigenvalues of the one-dimensional polynomial. This only occurs for the Lorentz-Joseph et al. scheme is $`\epsilon _\mathrm{s}=\epsilon _{\mathrm{}}`$, and $`q=2\omega /(1+\omega )`$, which is a resonant value we have already pointed out in the harmonic case for this scheme.
We shall not duplicate Tables 3 and 7 for two-dimensional models. If $`h_x=h_yh`$, condition $`q4`$ becomes $`kh/(\sqrt{2}c_{\mathrm{}})`$ and condition $`q2`$ becomes $`kh/(2c_{\mathrm{}})`$ in the physical variables. Besides, LorentzโJoseph et al. model which was leading to a lower bound on $`h`$ in the harmonic case, leads also to such a bound in the an-harmonic case. These are the only differences with Tables 3 and 7.
## 6 Conclusion
We have studied a class of FDโTD schemes for dispersive materials based on the Yee scheme for Maxwell equations and compared them from the stability point of view. This study was inspired by Petropoulos who performs the same analysis but using specific values for the physical and numerical constants and using numeric routines to locate eigenvalues of the amplification matrix. Here we have general results which gives you the constraint on numerical constants ($`k`$ and $`h`$) for any Debye or Lorentz material. Our results confirm those of Petropoulos.
For usual Debye media, both studied schemes are stable under the same conditions as the Yee scheme, ensuring also, if applied to optical waves, a fine discretization of the Debye equation. Among the studied schemes for Lorentz media, Kashiwa et al. model clearly ranks first as far as stability is concerned., Its stability condition is also that of the Yee scheme. However to take properly into account the Lorentz model, a smaller time step may have to be chosen, independently of stability issues. Such results have been proved for 1D and 2D models. The 3D case, which is much more tedious, is being studied and analogous results are expected.
|
warning/0506/astro-ph0506601.html
|
ar5iv
|
text
|
# Constraining the extra heating of the Diffuse Ionized Gas in the Milky Way
## 1 Introduction
Containing typically half the mass of ionized hydrogen in galaxies, the Diffuse Ionized Gas (DIG) is visible as an extended H$`\alpha `$ emitting layer in our Galaxy (e.g. the WHAMโsurvey) and in many other galaxies (see e.g. Tรผllmann & Dettmar tuellmannB (2000), Collins & Rand collins (2001), Otte et al. otte (2001), Hoopes & Walterbos hoopes (2003)). Studies of emission line ratios such as \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$ provide information about the physical conditions of the gas. Simple energy estimations (Reynolds reynolds90 (1990), Reynolds reynolds93 (1993)) favor O and early B stars to be responsible for most of the DIG. 3D models using various methods (e.g. Miller & Cox miller2 (1993), Dove & Shull dove (1994), Wood & Loeb wood2 (2000), Ciardi et al. ciardi (2002), Wood et al. wood (2004)) showed that it is possible for ionizing photons from O stars to penetrate from the midplane into the halo. Wood & Mathis wood3 (2004) noted that the line ratios increase with distance from the midplane due to the progressive hardening of the radiation. So far photoionization models made no specific attempt to model the trends of the line ratios. Models by Mathis mathis (2000), Domgรถrgen & Mathis domgoergen (1994), Sembach et al. sembach (2000), BlandโHawthorn et al. bland-hawthorn (1997) used volume average models to explain the observed data. The analytical approach by Haffner et al. haffner (1999), referred to as Haffner99 in the rest of this paper, treated the dependence of the line ratios with height. Haffner99 and its further application to other galaxies by e.g. Collins & Rand collins (2001), Otte et al. otte (2001), and Miller & Veilleux miller (2003) gave evidence that an additional heating source is needed in order to explain the rise of the line ratios with increasing distance $`z`$ from the midplane.
We are constructing photoionization models in order to examine the trends in the observed line ratios and if photoionization can heat up the gas sufficiently in order to explain the data. We introduce specific extra heating terms, extra means in addition to photoionization, and discuss their properties. In the following section 2 we introduce the observations of the Perseus Arm to which our models are compared. Section 3 deals with the model parameters and discusses geometry and sight line effects taken into account. The models are compared with the data in the next section. The last section summarizes the results.
## 2 Data of the Perseus Arm
We are using the data taken from the Wisconsin H$`\alpha `$ Mapper (WHAM) (e.g. Haffner99, Haffner et al. haffner01 (2003)) survey which were kindly provided by Ron Reynolds and Matt Haffner. The WHAMโsurvey mapped the northern sky in H$`\alpha `$ with declinations of $`\delta >30^{}`$. The Perseus Arm ($`35^{}<\delta <11^{}`$ and $`120^{}<l<150^{}`$) was additionally mapped in \[NII\]$`\lambda `$6583 and \[SII\]$`\lambda `$6716. At each pointing an averaged spectrum with a beam of $`1^{}`$ is measured with a velocity resolution of 12 km $`\mathrm{s}^1`$. The emission of the Perseus Arm can be separated in velocity space from the local emission (Haffner99) for galactic longitudes $`120^{}<\mathrm{l}<\mathrm{\hspace{0.17em}150}^{}`$. This was performed by integrating the line emission in the velocity range $`100`$ km $`\mathrm{s}^1<v<20`$ km $`\mathrm{s}^1`$, no line fitting was performed. We are using the intensity of the H$`\alpha `$ line as well as the line ratios \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$. The sensitivity limit of 0.1 Rayleighs<sup>1</sup><sup>1</sup>11 R = $`10^6/4\pi `$ photons $`\mathrm{cm}^2`$ $`\mathrm{s}^1`$ $`\mathrm{sr}^1`$ results in an observed vertical height of up to $`|z|`$ = 2 kpc assuming a distance to the arm of 2.5 kpc.
## 3 Model parameters
We use the spectral simulation code CLOUDY, version 96.00 (described by Ferland ferland3 (2002), ferland1 (2000), ferland2 (1998)) to model the DIG. CLOUDY determines the physical conditions by balancing the heating and cooling rates, so that the energy is conserved. The results of the models are compared to the observed emission line ratios \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$, and the gas temperature as derived from \[NII\]/H$`\alpha `$.
In order to realize a model describing the DIG certain parameters have to be specified: The ionizing spectrum of the source is a composition of three different stellar temperature: 56% from T = 35000 K, 12% from T = 40000 K, and 32% from T = 45000 K, as used in Mathis mathis (2000) and Wood et al. wood (2004). The WMbasic models (Pauldrach et al. pauldrach (2001)), which include NโLTE effects, Xโray emission from shocks within stellar winds, are used as the ionizing spectra. The luminosity of the source is chosen in such a way that the observed run of the H$`\alpha `$ intensity is matched, as shown in Figure 1.
The density structure is exponential, as derived from the observed H$`\alpha `$ intensity, with a scale height of 1 kpc and a midplane density of 0.2 $`\mathrm{cm}^3`$. The density is in clumps with a filling factor of 20%, i.e. only 1/5 of the volume is filled with this plasma. The geometry of the ionized gas is chosen to be a plane parallel layer, i.e. the ratio of the depth of the cloud to the distance of the illuminated face to the ionizing source is smaller than 1/10.
We are matching the observed H$`\alpha `$ intensity of the observations (Figure 1) by placing the illuminated face of the cloud at a zโheight of 1 scale height of the Lockmanโlayer (300pc) assuming a density law of $`n=0.1\mathrm{exp}(z/0.3)`$ as in Miller & Cox miller2 (1993). This is done after the model is calculated as otherwise the condition of plane parallel illumination of the cloud cannot be fulfilled. The information of the actual position of the ionizing stars are effectively removed, consistent with the picture of having the DIG being ionized by radiation leaking out of the Lockmanโlayer. The intensity gradient of H$`\alpha `$ is matched as well as the estimate of the hydrogen ionizing photon flux by Reynolds reynolds90 (1990)): $`\varphi _{DIG}5\times 10^6`$ hydrogen ionizing photons $`\mathrm{cm}^2\mathrm{s}^1`$. In order to compare the models on a common basis, which is important as extra heating is dominating for large zโheights, we choose to let the ionization structure be the same for all models (see Figure 5). This is in accordance with the idea that the extra heating affects only the temperature of the gas. The forbidden levels of nitrogen and sulphur can then be more easily excited by collisions with the electrons which in turn elevates the line ratios \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$. Figure 1 shows that the intensity gradient of H$`\alpha `$ varies only a little for the different models. The scale heights for the models are slightly different ($``$ 10%) for the models with an extra heating, which can be explained by assuming that the DIGโlayer is in pressure equilibrium. This effect was also noted by Wood & Mathis wood3 (2004). The ionization parameters (U) of all models lie in the narrow range between log(U) = -3.0 and -3.1.
The ISM composition of CLOUDY is used with N/H and S/H set to the values in Haffner99 (N/H = $`7.5\times 10^5`$, S/H = $`1.86\times 10^5`$). Graphite and silicate grains with the size distribution used for the ISM (see HAZY, Ferland ferland3 (2002)) are present in the gas and account for less than 10 % of the global heating. The inclusion of PAHs give only variations $`<`$ 3% for the line ratios and have only a small effect on the heating balance. The interaction with cosmic rays is taken into account as described in Ferland & Mushotsky ferland84 (1984).
A crucial factor is the consideration of the line of sight. The observations give information about the line ratios at different positions orthogonal to the source of the ionizing radiation. The correct local line ratios are represented by the ratios of the volume emission coefficients $`ฯต_V`$: $`\frac{I_1}{I_2}=\frac{{\scriptscriptstyle dzฯต_{V_1}}}{{\scriptscriptstyle dzฯต_{V_2}}}\frac{ฯต_{V_1}}{ฯต_{V_2}}`$. An important issue is the treatment of the line of sight to the Perseus Arm due to our position in the Milky Way as shown in Fig. 2. Each pointing of the WHAM survey contains contributions from different $`|z|`$ heights above the midplane ($`ฯต(z1)`$ to $`ฯต(z2)`$), this effect is taken into account by integrating over the particular sight line in the models. The distance to the Perseus Arm is assumed to be 2.5 kpc and the thickness of the arm to be 1 kpc as quoted in Haffner99. Moreover, the observations are the average of a beam size of $`1^{}`$ which means that we have to take into account an additional integration over different z heights, on average 45 pc. These effects of geometry and observational smearing are taken into consideration here for the first time. Fig 3 shows the effect of the beam smearing and line of sight geometry on the calculated line ratios for zโheights up to 2.5 kpc, which have to be considered for the observed heights of 1.8 kpc. These effect increase the โcorrectedโ line ratios by at most 15% and their trend is altered for high zโheights. The โcorrectedโ model without an extra heating seems to suggest that \[NII\]/H$`\alpha `$ is not increasing after about 1.6 kpc, but the โuncorrectedโ model shows a further increase due to the hardening of the radiation. The result of the gas temperature (section 3.2, Fig. 4) is even more influenced by this issue. The discussion of the line ratios therefore need to include an appropriate treatment of the line of sight geometry and the beam smearing to take these effects into account.
### 3.1 Extra heating
Additional heating to photoionization is included which is proportional to $`n^1`$ and $`n^0`$ in accordance with Reynolds et al. reynolds99 (1999), their factors $`G_1`$ and $`G_2`$ respectively. We choose rates in the same range as in their paper: $`G_1=1\times 10^{25}\mathrm{erg}\mathrm{cm}^3\mathrm{s}^1`$ and $`G_2=5\times 10^{27}\mathrm{erg}\mathrm{cm}^3\mathrm{s}^1`$. The heatingโcooling balance can then be written as either $`G_0+G_1/n_e=\mathrm{\Lambda }`$ or $`G_0+G_2/n_e^2=\mathrm{\Lambda }`$. The heating due to photoionization is given by $`G_0n_e^2`$ and the cooling by $`\mathrm{\Lambda }n_e^2`$. The inclusion of an extra heating source rises the gas temperature of the models, at the same time the ionization structure varies only slightly. The temperature of the CLOUDY models are calculated by heatingโcooling balance. The extra heating therefore increases the temperature, having more pronounced effects at larger z heights as the photoionization heating rate decreases like $`\mathrm{n}^2`$ as shown in Fig. 4. The graph is explained in more detail in the next section.
In our models the ionization structure is nearly unaffected by the inclusion of an extra heating source, sulphur is slightly more effected than nitrogen. This behavior is the basic assumption for the extra heating in Reynolds et al. reynolds99 (1999). However they are assuming a constant ratio of $`\mathrm{N}^+/\mathrm{N}`$, whereas the models show a dependence on z height, this is expected as the radiation gets progressively absorbed. Figure 5 shows ionization structure and the change dependent on the different heating rates. Hydrogen is nearly fully ionized throughout all models which is a basic characteristic of the DIG.
### 3.2 Comparison with Observations
In Fig. 6 and Fig. 7 the models differ by the type of extra heating, ranging between models without extra heating and a rate of $`G_1=1\times 10^{25}\mathrm{erg}\mathrm{cm}^3\mathrm{s}^1`$ and $`G_2=5\times 10^{27}\mathrm{erg}\mathrm{cm}^3\mathrm{s}^1`$. There was no fitting done in order to match the models with the observations. No โbestโfitโ exists and therefore the models are independent of the observational quality, individual spectral features, or small scale variations which cannot be reproduced with a smooth density distribution. As the data for small $`|z|`$ heights are contaminated with radiation from the midplane and dust absorption it is convenient to consider line ratios for $`|z|`$โheights above 0.8 kpc to be โpureโ DIG. This is also the range for which the H$`\alpha `$ scale height was determined. The models show lower values of \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$ for the lower z heights as doubly ionized nitrogen contributes 35% and doubly ionized sulphur even 80%, as seen in Fig. 5. As a consequence \[NII\] and \[SII\] are weaker. Figure 6 shows the development of the line ratios with $`|z|`$ height, both the data and the models show an increase with $`|z|`$. The trend in the line ratios is matched by the models even without an extra heating source.
The \[NII\]/H$`\alpha `$ line ratio above 1 kpc can be explained with the models including an extra heating rate. The modeled \[SII\]/H$`\alpha `$ ratio is up to a factor of two below the observations, however the theoretical uncertainty for the \[SII\] line is very high as dielectronic recombination is an important process in the DIG. As the corresponding recombination coefficients are not known (see discussion in Ferland et al. ferland98 (1998)), the results of the models have to be handled with care. Our models use the KLUDGE approximation (Ferland ferland3 (2002)). Models without dielectronic recombination have \[SII\]/H$`\alpha `$ decreased by 50%. If the rate is doubled then \[SII\]/H$`\alpha `$ is increased by 50%.
Figure 4 shows the gas temperature of the models, the increase of the line ratios with z height is due to the progressive hardening of the radiation as the photons go through the gas layer. In order to explain the observed line ratios an extra heating source is however needed which does not alter the general shape of the predicted line ratios but elevates the line ratios. As \[NII\] and \[SII\] are forbidden lines which get collisional excited by electrons, an increase in gas temperature increases the amount of electrons capable to excite the singly ionized nitrogen and sulphur ions which can then decay by emitting the emission lines in questions. The gas temperature is deducted from the observations through the relation - following Reynolds et al. reynolds99 (1999): $`\frac{I_{[NII]}}{I_{\mathrm{H}\alpha }}=1.84\times 10^5\left(\frac{\mathrm{N}^+}{N}\right)\left(\frac{\mathrm{H}^+}{H}\right)^1T_4^{0.39}\mathrm{exp}(2.18/T_4)`$. Assuming that $`N^+/N=H^+/H`$ and the abundance as stated in section 3 gives $`\frac{I_{[NII]}}{I_{\mathrm{H}\alpha }}=13.75T_4^{0.39}\mathrm{exp}(2.18/T_4)`$, $`T_4=T/10^4\mathrm{K}`$. We are not using the collision strength of singly ionized nitrogen from Reynolds et al. reynolds99 (1999): $`\mathrm{N}^+:\mathrm{\Omega }\left({}_{}{}^{3}P,^1D\right)=2.28T_4^{0.026}`$, Aller aller (1984), but the data from Stafford et al. stafford (1994): $`\mathrm{N}^+:\mathrm{\Omega }\left({}_{}{}^{3}P,^1D\right)=3.02T_4^{0.01}`$. This leads to the different coefficients and temperatures on average 250 K below the values of Reynolds et al. reynolds99 (1999). For singly ionized sulfur we use the data from Lanzafame et al. lanzafame93 (1993) instead of Aller aller (1984) as in Reynolds et al. reynolds99 (1999). The plot also shows the impact of the line of sight geometry and beam smearing as well as the consideration of the ionization structure of the models. As the line of sight leads to higher \[NII\]/H$`\alpha `$ ratios (up to 15%,) this effects, when accounted for, lower temperatures. The ionization structure, i.e. $`N^+/NH^+/H`$ is only valid for z $`>`$ 800 pc, leads to higher estimates of the temperature. The combined effect is the elevation of the derived temperature for z $`<`$ 800 pc and above that lower values by about 200 K. The models without an extra heating source have temperatures too low, an extra heating source independent of density shows very good agreement with the observed temperatures. The extra heating $`n^1`$ seems to best fit the data up to 800 pc, then the extra heating $`n^0`$ gives the best agreement for higher z values. The interpretation for the z heights $`<`$ 800 pc have to be handled with care as this region is contaminated by radiation from the midplane and therefore the part responsible for the DIG emission is difficult to estimate. Magnetic reconnection (e.g. Birk et al. birk (1998) ) or heating by cosmic rays through linear Landauโdamping (Lerche & Schlickeiser lerche (2001) ) are possible processes producing a heating independent of density. As photoelectric heating from dust grains is $`n^1`$, the data suggest that there is more dust present at zโheights below 800 pc than present in the models if this mechanism is responsible for the elevated temperature.
In Fig. 7 the line ratios \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$ are plotted against each other. Values of \[NII\]/H$`\alpha `$ greater than 1, which cannot be explained by classical HIIโregion calculations, are reached with models using an extra heating source. Also in this case the extra heating independent of density is able to produce higher \[NII\]/H$`\alpha `$ ratios than the other therefore matching better with the data. The limits of Haffner99 for two constant $`\mathrm{S}^+/\mathrm{S}`$ ratios (0.5 and 0.25) are also given. The ionization fractions of the models (see Fig. 5) are within these two limits for z $`>`$ 0.6 kpc. Together with the temperature plot (Fig. 4) the models match in all these direct and derived quantities with the data and agrees with the estimates of Haffner99 and Reynolds et al. reynolds99 (1999). The application to other galaxies shows that this diagram is also a valuable diagnostic for the chemical evolution (Elwert et al. elwert (2004)).
## 4 Summary
We have shown that the observed trend of the line ratios \[NII\]/H$`\alpha `$ and \[SII\]/H$`\alpha `$ above the Galactic plane can successfully be explained by photoionization models including extra heating and considering the line of sight geometry. The observed values need an extra heating source which strength lies at the lower end of the predicted values of Reynolds et al. reynolds99 (1999) due to our models. At high z heights (z $`>`$ 800 pc) an extra heating independent of density gives the best agreement with the data, whereas for smaller z heights an extra heating term $`n^1`$ gives better results concerning the temperature. There is an intrinsically increase in the line rations and the gas temperature due to the progressive hardening of the radiation. The extra heating terms are enhancing this trend and elevate the line ratios to the observed values. It is important to incorporate the observing geometry to the Perseus Arm into the models when comparing with the data. A discussion concerning models of observed line ratios in edgeโon galaxies is given in Elwert elwert03 (2003) and Elwert & Dettmar elwert05 (2005).
This work was supported by DFG through SFB 591 and through Deutsches Zentrum fรผr Luft- und Raumfahrt through grant 50 OR 9707. TE wants to thank Kenneth Wood and Ron Reynolds for helpful comments and enlightening discussions while writing the paper. We also want to thank the anonymous referee for making many very useful suggestions and comments which helped to improve the publication.
|
warning/0506/cond-mat0506040.html
|
ar5iv
|
text
|
# Quantum phase-space simulations of fermions and bosons
## Abstract
We introduce a unified Gaussian quantum operator representation for fermions and bosons. The representation extends existing phase-space methods to Fermi systems as well as the important case of Fermi-Bose mixtures. It enables simulations of the dynamics and thermal equilibrium states of many-body quantum systems from first principles. As an example, we numerically calculate finite-temperature correlation functions for the Fermi Hubbard model, with no evidence of the Fermi sign problem.
Calculating the quantum many-body physics of interacting Fermi systems is one of the great challenges in modern theoretical physics. In even the simplest cases, first-principles calculations are made difficult by the complexity of the fermionic wave-function, manifest notoriously in the Fermi sign problem. In previous quantum Monte Carlo (QMC) techniques, the sign problem appears as trajectories with negative weights, which contribute to a large sampling errorCeperley99 , together with large, computationally intensive determinants.
Fermion complexity issues appear in physical problems at all energy scales, from high-energy lattice QCD to the emerging area of ultra-cold atomic physics. Recent pioneering experiments in ultra-cold Fermi gases are capable of investigating fermion many-body physics in regimes of unprecedented experimental simplicity. This situation implies a substantial opportunity to develop and test novel first-principles theoretical methods for the investigation of correlations and dynamical effects.
Here we present a phase-space method for simulating many-body bosonGauss:Bosons ; PRL:Bosons and fermionGauss:Fermi-Bose ; Gauss:Fermions systems, based on a Gaussian operator expansion.
The method allows the treatment of dynamical and static problems at finite temperature. The expansion in the fermionic case represents *pairs* of Fermi operators. Since the pairs obey commutation relations, there are no anti-commutators causing sign problems, and no large determinant calculations as in some previous approaches.
The method is illustrated using the finite temperature Hubbard model, which is a well-known theory in condensed matter physics and high $`T_c`$ superconductors. The cases chosen have an acute sign problem using conventional QMC. The results are directly applicable to feasible experiments on ultra-cold fermions in an optical latticeFermilattice .
Like path-integral QMC, phase-space methods sample the many-body quantum density operator $`\widehat{\rho }`$. But rather than expressing the density operator in a position representation, one expands it in terms of an overcomplete basis of operators:
$$\widehat{\rho }(t)=P(\stackrel{}{\lambda },t)\widehat{\mathrm{\Lambda }}(\stackrel{}{\lambda })๐\stackrel{}{\lambda },$$
(1)
where $`P(\stackrel{}{\lambda },t)`$ is a positive probability distribution, $`\widehat{\mathrm{\Lambda }}`$ is an overcomplete operator basis for the class of density matrices being considered, and $`d\stackrel{}{\lambda }`$ is the integration measure for the generalized phase-space coordinate $`\stackrel{}{\lambda }`$. It is the overcompleteness of the basis which allows a positive representation of any physical density matrix in terms of Gaussian operators.
We define the operator basis $`\widehat{\mathrm{\Lambda }}\mathrm{\Omega }\widehat{\mathrm{\Lambda }}_+\widehat{\mathrm{\Lambda }}_{}`$ to be the product of Gaussian forms of bosonic ( $`\widehat{\mathrm{\Lambda }}_+`$) and fermionic ( $`\widehat{\mathrm{\Lambda }}_{}`$) creation and annihilation operators, over $`M_\pm `$ modes respectively. Here $`\mathrm{\Omega }`$ is an additional weighting factor. If $`\widehat{๐}`$ is a row vector of $`M_\pm `$ annihilation operators, and $`\widehat{๐}^{}`$ the corresponding column vector of creation operators, their commutation relations are: $`[\widehat{a}_k,\widehat{a}_j^{}]_{}=\delta _{kj}.`$ We use $``$ to indicate bosons (upper sign) or fermions (lower sign). For brevity, we restrict the present discussion to number-conserving systems. The most general Gaussian operator is then a generalized thermal density operator with a complex covariance:
$`\widehat{\mathrm{\Lambda }}_\pm (๐ง)`$ $`=`$ $`\left|๐\pm ๐ง\right|^1:\mathrm{exp}\left[\widehat{๐}\left(\left\{2๐\right\}\left[๐\pm ๐ง\right]^1\right)\widehat{๐}^{}\right]:`$
where the additional $`\left\{๐\right\}`$ in the exponent is bracketed to indicated that it only appears in the fermionic case. Normal ordering is defined as usual for Bose and Fermi systems, for example, $`:\widehat{a}_j\widehat{a}_i^{}:=\pm \widehat{a}_i^{}\widehat{a}_j=\pm \widehat{n}_{ij}`$. The $`M_\pm \times M_\pm `$ matrix $`๐ง`$ corresponds to a generalized thermal covariance.
Using these Gaussian operators in the density operator expansion of Eq. (1), one finds that operator expectation values become weighted moments of the distribution $`P`$, denoted as $`.._P`$. Thus the first- and second-order number correlations are
$`\widehat{๐ง}={\displaystyle \mathrm{\Omega }๐งP(\stackrel{}{\lambda },t)๐\stackrel{}{\lambda }}=๐ง_P,`$
$`\widehat{a}_i^{}\widehat{a}_j^{}\widehat{a}_j\widehat{a}_i=n_{ii}n_{jj}_P\pm n_{ij}n_{ji}_P.`$ (3)
As an illustration of the use of the unified representation, consider the canonical distribution of a Bose or Fermi field. The thermal state at temperature $`T=1/k_B\tau `$ can be cast into an imaginary time integro-differential equation:
$$\frac{d\widehat{\rho }}{d\tau }=\frac{1}{2}P(\stackrel{}{\lambda },t)[\widehat{H},\widehat{\mathrm{\Lambda }}(\stackrel{}{\lambda })]_+๐\stackrel{}{\lambda }.$$
(4)
To solve this, we first use identities derived in Gauss:Bosons ; Gauss:Fermions that describe the action of operators on the density operator as derivatives on elements of the Gaussian basis. After integrating Eq (4) by parts, we arrive at the following mappings:
$`\widehat{๐ง}\widehat{\rho }`$ $``$ $`๐งP\left\{{\displaystyle \frac{}{๐ง}}\left(๐\pm ๐ง\right)^T\right\}^T๐งP`$
$`\widehat{\rho }\widehat{๐ง}`$ $``$ $`๐งP\left\{{\displaystyle \frac{}{๐ง}}๐ง^T\right\}^T\left(๐\pm ๐ง\right)P.`$ (5)
The matrix derivative is here defined as $`(/๐ง)_{ij}=/n_{ij}`$. In the free-field case of $`\widehat{H}=_k\omega _k\widehat{n}_k`$, we arrive at a first-order Fokker-Planck equation for the distribution function $`P`$. This leads to deterministic equations for the mode occupations of form $`n_k/\tau =\omega _kn_k\left(1\pm n_k\right),`$ which can be integrated to get the well-known Bose-Einstein (Fermi-Dirac) distribution:
$$n_k=\frac{1}{\mathrm{exp}(\omega _k\tau )1}.$$
(6)
For systems of *i*nteracting particles, the unified representation gives nonlinear, stochastic phase-space equations, which must be solved numerically. Due to the non-uniqueness of the expansion, a careful choice of identities is mandatoryGauge to keep the nonlinear equations stable. We term this choice a stochastic gauge in analogy to the gauge fields in QED, since it results in freely chosen fields in the resulting stochastic equations. The choice in the fermionic case is especially large, since the fermionic anti-commutation relations result in a non-unique algebraic form of the Hamiltonian.
As an example, consider the Hubbard model, which is the simplest nontrivial model for strongly interacting electrons. It is an important system in condensed matter physics, with relevance to the theory of high-temperature superconductorsLinden92 . The Hamiltonian is:
$`H(\widehat{๐ง}_1,\widehat{๐ง}_1)`$ $`=`$ $`{\displaystyle \underset{ij,\sigma }{}}\left(t_{ij}+\mu \delta _{ij}\right)\widehat{n}_{ij,\sigma }`$ (7)
$`|U|{\displaystyle \underset{j}{}}:(\widehat{n}_{jj,1}s\widehat{n}_{jj,1})^2:/2`$
where $`\widehat{n}_{ij,\sigma }=\widehat{a}_{i,\sigma }^{}\widehat{a}_{j,\sigma }=\left\{\widehat{๐ง}_\sigma \right\}_{ij}`$. The coupling $`t_{ij}=t`$ if the $`i,j`$ correspond to nearest neighbour sites and is otherwise $`0`$. The index $`\sigma `$ denotes spin ($`\pm 1`$) and the indices $`i,j`$ label lattice location, and $`s=U/|U|=\pm 1`$. Traditional QMC methods for this problem have large sign problems in the repulsive case ($`s=1`$)Santos03 ; FettesMorgenstern00 .
Because of the way we have written the interaction term in Eq. (6) (which constitutes a kind of fermionic gauge choice), the Hubbard model maps to a set of real Stratonovich stochastic equations:
$`{\displaystyle \frac{d๐ง_\sigma }{d\tau }}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left\{\left(๐๐ง_\sigma \right)๐ป_\sigma ^{(1)}๐ง_\sigma +๐ง_\sigma ๐ป_\sigma ^{(2)}\left(๐๐ง_\sigma \right)\right\}.`$
Here we have introduced the stochastic propagation matrix:
$`T_{ij,\sigma }^{(r)}`$ $`=`$ $`t_{ij}+\delta _{ij}\left\{\mu \sigma ^{(s+1)/2}\xi _j^{(r)}\right\}`$ (9)
$`|U|\delta _{ij}\left\{sn_{jj\sigma }n_{jj\sigma }+{\displaystyle \frac{1}{2}}\right\}.`$
The real Gaussian noise $`\xi _j^{(r)}(\tau )`$ is defined by the correlations
$$\xi _j^{(r)}(\tau )\xi _j^{}^{(r^{})}(\tau ^{})=2|U|\delta (\tau \tau ^{})\delta _{jj^{}}\delta _{rr^{}}.$$
(10)
The weights for each trajectory evolve as physically expected for energy-weighted averages, with $`d\mathrm{\Omega }/d\tau =\mathrm{\Omega }H(๐ง_1,๐ง_1)`$. Because the equations for the phase-space variables $`n_{ij,\sigma }`$ are all real, the weights of all trajectories will remain positive. Thus the traditional manifestation of the sign problem is avoided, as there is no โdeterioration of the signโ from averaging over positive and negative weights. Furthermore, the mapping to real phase-space produces a stable set of equations, and thus there is no need to invoke additional gauge choices.
There is, however, the issue of spreading weights, which becomes serious for large lattice sizes and long simulations times. Physical quantities are weighted averages:
$$A(\tau )_P=\underset{p=1}{\overset{N_p}{}}\mathrm{\Omega }^{(p)}(\tau )A^{(p)}(\tau )/\underset{p=1}{\overset{N_p}{}}\mathrm{\Omega }^{(p)}(\tau ),$$
(11)
where $`N_p`$ is the total number of paths in the sample. A large spread in the weights makes a straightforward average very inefficient, as most paths in the ensemble may end up contributing very little to the final result. To increase efficiency, we instead use a simple branching algorithm adapted from Greenโs function Monte Carlo methodsTrivediCeperley90 , in which low-weights paths are deleted and high-weight paths are cloned, according to the rate:
$`m^{(p)}`$ $`=`$ $`\mathrm{Integer}\left[\xi +\mathrm{\Omega }^{(p)}/\overline{\mathrm{\Omega }}\right],`$ (12)
where $`\xi `$ is a random variable uniformly distributed on $`[0,1]`$ and where $`\overline{\mathrm{\Omega }}`$ is a reference weight, which is adapted to keep the number of paths $`N_p`$ under control. At branching, the weights are equalised and thereafter the clones evolve independently with spreading weights. To avoid biasing, the branching must occur sufficiently often to limit the diversity of weights at the branching times. For the results presented here, the branching algorithm is sufficient to control sampling error - other situations may require the use of more sophisticated importance sampling methodsLinden92 .
The stochastic phase-space equations are simulated by a robust semi-implicit algorithmDrummondMortimer91 , with an adaptive stepsize to overcome stiffness. Unlike Projector QMC methods, the Gaussian phase-space method can calculate any correlation function, at any temperature. Unlike Path Integral QMC, a single run generates results for a range of temperatures: longer simulation times correspond to lower temperatures. Strictly speaking, zero-temperature results are obtained only in the limit of long simulation times. In practice, however, one only has to run the simulation until the relevant correlation functions have plateaued.
Precision is of course limited by sampling error, but this can be reduced by several means. For example, one can a) include more trajectories in the sample, b) employ a more sophisticated branching/importance sampling technique to reduce the spread in weights, and c) make a better โstochastic gaugeโ choice to obtain phase-space equations with smaller sampling error for the correlations to be calculated.
Typical results for a $`16\times 16`$ lattice are shown in Figs (1) and (2), which plot the energy $`E`$ and second-order correlation function $`g_2`$, respectively, for different chemical potentials. The estimation of sampling error shown in the figures assumes independent samples, for simplicity. While this is liable to underestimate the error, especially for $`g_2`$ where there is also spatial averaging, it does indicate the approximate dependence on temperature. In particular, the sampling error remains well-controlled throughout the simulation, even away from half filling, where there is known to be a sign problem. A more detailed sampling error analysis will be given elsewhere.
In conclusion, we have presented a unified operator representation that is able to represent arbitrary physical states of bosons and fermions. By use of this representation, non-interacting systems can be mapped to deterministic phase-space equations, whereas systems with two-body interactions can be simulated by use of stochastic sampling methods, provided a suitable gauge is chosen to eliminate any boundary terms. For the example of the Hubbard model, we show how the thermal equilibrium problem can be mapped to a set of real, stable phase-space equations with positive weights. Bosonic problems can be solved in similar manner, and the method can also be used to simulate dynamics (although typically with an increased sampling error). Thus the one, unified method can solve both fermionic and bosonic problems, which makes it well suited to simulating Bose-Fermi mixtures, and to studying the BEC/BCS crossover.
|
warning/0506/hep-ph0506021.html
|
ar5iv
|
text
|
# Full ๐โข(๐ถ) corrections to ๐โบโข๐โปโ๐ฬ_๐โข(๐ฬ)ฬ_๐
## I Introduction
The Minimal Supersymmetric Standard Model (MSSM) provides the most attractive extension of the Standard Model (SM). Among other particles it includes supersymmetric partners of the fermions. These scalar states $`\stackrel{~}{f}_L`$, $`\stackrel{~}{f}_R`$ (sfermions) correspond to the two chirality states of each fermion $`f`$. The mass eigenstates $`\stackrel{~}{f}_1`$ and $`\stackrel{~}{f}_2`$ though are not identical with $`\stackrel{~}{f}_L`$, $`\stackrel{~}{f}_R`$ and are rather a linear combination of them. The mixing terms are proportional to the mass of the corresponding fermion. Hence the sfermions of the third generation play a special r$`\widehat{\mathrm{o}}`$le. As a consequence, one eigenstate ($`\stackrel{~}{f}_1`$) can be much lighter than the other one.
The sfermions, especially the strongly interacting ones ($`\stackrel{~}{t}_i,\stackrel{~}{b}_i`$), are likely to be detected at the LHC or the Tevatron. Nevertheless, to extract the fundamental parameters one must have a significant accuracy only obtainable at a linear collider. From sfermion pair production in $`e^+e^{}`$ collisions the sfermion mixing angle can be extracted. This is one of the reasons why it has been extensively studied phenomenologically exp . To match the expected precision of the linear collider, theory predictions must reach a similar accuracy. The effort to calculate higher order corrections to the sfermion production has begun by calculating the leading QCD, SUSY-QCD and Yukawa corrections QCD1 ; SUSY-QCD-A ; SUSY-QCD-H ; Yukawa . It was further shown that taking only the leading terms of the one-loop corrections is not sufficient and so also the full weak corrections were presented in letter ; hollik .
It is the aim of this paper to extend the existing weak corrections by including the full $`๐ช(\alpha )`$ contributions in a similar manner as in the case of the selectrons and the smuons in freitas . In addition, we present the full analytical results and all the details of the calculation for both the weak corrections letter and the QED contributions. Moreover, we generalize the results to include also the effects of polarization of the electron and positron beams. Apart from cross-sections, we calculate other observables such as the forward-backward and the left-right asymmetries as well.
Although we present the results in the form of cross-sections and asymmetries, we are well aware of the fact that the precise predictions have to be used for parameter extraction. As the definition of the parameters is no longer unique beyond the tree-level, there has been a recent proposal by the so-called SPA project (SUSY parameter analysis) which defines these parameters SPA . The SPA project also gives a firm base for calculating all sorts of observables (masses, decay widths, cross-sections etc.) and enables the development of tools for extracting the parameters.
The fundamental SUSY parameters in the SPA project are defined using the $`\overline{\mathrm{DR}}`$ (dimensional reduction) renormalization scheme at the scale $`Q=1\mathrm{T}\mathrm{e}\mathrm{V}`$. Specifying the renormalization scheme serves only to define the parameters uniquely and does not restrict the use of other schemes in different calculations. In this paper, we use an on-shell renormalization scheme. To use the parameters from the SPA project we have to translate them into the on-shell renormalization scheme. The results for any observable using different schemes (with correctly translated input parameters) must agree up to contributions of higher order.
The paper is organized as follows. In section II we give the formulae for the tree-level cross-section for polarized electron and positron beams. The calculation of the virtual corrections with a detailed discussion of the applied on-shell renormalization scheme are outlined in section III. All explicit analytic formulas needed for the calculation are given in the Appendices A, B, C and D. In section IV we work out the real radiative corrections where we include the Bremsstrahlung process $`\sigma (e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j\gamma )`$. In section V we present the numerical analysis with some results of the corrections. Section VI summarizes our conclusions.
## II Tree level
The sfermion mixing is described by the diagonalization of the sfermion mass matrix given in the left-right basis $`(\stackrel{~}{f}_L,\stackrel{~}{f}_R)`$ into the mass basis $`(\stackrel{~}{f}_1,\stackrel{~}{f}_2)`$, $`f=t,b`$ or $`\tau `$ GunionHaber ,
$`_{\stackrel{~}{f}}^{\mathrm{\hspace{0.17em}2}}`$ $`=`$ $`\left(\begin{array}{cc}m_{\stackrel{~}{f}_L}^{\mathrm{\hspace{0.17em}2}}& a_fm_f\\ a_fm_f& m_{\stackrel{~}{f}_R}^{\mathrm{\hspace{0.17em}2}}\end{array}\right)=\left(R^{\stackrel{~}{f}}\right)^{}\left(\begin{array}{cc}m_{\stackrel{~}{f}_1}^{\mathrm{\hspace{0.17em}2}}& 0\\ 0& m_{\stackrel{~}{f}_2}^{\mathrm{\hspace{0.17em}2}}\end{array}\right)R^{\stackrel{~}{f}},`$ (5)
where $`R_{i\alpha }^{\stackrel{~}{f}}`$ is a 2 x 2 rotation matrix with rotation angle $`\theta _{\stackrel{~}{f}}`$, which relates the mass eigenstates $`\stackrel{~}{f}_i`$, $`i=1,2`$, $`(m_{\stackrel{~}{f}_1}<m_{\stackrel{~}{f}_2})`$ to the weak eigenstates $`\stackrel{~}{f}_\alpha `$, $`\alpha =L,R`$, by $`\stackrel{~}{f}_i=R_{i\alpha }^{\stackrel{~}{f}}\stackrel{~}{f}_\alpha `$, with $`R_{11}^{\stackrel{~}{f}}=R_{22}^{\stackrel{~}{f}}=\mathrm{cos}\theta _{\stackrel{~}{f}}`$ and $`R_{12}^{\stackrel{~}{f}}=R_{21}^{\stackrel{~}{f}}=\mathrm{sin}\theta _{\stackrel{~}{f}}`$, and
$`m_{\stackrel{~}{f}_L}^{\mathrm{\hspace{0.17em}2}}`$ $`=`$ $`M_{\{\stackrel{~}{Q},\stackrel{~}{L}\}}^2+(I_f^{3L}e_fs_W^2)\mathrm{cos}2\beta m_Z^{\mathrm{\hspace{0.17em}2}}+m_f^2,`$ (6)
$`m_{\stackrel{~}{f}_R}^{\mathrm{\hspace{0.17em}2}}`$ $`=`$ $`M_{\{\stackrel{~}{U},\stackrel{~}{D},\stackrel{~}{E}\}}^2+e_fs_W^2\mathrm{cos}2\beta m_Z^{\mathrm{\hspace{0.17em}2}}+m_f^2,`$ (7)
$`a_f`$ $`=`$ $`A_f\mu (\mathrm{tan}\beta )^{2I_f^{3L}}.`$ (8)
$`M_{\stackrel{~}{Q}}`$, $`M_{\stackrel{~}{L}}`$, $`M_{\stackrel{~}{U}}`$, $`M_{\stackrel{~}{D}}`$ and $`M_{\stackrel{~}{E}}`$ are soft SUSY breaking masses, $`A_f`$ is the trilinear scalar coupling parameter, $`\mu `$ the higgsino mass parameter, $`\mathrm{tan}\beta =\frac{v_2}{v_1}`$ is the ratio of the vacuum expectation values of the two neutral Higgs doublet states , $`I_f^{3L}`$ denotes the third component of the weak isospin of the fermion $`f`$, $`e_f`$ the electric charge in terms of the elementary charge $`e`$, and $`s_W`$ is the sine of the Weinberg angle $`\theta _W`$.
The mass eigenvalues and the mixing angle are
$`m_{\stackrel{~}{f}_{1,2}}^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(m_{\stackrel{~}{f}_L}^2+m_{\stackrel{~}{f}_R}^2\sqrt{(m_{\stackrel{~}{f}_L}^2m_{\stackrel{~}{f}_R}^2)^2+4a_f^2m_f^2}\right),`$ (9)
$`\mathrm{cos}\theta _{\stackrel{~}{f}}`$ $`=`$ $`{\displaystyle \frac{a_fm_f}{\sqrt{(m_{\stackrel{~}{f}_L}^2m_{\stackrel{~}{f}_1}^2)^2+a_f^2m_f^2}}}(0\theta _{\stackrel{~}{f}}<\pi ),`$ (10)
and the mass of the sneutrino $`\stackrel{~}{\nu }_\tau `$ is given by $`m_{\stackrel{~}{\nu }_\tau }^2=M_{\stackrel{~}{L}}^2+\frac{1}{2}m_Z^2\mathrm{cos}2\beta `$.
The tree-level cross-section of $`e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j`$ for polarized electron and positron beams is given by
$`\sigma ^{\mathrm{tree}}`$ $`=`$ $`{\displaystyle \frac{1}{4}}(1P_{})(1+P_+)\sigma _L^{\mathrm{tree}}+{\displaystyle \frac{1}{4}}(1+P_{})(1P_+)\sigma _R^{\mathrm{tree}},`$ (11)
where $`P_{},P_+(1,1)`$ are the degrees of polarization of the electron and positron beams (e. g. $`P_{}(P_+)=0.8`$ means 80% of electrons (positrons) left polarized and 20 % unpolarized).
As we neglect the electron mass, we have only two terms contributing (out of 4 possible) where $`\sigma _L^{\mathrm{tree}}`$ is the tree-level cross-sections for $`e_R^+e_L^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j`$ (below referred to as the left part of the polarized cross-section) and $`\sigma _R^{\mathrm{tree}}`$ stands for $`e_L^+e_R^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j`$ (analogously referred to as the right part of the cross-section). They have the form
$`\sigma _{L,R}^{\mathrm{tree}}(e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j)`$ $`=`$ $`{\displaystyle \frac{N_C}{3}}{\displaystyle \frac{\kappa ^3(s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}{4\pi s^2}}\left(T_{L,R}^{\gamma \gamma }+T_{L,R}^{\gamma Z}+T_{L,R}^{ZZ}\right),`$ (12)
where
$`T_{L,R}^{\gamma \gamma }`$ $`=`$ $`{\displaystyle \frac{e^4e_f^2(\delta _{ij})^2}{s^2}}{\displaystyle \frac{1}{2}}K_{L,R}^2,`$ (13)
$`T_{L,R}^{\gamma Z}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}}{4s(sm_Z^2)}}C_{L,R}K_{L,R},`$ (14)
$`T_{L,R}^{ZZ}`$ $`=`$ $`{\displaystyle \frac{g_Z^4(a_{ij}^{\stackrel{~}{f}})^2}{32(sm_Z^2)^2}}C_{L,R}^2,`$ (15)
and $`\kappa (x,y,z)=\sqrt{(xyz)^24yz}`$.
Here we use $`K_{L,R}`$ and $`C_{L,R}`$ as the left- and right-handed couplings of the electron to the photon and $`Z`$-boson, respectively,
$$K_L=K_R=1,C_L=\frac{1}{2}+s_W^2,C_R=s_W^2.$$
(16)
The matrix elements $`a_{ij}^{\stackrel{~}{f}}`$ come from the coupling of $`Z\stackrel{~}{f}_i\stackrel{~}{f}_j`$,
$`a_{ij}^{\stackrel{~}{f}}`$ $`=`$ $`\left(\begin{array}{cc}4(I_f^{3L}\mathrm{cos}^2\theta _{\stackrel{~}{f}}s_W^2e_f)& 2I_f^{3L}\mathrm{sin}2\theta _{\stackrel{~}{f}}\\ 2I_f^{3L}\mathrm{sin}2\theta _{\stackrel{~}{f}}& 4(I_f^{3L}\mathrm{cos}^2\theta _{\stackrel{~}{f}}s_W^2e_f)\end{array}\right).`$ (19)
Apart from the tree-level cross-section we can calculate other observables such as the left-right asymmetry and the forward-backward asymmetry. They are defined by
$`A_{LR}={\displaystyle \frac{\sigma _L\sigma _R}{\sigma _L+\sigma _R}},A_{FB}={\displaystyle \frac{\sigma _F\sigma _B}{\sigma _F+\sigma _B}},`$ (20)
with
$$\sigma _F=_0^{2\pi }๐\phi _0^{\pi /2}\left(\frac{d\sigma }{d\mathrm{\Omega }}\right)d\mathrm{cos}\vartheta ,\sigma _B=_0^{2\pi }๐\phi _{\pi /2}^\pi \left(\frac{d\sigma }{d\mathrm{\Omega }}\right)d\mathrm{cos}\vartheta .$$
(21)
There is no lowest order (tree-level) contribution to the $`A_{FB}`$-asymmetry as the angle distribution is symmetric.
## III Virtual corrections
For a precision analysis of the sfermion production one has to include also higher order corrections. The calculation of the higher order corrections is performed analytically in the $`\overline{\mathrm{DR}}`$ scheme, adopting the $`\xi =1`$ โtHooft-Feynman gauge. All necessary ingredients of the analytical calculation are given in the Appendices. Furthermore, we neglect the electron mass wherever possible $`(m_e=0)`$. For the numerical evaluation of the loop integrals we use the packages LoopTools and FF loopFF . At the end the whole analytic result was checked with the result obtained using the computer algebra tools FeynArts and FormCalc feyn .
The virtual corrections receive contributions from vertex, self-energy and box diagrams depicted generically in Fig. 1 and explicitly in Figs. 2 and 3. All these contributions are summarized in the renormalized cross-section $`\sigma ^{\mathrm{ren}}`$.
The one-loop (renormalized) cross-section $`\sigma ^{\mathrm{ren}}`$ for polarized beams is expressed analogously to Eq. (11),
$`\sigma ^{\mathrm{ren}}(e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j)`$ $`=`$ $`{\displaystyle \frac{1}{4}}(1P_{})(1+P_+)\sigma _L^{\mathrm{ren}}+{\displaystyle \frac{1}{4}}(1+P_{})(1P_+)\sigma _R^{\mathrm{ren}},`$ (22)
where the left/right renormalized cross-sections are defined as
$`\sigma _{L,R}^{\mathrm{ren}}`$ $`=`$ $`\sigma _{L,R}^{\mathrm{tree}}+\mathrm{\Delta }\sigma _{L,R}^{\mathrm{QCD}}+\mathrm{\Delta }\sigma _{L,R}^{\mathrm{EW}}`$ (23)
with the symbol $`\mathrm{\Delta }`$ denoting UV-finite quantities.
The SUSY-QCD corrections ($`\mathrm{\Delta }\sigma ^{\mathrm{QCD}}`$) have already been calculated for the unpolarized case in SUSY-QCD-A ; SUSY-QCD-H . As the gluon part of $`\mathrm{\Delta }\sigma ^{\mathrm{QCD}}`$ is proportional to the tree-level cross-section, the polarized cross-sections are easily obtained using $`\sigma _{L,R}^{tree}`$ instead of $`\sigma ^{tree}`$. The gluino part of $`\mathrm{\Delta }\sigma ^{\mathrm{QCD}}`$ is treated analogously to $`\mathrm{\Delta }\sigma ^{V\stackrel{~}{f}}`$ (see Fig. 1).
We have already presented the unpolarized results for the electro-weak corrections ($`\mathrm{\Delta }\sigma ^{\mathrm{EW}}`$) in letter . In this paper, we give the result for polarized beams and also all formulas needed for the calculation.
The electroweak corrections can be split further into four UV-finite parts given in Fig. 1,
$`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{EW}}`$ $`=`$ $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\mathrm{e}}+\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\stackrel{~}{\mathrm{f}}}+\mathrm{\Delta }\sigma _{L,R}^{\mathrm{prop}}+\mathrm{\Delta }\sigma _{L,R}^{\mathrm{box}},`$ (24)
where $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\mathrm{e}}`$ and $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\stackrel{~}{\mathrm{f}}}`$ stand for the left/right part of the renormalized electron and sfermion vertex, $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{prop}}`$ and $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{box}}`$ for the left/right part of renormalized propagators and box contribution.
The renormalized electron vertex has the form
$`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\mathrm{e}}`$ $`=`$ $`{\displaystyle \frac{N_C}{3}}{\displaystyle \frac{\kappa ^3(s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}{4\pi s^2}}\left(\left(\mathrm{\Delta }T_{\gamma \gamma }^{Ve}\right)_{L,R}+\left(\mathrm{\Delta }T_{\gamma Z}^{Ve}\right)_{L,R}+\left(\mathrm{\Delta }T_{ZZ}^{Ve}\right)_{L,R}\right),`$ (25)
where
$`\left(\mathrm{\Delta }T_{\gamma \gamma }^{Ve}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{e^4e_f^2(\delta _{ij})^2}{s^2}}(\mathrm{\Delta }e_{L,R}K_{L,R}),`$ (26)
$`\left(\mathrm{\Delta }T_{\gamma Z}^{Ve}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}}{4s(sm_Z^2)}}(\mathrm{\Delta }e_{L,R}C_{L,R}+\mathrm{\Delta }a_{L,R}K_{L,R}),`$ (27)
$`\left(\mathrm{\Delta }T_{ZZ}^{Ve}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^4(a_{ij}^{\stackrel{~}{f}})^2}{16(sm_Z^2)^2}}(\mathrm{\Delta }a_{L,R}C_{L,R}).`$ (28)
$`\mathrm{\Delta }e_{L,R}`$ and $`\mathrm{\Delta }a_{L,R}`$ consist of 3 parts,
$`\mathrm{\Delta }e_{L,R}`$ $`=`$ $`\delta e_{L,R}^{(v)}+\delta e_{L,R}^{(w)}+\delta e_{L,R}^{(c)},`$ (29)
$`\mathrm{\Delta }a_{L,R}`$ $`=`$ $`\delta a_{L,R}^{(v)}+\delta a_{L,R}^{(w)}+\delta a_{L,R}^{(c)}.`$ (30)
$`\delta e_{L,R}^{(v)}`$, $`\delta a_{L,R}^{(v)}`$ correspond to the vertex corrections in Fig. 2, $`\delta e_{L,R}^{(w)}`$, $`\delta a_{L,R}^{(w)}`$ are the wave-function corrections, and $`\delta e_{L,R}^{(c)}`$, $`\delta a_{L,R}^{(c)}`$ correspond to the counterterms.
The renormalized sfermion vertex has a similar form,
$`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{V}\stackrel{~}{\mathrm{f}}}`$ $`=`$ $`{\displaystyle \frac{N_C}{3}}{\displaystyle \frac{\kappa ^3(s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}{4\pi s^2}}\left(\left(\mathrm{\Delta }T_{\gamma \gamma }^{V\stackrel{~}{f}}\right)_{L,R}+\left(\mathrm{\Delta }T_{\gamma Z}^{V\stackrel{~}{f}}\right)_{L,R}+\left(\mathrm{\Delta }T_{ZZ}^{V\stackrel{~}{f}}\right)_{L,R}\right),`$ (31)
where
$`\left(\mathrm{\Delta }T_{\gamma \gamma }^{V\stackrel{~}{f}}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{e^4e_f(\mathrm{\Delta }e_f)_{ij}}{s^2}}K_{L,R}^2,`$ (32)
$`\left(\mathrm{\Delta }T_{\gamma Z}^{V\stackrel{~}{f}}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2}{4s(sm_Z^2)}}K_{L,R}C_{L,R}((\mathrm{\Delta }e_f)_{ij}a_{ij}^{\stackrel{~}{f}}+\delta _{ij}(\mathrm{\Delta }a_f)_{ij}),`$ (33)
$`\left(\mathrm{\Delta }T_{ZZ}^{V\stackrel{~}{f}}\right)_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^4a_{ij}^{\stackrel{~}{f}}(\mathrm{\Delta }a_f)_{ij}}{16(sm_Z^2)^2}}C_{L,R}^2.`$ (34)
$`(\mathrm{\Delta }e_f)_{ij}`$ and $`(\mathrm{\Delta }a_f)_{ij}`$ can also be split into vertex corrections (see Fig. 2), wave-function corrections and counterterms,
$`(\mathrm{\Delta }e_f)_{ij}`$ $`=`$ $`(\delta e_f)_{ij}^{(v)}+(\delta e_f)_{ij}^{(w)}+(\delta e_f)_{ij}^{(c)},`$ (35)
$`(\mathrm{\Delta }a_f)_{ij}`$ $`=`$ $`(\delta a_f)_{ij}^{(v)}+(\delta a_f)_{ij}^{(w)}+(\delta a_f)_{ij}^{(c)}.`$ (36)
The diagrams contributing to the vertex corrections are shown in Fig. 2 and the explicit form of the corrections are given in Appendix A. The wave-function corrections and the counterterms to both vertices are listed in detail in sections III.1.1 and III.1.2. The $`(\delta e_f)_{ij}`$ and $`(\delta a_f)_{ij}`$ corresponding to gluino corrections can be found in SUSY-QCD-H .
The correction $`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{prop}}`$ which comes from inserting the self-energies of the $`\gamma `$ and $`Z`$-boson, in the propagator, see Fig. 3, can be expressed as
$`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{prop}}`$ $`=`$ $`{\displaystyle \frac{N_C}{3}}{\displaystyle \frac{\kappa ^3(s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}{4\pi s^2}}\times `$
$`\times \mathrm{\hspace{0.17em}2}\mathrm{}[({\displaystyle \frac{\widehat{\mathrm{\Pi }}_{\gamma \gamma }^T(s)}{s}})T_{L,R}^{\gamma \gamma }+({\displaystyle \frac{\widehat{\mathrm{\Pi }}_{\gamma \gamma }^T(s)}{s}}{\displaystyle \frac{\widehat{\mathrm{\Pi }}_{ZZ}^T(s)}{sm_Z^2}})T_{L,R}^{\gamma Z}+({\displaystyle \frac{\widehat{\mathrm{\Pi }}_{ZZ}^T(s)}{sm_Z^2}})T_{L,R}^{ZZ}`$
$`+\left(s_Wc_W{\displaystyle \frac{\widehat{\mathrm{\Pi }}_{Z\gamma }^T(s)}{s}}\right)((T_{Z\gamma }^\gamma )_{L,R}+(T_{ZZ}^\gamma )_{L,R})+\left({\displaystyle \frac{1}{s_Wc_W}}{\displaystyle \frac{\widehat{\mathrm{\Pi }}_{\gamma Z}^T(s)}{sm_Z^2}}\right)((T_{\gamma \gamma }^Z)_{L,R}+(T_{\gamma Z}^Z)_{L,R})],`$
where $`T_{L,R}^{\gamma \gamma }`$, $`T_{L,R}^{\gamma Z}`$, $`T_{L,R}^{ZZ}`$ are defined in Eqs. (13-15) and
$`(T_{Z\gamma }^\gamma )_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}}{4s(sm_Z^2)}}{\displaystyle \frac{1}{2}}K_{L,R}^2,`$ (38)
$`(T_{ZZ}^\gamma )_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^4(a_{ij}^{\stackrel{~}{f}})^2}{16(sm_Z^2)^2}}{\displaystyle \frac{1}{2}}C_{L,R}K_{L,R},`$ (39)
$`(T_{\gamma \gamma }^Z)_{L,R}`$ $`=`$ $`{\displaystyle \frac{e^4e_f^2(\delta _{ij})^2}{s^2}}{\displaystyle \frac{1}{2}}C_{L,R}K_{L,R},`$ (40)
$`(T_{\gamma Z}^Z)_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}}{4s(sm_Z^2)}}{\displaystyle \frac{1}{2}}C_{L,R}^2.`$ (41)
The $`\widehat{\mathrm{\Pi }}_{VV}^T(s)`$ in Eq. (III) are the transverse parts of the renormalized self-energies of the vector bosons $`\gamma `$ and $`Z`$. The unrenormalized self-energies are given in Appendix C.2 and the renormalization is done following Denner .
The box corrections are obtained by adding up the diagrams shown in Fig. 3 and are given by
$`\mathrm{\Delta }\sigma _{L,R}^{\mathrm{box}}`$ $`=`$ $`{\displaystyle \frac{N_C}{4}}{\displaystyle \frac{\kappa ^3(s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}{4\pi s^2}}{\displaystyle _0^\pi }T_{L,R}^{\mathrm{box}}\mathrm{sin}^2\vartheta \mathrm{d}\vartheta ,`$ (42)
where
$`T_{L,R}^{\mathrm{box}}`$ $`=`$ $`\left({\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{e^2e_f\delta _{ij}}{s}}{\displaystyle \frac{1}{2}}K_{L,R}+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g_Z^2a_{ij}^{\stackrel{~}{f}}}{4(sm_Z^2)}}{\displaystyle \frac{1}{2}}C_{L,R}\right)B_{L,R}.`$ (43)
The $`B_{L,R}`$ are the form-factors defined in Appendix B, where one can find the analytic expressions as well.
### III.1 Renormalization scheme
In order to make the result finite we have to introduce the wave-function renormalization constants and counterterms. We fix them following the on-shell renormalization scheme. The parameters already occurring in the Standard Model (SM) are renormalized according to Denner . We assume the CKM matrix to be diagonal and so have no flavour mixing among the SM fermions at one-loop level.
#### III.1.1 Wave-function renormalization
The wave-function corrections are due to a shift from unrenormalized (bare) fields to the renormalized (physical) ones. For the fields relevant here we have
$`\stackrel{~}{f}_i^0=(\delta _{ij}+\frac{1}{2}\delta Z_{ij})\stackrel{~}{f}_j,`$ $`\left(\begin{array}{c}f_L^0\\ f_R^0\end{array}\right)=\left(\begin{array}{cc}1+\frac{1}{2}\delta Z_L& 0\\ 0& 1+\frac{1}{2}\delta Z_R\end{array}\right)\left(\begin{array}{c}f_L\\ f_R\end{array}\right),`$ (50)
$`\left(\begin{array}{c}A_\mu ^0\\ Z_\mu ^0\end{array}\right)`$ $`=`$ $`\left(\begin{array}{cc}1+\frac{1}{2}\delta Z_{\gamma \gamma }& \frac{1}{2}\delta Z_{\gamma Z}\\ \frac{1}{2}\delta Z_{Z\gamma }& 1+\frac{1}{2}\delta Z_{ZZ}\end{array}\right)\left(\begin{array}{c}A_\mu \\ Z_\mu \end{array}\right).`$ (57)
The form of the corrections for the left vertex is
$`\delta e_{L,R}^{(w)}`$ $`=`$ $`(\delta Z_{L,R}+\frac{1}{2}\delta Z_{\gamma \gamma })K_{L,R}{\displaystyle \frac{1}{2}}{\displaystyle \frac{g_Z}{e}}\delta Z_{Z\gamma }C_{L,R},`$ (58)
$`\delta a_{L,R}^{(w)}`$ $`=`$ $`(\delta Z_{L,R}+\frac{1}{2}\delta Z_{ZZ})C_{L,R}{\displaystyle \frac{1}{2}}{\displaystyle \frac{e}{g_Z}}\delta Z_{\gamma Z}K_{L,R}.`$ (59)
The wave-function corrections for the right vertex are
$`(\delta e_f)_{ij}^{(w)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}e_f(\delta Z_{ij}+\delta Z_{ji})+{\displaystyle \frac{1}{2}}e_f\delta Z_{\gamma \gamma }\delta _{ij}+{\displaystyle \frac{1}{8s_Wc_W}}a_{ij}^{\stackrel{~}{f}}\delta Z_{Z\gamma },`$ (60)
$`(\delta a_f)_{ij}^{(w)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{k=1}{\overset{2}{}}}\left(\delta Z_{ki}a_{kj}^{\stackrel{~}{f}}+\delta Z_{kj}a_{ik}^{\stackrel{~}{f}}\right)+2s_Wc_We_f\delta _{ij}\delta Z_{\gamma Z}+{\displaystyle \frac{1}{2}}a_{ij}^{\stackrel{~}{f}}\delta Z_{ZZ},`$ (61)
where $`s_W=\mathrm{sin}\theta _W`$ and $`c_W=\mathrm{cos}\theta _W`$.
The wave-function renormalization constants are determined by imposing the on-shell renormalization conditions as in onshellren ; sche2 such that the on-shell masses are the real parts of the poles of the propagator and the fields are properly normalized,
$`\delta Z_{ii}`$ $`=`$ $`\mathrm{}\dot{\mathrm{\Pi }}_{ii}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_i}^2),\delta Z_{ij}={\displaystyle \frac{2\mathrm{}\mathrm{\Pi }_{ij}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_j}^2)}{m_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2}},\delta Z_{\gamma \gamma }=\mathrm{}\dot{\mathrm{\Pi }}_{\gamma \gamma }(0),`$ (62)
$`\delta Z_{ZZ}`$ $`=`$ $`\mathrm{}\dot{\mathrm{\Pi }}_{ZZ}(m_Z^2),\delta Z_{\gamma Z}={\displaystyle \frac{2\mathrm{}\mathrm{\Pi }_{\gamma Z}(m_Z^2)}{m_Z^2}},\delta Z_{Z\gamma }={\displaystyle \frac{2\mathrm{}\mathrm{\Pi }_{Z\gamma }(0)}{m_Z^2}},`$ (63)
$`\delta Z_L`$ $`=`$ $`\mathrm{}[\mathrm{\Pi }_L(m_e^2)m_e^2(\dot{\mathrm{\Pi }}_L(m_e^2)+\dot{\mathrm{\Pi }}_R(m_e^2))+{\displaystyle \frac{1}{2m_e}}(\mathrm{\Pi }_{SL}(m_e^2)\mathrm{\Pi }_{SR}(m_e^2))`$ (64)
$`m_e(\dot{\mathrm{\Pi }}_{SL}(m_e^2)+\dot{\mathrm{\Pi }}_{SR}(m_e^2))],`$
$`\delta Z_R`$ $`=`$ $`\mathrm{}[\mathrm{\Pi }_R(m_e^2)m_e^2(\dot{\mathrm{\Pi }}_R(m_e^2)+\dot{\mathrm{\Pi }}_L(m_e^2))+{\displaystyle \frac{1}{2m_e}}(\mathrm{\Pi }_{SR}(m_e^2)\mathrm{\Pi }_{SL}(m_e^2))`$ (65)
$`m_e(\dot{\mathrm{\Pi }}_{SR}(m_e^2)+\dot{\mathrm{\Pi }}_{SL}(m_e^2))],`$
where $`\dot{\mathrm{\Pi }}(m^2)=\left[\frac{}{k^2}\mathrm{\Pi }(k^2)\right]_{k^2=m^2}`$. We use the self-energies given in Appendix C and in chrislet where we adopted the conventions from.
A remark should be made at this point. We include the wave-functions renormalization constants of the vector bosons although they are not external particles. By introducing them into the wave-function renormalization of the vertices, we have additional checks that can be made. First of all, the renormalization constants of the vector bosons must drop out in the final result. Secondly, the vertex corrections and the propagators can be both made UV-finite separately.
#### III.1.2 Counterterms
The counterterms come from the shift from the bare to the physical parameters in the lagrangian. It includes the shifting of $`e,m_W,m_Z,\theta _{\stackrel{~}{f}}`$ defined by
$`e^0=e+\delta e,m_W^0=m_W+\delta m_W,m_Z^0=m_Z+\delta m_Z,\theta _{\stackrel{~}{f}}^0=\theta _{\stackrel{~}{f}}+\delta \theta _{\stackrel{~}{f}}.`$ (66)
The counterterm contributions for both vertices are
$`\delta e_{L,R}^{(c)}`$ $`=`$ $`{\displaystyle \frac{\delta e}{e}}K_{L,R},`$ (67)
$`\delta a_{L,R}^{(c)}`$ $`=`$ $`\left[{\displaystyle \frac{\delta e}{e}}\left({\displaystyle \frac{\delta m_W}{m_W}}{\displaystyle \frac{\delta m_Z}{m_Z}}\right)+{\displaystyle \frac{12s_W}{t_W^2}}\left({\displaystyle \frac{\delta m_W}{m_W}}{\displaystyle \frac{\delta m_Z}{m_Z}}\right)\right]C_{L,R},`$ (68)
$`(\delta e_f)_{ij}^{(c)}`$ $`=`$ $`{\displaystyle \frac{\delta e}{e}}e_f\delta _{ij},`$ (69)
$`(\delta a_f)_{ij}^{(c)}`$ $`=`$ $`\left[{\displaystyle \frac{\delta e}{e}}+{\displaystyle \frac{c_W^2s_W^2}{s_W^2}}\left({\displaystyle \frac{\delta m_W}{m_W}}{\displaystyle \frac{\delta m_Z}{m_Z}}\right)\right]a_{ij}^{\stackrel{~}{f}}+8e_fc_W^2\left({\displaystyle \frac{\delta m_W}{m_W}}{\displaystyle \frac{\delta m_Z}{m_Z}}\right)\delta _{ij},`$ (70)
where the contributions containing $`\delta \theta _{\stackrel{~}{f}}`$ were intentionally left out and will be discussed below.
#### III.1.3 Renormalization of electric charge
The standard on-shell input value for the electric charge is the one in the Thomson limit $`\alpha e^2/(4\pi )=1/137.036`$. This corresponds to a counterterm
$`{\displaystyle \frac{\delta e}{e}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\delta Z_{\gamma \gamma }{\displaystyle \frac{s_W}{2c_W}}\delta Z_{Z\gamma }.`$ (71)
In this way of fixing the electric charge has a significant theoretical uncertainty coming from the light quarks which we circumvent by using as input parameter for $`\alpha `$ the $`\overline{\mathrm{MS}}`$ value at the $`Z`$-pole, $`\alpha \alpha (m_Z)|_{\overline{\mathrm{MS}}}=e^2/(4\pi )`$. The counterterm then is given by chrislet ; 0111303 ; wcharge
$`{\displaystyle \frac{\delta e}{e}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{e^2}{6}}[\mathrm{\hspace{0.17em}4}{\displaystyle \underset{f}{}}N_C^fe_f^2(\mathrm{\Delta }+\mathrm{log}{\displaystyle \frac{Q^2}{x_f^2}})+{\displaystyle \underset{\stackrel{~}{f}}{}}{\displaystyle \underset{m=1}{\overset{2}{}}}N_C^fe_f^2(\mathrm{\Delta }+\mathrm{log}{\displaystyle \frac{Q^2}{m_{\stackrel{~}{f}_m}^2}})`$ (72)
$`+\mathrm{\hspace{0.17em}4}{\displaystyle \underset{k=1}{\overset{2}{}}}(\mathrm{\Delta }+\mathrm{log}{\displaystyle \frac{Q^2}{m_{\stackrel{~}{\chi }_k^+}^2}})+(\mathrm{\Delta }+\mathrm{log}{\displaystyle \frac{Q^2}{m_{H^+}^2}})21(\mathrm{\Delta }+\mathrm{log}{\displaystyle \frac{Q^2}{m_W^2}})],`$
with $`x_f=m_Zm_f<m_Z`$ and $`x_t=m_t`$. $`N_C^f`$ is the color factor, $`N_C^f=1,3`$ for (s)leptons and (s)quarks, respectively. $`\mathrm{\Delta }`$ denotes the UV divergence factor, $`\mathrm{\Delta }=2/ฯต\gamma +\mathrm{log}4\pi `$.
#### III.1.4 Renormalization of $`m_W`$ and $`m_Z`$
The masses of the $`Z`$-boson and the $`W`$-boson are fixed as the physical (pole) masses, i. e.
$`\delta m_Z^2=\mathrm{}\mathrm{\Pi }_{ZZ}^T(m_Z^2),\delta m_W^2=\mathrm{}\mathrm{\Pi }_{WW}^T(m_W^2),`$ (73)
where
$`{\displaystyle \frac{\delta m_Z}{m_Z}}={\displaystyle \frac{1}{2}}{\displaystyle \frac{\delta m_Z^2}{m_Z^2}},{\displaystyle \frac{\delta m_W}{m_W}}={\displaystyle \frac{1}{2}}{\displaystyle \frac{\delta m_W^2}{m_W^2}}.`$ (74)
The formulas for the vector boson self-energies $`\mathrm{\Pi }_{WW}^T(m_W^2)`$ and $`\mathrm{\Pi }_{ZZ}^T(m_Z^2)`$ are given in Appendix C and in chrislet . The counterterms for the intermediate boson masses are used to determine the Weinberg angle fixing according to Sirlin .
#### III.1.5 Renormalization of $`\theta _{\stackrel{~}{f}}`$
The counterterm of the sfermion mixing angle, $`\delta \theta _{\stackrel{~}{f}}`$, is fixed such that it cancels the anti-hermitian part of the sfermion wave-function corrections Yukawa ; guasch ,
$`\delta \theta _{\stackrel{~}{f}}`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(\delta Z_{12}\delta Z_{21}\right)={\displaystyle \frac{1}{2\left(m_{\stackrel{~}{f}_1}^2m_{\stackrel{~}{f}_2}^2\right)}}\mathrm{}\left(\mathrm{\Pi }_{12}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_2}^2)+\mathrm{\Pi }_{21}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_1}^2)\right).`$ (75)
Including the terms proportional to $`\delta \theta _{\stackrel{~}{f}}`$ in Eq. (70) is equivalent to symmetrizing the off-diagonal sfermion wave-function corrections in Eq. (62) as sche2 ; sche3
$`\delta Z_{12}=\delta Z_{21}={\displaystyle \frac{\mathrm{}\mathrm{\Pi }_{12}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_2}^2)\mathrm{}\mathrm{\Pi }_{21}^{\stackrel{~}{f}}(m_{\stackrel{~}{f}_1}^2)}{m_{\stackrel{~}{f}_1}^2m_{\stackrel{~}{f}_2}^2}}.`$ (76)
This fixing of the counterterm for the mixing angle is analogous to the renormalization of the CKM matrix in CKM and similarly has to be made gauge-independent. It was shown in yamada that this can be avoided or, equivalently, the result in the $`\xi =1`$ gauge can be regarded as the gauge-independent one.
## IV Real photon corrections
Similarly to the QCD case where the cross-section was IR-divergent due to massless gluons QCD1 ; QCD2 ; SUSY-QCD-A ; SUSY-QCD-H , the one-loop cross-section $`\sigma ^{\mathrm{ren}}(e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j)`$ is IR-divergent owing to the diagrams with photon exchange where the photon mass is zero. This is remedied by introducing a small mass $`\lambda `$ and including also the Bremsstrahlung process i. e. $`\sigma (e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j\gamma )`$, see Fig. 4. Summing these two contributions yields an IR-finite result for the physical value $`\lambda =0`$,
$`\sigma ^{\mathrm{corr}}(e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j)`$ $`=`$ $`\sigma ^{\mathrm{ren}}(e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j)+\sigma (e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j\gamma ).`$ (77)
To calculate the radiative cross-section $`\sigma (e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j\gamma )`$ we use the phase-space splicing method slicing1 which splits the bremsstrahlung phase-space into 3 regions. The corresponding 3 parts are
$`\sigma (e^+e^{}\stackrel{~}{f}_i\overline{\stackrel{~}{f}}_j\gamma )`$ $`=`$ $`\sigma ^{\mathrm{soft}}(\lambda ,\mathrm{\Delta }E)+\sigma ^{\mathrm{hard}}(\mathrm{\Delta }E,\mathrm{\Delta }\theta )+\sigma ^{\mathrm{coll}}(\mathrm{\Delta }E,\mathrm{\Delta }\theta ).`$ (78)
In our calculation, we used a soft-photon approximation ($`\sigma ^{\mathrm{soft}}`$) to reproduce the divergence pattern correctly. However, this approximation introduces a cut $`\mathrm{\Delta }E`$ on the energy of the radiated photon. The dependence on the cut $`\mathrm{\Delta }E`$ drops out if we include the full $`23`$ process ($`\sigma ^{\mathrm{hard}}`$). In order to get simpler expressions for $`\sigma ^{\mathrm{hard}}`$ we neglect the electron mass but then a collinear divergence occurs when the photon is radiated in the direction of the electron and positron beams. This collinear divergence can be regulated by introducing yet another approximation ($`\sigma ^{\mathrm{coll}}`$) for the above mentioned phase-space region. Another cut $`\mathrm{\Delta }\theta `$ is hereby introduced. After summing the 3 contributions the result must be independent of both the cuts and has to cancel the IR-divergence of the one-loop cross-section. This is the ultimate test we have made at the end of the calculation.
### IV.1 Soft-photon approximation
The soft-photon approximation supposes that the 4-momentum of the photon is small compared to other momenta (for details see e. g.Denner ). Using this assumption the differential cross-section $`\left(\frac{\mathrm{d}\sigma }{\mathrm{d}\mathrm{\Omega }}\right)_{\mathrm{soft}}`$ is proportional to the tree-level differential cross-section. The full cross-section for polarized beams is
$`\sigma ^{\mathrm{soft}}(\lambda ,\mathrm{\Delta }E)`$ $`=`$ $`{\displaystyle \frac{1}{4}}(1P_{})(1+P_+)\sigma _L^{\mathrm{soft}}+{\displaystyle \frac{1}{4}}(1+P_{})(1P_+)\sigma _R^{\mathrm{soft}},`$ (79)
where
$`\sigma _{L,R}^{\mathrm{soft}}`$ $`=`$ $`{\displaystyle \left(\frac{\mathrm{d}\sigma _{L,R}}{\mathrm{d}\mathrm{\Omega }}\right)_{\mathrm{soft}}d\mathrm{\Omega }}={\displaystyle \left(\frac{\mathrm{d}\sigma _{L,R}}{\mathrm{d}\mathrm{\Omega }}\right)_{\mathrm{tree}}\delta _sd\mathrm{\Omega }}.`$ (80)
The factor $`\delta _s`$ is defined as
$`\delta _s`$ $`=`$ $`{\displaystyle \frac{\alpha }{4\pi ^2}}\left(I_{p_1^2}+I_{p_2^2}2I_{p_1p_2}+e_f^2(I_{k_1^2}+I_{k_2^2}2I_{k_1k_2})+2e_f(I_{p_1k_1}+I_{p_2k_2}I_{p_1k_2}I_{p_2k_1})\right),`$ (81)
where the integrals $`I_{ab}`$ are defined in Denner and were worked out e. g. in tHooft . The explicit formula for $`\delta _s`$ can be found in Appendix D.
### IV.2 Hard and collinear photon radiation
The cross-section for the full bremsstrahlung process $`e^+(p_2)e^{}(p_1)\stackrel{~}{f}_i(k_1)\overline{\stackrel{~}{f}}_j(k_2)\gamma (k_3)`$ is given by
$`\sigma ^{\mathrm{hard}}(\mathrm{\Delta }E,\mathrm{\Delta }\theta )`$ $`=`$ $`{\displaystyle \frac{1}{2s}}{\displaystyle \frac{1}{8(2\pi )^4}}{\displaystyle ||^2dk_1^0dk_3^0d\eta \mathrm{d}\mathrm{cos}\theta },`$ (82)
where the cuts $`\mathrm{\Delta }E`$ and $`\mathrm{\Delta }\theta `$ appear in the integration bounds of $`\mathrm{d}k_3^0`$ and $`\mathrm{d}\mathrm{cos}\theta `$. The angle $`\eta `$ is defined as in feyn . The explicit form of the squared matrix element is given in Appendix D and the integral is evaluated numerically using the routines from the CUBA library cuba .
As we have neglected the electron mass in the calculation of $`\sigma ^{\mathrm{hard}}`$, we have to take an another approach in the collinear region of the phase-space. We follow the approach of slicing1 ; slicing2 and get for the collinear cross-section the following expression ,
$`\sigma ^{\mathrm{coll}}(\mathrm{\Delta }E,\mathrm{\Delta }\theta )`$ $`=`$ $`{\displaystyle \frac{1}{4}}[(1P_{})(1P_+)\sigma _{LL}^{\mathrm{coll}}+(1P_{})(1+P_+)\sigma _{LR}^{\mathrm{coll}}`$ (83)
$`+(1+P_{})(1P_+)\sigma _{RL}^{\mathrm{coll}}+(1+P_{})(1+P_+)\sigma _{RR}^{\mathrm{coll}}],`$
where all polarization states $`(\sigma _{LL}^{\mathrm{coll}},\sigma _{LR}^{\mathrm{coll}},\sigma _{RL}^{\mathrm{coll}},\sigma _{RR}^{\mathrm{coll}})`$ appear. This is due to the radiation of an additional photon and the fact that the electron is massive ($`\sigma _{LR}`$ stands for the cross-section with left-handed electrons and right-handed positrons etc.). The single polarization states are given by
$`\sigma _{LL,RR}^{\mathrm{coll}}`$ $`=`$ $`{\displaystyle \frac{e^2}{8\pi ^2}}{\displaystyle _{x_{min}}^{x_{max}}}dxx\left[\sigma _L^{\mathrm{tree}}\left((1x)s\right)+\sigma _R^{\mathrm{tree}}\left((1x)s\right)\right],`$ (84)
$`\sigma _{LR,RL}^{\mathrm{coll}}`$ $`=`$ $`{\displaystyle \frac{e^2}{8\pi ^2}}{\displaystyle _{x_{min}}^{x_{max}}}dx\mathrm{\hspace{0.33em}2}{\displaystyle \frac{x^22x+2}{x}}\left[\mathrm{log}\left({\displaystyle \frac{s\mathrm{\Delta }\theta ^2}{4m_e^2}}\right)1\right]\sigma _{L,R}^{\mathrm{tree}}\left((1x)s\right),`$ (85)
with
$`x_{min}=\mathrm{\hspace{0.17em}2}\mathrm{\Delta }E/\sqrt{s},x_{max}=\mathrm{\hspace{0.17em}1}{\displaystyle \frac{(m_{\stackrel{~}{f}_i}+m_{\stackrel{~}{f}_j})^2}{s}}.`$ (86)
After including all the above-mentioned contributions we arrive at a cut-independent result.
### IV.3 Higher order corrections
Substantial correction from the collinear photon radiation is due to the smallness of the electron mass compared to a typical energy scale in the process. This effect is such that to reach the collider precision one has to include also the leading higher order corrections (i. e. beyond $`๐ช(\alpha )`$). Owing to the mass-factorization theorem, one can factorize the corrections in the leading-log (LL) approximation as
$`{\displaystyle ๐\sigma ^{\mathrm{tree}}}+{\displaystyle ๐\sigma ^{\mathrm{LL}}}`$ $`=`$ $`{\displaystyle _0^1}๐x_1{\displaystyle _0^1}๐x_2\mathrm{\Gamma }_{ee}^{\mathrm{LL}}(x_1,Q^2)\mathrm{\Gamma }_{ee}^{\mathrm{LL}}(x_2,Q^2){\displaystyle ๐\sigma ^{\mathrm{tree}}(x_1p_1,x_2p_2)}.`$ (87)
where $`x_1`$, $`x_2`$ are the momentum fractions of the electron and the positron carried after the radiation of the photon(s).
The $`\mathrm{\Gamma }_{ee}^{\mathrm{LL}}(x,Q^2)`$ is the leading-log structure function up to $`๐ช(\alpha ^3)`$, given in ref.Sk90 ,
$`\mathrm{\Gamma }_{ee}^{\mathrm{LL}}(x,Q^2)`$ $`=`$ $`{\displaystyle \frac{\mathrm{exp}(\frac{1}{2}\beta \gamma _E+\frac{3}{8}\beta )}{\mathrm{\Gamma }(1+\frac{\beta }{2})}}{\displaystyle \frac{\beta }{2}}(1x)^{\frac{\beta }{2}1}`$ (88)
$`{\displaystyle \frac{\beta }{4}}(1+x)+{\displaystyle \frac{\beta ^2}{16}}\left(2(1+x)\mathrm{log}(1x){\displaystyle \frac{2\mathrm{log}x}{1x}}+{\displaystyle \frac{3}{2}}(1+x)\mathrm{log}x{\displaystyle \frac{x}{2}}{\displaystyle \frac{5}{2}}\right)`$
$`+{\displaystyle \frac{\beta ^3}{8}}[{\displaystyle \frac{1}{2}}(1+x)({\displaystyle \frac{9}{32}}{\displaystyle \frac{\pi ^2}{12}}+{\displaystyle \frac{3}{4}}\mathrm{log}(1x)+{\displaystyle \frac{1}{2}}\mathrm{log}^2(1x){\displaystyle \frac{1}{4}}\mathrm{log}x\mathrm{log}(1x)`$
$`+{\displaystyle \frac{1}{16}}\mathrm{log}^2x{\displaystyle \frac{1}{4}}\mathrm{Li}_2(1x))+{\displaystyle \frac{1}{2}}{\displaystyle \frac{1+x^2}{1x}}({\displaystyle \frac{3}{8}}\mathrm{log}x+{\displaystyle \frac{1}{12}}\mathrm{log}^2x{\displaystyle \frac{1}{2}}\mathrm{log}x\mathrm{log}(1x))`$
$`{\displaystyle \frac{1}{4}}(1x)(\mathrm{log}(1x)+{\displaystyle \frac{1}{4}})+{\displaystyle \frac{1}{32}}(53x)\mathrm{log}x],`$
with the gamma function $`\mathrm{\Gamma }`$, the Euler constant $`\gamma _E0.577216`$, and $`\beta =\frac{2\alpha }{\pi }(\mathrm{log}\frac{Q^2}{m_e^2}1)`$. For the free scale $`Q^2`$ we take the typical energy of the process $`s`$. The soft-photon contributions were summed up to all orders in the perturbation series.
The structure function (88) contains not only the higher orders beginning with $`๐ช(\alpha ^2)`$ but also parts of terms $`๐ช(\alpha )`$ already included elsewhere. To avoid double counting we subtract these terms as in slicing2 .
## V Numerical analysis
In contrast to letter , we do not attempt to make a scan over a large area of MSSM parameter space but rather consider only one benchmark point in the numerical analysis. It is the SPS1aโ point we use as input which is defined in the SPA project SPA . The point is chosen such that it satisfies all the precision data and both the bounds for the masses of the SUSY particles and the bounds from cosmology.
The input parameters for the SPS1aโ point are defined in the $`\overline{\mathrm{DR}}`$ scheme at the scale $`Q=1\mathrm{T}\mathrm{e}\mathrm{V}`$. As we use the on-shell renormalization scheme, we have to transform the SPS1aโ input parameters $`๐ซ`$ into on-shell parameters $`๐ซ^{\mathrm{OS}}`$. This transformation is simply performed by subtracting the corresponding counterterms i. e. $`๐ซ^{\mathrm{OS}}=๐ซ(Q)\delta ๐ซ(Q)`$ and the results for the relevant parameters are listed in Table 1. All other parameters do not enter in the calculation at tree-level and so the differences when using the on-shell or the $`\overline{\mathrm{DR}}`$ value are of a higher order. A further remark is necessary here. One of the parameters not entering the tree-level directly is the infamous $`A_b`$ parameter. Fortunately, the parameter set taken here causes all the on-shell input parameters to be insensitive to the problems of the $`A_b`$ on-shell definition.
The Figs. 5-8 show the total cross-sections for the pair production of the sfermions of the third generation. In general, we show the complete corrections and the tree-level where the tree-level is defined according to the SPA project. According to the SPA project, all masses are taken on-shell and all parameters in the couplings are given in the $`\overline{\mathrm{DR}}`$ scheme. The virtue of using this tree-level definition is that not only the total corrected cross-sections are directly comparable to other calculations using the SPA conventions but one can also compare the relative corrections.
For each sfermion type we show an unpolarized case (left) and a case where the beams are polarized (right). We take two sets of polarizations, either $`P_{}=0.8`$ and $`P_+=0.6`$ or $`P_{}=0.8`$ and $`P_+=0.6`$. The difference to the earlier calculations letter ; hollik are the QED corrections which give a negative contribution near the threshold due to the known soft-photon behaviour. The QED corrections are substantial (as can be checked when comparing the results of this paper with those of letter ) and cannot to be neglected. The plots on the right-hand side of Fig. 5-8 show the effect of beams polarization on the radiative corrections. Polarization and its effects are best seen in other observables which we discuss in the following.
The Figs. 9-12 show the left-right and forward-backward asymmetries for different final states as defined in Eq. (20). Owing to the fact that at tree level there is only a s-channel contribution the $`\sqrt{s}`$-dependence drops out in the left-right asymmetry, making it to a good approximation constant. The $`\sqrt{s}`$-dependence is then a result of the one-loop corrections. Notice that the corrections are substantial especially in the $`\stackrel{~}{t}_1\stackrel{~}{t}_2`$, $`\stackrel{~}{b}_1\stackrel{~}{b}_2`$, $`\stackrel{~}{\tau }_1\stackrel{~}{\tau }_2`$, as well as in the $`\stackrel{~}{\nu }_\tau \stackrel{~}{\nu }_\tau `$ channel, where there is only a $`Z`$ exchange at tree-level.
As we have already mentioned, there is no tree-level contribution to the forward-backward asymmetry and thus the asymmetry is loop-induced. In the calculation of the forward-backward asymmetry at one-loop one has to define the forward direction, in particular for the contributions coming from the photon radiation . We define it by $`\sigma _F\sigma (\mathrm{cos}\theta _{\stackrel{}{p_1}\stackrel{}{k_{1,2}}}0)`$ where $`\theta _{\stackrel{}{p_1}\stackrel{}{k_{1,2}}}`$ is the angle between the incoming electron and the outgoing sfermion with negative isospin. As an additional feature, we also show the forward-backward asymmetry for polarized beams where the polarizations are $`P_{}=0.8`$ and $`P_+=0.6`$. In general, one sees that the asymmetries receive sizeable corrections and thus justify the higher-order calculation.
## VI Conclusion
We have calculated the full $`๐ช(\alpha )`$ corrections to stop, sbottom, stau and tau-sneutrino production in the MSSM. We have presented the details of our analytical calculation which was also checked by the computer algebra tools FeynArts and FormCalc feyn . The results extend our previous calculations SUSY-QCD-H ; Yukawa ; letter by including also QED contributions and the real photon radiation. We have also used the structure function approach slicing2 to include some higher order effects. Moreover, the whole calculation was extended to the case of polarized $`e^\pm `$-beams.
In the numerical analysis, we have studied only one specific scenario based on the SPS1aโ benchmark point defined in the SPA project. We have transformed the input parameters into the on-shell renormalization scheme which we have used throughout the paper. The numerical results show the total cross-sections and asymmetries with the effect of the $`๐ช(\alpha )`$ corrections. These are found to be sizeable (in some cases up to 15% and larger), and in particular the forward-backward asymmetry is only due to higher order corrections.
Acknowledgements
We thank W. รller for useful discussions. The authors acknowledge support from EU under the HPRN-CT-2000-00149 network programme and the โFonds zur Fรถrderung der wissenschaftlichen Forschungโ of Austria, project No. P16592-N02.
## Appendix A Vertex corrections
Here we give the explicit form of the electroweak contributions to the vertex corrections which are depicted in Fig. 2. For SUSY-QCD contributions we refer to SUSY-QCD-H . All couplings used in this paper can be found in chrislet .
The vertex corrections $`(\delta e_f)_{ij}^{(v)}`$ and $`(\delta a_f)_{ij}^{(v)}`$ (or $`\delta e_{L,R}^{(v)}`$ and $`\delta a_{L,R}^{(v)}`$) originate from the diagrams in Fig. 2 with $`\gamma `$, $`Z`$-boson exchange, respectively.
The vertex corrections to the right vertex $`(\delta e_f)_{ij}^{(v)}`$ and $`(\delta a_f)_{ij}^{(v)}`$ are defined on the amplitude level from the corresponding diagrams. The general form of the amplitude is given by
$$=\frac{i}{(4\pi )^2}\frac{X_V}{sm_V^2}\overline{v}(p_2)\gamma ^\mu (Y_L^VP_L+Y_R^VP_R)u(p_1)\left[A_{ij}(k_1k_2)_\mu +B_{ij}(k_1+k_2)_\mu \right].$$
(89)
For $`X_V`$ and $`Y_{L,R}^V`$ with $`V=\gamma ,Z`$ we have
$`X_\gamma =e,Y_{L,R}^\gamma =K_{L,R},X_Z=g_Z,Y_{L,R}^Z=C_{L,R}.`$ (90)
We use the form-factor $`A_{ij}`$ (the other form-factor vanishes in $`||^2`$) to define the vertex correction $`(\delta e_f)_{ij}^{(v)}`$ and $`(\delta a_f)_{ij}^{(v)}`$ as
$$(\delta e_f)_{ij}^{(v)}=\frac{1}{(4\pi )^2}\frac{1}{e}A_{ij},(\delta a_f)_{ij}^{(v)}=\frac{1}{(4\pi )^2}\frac{4c_W}{g}A_{ij}.$$
(91)
The explicit formulas for $`(\delta e_f)_{ij}^{(v)}`$ and $`(\delta a_f)_{ij}^{(v)}`$ are given below in the Appendices A.1 and A.2.
The vertex corrections to the left vertex, $`\delta e_{L,R}^{(v)}`$ and $`\delta a_{L,R}^{(v)}`$, include corrections to the two chiral parts of the electron photon/$`Z`$-boson vertex. The generic form is
$`={\displaystyle \frac{i}{(4\pi )^2}}{\displaystyle \frac{(X_V)_{ij}}{sm_V^2}}\overline{v}(p_2)[A_L\gamma ^\mu P_L+A_R\gamma ^\mu P_R+B_LP_L(p_1p_2)^\mu +B_RP_R(p_1p_2)^\mu `$
$`+C_LP_L(p_1+p_2)^\mu +C_RP_R(p_1+p_2)^\mu ]u(p_1)(k_1k_2)_\mu ,`$ (92)
where $`(X_V)_{ij}`$ ($`V=\gamma ,Z`$) stands for
$$(X_\gamma )_{ij}=ee_f\delta _{ij},(X_Z)_{ij}=\frac{g_Z}{4}a_{ij}^f.$$
(93)
From this generic structure only the form-factors $`A_{L,R}`$ survive. We define the vertex corrections as
$$\delta e_{L,R}^{(v)}=\frac{1}{(4\pi )^2}\frac{1}{e}A_{L,R},\delta a_{L,R}^{(v)}=\frac{1}{(4\pi )^2}\frac{c_W}{g}A_{L,R}.$$
(94)
The contributions to the vertex corrections $`\delta e_{L,R}^{(v)}`$ and $`\delta a_{L,R}^{(v)}`$ are given in the Appendices A.3 and A.4.
### A.1 Corrections to $`๐ธ\stackrel{\mathbf{~}}{๐}_๐\stackrel{\mathbf{~}}{๐}_๐`$ vertex
The vertex correction $`(\delta e_f)_{ij}^{(v)}`$ is composed of contributions from different classes of diagrams as follows,
$$(\delta e_f)_{ij}^{(v)}=(\delta e_f)_{ij}^{(v,\stackrel{~}{\chi })}+(\delta e_f)_{ij}^{(v,SSS)}+(\delta e_f)_{ij}^{(v,V\stackrel{~}{f}\stackrel{~}{f})}+(\delta e_f)_{ij}^{(v,SSV+SVS)}+(\delta e_f)_{ij}^{(v,SVV)}+(\delta e_f)_{ij}^{(v,SV)}.$$
(95)
In the following we use the standard two- and three-point functions $`B_i`$ and $`C_i`$ from PaVe in the conventions of Denner . We introduce the following standard set of arguments $`C_iC_i(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,M_0^2,M_1^2,M_2^2)`$ to be used in the generic functions $`๐ฎ`$.
The first contribution coming from the exchange of one or two gauginos is
$`(\delta e_f)_{ij}^{(v,\stackrel{~}{\chi })}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k,l=1}{\overset{2}{}}}๐ฎ_{ij}^{FFF}(m_f^{},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_l^+};2I_f^{3L}\delta _{kl},2I_f^{3L}\delta _{kl},k_{ik}^{\stackrel{~}{f}},l_{ik}^{\stackrel{~}{f}},l_{jl}^{\stackrel{~}{f}},k_{jl}^{\stackrel{~}{f}})`$ (96)
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{4}{}}}๐ฎ_{ij}^{FFF}(m_{\stackrel{~}{\chi }_k^0},m_f,m_f;e_f,e_f,b_{ik}^{\stackrel{~}{f}},a_{ik}^{\stackrel{~}{f}},a_{jk}^{\stackrel{~}{f}},b_{jk}^{\stackrel{~}{f}})`$
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}๐ฎ_{ij}^{FFF}(m_{\stackrel{~}{\chi }_k^+},m_f^{},m_f^{};e_f^{},e_f^{},k_{ik}^{\stackrel{~}{f}},l_{ik}^{\stackrel{~}{f}},l_{jk}^{\stackrel{~}{f}},k_{jk}^{\stackrel{~}{f}}),`$
with the generic vertex function
$`๐ฎ_{ij}^{FFF}(M_0,M_1,M_2;g_0^R,g_0^L,g_1^R,g_1^L,g_2^R,g_2^L)=M_0M_2\left(g_0^Lg_1^Lg_2^L+g_0^Rg_1^Rg_2^R\right)\left(C_0+C_1+C_2\right)`$ (97)
$`+M_0M_1\left(g_0^Lg_1^Rg_2^R+g_0^Rg_1^Lg_2^L\right)\left(C_0+C_1+C_2\right)+M_1M_2\left(g_0^Lg_1^Rg_2^L+g_0^Rg_1^Lg_2^R\right)\left(C_1+C_2\right)`$
$`+\left(g_0^Lg_1^Lg_2^R+g_0^Rg_1^Rg_2^L\right)\left(B_0(s,M_1^2,M_2^2)+M_0^2\left(2C_0+C_1+C_2\right)+m_{\stackrel{~}{f}_i}^2C_1+m_{\stackrel{~}{f}_j}^2C_2\right).`$
The corrections due to graphs with 3 scalar particles in the loop are given by
$`(\delta e_f)_{ij}^{(v,SSS)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\mathrm{\hspace{0.17em}2}I_f^{3L}{\displaystyle \underset{k,m=1}{\overset{2}{}}}G_{imk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}G_{jmk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m^{}}^2,m_{H_k^+}^2,m_{H_k^+}^2)`$ (98)
$`{\displaystyle \frac{1}{(4\pi )^2}}e_f{\displaystyle \underset{k,m=1}{\overset{2}{}}}G_{imk}^{\stackrel{~}{f}}G_{mjk}^{\stackrel{~}{f}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{H_k^0}^2,m_{\stackrel{~}{f}_m}^2,m_{\stackrel{~}{f}_m}^2)`$
$`{\displaystyle \frac{1}{(4\pi )^2}}e_f^{}{\displaystyle \underset{k,m=1}{\overset{2}{}}}G_{imk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}G_{jmk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{H_k^+}^2,m_{\stackrel{~}{f}_m^{}}^2,m_{\stackrel{~}{f}_m^{}}^2).`$
The graphs with one vector particle ($`\gamma ,Z^0,W^+`$) and two sfermions in the loop yield
$`(\delta e_f)_{ij}^{(v,V\stackrel{~}{f}\stackrel{~}{f})}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}e^2e_f^3\delta _{ij}๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(\lambda ,m_{\stackrel{~}{f}_i},m_{\stackrel{~}{f}_i})+{\displaystyle \frac{1}{(4\pi )^2}}g_Z^2e_f{\displaystyle \underset{m=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}z_{mj}^{\stackrel{~}{f}}๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(m_Z,m_{\stackrel{~}{f}_m},m_{\stackrel{~}{f}_m})`$ (99)
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g^2}{2}}e_f^{}R_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(m_W,m_{\stackrel{~}{f}_m^{}},m_{\stackrel{~}{f}_m^{}}),`$
where $`๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(\mathrm{})`$ is a short form for
$`๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(M_0,M_1,M_2)=4C_{00}+\left(M_0^22s+2m_{\stackrel{~}{f}_i}^2+2m_{\stackrel{~}{f}_j}^2\right)\left(C_0+C_1+C_2\right)`$
$`+\left(2m_{\stackrel{~}{f}_i}^2+2m_{\stackrel{~}{f}_j}^2s\right)\left(C_1+C_2+C_{11}+2C_{12}+C_{22}\right)+\left(m_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2\right)\left(C_1C_2+C_{11}C_{22}\right).`$
From the diagrams with one $`W^+`$-boson, one Goldstone boson $`G^+`$ and a sfermion we obtain
$`(\delta e_f)_{ij}^{(v,SSV+SVS)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{gm_W}{2\sqrt{2}}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{i1}^{\stackrel{~}{f}}G_{jm2}^{\stackrel{~}{f}\stackrel{~}{f}^{}}+R_{j1}^{\stackrel{~}{f}}G_{im2}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\right)R_{m1}^{\stackrel{~}{f}^{}}`$ (101)
$`\times \left(C_0C_1C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m^{}}^2,m_W^2,m_W^2),`$
and with the generic scalarโvectorโvector vertex function
$`๐ฎ_{ij}^{SVV}(M_0,M_1,M_2)=2B_0(s,M_1^2,M_2^2)2C_{00}+\left({\displaystyle \frac{s}{2}}+2M_0^2\right)C_0s\left(C_1+C_2\right)`$ (102)
$`{\displaystyle \frac{1}{2}}\left(m_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2\right)\left(C_1C_2+C_{11}C_{22}\right){\displaystyle \frac{1}{2}}\left(2m_{\stackrel{~}{f}_i}^2+2m_{\stackrel{~}{f}_j}^2s\right)\left(C_{11}+2C_{12}+C_{22}\right)`$
the correction due to the exchange of two $`W^+`$-bosons and one sfermion reads
$`(\delta e_f)_{ij}^{(v,SVV)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}g^2I_f^{3L}R_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2๐ฎ_{ij}^{SVV}(m_{\stackrel{~}{f}_m^{}},m_W,m_W).`$ (103)
The contributions from one sfermion and one vector particle ($`\gamma ,Z^0,W^+`$) can be expressed as
$`(\delta e_f)_{ij}^{(v,SV)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\mathrm{\hspace{0.17em}2}e^2e_f^3\delta _{ij}\left(2B_0+B_1\right)(m_{\stackrel{~}{f}_i}^2,\lambda ^2,m_{\stackrel{~}{f}_i}^2)`$ (104)
$`{\displaystyle \frac{1}{(4\pi )^2}}g_Z^2e_f{\displaystyle \underset{m=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}z_{jm}^{\stackrel{~}{f}}(2B_0+B_1)(m_{\stackrel{~}{f}_i}^2,m_Z^2,m_{\stackrel{~}{f}_m}^2)+(ij)`$
$`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g^2}{4}}Y_L^fR_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2(2B_0+B_1)(m_{\stackrel{~}{f}_i}^2,m_W^2,m_{\stackrel{~}{f}_m^{}}^2)+(ij)`$
with $`Y_L^f=2(I_f^{3L}e_f)`$. The symbol $`(ij)`$ denotes the previous term with the indices $`i`$ and $`j`$ interchanged.
### A.2 Corrections to $`๐^\mathrm{๐}\stackrel{\mathbf{~}}{๐}_๐\stackrel{\mathbf{~}}{๐}_๐`$ vertex
The corrections to the $`Z^0\stackrel{~}{f}_i\stackrel{~}{f}_j`$ vertex have the same components as in Eq. (95). Using the same abbreviations for the generic vertex functions as in the previous section we get for the single contributions:
$`(\delta a_f)_{ij}^{(v,\stackrel{~}{\chi })}`$ $`=`$ $`{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \underset{k,l=1}{\overset{4}{}}}๐ฎ_{ij}^{FFF}(m_f,m_{\stackrel{~}{\chi }_k^0},m_{\stackrel{~}{\chi }_l^0};O_{kl}^{{}_{}{}^{\prime \prime }R},O_{kl}^{{}_{}{}^{\prime \prime }L},b_{ik}^{\stackrel{~}{f}},a_{ik}^{\stackrel{~}{f}},a_{jl}^{\stackrel{~}{f}},b_{jl}^{\stackrel{~}{f}})`$ (105)
$`{\displaystyle \frac{4}{(4\pi )^2}}\mathrm{\hspace{0.17em}2}I_f^{3L}{\displaystyle \underset{k,l=1}{\overset{2}{}}}๐ฎ_{ij}^{FFF}(m_f^{},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_l^+};\stackrel{~}{O}_{kl}^{}_{}{}^{}R,\stackrel{~}{O}_{kl}^{}_{}{}^{}L,k_{ik}^{\stackrel{~}{f}},l_{ik}^{\stackrel{~}{f}},l_{jl}^{\stackrel{~}{f}},k_{jl}^{\stackrel{~}{f}})`$
$`+{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{4}{}}}๐ฎ_{ij}^{FFF}(m_{\stackrel{~}{\chi }_k^0},m_f,m_f;C_R^f,C_L^f,b_{ik}^{\stackrel{~}{f}},a_{ik}^{\stackrel{~}{f}},a_{jk}^{\stackrel{~}{f}},b_{jk}^{\stackrel{~}{f}})`$
$`+{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}๐ฎ_{ij}^{FFF}(m_{\stackrel{~}{\chi }_k^+},m_f^{},m_f^{};C_R^f^{},C_L^f^{},k_{ik}^{\stackrel{~}{f}},l_{ik}^{\stackrel{~}{f}},l_{jk}^{\stackrel{~}{f}},k_{jk}^{\stackrel{~}{f}}),`$
with $`\stackrel{~}{O}_{kl}^{{}_{}{}^{}L/R}=O_{kl}^{{}_{}{}^{}L/R}`$ for up-type sfermions (up-squarks and sneutrinos) and $`\stackrel{~}{O}_{kl}^{{}_{}{}^{}L/R}=O_{kl}^{{}_{}{}^{}R/L}`$ for down-type sfermions (down-squarks and sleptons),
$`(\delta a_f)_{ij}^{(v,SSS)}`$ $`=`$ $`{\displaystyle \frac{2i}{(4\pi )^2}}{\displaystyle \underset{k,m=1}{\overset{2}{}}}{\displaystyle \underset{l=3}{\overset{4}{}}}R_{k,l2}(\beta \alpha )G_{imk}^{\stackrel{~}{f}}G_{mjl}^{\stackrel{~}{f}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m}^2,m_{H_k^0}^2,m_{H_l^0}^2)`$ (106)
$`{\displaystyle \frac{2i}{(4\pi )^2}}{\displaystyle \underset{k=3}{\overset{4}{}}}{\displaystyle \underset{l,m=1}{\overset{2}{}}}R_{l,k2}(\beta \alpha )G_{imk}^{\stackrel{~}{f}}G_{mjl}^{\stackrel{~}{f}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m}^2,m_{H_k^0}^2,m_{H_l^0}^2)`$
$`{\displaystyle \frac{4}{(4\pi )^2}}I_f^{3L}\mathrm{cos}2\theta _W{\displaystyle \underset{k,m=1}{\overset{2}{}}}G_{imk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}G_{jmk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m^{}}^2,m_{H_k^+}^2,m_{H_k^+}^2)`$
$`{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \underset{k,m,n=1}{\overset{2}{}}}z_{mn}^{\stackrel{~}{f}}G_{imk}^{\stackrel{~}{f}}G_{njk}^{\stackrel{~}{f}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{H_k^0}^2,m_{\stackrel{~}{f}_m}^2,m_{\stackrel{~}{f}_n}^2)`$
$`{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \underset{k,m,n=1}{\overset{2}{}}}z_{mn}^{\stackrel{~}{f}^{}}G_{imk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}G_{jnk}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\left(C_0+C_1+C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{H_k^+}^2,m_{\stackrel{~}{f}_m^{}}^2,m_{\stackrel{~}{f}_n^{}}^2),`$
$`(\delta a_f)_{ij}^{(v,V\stackrel{~}{f}\stackrel{~}{f})}`$ $`=`$ $`{\displaystyle \frac{4}{(4\pi )^2}}e^2e_f^2z_{ij}^{\stackrel{~}{f}}๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(\lambda ,m_{\stackrel{~}{f}_i},m_{\stackrel{~}{f}_j})+{\displaystyle \frac{4}{(4\pi )^2}}g_Z^2{\displaystyle \underset{m,n=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}z_{mn}^{\stackrel{~}{f}}z_{nj}^{\stackrel{~}{f}}๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(m_Z,m_{\stackrel{~}{f}_m},m_{\stackrel{~}{f}_n})`$ (107)
$`+{\displaystyle \frac{4}{(4\pi )^2}}{\displaystyle \frac{g^2}{2}}R_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m,n=1}{\overset{2}{}}}R_{m1}^{\stackrel{~}{f}^{}}R_{n1}^{\stackrel{~}{f}^{}}๐ฎ_{ij}^{V\stackrel{~}{f}\stackrel{~}{f}}(m_W,m_{\stackrel{~}{f}_m^{}},m_{\stackrel{~}{f}_n^{}}),`$
$`(\delta a_f)_{ij}^{(v,SSV+SVS)}`$ $`=`$ (108)
$`{\displaystyle \frac{1}{(4\pi )^2}}\sqrt{2}gs_W^2m_W{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{i1}^{\stackrel{~}{f}}G_{jm2}^{\stackrel{~}{f}\stackrel{~}{f}^{}}+R_{j1}^{\stackrel{~}{f}}G_{im2}^{\stackrel{~}{f}\stackrel{~}{f}^{}}\right)R_{m1}^{\stackrel{~}{f}^{}}\left(C_0C_1C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m^{}}^2,m_W^2,m_W^2)`$
$`+{\displaystyle \frac{1}{(4\pi )^2}}\mathrm{\hspace{0.17em}2}g_Zm_Z{\displaystyle \underset{k,m=1}{\overset{2}{}}}R_{k_2}(\beta \alpha )z_{im}^{\stackrel{~}{f}}G_{mjk}^{\stackrel{~}{f}}\left(C_0C_1C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m}^2,m_Z^2,m_{H_k^0}^2)`$
$`+{\displaystyle \frac{1}{(4\pi )^2}}\mathrm{\hspace{0.17em}2}g_Zm_Z{\displaystyle \underset{k,m=1}{\overset{2}{}}}R_{k_2}(\beta \alpha )z_{mj}^{\stackrel{~}{f}}G_{imk}^{\stackrel{~}{f}}\left(C_0C_1C_2\right)(m_{\stackrel{~}{f}_i}^2,s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_m}^2,m_{H_k^0}^2,m_Z^2),`$
$`(\delta a_f)_{ij}^{(v,SVV)}`$ $`=`$ $`{\displaystyle \frac{4}{(4\pi )^2}}g^2c_W^2I_f^{3L}R_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2๐ฎ_{ij}^{SVV}(m_{\stackrel{~}{f}_m^{}},m_W,m_W),`$ (109)
$`(\delta a_f)_{ij}^{(v,SV)}`$ $`=`$ $`{\displaystyle \frac{4}{(4\pi )^2}}e^2e_f^2z_{ij}^{\stackrel{~}{f}}(2B_0+B_1)(m_{\stackrel{~}{f}_i}^2,\lambda ^2,m_{\stackrel{~}{f}_i}^2)+(ij)`$ (110)
$`{\displaystyle \frac{4}{(4\pi )^2}}g_Z^2{\displaystyle \underset{m=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}\stackrel{~}{z}_{jm}^{\stackrel{~}{f}}(2B_0+B_1)(m_{\stackrel{~}{f}_i}^2,m_Z^2,m_{\stackrel{~}{f}_m}^2)+(ij)`$
$`+{\displaystyle \frac{1}{(4\pi )^2}}e^2Y_L^fR_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2(2B_0+B_1)(m_{\stackrel{~}{f}_i}^2,m_W^2,m_{\stackrel{~}{f}_m^{}}^2)+(ij).`$
### A.3 Corrections to $`๐ธ๐^\mathbf{+}๐^{\mathbf{}}`$ vertex
In the following we list the analytic formulas of the vertex corrections to the electronโpositronโphoton vertex. We only give the right-handed coefficients of the generic vertex functions, $`_R^{(\mathrm{})}`$ as the coefficients $`_L^{(\mathrm{})}`$ can be obtained by exchanging the indices $`R`$ and $`L`$, i. e. $`_L^{(\mathrm{})}=_R^{(\mathrm{})}(RL)`$.
In the remaining vertex corrections we use the standard set of arguments for the whole class of $`C`$-functions $`CC(m_e^2,s,m_e^2,M_0^2,M_1^2,M_2^2)`$.
The vertex correction $`\delta e_{L,R}^{(v)}`$ is split into the following classes:
$$\delta e_{L,R}^{(v)}=\delta e_{L,R}^{(v,\stackrel{~}{f}\stackrel{~}{\chi }\stackrel{~}{\chi })}+\delta e_R^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}+\delta e_{L,R}^{(v,VFF)}+\delta e_{L,R}^{(v,FVV)}$$
(111)
The contribution from one sfermion and two gauginos in the loop is given by
$`\delta e_{L,R}^{(v,\stackrel{~}{f}\stackrel{~}{\chi }\stackrel{~}{\chi })}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}_{L,R}^{SFF}(m_{\stackrel{~}{\nu }_e},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_k^+};1,1,l_{1k}^{\stackrel{~}{\nu }_e},k_{1k}^{\stackrel{~}{\nu }_e},k_{1k}^{\stackrel{~}{\nu }_e},l_{1k}^{\stackrel{~}{\nu }_e}),`$ (112)
where we have used the generic vertex function
$`_R^{SFF}(M_0,M_1,M_2;g_0^R,g_0^L,g_1^R,g_1^L,g_2^R,g_2^L)=g_0^Lg_1^Lg_2^R\left(2C_{00}B_0(s,M_1^2,M_2^2)\right)`$
$`\left(g_0^Rh_1^{RL}h_2^{LR}g_0^Lg_1^Lg_2^RM_0^2\right)C_0\left(g_0^Rg_1^Rh_2^{LR}g_0^Lg_2^Rh_1^{LR}\right)m_eC_1+\left(g_0^Lg_1^Lh_2^{RL}g_0^Rg_2^Lh_1^{RL}\right)m_eC_2`$
and the abbreviations (no sum over $`i`$) $`h_i^{jk}=g_i^jm_e+g_i^kM_i`$ for $`i=1,2`$ and $`(j,k)=L,R`$.
The corrections due to the exchange of one gaugino and two sfermions are
$`\delta e_R^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}`$ $`=`$ $`{\displaystyle \frac{2}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{4}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(b_{mk}^{\stackrel{~}{e}}\right)^2C_{00}(m_e^2,s,m_e^2,m_{\stackrel{~}{\chi }_k^0}^2,m_{\stackrel{~}{e}_m}^2,m_{\stackrel{~}{e}_m}^2),`$ (114)
and $`\delta e_L^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}=\delta e_R^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}(b_{mk}^{\stackrel{~}{e}}a_{mk}^{\stackrel{~}{e}})`$.
Using the generic vertex function for one vector particle and two fermions in the loop,
$`_R^{VFF}(M_0,M_1,M_2;g_0^R,g_0^L,g_1^R,g_1^L,g_2^R,g_2^L)=2[g_0^Rg_1^Rg_2^R(2C_{00}B_0(s,M_1^2,M_2^2)m_e^2(C_1+C_2)`$
$`M_0^2C_0+(s2m_e^2)(C_0+C_1+C_2)+{\displaystyle \frac{r}{2}})+g_0^Lg_1^Rg_2^RM_1M_2C_0g_0^Lg_1^Lg_2^Lm_e^2(C_0+C_1+C_2)],`$
with $`r=0`$ in the $`\overline{\mathrm{DR}}`$ renormalization scheme we get the corrections stemming from one vector boson ($`\gamma ,Z^0`$) and two electrons given by
$`\delta e_{L,R}^{(v,VFF)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}e^2_{L,R}^{VFF}(\lambda ,m_e,m_e;1,1,1,1,1,1)`$ (116)
$`+{\displaystyle \frac{1}{(4\pi )^2}}g_Z^2_{L,R}^{VFF}(m_Z,m_e,m_e;1,1,C_R^e,C_L^e,C_R^e,C_L^e).`$
For the graphs with one electron-neutrino and two $`W`$-bosons we obtain
$`\delta e_{L,R}^{(v,FVV)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g^2}{2}}_{L,R}^{FVV}(0,m_W,m_W;1,0,1,0,1),`$ (117)
where we have used the function
$`_R^{FVV}(M_0,M_1,M_2;g_0,g_1^R,g_1^L,g_2^R,g_2^L)=g_0[g_1^Rg_2^R(2B_0(s,M_1^2,M_2^2)r+4C_{00}+2M_0^2C_0`$ (118)
$`+(5m_e^22s)(C_1+C_2))+3(g_1^Lg_2^R+g_1^Rg_2^L)m_eM_0C_0+3g_1^Lg_2^Lm_e^2(C_1+C_2)].`$
### A.4 Corrections to $`๐^\mathrm{๐}๐^\mathbf{+}๐^{\mathbf{}}`$ vertex
In the following we list the single contributions to the electronโpositronโ$`Z^0`$ vertex. The generic vertex functions used in this section can be looked up in Appendix A.3.
$`\delta a_{L,R}^{(v,\stackrel{~}{f}\stackrel{~}{\chi }\stackrel{~}{\chi })}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k,l=1}{\overset{4}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}_{L,R}^{SFF}(m_{\stackrel{~}{e}_m},m_{\stackrel{~}{\chi }_k^0},m_{\stackrel{~}{\chi }_l^0};O_{kl}^{{}_{}{}^{\prime \prime }R},O_{kl}^{{}_{}{}^{\prime \prime }L},a_{mk}^{\stackrel{~}{e}},b_{mk}^{\stackrel{~}{e}},b_{ml}^{\stackrel{~}{e}},a_{ml}^{\stackrel{~}{e}})`$ (119)
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k,l=1}{\overset{2}{}}}_{L,R}^{SFF}(m_{\stackrel{~}{\nu }_e},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_l^+};O_{kl}^{}_{}{}^{}L,O_{kl}^{}_{}{}^{}R,l_{1k}^{\stackrel{~}{\nu }_e},k_{1k}^{\stackrel{~}{\nu }_e},k_{1l}^{\stackrel{~}{\nu }_e},l_{1l}^{\stackrel{~}{\nu }_e}),`$
$`\delta a_R^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}`$ $`=`$ $`{\displaystyle \frac{2}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{4}{}}}{\displaystyle \underset{m,n=1}{\overset{2}{}}}z_{mn}^{\stackrel{~}{f}}b_{mk}^{\stackrel{~}{e}}b_{nk}^{\stackrel{~}{e}}C_{00}(m_e^2,s,m_e^2,m_{\stackrel{~}{\chi }_k^0}^2,m_{\stackrel{~}{e}_m}^2,m_{\stackrel{~}{e}_n}^2)`$ (120)
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}\left(k_{1k}^{\stackrel{~}{\nu }_e}\right)^2C_{00}(m_e^2,s,m_e^2,m_{\stackrel{~}{\chi }_k^0}^2,m_{\stackrel{~}{\nu }_e}^2,m_{\stackrel{~}{\nu }_e}^2),`$
and $`\delta a_L^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}=\delta a_R^{(v,\stackrel{~}{\chi }\stackrel{~}{f}\stackrel{~}{f})}(b_{\mathrm{}}^{\stackrel{~}{e}}a_{\mathrm{}}^{\stackrel{~}{e}},k_{\mathrm{}}^{\stackrel{~}{\nu }_e}l_{\mathrm{}}^{\stackrel{~}{\nu }_e})`$.
$`\delta a_{L,R}^{(v,VFF)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}e^2_{L,R}^{VFF}(\lambda ,m_e,m_e;C_R^e,C_L^e,1,1,1,1)`$ (121)
$`+{\displaystyle \frac{1}{(4\pi )^2}}g_Z^2_{L,R}^{VFF}(m_Z,m_e,m_e;C_R^e,C_L^e,C_R^e,C_L^e,C_R^e,C_L^e)`$
$`+{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g^2}{4}}_{L,R}^{VFF}(m_W,0,0;0,1,0,1,0,1),`$
$`\delta a_{L,R}^{(v,FVV)}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{g^2c_W^2}{2}}_{L,R}^{FVV}(0,m_W,m_W;1,0,1,0,1).`$ (122)
## Appendix B Box contributions
In this section we give the explicit form of the radiative corrections which stem from box diagrams with two different topologies. The matrix element is parameterized as
$`_{\mathrm{box}}`$ $`=`$ $`{\displaystyle \frac{i}{(4\pi )^2}}\overline{v}(p_2)\left[A_LP_L+A_RP_R+B_L{\displaystyle \frac{1}{2}}(\overline{)}k_1\overline{)}k_2)P_L+B_R{\displaystyle \frac{1}{2}}(\overline{)}k_1\overline{)}k_2)P_R\right]u(p_1),`$ (123)
where the form-factors $`A_{L,R}`$ do not contribute to the squared matrix element. The form-factors $`B_{L,R}`$ depend on the Mandelstam variables $`t`$ and $`u`$ which are defined as
$`t`$ $`=`$ $`{\displaystyle \frac{1}{2}}(m_{\stackrel{~}{f}_i}^2+m_{\stackrel{~}{f}_j}^2s)+{\displaystyle \frac{1}{2}}\kappa (s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)\mathrm{cos}\vartheta ,`$ (124)
$`u`$ $`=`$ $`2m_e^2+m_{\stackrel{~}{f}_i}^2+m_{\stackrel{~}{f}_j}^2st.`$ (125)
The single contributions to the form-factors
$`B_{L,R}`$ $`=`$ $`B_{L,R}^{\gamma \gamma }+B_{L,R}^{\gamma Z}+B_{L,R}^{ZZ}+B_{L,R}^{WW}+B_{L,R}^{\stackrel{~}{\chi }^+\stackrel{~}{\chi }^+}+B_{L,R}^{\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0}`$ (126)
correspond to the diagrams with two vector bosons, where the particles in the loop are indicated by a superscript, and similarly $`B_{L,R}^{\stackrel{~}{\chi }^+\stackrel{~}{\chi }^+}`$ and $`B_{L,R}^{\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0}`$ denote the contributions from charginos and neutralinos, respectively.
### B.1 Vector bosons in the loop
In the case of two vector bosons, we use the generic functions
$`^{VV}(M_0,M_1,M_2,M_3)`$ $`=`$ $`4C_0+C_2+4M_0^2D_0+M_0^2D_3+4\left[(m_{\stackrel{~}{f}_j}^2u)D_1sD_1+(tm_{\stackrel{~}{f}_j}^2)D_2+tD_3\right]`$
for a diagram with vector bosons $`V`$ in the loop and
$`^{VVx}(M_0,M_1,M_2,M_3)`$ $`=`$ $`\left(4C_0^x+C_1^x+4M_0^2D_0^x+4(um_{\stackrel{~}{f}_j}^2)D_1^x+4(um_{\stackrel{~}{f}_i}^2)D_2^x+(M_0^2+4u)D_3^x\right)`$
for the corresponding crossed diagram. In case of a $`W`$-boson there is no crossed diagram and we use $`^{VV}`$ or $`^{VVx}`$ depending on the charge of the final state particle.
The scalar three-point and four-point functions used above have a standard set of arguments defined as
$`\begin{array}{cccccc}\hfill C_i& =\hfill & C_i(s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_i}^2,M_1^2,M_2^2,M_3^2),\hfill & \hfill D_i& =\hfill & D_i(m_e^2,s,m_{\stackrel{~}{f}_j}^2,t,m_e^2,m_{\stackrel{~}{f}_i}^2,M_0^2,M_1^2,M_2^2,M_3^2),\hfill \\ \hfill C_i^x& =\hfill & C_i(m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_i}^2,s,M_1^2,M_3^2,M_2^2),\hfill & \hfill D_i^x& =\hfill & D_i(m_e^2,s,m_{\stackrel{~}{f}_i}^2,u,m_e^2,m_{\stackrel{~}{f}_j}^2,M_0^2,M_1^2,M_2^2,M_3^2).\hfill \end{array}`$ (131)
The contributions from 2 photons, one photon and one $`Z`$-boson, 2 $`Z`$-bosons and 2 $`W`$-bosons are
$`B_{L,R}^{\gamma \gamma }`$ $`=`$ $`e^4e_f^2\delta _{ij}\left(^{VV}+^{VVx}\right)(m_e,0,0,m_{\stackrel{~}{f}_i}),`$ (132)
$`B_{L,R}^{\gamma Z}`$ $`=`$ $`e^2e_fg_Z^2C_{L,R}{\displaystyle \underset{m=1}{\overset{2}{}}}\delta _{im}z_{mj}^{\stackrel{~}{f}}\left[^{VV}(m_e,0,m_Z,m_{\stackrel{~}{f}_m})+^{VVx}(m_e,m_Z,0,m_{\stackrel{~}{f}_m})\right]`$ (133)
$`e^2e_fg_Z^2C_{L,R}{\displaystyle \underset{m=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}\delta _{mj}\left[^{VV}(m_e,m_Z,0,m_{\stackrel{~}{f}_m})+^{VVx}(m_e,0,m_Z,m_{\stackrel{~}{f}_m})\right],`$
$`B_{L,R}^{ZZ}`$ $`=`$ $`g_Z^4C_{L,R}^2{\displaystyle \underset{m=1}{\overset{2}{}}}z_{im}^{\stackrel{~}{f}}z_{mj}^{\stackrel{~}{f}}\left(^{VV}+^{VVx}\right)(m_e,m_Z,m_Z,m_{\stackrel{~}{f}_m}),`$ (134)
$`B_L^{WW}`$ $`=`$ $`{\displaystyle \frac{g^4}{4}}R_{i1}^{\stackrel{~}{f}}R_{j1}^{\stackrel{~}{f}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left(R_{m1}^{\stackrel{~}{f}^{}}\right)^2^{VV(x)}(m_e,m_W,m_W,m_{\stackrel{~}{f}_m^{}}),B_R^{WW}=0,`$ (135)
where $`^{VV(x)}`$ denotes $`^{VVx}`$ for up-type sfermions and $`^{VV}`$ for down-type sfermions in the final state.
### B.2 Scalars and fermions in the loop
In analogy to the case with vector bosons in the loop we define the following generic function for box diagrams with fermions and sfermions in the loop
$`_L^{FF}(M_0,M_1,M_2,M_3;g_0^R,g_0^L,g_1^R,g_1^L,g_2^R,g_2^L,g_3^R,g_3^L)=`$
$`g_0^Lg_1^Rg_2^Lg_3^R\left(C_0(s,m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_i}^2,M_1^2,M_2^2,M_3^2)+M_0^2D_0(M_3^2t)D_3\right)`$
$`M_1M_2g_0^Lg_1^Lg_2^Rg_3^R(D_0+D_3)M_1M_3g_0^Lg_1^Lg_2^Lg_3^RD_3M_2M_3g_0^Lg_1^Rg_2^Rg_3^RD_3,`$ (136)
and for the crossed counterpart
$`_L^{FFx}(M_0,M_1,M_2,M_3;g_0^R,g_0^L,g_1^R,g_1^L,g_2^R,g_2^L,g_3^R,g_3^L)=`$
$`g_0^Lg_1^Rg_2^Lg_3^R\left(C_0(m_{\stackrel{~}{f}_j}^2,m_{\stackrel{~}{f}_i}^2,s,M_1^2,M_3^2,M_2^2)+M_0^2D_0^x(M_3^2u)D_3^x\right)`$
$`M_1M_2g_0^Lg_1^Lg_2^Rg_3^R(D_0^x+D_3^x)M_1M_3g_0^Lg_1^Lg_2^Lg_3^RD_3^xM_2M_3g_0^Lg_1^Rg_2^Rg_3^RD_3^x,`$ (137)
with $`_R^{FF(x)}=_L^{FF(x)}(LR)`$.
As in the case of two $`W`$-bosons in the loop, for the graphs with charginos we use either
$`_{L,R}^{\stackrel{~}{\chi }^+\stackrel{~}{\chi }^+}`$ $`=`$ $`{\displaystyle \underset{k,l=1}{\overset{2}{}}}_{L,R}^{FFx}(m_{\stackrel{~}{\nu }_e},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_l^+},m_f^{};k_{1k}^{\stackrel{~}{\nu }_e},l_{1k}^{\stackrel{~}{\nu }_e},l_{jk}^{\stackrel{~}{f}},k_{jk}^{\stackrel{~}{f}},k_{il}^{\stackrel{~}{f}},l_{il}^{\stackrel{~}{f}},l_{1l}^{\stackrel{~}{\nu }_e},k_{1l}^{\stackrel{~}{\nu }_e})`$ (138)
for up-type sfermions or
$`_{L,R}^{\stackrel{~}{\chi }^+\stackrel{~}{\chi }^+}`$ $`=`$ $`{\displaystyle \underset{k,l=1}{\overset{2}{}}}_{L,R}^{FF}(m_{\stackrel{~}{\nu }_e},m_{\stackrel{~}{\chi }_k^+},m_{\stackrel{~}{\chi }_l^+},m_f^{};k_{1k}^{\stackrel{~}{\nu }_e},l_{1k}^{\stackrel{~}{\nu }_e},k_{ik}^{\stackrel{~}{f}},l_{ik}^{\stackrel{~}{f}},l_{jl}^{\stackrel{~}{f}},k_{jl}^{\stackrel{~}{f}},l_{1l}^{\stackrel{~}{\nu }_e},k_{1l}^{\stackrel{~}{\nu }_e})`$ (139)
for down-type sfermions in the final state.
The contributions from 2 neutralinos in the loop have the following explicit form:
$`_{L,R}^{\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0}`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{4}{}}}{\displaystyle \underset{l=1}{\overset{4}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}_{L,R}^{FF}(m_{\stackrel{~}{e}},m_{\stackrel{~}{\chi }_k^0},m_{\stackrel{~}{\chi }_l^0},m_f;b_{mk}^{\stackrel{~}{e}},a_{mk}^{\stackrel{~}{e}},b_{ik}^{\stackrel{~}{f}},a_{ik}^{\stackrel{~}{f}},a_{jl}^{\stackrel{~}{f}},b_{jl}^{\stackrel{~}{f}},a_{ml}^{\stackrel{~}{\nu }_e},b_{ml}^{\stackrel{~}{\nu }_e})`$
$`+`$ $`{\displaystyle \underset{k=1}{\overset{4}{}}}{\displaystyle \underset{l=1}{\overset{4}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}_{L,R}^{FFx}(m_{\stackrel{~}{e}},m_{\stackrel{~}{\chi }_k^0},m_{\stackrel{~}{\chi }_l^0},m_f;b_{mk}^{\stackrel{~}{e}},a_{mk}^{\stackrel{~}{e}},a_{jk}^{\stackrel{~}{f}},b_{jk}^{\stackrel{~}{f}},b_{il}^{\stackrel{~}{f}},a_{il}^{\stackrel{~}{f}},a_{ml}^{\stackrel{~}{e}},b_{ml}^{\stackrel{~}{e}})`$ (140)
## Appendix C Self-energies
Here we give the explicit form of the self-energies needed for the computation of some wave-function renormalization constants and various counterterms. We omit the sfermion self-energies already given in chrislet . All fermion, sfermion and vector self-energy diagrams are shown in Figs. 3 and 13.
### C.1 Fermion self-energies
In our notation, the fermion self-energy is defined as
with
$`\mathrm{\Pi }(k)`$ $`=`$ $`\overline{)}kP_L\mathrm{\Pi }^L(k^2)+\overline{)}kP_R\mathrm{\Pi }^R(k^2)+\mathrm{\Pi }^{SL}(k^2)P_L+\mathrm{\Pi }^{SR}(k^2)P_R.`$ (141)
Below we list the contributions to the left- and right-handed parts $`\mathrm{\Pi }^{L,R}`$ and $`\mathrm{\Pi }^{SL,SR}`$ from the single diagrams. The form-factor $`\mathrm{\Pi }`$ is defined as a sum of the contributions coming from the diagrams in Fig. 13.
$`\mathrm{\Pi }=\mathrm{\Pi }^{eH_k^0}+\mathrm{\Pi }^{\nu _eH_k^+}+\mathrm{\Pi }^{\stackrel{~}{e}\stackrel{~}{\chi }^0}+\mathrm{\Pi }^{\stackrel{~}{\nu }_e\stackrel{~}{\chi }^+}+\mathrm{\Pi }^{e\gamma }+\mathrm{\Pi }^{eZ^0}+\mathrm{\Pi }^{\nu _eW^+}.`$ (142)
We give the full formulas for the electron self-energy without neglecting the electron mass (although it is being neglected in the actual calculation).
Note that for quarks and leptons (contrary to charginos), the left- and right-handed scalar parts of $`\mathrm{\Pi }(k)`$ are equal, i. e. $`\mathrm{\Pi }^{SL}(k)=\mathrm{\Pi }^{SR}(k)`$.
$`\mathrm{\Pi }^{eH_k^0}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\left[\overline{)}k\left({\displaystyle \underset{l=1}{\overset{2}{}}}(s_l^e)^2B_1+{\displaystyle \underset{l=3}{\overset{4}{}}}(s_l^e)^2B_1\right)+{\displaystyle \underset{l=1}{\overset{4}{}}}(s_l^e)^2m_eB_0\right](k^2,m_e^2,m_{H_l^0}^2),`$ (143)
$`\mathrm{\Pi }^{\nu _eH_k^+}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\overline{)}kP_R{\displaystyle \underset{l=1}{\overset{2}{}}}(y_l^e)^2B_1(k^2,0,m_{H_l^+}^2),`$ (144)
$`\mathrm{\Pi }^{\stackrel{~}{e}\stackrel{~}{\chi }^0}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{l=1}{\overset{4}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left[\overline{)}kP_L(a_{ml}^{\stackrel{~}{e}})^2B_1+\overline{)}kP_R(b_{ml}^{\stackrel{~}{e}})^2B_1m_{\stackrel{~}{\chi }_l^0}a_{ml}^{\stackrel{~}{e}}b_{ml}^{\stackrel{~}{e}}B_0\right](k^2,m_{\stackrel{~}{\chi }_l^0}^2,m_{\stackrel{~}{e}_m}^2),`$
$`\mathrm{\Pi }^{\stackrel{~}{\nu }_e\stackrel{~}{\chi }^+}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{l=1}{\overset{2}{}}}{\displaystyle \underset{m=1}{\overset{2}{}}}\left[\overline{)}kP_L(l_{ml}^{\stackrel{~}{\nu }_e})^2B_1+\overline{)}kP_R(k_{ml}^{\stackrel{~}{\nu }_e})^2B_1m_{\stackrel{~}{\chi }_l^+}k_{ml}^{\stackrel{~}{\nu }_e}l_{ml}^{\stackrel{~}{\nu }_e}B_0\right](k^2,m_{\stackrel{~}{\chi }_l^+}^2,m_{\stackrel{~}{\nu }_e}^2),`$
$`\mathrm{\Pi }^{e\gamma }(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\left[\overline{)}k\mathrm{\hspace{0.17em}2}e^2\left(B_1(k^2,m_e^2,\lambda ^2)+{\displaystyle \frac{r}{2}}\right)+4e^2m_e\left(B_0(k^2,m_e^2,\lambda ^2){\displaystyle \frac{r}{2}}\right)\right],`$ (147)
$`\mathrm{\Pi }^{eZ^0}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\left[\overline{)}k\mathrm{\hspace{0.17em}2}g_Z^2(C_{L,R}^e)^2\left(B_1(k^2,m_e^2,\lambda ^2)+{\displaystyle \frac{r}{2}}\right)+4g_Z^2m_eC_L^eC_R^e\left(B_0(k^2,m_e^2,\lambda ^2){\displaystyle \frac{r}{2}}\right)\right],`$
$`\mathrm{\Pi }^{\nu _eW^+}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\left[\overline{)}kP_Lg^2\left(B_1(k^2,m_e^2,\lambda ^2)+{\displaystyle \frac{r}{2}}\right)\right].`$ (149)
### C.2 Vector self-energies
Here we give the explicit form of the general gauge boson self-energies (the transverse parts only) which are then applied to the cases of the photon and the $`Z`$-boson (and their mixing). The corresponding couplings are given in a table after each generic formula. We do not list the contributions to the counterterms of the $`Z`$\- and $`W`$-bosons as they can be found in chrislet .
The self-energy of a vector boson is defined as follows
The transverse part of the self-energy consists of the following parts:
$`\mathrm{\Pi }_{VV}^T`$ $`=`$ $`\left(\mathrm{\Pi }_{VV}^T\right)^{ff}+\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0}+\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{\chi }^\pm \stackrel{~}{\chi }^{}}+\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{f}\stackrel{~}{f}}+\left(\mathrm{\Pi }_{VV}^T\right)^{H^\pm H^\pm }+\left(\mathrm{\Pi }_{VV}^T\right)^{H^0H^0}`$ (150)
$`+\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{f}}+\left(\mathrm{\Pi }_{VV}^T\right)^{H^0}+\left(\mathrm{\Pi }_{VV}^T\right)^{H^\pm }+\left(\mathrm{\Pi }_{VV}^T\right)^{Z^0H^0}+\left(\mathrm{\Pi }_{VV}^T\right)^{W^\pm H^{}}+\left(\mathrm{\Pi }_{VV}^T\right)^{W^\pm W^{}}`$
$`+\left(\mathrm{\Pi }_{VV}^T\right)^{W^\pm }+\left(\mathrm{\Pi }_{VV}^T\right)^{2\mathrm{F}\mathrm{P}\mathrm{g}\mathrm{h}\mathrm{o}\mathrm{s}\mathrm{t}\mathrm{s}}`$
For the contributions with a fermion loop we define the following generic function:
$`๐_{VV}(k^2,m_1,m_2;g_1^L,g_1^R,g_2^L,g_2^R)`$ $`=`$ $`2[(g_1^Lg_2^L+g_1^Rg_2^R)(k^2B_1+m_1^2B_0+A_0(m_2^2)2B_{00})`$ (151)
$`+(g_1^Rg_2^L+g_1^Lg_2^R)m_1m_2B_0\left]\right(k^2,m_1^2,m_2^2)`$
Using the generic function we can write the 2 fermion, 2 neutralino and the 2 chargino contributions as
$`\left(\mathrm{\Pi }_{VV}^T\right)^{ff}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{f}{}}N_C^f๐_{VV}(k^2,m_f,m_f;g_1^L,g_1^R,g_2^L,g_2^R),`$ (152)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{\chi }^0\stackrel{~}{\chi }^0}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \frac{1}{2}}{\displaystyle \underset{k,l=1}{\overset{4}{}}}๐_{VV}(k^2,m_{\stackrel{~}{\chi }_l^0},m_{\stackrel{~}{\chi }_k^0};g_1^L,g_1^R,g_2^L,g_2^R),`$ (153)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{\chi }^\pm \stackrel{~}{\chi }^{}}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k,l=1}{\overset{2}{}}}๐_{VV}(k^2,m_{\stackrel{~}{\chi }_l^+},m_{\stackrel{~}{\chi }_k^+};g_1^L,g_1^R,g_2^L,g_2^R).`$ (154)
The next set of contributions are the ones with 2 scalar particles in the loop. For this set we introduce
$`_{VV}(k^2,m_1^2,m_2^2;g_1,g_2)`$ $`=`$ $`4g_1g_2B_{00}(k^2,m_1^2,m_2^2)`$ (155)
and get for the sfermion, neutral and charged Higgs in the loop the following forms:
$`\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{f}\stackrel{~}{f}}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{f}{}}N_C^f{\displaystyle \underset{m,n=1}{\overset{2}{}}}_{VV}(k^2,m_{\stackrel{~}{f}_n}^2,m_{\stackrel{~}{f}_m}^2;g_1,g_2)`$ (156)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{H_k^0H_l^0}(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}{\displaystyle \underset{l=3}{\overset{4}{}}}_{VV}(k^2,m_{H_k^0}^2,m_{H_l^0}^2;g_1,g_2)`$ (157)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{H_k^\pm H_k^\pm }(k^2)`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}_{VV}(k^2,m_{H_k^+}^2,m_{H_k^+}^2;g_1,g_2)`$ (158)
The next class are the self-energies with a single scalar particle in the loop for which we use the generic form
$`๐_{VV}(m^2;g_1)`$ $`=`$ $`g_1A_0\left(m^2\right).`$ (159)
The diagrams with 1 sfermion, 1 neutral or charged boson can be written as
$`\left(\mathrm{\Pi }_{VV}^T\right)^{\stackrel{~}{f}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{f}{}}N_C^f{\displaystyle \underset{m=1}{\overset{2}{}}}๐_{VV}(m_{\stackrel{~}{f}_m}^2;g_1),`$ (160)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{H_k^0}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{4}{}}}๐_{VV}(m_{H_k^0}^2;g_1),`$ (161)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{H_k^+}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}๐_{VV}(m_{H_k^+}^2;g_1).`$ (162)
The diagrams with a vector and a scalar particle in the loop use the simple generic form
$`๐_{VV}(k^2,m_1^2,m_2^2;g_1,g_2)`$ $`=`$ $`g_1g_2B_0(k^2,m_1^2,m_2^2)`$ (163)
and give
$`\left(\mathrm{\Pi }_{VV}^T\right)^{H_k^0Z}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}{\displaystyle \underset{k=1}{\overset{2}{}}}๐_{VV}(k^2,m_{H_k^0}^2,m_Z^2;g_1,g_2),`$ (164)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{G^\pm W^{}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}๐_{VV}(k^2,m_{G^+}^2,m_W^2;g_1,g_2).`$ (165)
The remaining 3 contributions comprising of 2 $`W`$-bosons, 2 FP ghosts and a single $`W`$-boson in the loop have the following explicit forms:
$`\left(\mathrm{\Pi }_{VV}^T\right)^{W^+W^{}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}g_1g_2[10B_{00}+5k^2B_0+2k^2B_1+2A_0\left(m_W^2\right)`$ (166)
$`+2m_W^2B_0+r({\displaystyle \frac{2}{3}}k^24m_W^2)\left]\right(k^2,m_W^2,m_W^2),`$
$`\left(\mathrm{\Pi }_{VV}^T\right)^{W^\pm }`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}g_1\left[3A_0\left(m_W^2\right)2rm_W^2\right],`$ (167)
$`\left(\mathrm{\Pi }_{VV}^T\right)^{2\mathrm{F}\mathrm{P}\mathrm{g}\mathrm{h}\mathrm{o}\mathrm{s}\mathrm{t}\mathrm{s}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}g_1g_2B_{00}(k^2,m_W^2,m_W^2).`$ (168)
## Appendix D Bremsstrahlung integrals
### D.1 Soft photon integral
Using the kinematics of the process we get for $`\delta _s`$ the explicit form
$`\delta _s`$ $`=`$ $`{\displaystyle \frac{\alpha }{\pi }}\{(1+\mathrm{log}{\displaystyle \frac{m_e^2}{s}})\mathrm{log}{\displaystyle \frac{4(\mathrm{\Delta }E)^2}{\lambda ^2}}+\mathrm{log}{\displaystyle \frac{m_e^2}{s}}+{\displaystyle \frac{1}{2}}\mathrm{log}^2{\displaystyle \frac{m_e^2}{s}}+{\displaystyle \frac{\pi ^2}{3}}`$
$`+e_f^2[{\displaystyle \frac{sm_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2}{\kappa }}({\displaystyle \frac{1}{2}}\mathrm{log}{\displaystyle \frac{d_i^{}}{d_i^+}}{\displaystyle \frac{d_j^{}}{d_j^+}}\mathrm{log}{\displaystyle \frac{4(\mathrm{\Delta }E)^2}{\lambda ^2}}+{\displaystyle \frac{1}{4}}\mathrm{log}^2{\displaystyle \frac{d_i^{}}{d_i^+}}+{\displaystyle \frac{1}{4}}\mathrm{log}^2{\displaystyle \frac{d_j^{}}{d_j^+}}+\mathrm{Li}_2{\displaystyle \frac{2\kappa }{d_i^+}}+\mathrm{Li}_2{\displaystyle \frac{2\kappa }{d_j^+}})`$
$`+\mathrm{log}{\displaystyle \frac{4(\mathrm{\Delta }E)^2}{\lambda ^2}}+{\displaystyle \frac{c_i}{2\kappa }}\mathrm{log}{\displaystyle \frac{d_i^{}}{d_i^+}}+{\displaystyle \frac{c_j}{2\kappa }}\mathrm{log}{\displaystyle \frac{d_j^{}}{d_j^+}}]+e_f[\mathrm{log}{\displaystyle \frac{(m_{\stackrel{~}{f}_i}^2t)(m_{\stackrel{~}{f}_j}^2t)}{(m_{\stackrel{~}{f}_i}^2u)(m_{\stackrel{~}{f}_j}^2u)}}\mathrm{log}{\displaystyle \frac{4(\mathrm{\Delta }E)^2}{\lambda ^2}}`$
$`+\mathrm{Li}_2\left(1{\displaystyle \frac{d_i^+}{2(m_{\stackrel{~}{f}_i}^2u)}}\right)+\mathrm{Li}_2\left(1{\displaystyle \frac{d_i^{}}{2(m_{\stackrel{~}{f}_i}^2u)}}\right)+\mathrm{Li}_2\left(1{\displaystyle \frac{d_j^+}{2(m_{\stackrel{~}{f}_j}^2u)}}\right)+\mathrm{Li}_2\left(1{\displaystyle \frac{d_j^{}}{2(m_{\stackrel{~}{f}_j}^2u)}}\right)`$
$`\mathrm{Li}_2(1{\displaystyle \frac{d_i^+}{2(m_{\stackrel{~}{f}_i}^2t)}})\mathrm{Li}_2(1{\displaystyle \frac{d_i^{}}{2(m_{\stackrel{~}{f}_i}^2t)}})\mathrm{Li}_2(1{\displaystyle \frac{d_j^+}{2(m_{\stackrel{~}{f}_j}^2t)}})\mathrm{Li}_2(1{\displaystyle \frac{d_j^{}}{2(m_{\stackrel{~}{f}_j}^2t)}})]\},`$
where $`s,t`$ and $`u`$ are the Mandelstam variables, $`d_i^\pm =c_i\pm \kappa `$ with $`c_i`$ and $`\kappa `$ defined as
$`\begin{array}{ccc}& \kappa =\kappa (s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)=\sqrt{\lambda (s,m_{\stackrel{~}{f}_i}^2,m_{\stackrel{~}{f}_j}^2)}=\sqrt{(sm_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2)^24m_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2},& \\ & c_i=s+m_{\stackrel{~}{f}_i}^2m_{\stackrel{~}{f}_j}^2,c_j=s+m_{\stackrel{~}{f}_j}^2m_{\stackrel{~}{f}_i}^2.& \end{array}`$ (172)
### D.2 Hard photon integrals
The squared matrix element for the hard photon radiation can be split into 3 parts,
$$|^{\mathrm{hard}}|^2=|^\mathrm{e}|^2+|^{\stackrel{~}{\mathrm{f}}}|^2+2\mathrm{}(^{\stackrel{~}{\mathrm{f}}}^\mathrm{e}),$$
(173)
where $`|^\mathrm{X}|^2`$ stands for the part of the amplitude where the photon is radiated off the particle indicated.
The squared matrix part corresponding to the photon being radiated from the electron or positron has the form
$$|^\mathrm{e}|^2=\frac{1}{4}(1P_{})(1+P_+)|^\mathrm{e}|_L^2+\frac{1}{4}(1+P_{})(1P_+)|^\mathrm{e}|_R^2.$$
(174)
The chiral $`L,R`$ parts are
$`|^\mathrm{e}|_{L,R}^2`$ $`=`$ $`N_Ce^2[(T_\mathrm{e}^{\gamma \gamma })_{L,R}+(T_\mathrm{e}^{\gamma Z})_{L,R}+(T_\mathrm{e}^{ZZ})_{L,R}](_1+_1(p_1p_2)+_2),`$ (175)
where
$`(T_\mathrm{e}^{\gamma \gamma })_{L,R}`$ $`=`$ $`{\displaystyle \frac{e^4e_f^2(\delta _{ij})^2}{s_{\mathrm{red}}^2}}K_{L,R}^2,`$ (176)
$`(T_\mathrm{e}^{\gamma Z})_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}}{2s_{\mathrm{red}}(s_{\mathrm{red}}m_Z^2)}}C_{L,R}K_{L,R},`$ (177)
$`(T_\mathrm{e}^{ZZ})_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^4(a_{ij}^{\stackrel{~}{f}})^2}{16(s_{\mathrm{red}}m_Z^2)^2}}C_{L,R}^2,`$ (178)
with $`s_{\mathrm{red}}=(p_1+p_2k_3)^2`$.
The functions $`_1`$ and $`_2`$ contain only scalar products of the external momenta and are defined as
$`_1`$ $`=`$ $`{\displaystyle \frac{1}{(p_1.k_3)}}[4(k_1.k_3)(k_1.p_2)+4(k_1.p_2)(k_2.k_3)+4(k_1.k_3)(k_2.p_2)4(k_2.k_3)(k_2.p_2)`$ (179)
$`+2m_{\stackrel{~}{f}_i}^2(k_3.p_2)+2m_{\stackrel{~}{f}_j}^2(k_3.p_2)4(k_1.k_2)(k_3.p_2)],`$
$`_2`$ $`=`$ $`{\displaystyle \frac{(p_1.p_2)}{(p_1.k_3)(p_2.k_3)}}[2(k_1.k_3)(k_1.p_1)+2(k_1.k_3)(k_1.p_2)4(k_1.p_1)(k_1.p_2)2(k_1.p_1)(k_2.k_3)`$ (180)
$`2(k_1.p_2)(k_2.k_3)2(k_1.k_3)(k_2.p_1)+4(k_1.p_2)(k_2.p_1)+2(k_2.k_3)(k_2.p_1)`$
$`2(k_1.k_3)(k_2.p_2)+4(k_1.p_1)(k_2.p_2)+2(k_2.k_3)(k_2.p_2)4(k_2.p_1)(k_2.p_2)+2m_{\stackrel{~}{f}_i}^2(p_1.p_2)`$
$`+2m_{\stackrel{~}{f}_j}^2(p_1.p_2)4(k_1.k_2)(p_1.p_2)]{\displaystyle \frac{1}{(p_1.k_3)}}[2(k_1.p_1)^2+2(k_1.p_1)(k_1.p_2)`$
$`+4(k_1.p_1)(k_2.p_1)2(k_1.p_2)(k_2.p_1)2(k_2.p_1)^22(k_1.p_1)(k_2.p_2)+2(k_2.p_1)(k_2.p_2)`$
$`2m_{\stackrel{~}{f}_i}^2(p_1.p_2)2m_{\stackrel{~}{f}_j}^2(p_1.p_2)+4(k_1.k_2)(p_1.p_2)]{\displaystyle \frac{1}{(p_2.k_3)}}[2(k_1.p_2)^2`$
$`+2(k_1.p_2)(k_1.p_1)+4(k_1.p_2)(k_2.p_2)2(k_1.p_1)(k_2.p_2)2(k_2.p_2)^22(k_1.p_2)(k_2.p_1)`$
$`+2(k_2.p_2)(k_2.p_1)2m_{\stackrel{~}{f}_i}^2(p_1.p_2)2m_{\stackrel{~}{f}_j}^2(p_1.p_2)+4(k_1.k_2)(p_1.p_2)].`$
The radiation off the sfermion can be written as
$$|^{\stackrel{~}{\mathrm{f}}}|^2=\frac{1}{4}(1P_{})(1+P_+)|^{\stackrel{~}{\mathrm{f}}}|_L^2+\frac{1}{4}(1+P_{})(1P_+)|^{\stackrel{~}{\mathrm{f}}}|_R^2,$$
(181)
where
$`|^{\stackrel{~}{\mathrm{f}}}|_{L,R}^2=N_Ce^2e_f^2\mathrm{\hspace{0.17em}2}s\left(T_{L,R}^{\gamma \gamma }+T_{L,R}^{\gamma Z}+T_{L,R}^{ZZ}\right)\left(g^{\mu \nu }{\displaystyle \frac{2}{s}}(p_1^\mu p_2^\nu +p_1^\nu p_2^\mu )\right)T_{\mu \rho }T_\nu ^\rho .`$ (182)
The tensor $`T_{\mu \nu }`$ is defined as
$$T_{\mu \nu }=\frac{1}{2k_1.k_3}(k_1k_2+k_3)_\mu (2k_1+k_3)_\nu \frac{1}{2k_2.k_3}(k_1k_2k_3)_\mu (2k_2+k_3)_\nu 2g_{\mu \nu }.$$
(183)
The interference term of the squared hard photon amplitude is
$`2\mathrm{}(^{\stackrel{~}{\mathrm{f}}}^\mathrm{e})`$ $`=`$ $`{\displaystyle \frac{1}{4}}(1P_{})(1+P_+)\mathrm{\hspace{0.17em}2}\mathrm{}(^{\stackrel{~}{\mathrm{f}}}^\mathrm{e})_L+{\displaystyle \frac{1}{4}}(1+P_{})(1P_+)\mathrm{\hspace{0.17em}2}\mathrm{}(^{\stackrel{~}{\mathrm{f}}}^\mathrm{e})_R.`$ (184)
The chiral $`L,R`$ parts are
$`2\mathrm{}(^{\stackrel{~}{\mathrm{f}}}^{\mathrm{e}+})_{L,R}=N_Ce^2[(T_{\mathrm{int}}^{\gamma \gamma })_{L,R}+(T_{\mathrm{int}}^{\gamma Z})_{L,R}+(T_{\mathrm{int}}^{ZZ})_{L,R}]\times `$
$`[_3_3(p_1p_2)_3(k_1k_2,m_{\stackrel{~}{f}_i}m_{\stackrel{~}{f}_j})+_3(p_1p_2,k_1k_2,m_{\stackrel{~}{f}_i}m_{\stackrel{~}{f}_j})`$
$`+_4+_4(p_1p_2,k_1k_2,m_{\stackrel{~}{f}_i}m_{\stackrel{~}{f}_j})],`$ (185)
where
$`(T_{\mathrm{int}}^{\gamma \gamma })_{L,R}`$ $`=`$ $`{\displaystyle \frac{e^4e_f^2(\delta _{ij})^2}{ss_{\mathrm{red}}}}K_{L,R}^2,`$ (186)
$`(T_{\mathrm{int}}^{\gamma Z})_{L,R}`$ $`=`$ $`g_Z^2e^2e_fa_{ij}^{\stackrel{~}{f}}\delta _{ij}\left[{\displaystyle \frac{1}{4s_{\mathrm{red}}(sm_Z^2)}}+{\displaystyle \frac{1}{4s(s_{\mathrm{red}}m_Z^2)}}\right]C_{L,R}K_{L,R},`$ (187)
$`(T_{\mathrm{int}}^{ZZ})_{L,R}`$ $`=`$ $`{\displaystyle \frac{g_Z^4(a_{ij}^{\stackrel{~}{f}})^2}{16(sm_Z^2)(s_{\mathrm{red}}m_Z^2)}}C_{L,R}^2,`$ (188)
with $`s_{\mathrm{red}}=(p_1+p_2k_3)^2`$.
The functions $`_3`$ and $`_4`$ are given by
$`_3`$ $`=`$ $`{\displaystyle \frac{2}{(p_1.k_3)(k_1.k_3)}}[4(k_1.k_3)(k_1.p_1)(k_1.p_2)+4(k_1.p_1)^2(k_1.p_2)+2(k_1.p_1)(k_1.p_2)(k_2.k_3)`$ (189)
$`+2(k_1.k_3)(k_1.p_2)(k_2.p_1)4(k_1.p_1)(k_1.p_2)(k_2.p_1)+4(k_1.k_3)(k_1.p_1)(k_2.p_2)4(k_1.p_1)^2(k_2.p_2)`$
$`2(k_1.p_1)(k_2.k_3)(k_2.p_2)2(k_1.k_3)(k_2.p_1)(k_2.p_2)+4(k_1.p_1)(k_2.p_1)(k_2.p_2)+m_{\stackrel{~}{f}_i}^2(k_1.p_2)(k_3.p_1)`$
$`m_{\stackrel{~}{f}_j}^2(k_1.p_2)(k_3.p_1)2(k_1.k_3)(k_1.p_2)(k_3.p_1)+4(k_1.p_1)(k_1.p_2)(k_3.p_1)+2(k_1.p_2)(k_2.k_3)(k_3.p_1)`$
$`2(k_1.p_2)(k_2.p_1)(k_3.p_1)2m_{\stackrel{~}{f}_i}^2(k_2.p_2)(k_3.p_1)+2(k_1.k_2)(k_2.p_2)(k_3.p_1)4(k_1.p_1)(k_2.p_2)(k_3.p_1)`$
$`+2(k_2.p_1)(k_2.p_2)(k_3.p_1)+(k_1.p_2)(k_3.p_1)^2(k_2.p_2)(k_3.p_1)^2+m_{\stackrel{~}{f}_i}^2(k_1.p_1)(k_3.p_2)`$
$`+m_{\stackrel{~}{f}_j}^2(k_1.p_1)(k_3.p_2)2(k_1.k_2)(k_1.p_1)(k_3.p_2)2(k_1.k_3)(k_1.p_1)(k_3.p_2)+2(k_1.p_1)^2(k_3.p_2)`$
$`+2(k_1.k_3)(k_2.p_1)(k_3.p_2)2(k_1.p_1)(k_2.p_1)(k_3.p_2)+2m_{\stackrel{~}{f}_i}^2(k_3.p_1)(k_3.p_2)2(k_1.k_2)(k_3.p_1)(k_3.p_2)`$
$`+(k_1.p_1)(k_3.p_1)(k_3.p_2)(k_2.p_1)(k_3.p_1)(k_3.p_2)+m_{\stackrel{~}{f}_i}^2(k_1.k_3)(p_1.p_2)+m_{\stackrel{~}{f}_j}^2(k_1.k_3)(p_1.p_2)`$
$`2(k_1.k_2)(k_1.k_3)(p_1.p_2)+2(k_1.k_3)^2(p_1.p_2)2m_{\stackrel{~}{f}_i}^2(k_1.p_1)(p_1.p_2)2m_{\stackrel{~}{f}_j}^2(k_1.p_1)(p_1.p_2)`$
$`+4(k_1.k_2)(k_1.p_1)(p_1.p_2)2(k_1.k_3)(k_1.p_1)(p_1.p_2)2(k_1.k_3)(k_2.k_3)(p_1.p_2)`$
$`+2(k_1.p_1)(k_2.k_3)(p_1.p_2)m_{\stackrel{~}{f}_i}^2(k_3.p_1)(p_1.p_2)m_{\stackrel{~}{f}_j}^2(k_3.p_1)(p_1.p_2)+2(k_1.k_2)(k_3.p_1)(p_1.p_2)`$
$`(k_1.k_3)(k_3.p_1)(p_1.p_2)+(k_2.k_3)(k_3.p_1)(p_1.p_2)],`$
$`_4`$ $`=`$ $`{\displaystyle \frac{8}{((p_1.k_3))}}[(k_1.p_2)(k_3.p_1)+(k_2.p_2)(k_3.p_1)+(k_1.p_1)(k_3.p_2)(k_2.p_1)(k_3.p_2)`$ (190)
$`(k_1.k_3)(p_1.p_2)+(k_2.k_3)(p_1.p_2)].`$
|
warning/0506/hep-ph0506075.html
|
ar5iv
|
text
|
# Can one detect new physics in ๐ผ=0 and/or ๐ผ=2 contributions to the decays ๐ตโ๐โข๐?
## I Introduction
The purpose of $`B`$-physics experiments is the detection of new physics. Because CP violation appears in the Standard Model (SM) through one single irremovable phase in the Cabibbo-Kobayashi-Maskawa (CKM) matrix CKM , early strategies involved determining the various incarnations of this phase ($`\beta `$, $`\gamma `$, or $`\alpha \pi \beta \gamma `$), looking for discrepancies. Several techniques were proposed to sidestep the need to deal with the amplitude magnitudes and with the CP-even strong phases, since these are affected by uncertain hadronic matrix elements โ reviews can be found, for example, in BLS ; CKMfitter-04 ; Prague04 .
In one such proposal, due to Gronau and London, one uses the isospin symmetry between different $`B\pi \pi `$ decays GL . Their proposal can be worded in several different ways. We may take it as a measurement of $`\beta +\gamma `$, to be compared with the values allowed for this quantity by current CKM constraints on the Wolfenstein $`\rho `$$`\eta `$ plane Wolf83 ; we may use the measurement of $`\beta `$ from $`B_d\psi K`$ decays, and view this as a measurement of $`\gamma `$; or, one may take $`\gamma _{\mathrm{ckm}}`$ and $`\beta _{\mathrm{ckm}}`$ from the fit to the $`\rho `$$`\eta `$ plane, looking for inconsistencies in the overall fit of the SM parameters (including all CP-odd and CP-even quantities) to the experimental observables in $`B\pi \pi `$ decays.
In this article, we follow the last approach with respect to the weak phases (dropping the subscript โckmโ), but we will consider the most general type of new physics that could affect these decays. Our objective is to find which types of new physics can be probed in $`B\pi \pi `$ decays without making any assumptions about the hadronic matrix elements of the SM contributions to these decays, and which cannot. We show that:
1. there are only two probes of new physics in $`I=2`$ contributions: one probes the presence of a new weak phase in $`A_2`$; the other compares the value of $`\gamma _{\pi \pi }`$ extracted from the isospin analysis with that obtained independently through CKM unitarity or some other decay;
2. one cannot probe for new physics in $`I=0`$ contributions.
We show how these conclusions follow simply from the โreparametrization invarianceโ introduced by two of us (Botella and Silva) in reparametrization . In addition, if a new weak phase in $`A_2`$ is seen, we show that it is possible to measure the new-physics parameters using independent determinations of the weak phases.
In section II, we explain the generic features of โreparametrization invarianceโ relevant for this problem. In section III, we perform a general analysis of the $`B\pi \pi `$ decays valid in the presence of new physics and we prove that the conclusions announced above follow simply from reparametrization invariance. In section IV we perform a fit of the relevant new-physics parameters to the current experimental data. These constraints on new physics do not depend on any assumptions about the SM contributions, which are also independently extracted from our fit. We present our conclusions in section V.
## II Consequences of reparametrization invariance
Let us consider the decay of a $`B`$ meson into some specific final state $`f`$. For the moment, $`B`$ stands for $`B^+`$, $`B_d^0`$ or $`B_s^0`$. When discussing generic features of the decay amplitudes without reference to any particular model, it has become commonplace to parametrize the decay amplitudes as
$`A_f`$ $`=`$ $`M_1e^{i\varphi _{A1}}e^{i\delta _1}+M_2e^{i\varphi _{A2}}e^{i\delta _2},`$ (1)
$`\overline{A}_{\overline{f}}`$ $`=`$ $`M_1e^{i\varphi _{A1}}e^{i\delta _1}+M_2e^{i\varphi _{A2}}e^{i\delta _2},`$ (2)
where $`\varphi _{A1}`$ and $`\varphi _{A2}`$ are two CP-odd weak phases; $`M_1`$ and $`M_2`$ are the magnitudes of the corresponding terms; and $`\delta _1`$ and $`\delta _2`$ are the corresponding CP-even strong phases detail . These expressions apply to the decays of a (neutral or charged) $`B`$ meson into the final state $`f`$ and the charge-conjugated decay, respectively. For the decay of a neutral $`B`$ meson into a CP eigenstate with CP eigenvalue $`\eta _f=\pm 1`$, the right-hand-side of Eq. (2) appears multiplied by $`\eta _f`$.
As shown in reference reparametrization , the fact that any third weak phase may be written in terms of the first two means that one may write any amplitude, with an arbitrary number $`N`$ of distinct weak phases, in terms of only two. Indeed,
$$A_f=\stackrel{~}{M}_1e^{i\varphi _{A1}}e^{i\stackrel{~}{\delta }_1}+\stackrel{~}{M}_2e^{i\varphi _{A2}}e^{i\stackrel{~}{\delta }_2}+\underset{k=3}{\overset{N}{}}\stackrel{~}{M}_ke^{i\varphi _{Ak}}e^{i\stackrel{~}{\delta }_k}$$
(3)
and
$$\overline{A}_{\overline{f}}=\stackrel{~}{M}_1e^{i\varphi _{A1}}e^{i\stackrel{~}{\delta }_1}+\stackrel{~}{M}_2e^{i\varphi _{A2}}e^{i\stackrel{~}{\delta }_2}+\underset{k=3}{\overset{N}{}}\stackrel{~}{M}_ke^{i\varphi _{Ak}}e^{i\stackrel{~}{\delta }_k}$$
(4)
may be written as in Eqs. (1) and (2), respectively, through the choices
$`M_1e^{i\delta _1}`$ $`=`$ $`\stackrel{~}{M}_1e^{i\stackrel{~}{\delta }_1}+{\displaystyle \underset{k=3}{\overset{N}{}}}a_k\stackrel{~}{M}_ke^{i\stackrel{~}{\delta }_k},`$
$`M_2e^{i\delta _2}`$ $`=`$ $`\stackrel{~}{M}_2e^{i\stackrel{~}{\delta }_2}+{\displaystyle \underset{k=3}{\overset{N}{}}}b_k\stackrel{~}{M}_ke^{i\stackrel{~}{\delta }_k},`$ (5)
with
$`a_k`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}(\varphi _{Ak}\varphi _{A2})}{\mathrm{sin}(\varphi _{A1}\varphi _{A2})}},`$
$`b_k`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}(\varphi _{Ak}\varphi _{A1})}{\mathrm{sin}(\varphi _{A2}\varphi _{A1})}}.`$ (6)
Notice that, in addition, the phases $`\varphi _{A1}`$ and $`\varphi _{A2}`$ may be chosen completely at will. This property, which we refer to as โreparametrization invarianceโ, has very unusual consequences, which were explored at length in reparametrization .
Sometimes it is useful to consider the sums of all new contributions to $`B`$ and $`\overline{B}`$ decays,
$`N`$ $`=`$ $`{\displaystyle \underset{k=3}{\overset{N}{}}}\stackrel{~}{M}_ke^{i\varphi _{Ak}}e^{i\stackrel{~}{\delta }_k},`$
$`\overline{N}`$ $`=`$ $`{\displaystyle \underset{k=3}{\overset{N}{}}}\stackrel{~}{M}_ke^{i\varphi _{Ak}}e^{i\stackrel{~}{\delta }_k}.`$ (7)
With this notation, the proof that we may use only two weak phases as our basis follows simply from
$`N`$ $`=`$ $`N_{\varphi _{A1}}e^{i\varphi _{A1}}+N_{\varphi _{A2}}e^{i\varphi _{A2}},`$ (8)
$`\overline{N}`$ $`=`$ $`N_{\varphi _{A1}}e^{i\varphi _{A1}}+N_{\varphi _{A2}}e^{i\varphi _{A2}},`$ (9)
where
$`N_{\varphi _{A1}}`$ $`=`$ $`{\displaystyle \frac{Ne^{i\varphi _{A2}}\overline{N}e^{i\varphi _{A2}}}{2i\mathrm{sin}(\varphi _{A1}\varphi _{A2})}}{\displaystyle \underset{k=3}{\overset{N}{}}}a_k\stackrel{~}{M}_ke^{i\stackrel{~}{\delta }_k},`$
$`N_{\varphi _{A2}}`$ $`=`$ $`{\displaystyle \frac{Ne^{i\varphi _{A1}}\overline{N}e^{i\varphi _{A1}}}{2i\mathrm{sin}(\varphi _{A2}\varphi _{A1})}}{\displaystyle \underset{k=3}{\overset{N}{}}}b_k\stackrel{~}{M}_ke^{i\stackrel{~}{\delta }_k}.`$ (10)
Notice that, as required, the same complex numbers $`N_{\varphi _{A1}}`$ and $`N_{\varphi _{A2}}`$ appear in Eqs. (8) and (9). Said otherwise, $`N_{\varphi _{A1}}`$ and $`N_{\varphi _{A2}}`$ carry only magnitudes and CP-even phases, since the CP-odd phases, $`\varphi _{A1}`$ and $`\varphi _{A2}`$, have been factored out explicitly in Eqs. (8) and (9).
## III Parametrizing the $`B\pi \pi `$ decay amplitudes
We may parametrize the $`B\pi \pi `$ decay amplitudes according to the isospin of the final state as
$`\sqrt{2}A(B^+\pi ^+\pi ^0)=\sqrt{2}A_{+0}`$ $`=`$ $`3A_2,`$
$`A(B^0\pi ^+\pi ^{})=A_+`$ $`=`$ $`A_2+A_0,`$
$`\sqrt{2}A(B^0\pi ^0\pi ^0)=\sqrt{2}A_{00}`$ $`=`$ $`2A_2A_0,`$ (11)
and
$`\sqrt{2}A(B^{}\pi ^{}\pi ^0)=\sqrt{2}\overline{A}_{+0}`$ $`=`$ $`3\overline{A}_2,`$
$`A(\overline{B^0}\pi ^+\pi ^{})=\overline{A}_+`$ $`=`$ $`\overline{A}_2+\overline{A}_0,`$
$`\sqrt{2}A(\overline{B^0}\pi ^0\pi ^0)=\sqrt{2}\overline{A}_{00}`$ $`=`$ $`2\overline{A}_2\overline{A}_0.`$ (12)
In writing Eqs. (11) and (12), some coefficients and signs have been absorbed into the definition of the amplitudes for $`I=0`$ ($`A_0`$ and $`\overline{A}_0`$) and $`I=2`$ ($`A_2`$ and $`\overline{A}_2`$); this choice is not universal and great care should be exercised when comparing with other sources.
The right-hand-sides of Eqs. (11) and (12) contain seven independent parameters: four magnitudes ($`|A_2|`$, $`|\overline{A}_2|`$, $`|A_0|`$, and $`|\overline{A}_0|`$); and three relative phases ($`\overline{\delta }_2\delta _2`$, $`\overline{\delta }_0\delta _0`$, and $`\delta _2\delta _0`$). An overall phase can be rotated away. These seven quantities may be extracted from experiments detecting the average branching ratios ($`B_{+0}`$, $`B_+`$, and $`B_{00}`$), the direct CP violation ($`C_{+0}`$, $`C_+`$, and $`C_{00}`$), and the interference CP violation ($`S_+`$ and $`S_{00}`$) of $`B\pi \pi `$ decays, where the sub-indices refer to the charges of the physical pions in the final state. It turns out that $`S_{00}`$ may be written as a function of the other observables, up to discrete ambiguities. Therefore, there are seven independent measurements in $`B\pi \pi `$ decays, allowing the determination of the seven physical parameters present on the right-hand-sides of Eqs. (11) and (12).
A different decomposition is sometimes utilized within the SM. This is related to a diagrammatic analysis and it involves two weak phases ($`\beta `$ and $`\gamma `$) which appear naturally within the SM:
$`\sqrt{2}A_{+0}`$ $`=`$ $`\left(T+C\right)e^{i\gamma },`$
$`A_+`$ $`=`$ $`Te^{i\gamma }+Pe^{i\beta },`$
$`\sqrt{2}A_{00}`$ $`=`$ $`Ce^{i\gamma }Pe^{i\beta }.`$ (13)
Here $`T`$, $`C`$, and $`P`$ contain only magnitudes and CP-even (strong) phases. Similar relations hold for the conjugated (barred) amplitudes, by changing the signs of the CP-odd phases $`\gamma `$ and $`\beta `$. The relation between the two decompositions is
$`A_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(T+C)e^{i\gamma },`$
$`\overline{A}_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(T+C)e^{i\gamma },`$
$`A_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2TC)e^{i\gamma }+Pe^{i\beta },`$
$`\overline{A}_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2TC)e^{i\gamma }+Pe^{i\beta }.`$ (14)
For simplicity, in writing Eqs. (14) we have neglected the SM electroweak penguin contributions, but these can be included in a straightforward way by shifting gamma roughly by $`1.5^{}`$, following references PEW .
The impact of a generic new-physics model in $`B\pi \pi `$ decays will show up in both $`I=0`$ and $`I=2`$ amplitudes, with a variety of weak phases. This can be parametrized as
$`A_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(T+C)e^{i\gamma }+N_2,`$
$`\overline{A}_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(T+C)e^{i\gamma }+\overline{N}_2,`$
$`A_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2TC)e^{i\gamma }+Pe^{i\beta }+N_0,`$
$`\overline{A}_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2TC)e^{i\gamma }+Pe^{i\beta }+\overline{N}_0,`$ (15)
where $`N_0`$, $`\overline{N}_0`$, $`N_2`$, and $`\overline{N}_2`$ are complex numbers. We may use the consequences of reparametrization invariance in Eqs. (8)โ(10) in order to rewrite Eqs. (15) as
$`A_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(t+c)e^{i\gamma }+N_{2,o},`$
$`\overline{A}_2`$ $`=`$ $`{\displaystyle \frac{1}{3}}(t+c)e^{i\gamma }+N_{2,o},`$
$`A_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2tc)e^{i\gamma }+pe^{i\beta },`$
$`\overline{A}_0`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2tc)e^{i\gamma }+pe^{i\beta }.`$ (16)
Here
$`t+c`$ $`=`$ $`T+C+3N_{2,\gamma },`$ (17)
$`2tc`$ $`=`$ $`2TC+3N_{0,\gamma },`$ (18)
$`p`$ $`=`$ $`P+N_{0,\beta },`$ (19)
where
$`N_{2,\gamma }`$ $`=`$ $`i{\displaystyle \frac{\overline{N}_2N_2}{2\mathrm{sin}\gamma }},`$
$`N_{2,o}`$ $`=`$ $`{\displaystyle \frac{\overline{N}_2+N_2}{2}}i{\displaystyle \frac{\overline{N}_2N_2}{2\mathrm{tan}\gamma }},`$
$`N_{0,\gamma }`$ $`=`$ $`{\displaystyle \frac{\overline{N}_0+N_0}{2}}{\displaystyle \frac{\mathrm{sin}\beta }{\mathrm{sin}(\beta +\gamma )}}+i{\displaystyle \frac{\overline{N}_0N_0}{2}}{\displaystyle \frac{\mathrm{cos}\beta }{\mathrm{sin}(\beta +\gamma )}},`$
$`N_{0,\beta }`$ $`=`$ $`{\displaystyle \frac{\overline{N}_0+N_0}{2}}{\displaystyle \frac{\mathrm{sin}\gamma }{\mathrm{sin}(\beta +\gamma )}}i{\displaystyle \frac{\overline{N}_0N_0}{2}}{\displaystyle \frac{\mathrm{cos}\gamma }{\mathrm{sin}(\beta +\gamma )}},`$ (20)
are obtained from Eqs. (8)โ(10) with $`\{\varphi _{A1},\varphi _{A2}\}=\{\gamma ,0\}`$ for the $`I=2`$ contributions, and with $`\{\varphi _{A1},\varphi _{A2}\}=\{\gamma ,\beta \}`$ for the $`I=0`$ contributions.
We stress that our choice of $`\{\varphi _{A1},\varphi _{A2}\}=\{\gamma ,0\}`$ for the $`I=2`$ contributions is not mandatory. We could equally well have chosen a more general basis $`\{\varphi _{A1},\varphi _{A2}\}=\{\gamma ,\varphi \}`$, as long as the phase $`\varphi `$ was known and did not have to be fitted for general . For example, we could take $`\varphi =5^{}`$, or $`\varphi =10^{}`$, or even $`\varphi =\beta `$, with $`\beta `$ determined from $`B_d\psi K`$ decays.
The main results of our paper arise by comparing Eqs. (16), valid in the presence of generic new-physics contributions to $`B\pi \pi `$ decays, with Eqs. (14), valid within the SM. First, we notice that the expressions for $`A_0`$ and $`\overline{A}_0`$ have exactly the same form in Eqs. (14) and in Eqs. (16). This means that, without specific assumptions made about the hadronic matrix elements involved in the SM contributions $`T`$, $`C`$, and $`P`$, the measurements of $`A_0`$ and $`\overline{A}_0`$ cannot be used to test for the presence of new physics in $`I=0`$ (or lack thereof). This is one of our main points. It is impossible to detect new physics in $`I=0`$ without specific assumptions about the hadronic matrix elements involved in the SM contributions. Note that the impossibility of detecting $`I=0`$ new physics has long been suspected; reparametrization invariance offers a proof of this fact.
Conversely, if one makes assumptions about the quantities involved in the SM contributions $`2TC`$ and/or $`P`$, then the deviations $`(2tc)_{\mathrm{exp}}(2TC)`$ and $`p_{\mathrm{exp}}P`$ can indeed be used to probe the $`I=0`$ contributions $`N_{0,\gamma }`$ and $`N_{0,\beta }`$, respectively. This contradicts an analysis performed earlier by two of us (Baek and London) in references BL1 ; BL2 . The imprecision had to do with a very subtle question related to rephasing. It is only in the language of reparametrization invariance that this issue becomes simple to understand, illustrating how powerful reparametrization invariance is as a tool to organize the new-physics contributions.
Second, we notice that the expressions for $`A_2`$ and $`\overline{A}_2`$ do not have the same form in Eqs. (14) and in Eqs. (16). One piece of the new-physics contribution, $`N_{2,\gamma }`$, can indeed be reabsorbed into the definition of $`t+c`$, as in Eq. (17). (As with the $`I=0`$ contributions, the presence of the new $`I=2`$ contribution $`N_{2,\gamma }`$ may only be tested for under specific assumptions for the SM contributions to $`T+C`$.) But the other piece, $`N_{2,o}`$, cannot be reabsorbed by a redefinition of SM-like parameters. This means that the presence of some types of new physics in $`I=2`$ can be detected, even without specific assumptions made about the hadronic matrix elements involved in the SM contributions $`T`$ and $`C`$. Because $`N_{2,o}`$ is a complex number, we expect two such tests; these are related with the magnitude of $`N_{2,o}`$, and (once this magnitude is nonzero) with the difference between its (strong) phase and that of $`t+c`$.
To understand the first test, let us start by considering the case in which the (strong) phase of $`N_{2,o}`$ coincides with that of $`t+c`$, $`\delta _{t+c}`$. In that case the $`I=2`$ amplitudes may be written as
$`A_2`$ $`=`$ $`e^{i\delta _{t+c}}\left[{\displaystyle \frac{1}{3}}|t+c|e^{i\gamma }+|N_{2,o}|\right]=e^{i\delta _{t+c}}e^{i\gamma _{\pi \pi }}|A_2|,`$
$`\overline{A}_2`$ $`=`$ $`e^{i\delta _{t+c}}\left[{\displaystyle \frac{1}{3}}|t+c|e^{i\gamma }+|N_{2,o}|\right]=e^{i\delta _{t+c}}e^{i\gamma _{\pi \pi }}|A_2|,`$ (21)
where
$$\mathrm{tan}\gamma _{\pi \pi }=\frac{\mathrm{sin}\gamma }{\mathrm{cos}\gamma +3\frac{|N_{2,o}|}{|t+c|}}.$$
(22)
This type of new physics will be seen as a difference between the phase $`\gamma _{\pi \pi }`$ obtained from the isospin analysis of $`B\pi \pi `$ decays and the phase $`\gamma _{\mathrm{ckm}}`$ obtained from the current CKM constraints on the Wolfenstein $`\rho `$$`\eta `$ plane. Naturally, this signal of new physics disappears as $`N_{2,o}`$ vanishes. Moreover, in this case, because the same $`|A_2|`$ appears on both lines of Eq. (21), $`|\overline{A}_{+0}|^2|A_{+0}|^2|\overline{A}_2|^2|A_2|^2=0`$, and there is no direct CP violation in $`B^\pm \pi ^\pm \pi ^0`$ decays. So, the (one) test of new physics possible when $`C_{+0}=0`$ is
$$\left|\frac{N_{2,o}}{t+c}\right|=\frac{\mathrm{sin}(\gamma _{\mathrm{ckm}}\gamma _{\pi \pi })}{3\mathrm{sin}\gamma _{\pi \pi }}.$$
(23)
The second test on $`N_{2,o}`$ arises if it carries a strong phase which differs from $`\delta _{t+c}`$. In that case $`|\overline{A}_2|`$ differs from $`|A_2|`$, and this will be reflected in the appearance of direct CP violation in $`B^\pm \pi ^\pm \pi ^0`$ decays.
In both cases, if we take the values of $`\gamma `$ and $`\beta `$ from independent measurements, the number of observables in $`B\pi \pi `$ decays is equal to the number of theoretical parameters. Thus, it is not only possible to detect a nonzero $`N_{2,o}`$; one can also measure its parameters. Up to now, this has not been realized; as above, it is only by using reparametrization invariance that one sees this.
We conclude that there are only two independent tests for new physics in $`B\pi \pi `$ decays which do not depend on hadronic estimates for the SM contributions. New physics in $`I=0`$ contributions and $`N_{2,\gamma }`$ pieces in $`I=2`$ cannot be tested for. In contrast, $`N_{2,o}`$ contributions can be tested for, and they appear as $`\gamma _{\pi \pi }\gamma _{\mathrm{ckm}}0`$, or $`C_{+0}0`$. In addition, if the weak phases are assumed to be known independently, one can measure the parameters of $`N_{2,o}`$. Further tests and measurements are possible if one makes specific assumptions about the hadronic matrix elements of the SM.
## IV Constraining new-physics contributions with current data
The present $`B\pi \pi `$ measurements are detailed in Table 1.
The phase $`\beta `$ is taken from the measurements of interference CP violation in $`B\psi K`$ decays: $`\mathrm{sin}2\beta =0.725\pm 0.037`$ beta . Thus, $`2\beta `$ is determined up to a twofold ambiguity. We assume that $`\beta 23.5^{}`$, in agreement with the SM. The value of $`\gamma `$ is taken from independent measurements CKMfitter . For the purposes of the fit, we assume symmetric errors, and take $`\gamma =(58.2\pm 6.0)^{}`$.
Using the independent determinations of the SM CP phases, along with the latest $`B\pi \pi `$ measurements, we obtain the values for the isospin amplitudes. The fit to present data yields four solutions, presented in Table 2.
We get $`\chi _{min}^2/d.o.f.=0.0049/0`$, which is larger than expected. This occurs because the current data are slightly inconsistent with the isospin $`\{A_0,A_2\}`$ description. Indeed, we have for the central values
$$\mathrm{cos}(\delta _2\delta _0)=\frac{\frac{2}{3}|A_{+0}|^2+|A_+|^22|A_{00}|^2}{2\sqrt{2}|A_{+0}||A_0|}=1.07,$$
(24)
where $`|A_0|`$ is given by
$$|A_0|^2=\frac{2}{3}\left(\frac{2}{3}|A_{+0}|^2+|A_+|^2+|A_{00}|^2\right).$$
(25)
This explains why our fit gives the same values for $`\delta _2`$ and $`\delta _0`$.
We now wish to perform the fit in the notation of diagrammatic amplitudes. Using the rephasing freedom to set $`\mathrm{arg}N_{2,0}=0`$, we obtain the results in Table 3.
We get $`\chi _{min}^2=0.0049`$.
The results in Table 3 are related to those in Table 2 through
$`p`$ $`=`$ $`{\displaystyle \frac{\overline{A}_0e^{i\gamma }A_0e^{i\gamma }}{2i\mathrm{sin}(\beta +\gamma )}},`$
$`t`$ $`=`$ $`{\displaystyle \frac{\overline{A}_2A_2}{2i\mathrm{sin}\gamma }}{\displaystyle \frac{\overline{A}_0e^{i\beta }A_0e^{i\beta }}{2i\mathrm{sin}(\beta +\gamma )}},`$
$`c`$ $`=`$ $`2{\displaystyle \frac{\overline{A}_2A_2}{2i\mathrm{sin}\gamma }}+{\displaystyle \frac{\overline{A}_0e^{i\beta }A_0e^{i\beta }}{2i\mathrm{sin}(\beta +\gamma )}},`$
$`N_{2,0}`$ $`=`$ $`{\displaystyle \frac{\overline{A}_2e^{i\gamma }A_2e^{i\gamma }}{2i\mathrm{sin}\gamma }}.`$ (26)
One could be worried by the fact that we have used the rephasing freedom in order to set $`\overline{\delta }_0=0`$ when obtaining Table 2, while we have used the rephasing freedom in order to set $`\mathrm{arg}N_{2,0}=0`$ in obtaining Table 3. Nevertheless, both Tables contain only rephasing-invariant quantities which, therefore, can be related. It is easy to see how the rephasing freedom drops out from Eqs. (26) when one relates rephasing-invariant quantities in both parametrizations.
We have also performed the fit of the current experimental data to the SM, obtained by setting $`N_{2,0}=0`$. The results are listed in Table 4.
We find $`\chi _{min}^2/d.o.f.=0.296/2`$, meaning that, if one waives any predictions for the hadronic matrix elements, then the SM provides an excellent fit to the current data.
Notice that Table 4 only has one solution, while Table 3 had four. The reason is the following: in the SM $`\overline{A}_2=A_2e^{2i\gamma }`$, or, in term of rephasing invariant quantities,
$$|\overline{A}_2|e^{i(\overline{\delta }_2\overline{\delta }_0)}=|A_2|e^{i(\delta _2\overline{\delta }_0)}e^{2i\gamma }.$$
(27)
We can see that the third solution in Table 3 is the one which best satisfies Eq. (27), giving the smallest $`\chi ^2`$ of all.
## V Conclusions
We have considered the most general new-physics contributions to the $`I=0`$ and $`I=2`$ amplitudes in $`B\pi \pi `$ decays, which involve 4 new complex parameters $`N_0`$, $`\overline{N}_0`$, $`N_2`$, and $`\overline{N}_2`$. We have shown that $`N_0`$ and $`\overline{N}_0`$ may be absorbed by a redefinition of the SM contributions to $`B\pi \pi `$ decays, as can $`N_{2,\gamma }`$, c.f. Eqs. (17)โ(19). This means that new-physics contributions of this type โ and in particular, all new-physics contributions to $`I=0`$cannot be detected unless specific ranges are taken for the SM contributions. In contrast, $`N_{2,o}`$ allows for two tests for the new physics, related to $`C_{+0}`$ and $`\gamma _{\pi \pi }\gamma _{\mathrm{ckm}}`$. These are the only two probes of new physics in $`B\pi \pi `$ decays which do not involve estimates of the SM hadronic matrix elements. Furthermore, if one takes values for the weak phases from independent determinations, the $`B\pi \pi `$ observables allow one to measure the $`N_{2,o}`$ parameters. We have shown that all of these conclusions follow simply from the reparametrization invariance introduced in reparametrization , thus illustrating the power of this concept in providing a clear organization of the new-physics contributions.
###### Acknowledgements.
We thank Y. Nir, G. Raz, and L. Wolfenstein for discussions. J. P. S. is extremely grateful to Y. Nir and to the Department of Particle Physics of the Weizmann Institute of Science for their excellent hospitality, while portions of the this work were made. The work of S. B. and D. L. is supported by NSERC of Canada. F. J. B. is partially supported by the spanish M. E. C. under FPA2002-00612 and HP2003-0079 (โAccion Integrada hispano-portuguesaโ). J. P. S. is supported in part by the Portuguese Fundaรงรฃo para a Ciรชncia e a Tecnologia (FCT) under the contract CFTP-Plurianual (777), and through the project POCTI/37449/FNU/2001, approved by the Portuguese FCT and POCTI, and co-funded by FEDER.
|
warning/0506/cond-mat0506777.html
|
ar5iv
|
text
|
# First-principles study of epitaxial strain in perovskites
## I Introduction
Ferroelectrics are insulating solids of technological importance because of their ability to maintain an electric polarization that can be reoriented by the application of an electric field.Lines1977book This property lends itself to technological applications including microelectronic devices and computer memories. Among ferroelectrics, perovskites constitute a subclass that has been of theoretical and experimental interest since the discovery in 1945 of its first member, barium titanate (BaTiO<sub>3</sub>). This interest is motivated in part by the relative simplicity of their cubic crystalline phase. For a perovskite of general formula ABO<sub>3</sub>, this structure contains cations A at the cube corners, a cation B at the center of the cube, and oxygen atoms at the center of the cube faces forming a regular octahedron, as depicted in Fig. 1(a). Typically, perovskites are found in this cubic paraelectric phase at high temperature; as the temperature is reduced, symmetry-lowering distortions to other phases, including ferroelectric ones, may occur.
The electronics industryโs demands for smaller components have made thin ferroelectric films the subject of recent attention.Ahn2004S ; Dawber2005 Experimentally, it is found that the properties of ferroelectrics in thin-film form generally differ significantly from those in the bulk. While many factors are expected to contribute to these differences, it has been shown that the properties of perovskite thin films are strongly influenced by the magnitude of the epitaxial strain resulting from lattice-matching the film to the substrate, known as misfit strain or epitaxial strain.
Previous theoretical studies have isolated the effects of epitaxial strain on the structure and properties of films by imposing the epitaxial constraint on the in-plane lattice vectors of a periodic bulk sample. Using a phenomenological Landau-Devonshire model, Pertsev, Zembilgotov and TagantsevPertsev1998PRL introduced the concept of mapping the equilibrium structure of a ferroelectric perovskite material versus temperature and misfit strain, thus producing a phase diagram of the observable epitaxial phases. Given the importance of strain in determining the properties of these films, these diagrams have proven to be of enormous interest to experimentalists seeking to interpret the results of experiments on epitaxial thin films and heterostructures. This phenomenological approach should give excellent results in the temperature/strain regime in which the model parameters were fitted (usually near the bulk ferroelectric transition) but will generally be less accurate when extrapolated to other regimes. In particular, Figure 1 of Ref. Dieguez2004PRB, shows that two different sets of parameters can give two quite different phase diagrams. Furthermore, it is only possible to study materials for which all the needed experimental information is available.
In previous work,Dieguez2004PRB ; Antons2005PRB ; Bungaro2004PRB we have examined the effects of epitaxial strain with an analogous, but fully first-principles, approach. Specifically, we presented density-functional theory (DFT)Hohenberg1964PR calculations for the structure and properties of BaTiO<sub>3</sub>, PbTiO<sub>3</sub> and SrTiO<sub>3</sub> with varying in-plane strain, fully relaxing all structural degrees of freedom consistent with uniform distortions (that is, retaining the five-atom unit cell). From this, we obtained zero-temperature phase diagrams that complement the phenomenological results of Pertsev and coworkers. In this paper, we show that these phase diagrams can be reproduced using the first-principles energy parameterization of King-Smith and Vanderbilt, and give results for an additional five perovskite oxides: CaTiO<sub>3</sub>, KNbO<sub>3</sub>, NaNbO<sub>3</sub>, PbZrO<sub>3</sub>, and BaZrO<sub>3</sub>. This approach greatly decreases the computational effort involved in computing the phase diagram and readily allows the inclusion of nonzero external stress.Emelyanov2002PRB Moreover, the parameters can be fully specified in a compact table, and the functional form of the energy is suitable for analytical computations and conceptual interpretation, leading to a classification of possible stress-strain phase diagrams. These results can be used for predictions of the structure of epitaxially strained thin films grown on substrates with square symmetry, and for the design of novel perovskite strained-layer superlattices with two or more components.
The paper is organized as follows. In Sec. II we describe the extension of the KSV method to study the effects of the epitaxial strain constraint and external stress, and the first-principles calculations used to obtain the KSV parameters. Section III presents the results for the eight perovskite oxides considered, including the sequence of phase transitions with varying misfit strain at zero stress, and phase diagrams in which we show the most stable phase for given misfit strain and external stress. In Sec. IV, we review the approximations and discuss how to use the present results for the prediction of the structures and polarization of thin films and superlattices. Finally, in Sec. V we present our conclusions.
## II Method
### II.1 Formalism
The starting point of this analysis is the parameterized total-energy expression presented by King-Smith and Vanderbilt in Ref. KingSmith1994PRB, . This is a Taylor expansion around the cubic perovskite structure in terms of the six independent components $`\eta _i`$ of the strain tensor ($`i`$ is a Voigt index, $`i=`$1-6) and the three Cartesian components $`u_\alpha `$ ($`\alpha =x,y,z`$) describing the amplitude of the soft mode defined by the pattern of eigen-displacements associated with the smallest eigenvalue of the (zone-center) force-constant matrix. The arrows in Fig. 1 indicate a typical displacement pattern associated with this mode, greatly magnified to allow its visualization.
The contributions to the energy (per unit cell) can divided into terms arising from pure strain and from pure soft-mode amplitude, and an interaction term,
$$E=E^{\mathrm{elas}}(\{\eta _i\})+E^{\mathrm{soft}}(\{u_\alpha \})+E^{\mathrm{int}}(\{\eta _i\},\{u_\alpha \}),$$
(1)
with the zero of the energy corresponding to the cubic structure. For crystals with cubic symmetry the strain energy is given, correct to second order in the strains, by
$`E^{\mathrm{elas}}(\{\eta _i\})`$ $`=`$ $`{\displaystyle \frac{1}{2}}B_{11}(\eta _1^2+\eta _2^2+\eta _3^2)`$ (2)
$`+B_{12}(\eta _1\eta _2+\eta _2\eta _3+\eta _3\eta _1)`$
$`+{\displaystyle \frac{1}{2}}B_{44}(\eta _4^2+\eta _5^2+\eta _6^2),`$
where $`B_{11}`$, $`B_{12}`$, and $`B_{44}`$ are related to the elastic constants of the crystal by factors of the cell volume. The soft-mode energy given in Ref. KingSmith1994PRB, contains terms up to fourth-order in the soft-mode amplitude,
$$E^{\mathrm{soft}}(\{u_\alpha \})=\kappa u^2+\alpha u^4+\gamma (u_x^2u_y^2+u_y^2u_z^2+u_z^2u_x^2),$$
(3)
where $`u^2=u_x^2+u_y^2+u_z^2`$, $`\kappa `$ is twice the soft-mode eigenvalue, and $`\alpha `$ and $`\gamma `$ are the two independent symmetry-allowed fourth-order coefficients describing the cubic anisotropy. Finally, the interaction between the strains and the soft-mode amplitude is given by
$`E^{\mathrm{int}}(\{\eta _i\},\{u_\alpha \})`$ $`=`$ $`{\displaystyle \frac{1}{2}}B_{1xx}(\eta _1u_x^2+\eta _2u_y^2+\eta _3u_z^2)`$ (4)
$`+{\displaystyle \frac{1}{2}}B_{1yy}[\eta _1(u_y^2+u_z^2)+\eta _2(u_z^2+u_x^2)`$
$`+\eta _3(u_x^2+u_y^2)]`$
$`+B_{4yz}(\eta _4u_yu_z+\eta _5u_zu_x+\eta _6u_xu_y),`$
where $`B_{1xx}`$, $`B_{1yy}`$, and $`B_{4yz}`$ are the phonon-strain interaction coefficients. All the coefficients in these three parts of the total-energy expression can be obtained from first-principles calculations on a series of distorted structures as described in Ref. KingSmith1994PRB, and in the next subsection.
In this paper we will be concerned with the effects of strain on a film grown epitaxially on a substrate with square symmetry. The epitaxial strain constraint imposed by the substrate is
$$\eta _1=\eta _2=\overline{\eta },$$
(5)
$$\eta _6=0,$$
(6)
where $`\overline{\eta }`$ is the misfit strain between the minimum-energy cubic structure of the film material and the substrate.
In the case of epitaxy, where strain elements $`\eta _1`$, $`\eta _2`$ and $`\eta _6`$ are constrained while the others are not, it is useful to introduce a mixed stress-strain elastic enthalpy $`G=E\sigma _3\eta _3\sigma _4\eta _4\sigma _5\eta _5`$ whose natural variables are $`u_x,u_y,u_z,\eta _1,\eta _2,\eta _6,\sigma _3,\sigma _4`$ and $`\sigma _5`$. Specializing to our case in which $`\eta _1=\eta _2=\overline{\eta }`$, $`\eta _6=0`$, and assuming that the shear stresses $`\sigma _4`$ and $`\sigma _5`$ vanish, we define an effective elastic enthalpy given by
$$\stackrel{~}{G}=E\sigma _3\eta _3$$
(7)
whose natural variables are $`u_x,u_y,u_z,\overline{\eta }`$ and $`\sigma _3`$. Using Eqs. (2-4) and minimizing Eq. (7) with respect to $`\eta _3`$, $`\eta _4`$ and $`\eta _5`$ yields
$`\eta _3`$ $`=`$ $`{\displaystyle \frac{1}{B_{11}}}[\sigma _32B_{12}\overline{\eta }`$ (8)
$`{\displaystyle \frac{1}{2}}B_{1xx}u_z^2{\displaystyle \frac{1}{2}}B_{1yy}(u_x^2+u_y^2)]`$
$`\eta _4`$ $`=`$ $`{\displaystyle \frac{B_{4yz}}{B_{44}}}u_yu_z,`$ (9)
$`\eta _5`$ $`=`$ $`{\displaystyle \frac{B_{4yz}}{B_{44}}}u_zu_x.`$ (10)
Substituting these expressions back into Eq. (7), we express $`\stackrel{~}{G}`$ in terms of its natural variables as
$`\stackrel{~}{G}`$ $`=`$ $`(A_{\overline{\eta }\overline{\eta }}\overline{\eta }^2+A_{\overline{\eta }\sigma }\overline{\eta }\sigma +A_{\sigma \sigma }\sigma ^2)`$ (11)
$`+(B_{\overline{\eta }}\overline{\eta }+B_\sigma \sigma +B)u_{xy}^2`$
$`+(C_{\overline{\eta }}\overline{\eta }+C_\sigma \sigma +C)u_z^2`$
$`+Du_{xy}^4+Eu_z^4+Fu_{xy}^2u_z^2`$
$`+Hu_{xy}^4\mathrm{sin}^2\theta \mathrm{cos}^2\theta .`$
Here we have simplified the notation by replacing $`\sigma _3`$ by $`\sigma `$. Also, the two soft-mode amplitude components are represented in polar coordinates as
$`u_x`$ $`=`$ $`u_{xy}\mathrm{cos}\theta ,`$ (12)
$`u_y`$ $`=`$ $`u_{xy}\mathrm{sin}\theta .`$ (13)
The coefficients in $`\stackrel{~}{G}`$ are expressed in terms of the KSV parameters as follows:
$`A_{\overline{\eta }\overline{\eta }}=B_{11}+B_{12}2{\displaystyle \frac{B_{12}^2}{B_{11}}},`$ (14)
$`A_{\overline{\eta }\sigma }=2{\displaystyle \frac{B_{12}}{B_{11}}},`$ (15)
$`A_{\sigma \sigma }={\displaystyle \frac{1}{2B_{11}}},`$ (16)
$`B_{\overline{\eta }}={\displaystyle \frac{B_{1xx}+B_{1yy}}{2}}{\displaystyle \frac{B_{12}}{B_{11}}}B_{1yy},`$ (17)
$`B_\sigma ={\displaystyle \frac{B_{1yy}}{2B_{11}}},`$ (18)
$`B=\kappa ,`$ (19)
$`C_{\overline{\eta }}=B_{1yy}{\displaystyle \frac{B_{12}}{B_{11}}}B_{1xx},`$ (20)
$`C_\sigma ={\displaystyle \frac{B_{1xx}}{2B_{11}}},`$ (21)
$`C=\kappa ,`$ (22)
$`D=\alpha {\displaystyle \frac{1}{8}}{\displaystyle \frac{B_{1yy}^2}{B_{11}}},`$ (23)
$`E=\alpha {\displaystyle \frac{1}{8}}{\displaystyle \frac{B_{1xx}^2}{B_{11}}},`$ (24)
$`F=2\alpha +\gamma {\displaystyle \frac{1}{4}}{\displaystyle \frac{B_{1xx}B_{1yy}}{B_{11}}}{\displaystyle \frac{B_{1xx}^2}{B_{11}}},`$ (25)
$`H=\gamma .`$ (26)
For a given set of coefficients in the potential $`\stackrel{~}{G}`$ of Eq. (11), we can predict the phase diagram as a function of misfit strain $`\overline{\eta }`$ and the normal external stress $`\sigma `$ by minimizing $`\stackrel{~}{G}`$ to find the values of the ground-state soft-mode amplitude components. For a fourth order theory, like the present KSV expression, the entire optimization process can be done analytically, since it is possible to compute first and second derivatives of $`\stackrel{~}{G}`$ and to do a stability analysis of the various possible phases, classified by the nature of the minimum-energy soft-mode vector. For example, a paraelectric $`p`$ phase similar to the cubic phase can appear if the potential is minimized for $`๐ฎ=0`$, but relaxation along $`\widehat{z}`$ will occur, making the cell tetragonal, as shown in Fig 1(b). The classification, following Pertsev and coworkers,Pertsev1998PRL is given in table 1. Expressions for the elastic enthalpy of a given phase as a function of misfit strain can be obtained by minimizing Eq. (11) with the appropriate constraint on $`๐ฎ`$. For example, the elastic enthalpies for the $`p`$, $`c`$, $`a`$ and $`aa`$ phases are
$`\stackrel{~}{G}_p`$ $`=`$ $`A_{\overline{\eta }\overline{\eta }}\overline{\eta }^2+A_{\overline{\eta }\sigma }\overline{\eta }\sigma +A_{\sigma \sigma }\sigma ^2,`$ (27)
$`\stackrel{~}{G}_c`$ $`=`$ $`\stackrel{~}{G}_p{\displaystyle \frac{(C+C_{\overline{\eta }}\overline{\eta }+C_\sigma \sigma )^2}{4E}},`$ (28)
$`\stackrel{~}{G}_a`$ $`=`$ $`\stackrel{~}{G}_p{\displaystyle \frac{(B+B_{\overline{\eta }}\overline{\eta }+B_\sigma \sigma )^2}{4D}},`$ (29)
$`\stackrel{~}{G}_{aa}`$ $`=`$ $`\stackrel{~}{G}_p{\displaystyle \frac{(B+B_{\overline{\eta }}\overline{\eta }+B_\sigma \sigma )^2}{4D+H}}.`$ (30)
The process of generating the stress-strain diagrams is thus extremely rapid in comparison with a full DFT analysis, once the KSV parameters have been obtained from first principles calculations as described in the next section. Justification of the approximations involved has been presented in Ref. KingSmith1994PRB, , and will be discussed further in Sec. IV.
### II.2 First-principles calculations of the coefficients
In Table V of their paper,KingSmith1994PRB King-Smith and Vanderbilt report the computed coefficients to be used in their model. We have now repeated calculations analogous to theirs, taking advantage of the increase in computational power that has taken place during the last ten years to push the boundaries of the numerical approximations that control the accuracy of the first-principles calculations. We used the same local-density approximationKohn1965PR ; Ceperley1980PRB DFT methodology and the same ultrasoft pseudopotentials that they used. Vanderbilt1990PRB In our case, the plane-wave kinetic-energy cutoff has been raised to 50 Ry, and the Monkhorst-PackMonkhorst1979PRB k-point mesh is finer, containing $`8\times 8\times 8`$ points.
The lattice parameters and the soft-mode eigenvector components calculated for the eight perovskites under study are reported in Table 2. These values are quite similar to those reported in Tables IV and VII of Ref. KingSmith1994PRB, , where a comparison with experimental and other theoretical data can also be found.
Following the prescription given by King-Smith and Vanderbilt,KingSmith1994PRB we found the updated KSV parameters displayed in Table 3. The two final columns of this table contain the values of $`\alpha ^{}`$ and $`\gamma ^{}`$, the coupling constants whose values determine the symmetry of the low-temperature phase for bulk perovskites gammaprime (see Ref. KingSmith1994PRB, ). Most of the parameters are only slightly different from those given in Table V of Ref. KingSmith1994PRB, , and we have found that the predictions that both sets give for the epitaxial films are qualitatively the same (note, however, that cubic SrTiO<sub>3</sub> changes from being marginally unstable to marginally stable). The differences are mainly related to the use of finer grids in our case when performing the fast Fourier transforms required in the calculations. On the other hand, the improvements in plane-wave cutoff and k-point mesh have a quite small impact on the values of the coefficients.
For our discussion of the effects of epitaxial strain on the polarization, it is useful to have an expansion not just of the energy, but also of the polarization, in terms of the soft-mode amplitude. The linearized expression
$$P_z=\frac{e}{\mathrm{\Omega }}z^{}u_z$$
(31)
is adequate for most purposes. Here $`e`$ is the absolute value of the electron charge, $`\mathrm{\Omega }`$ is the unit cell volume of the perovskite, and $`z^{}`$ is the Born effective charge of the soft mode, given in terms of the soft-mode eigenvectors of Table 2 by
$$z^{}=Z_\mathrm{A}^{}\xi _\mathrm{A}+Z_\mathrm{B}^{}\xi _\mathrm{B}+2Z_{\mathrm{O}_{1,2}}^{}\xi _{\mathrm{O}_{1,2}}+Z_{\mathrm{O}_3}^{}\xi _{\mathrm{O}_3}.$$
(32)
We take the Born effective charges $`Z^{}`$ for each atom to have their values in the cubic structure as calculated by Zhong, King-Smith, and Vanderbilt.Zhong1994PRL Using these together with the eigenvectors reported in Table 2 we obtain the following values for $`z^{}`$: 9.94 (BaTiO<sub>3</sub>), 8.65 (SrTiO<sub>3</sub>), 7.03 (CaTiO<sub>3</sub>), 11.06 (KNbO<sub>3</sub>), 9.14 (NaNbO<sub>3</sub>), 9.40 (PbTiO<sub>3</sub>), 6.29 (PbZrO<sub>3</sub>), and 5.69 (BaZrO<sub>3</sub>). This approximate expression neglects the possible dependence of the Born effective charges upon the strain.
## III Results
### III.1 Calculations at zero external stress
We first consider the case in which the external perpendicular stress $`\sigma `$ vanishes. Table 4 shows the sequence of transitions that occurs for each of the eight perovskites studied as the misfit strain increases, and the values of strain at which the transitions occur.
The observed sequences of phases can be understood with the help of Fig. 2, which illustrates the types of elastic enthalpy behaviors that we observe for the materials considered. At the strains at which a transition occurs from one phase to another, the energy curves join smoothly, indicating that these transitions are of second order. It can be shown analytically that this is indeed the case for the KSV model, and that at the symmetry-breaking transitions (c-r, aa-r, a-ac, p-c, p-a, or p-aa) the higher-symmetry phase becomes unstable. For all compounds considered, we see that for sufficiently high compressive strains, the lowest energy phase is always the $`c`$ phase, in which the atomic displacements and therefore the polarization point in the direction, perpendicular to the substrate. On the other hand, for sufficiently high tensile strains, we obtain a phase in which the polarization lies in the substrate plane, pointing along ($`a`$ phase) for for BaZrO<sub>3</sub> or along ($`aa`$ phase) for the rest of the compounds. The in-plane orientation is determined by the sign of $`H`$ (i.e., of $`\gamma `$), which is positive for BaZrO<sub>3</sub> and negative for the seven other compounds. In the intermediate strain regime, three different behaviors are found. For BaTiO<sub>3</sub>, KNbO<sub>3</sub>, NaNbO<sub>3</sub>, and PbZrO<sub>3</sub>, an $`r`$ phase appears between the $`c`$ and $`aa`$ phases, as in Fig. 2(a). The fact that these perovskites crystallize in the $`r`$ phase at low absolute values of strain is not surprising, since this phase is the most similar to the rhombohedral phase that they adopt as the bulk ground state according to the KSV theory (see Table VI of Ref. KingSmith1994PRB, ). For SrTiO<sub>3</sub> and BaZrO<sub>3</sub> it is the paraelectric $`p`$ phase that appears at intermediate strains, as in Fig. 2(b). For these two compounds, no $`r`$ or $`ac`$ phase appears. Instead, the polarization along continuously goes to zero as the strain becomes less compressive, and only reappears in the $`xy`$ plane once the tensile strain reaches some given value, continuously growing from then on. The epitaxial paraelectric $`p`$ phase is the analog of the bulk cubic phase (see Fig. 1(b)), which is the ground state predicted by the KSV theory with the parameters of Table 3 for both bulk SrTiO<sub>3</sub> and bulk BaZrO<sub>3</sub>. Finally, for CaTiO<sub>3</sub> and PbTiO<sub>3</sub> yet another behavior is obtained, as shown in Fig. 2(c). At intermediate strains, the rhombohedral phase is the lowest energy single phase. However, partly because of its inverted-parabola elastic enthalpy curve, the common tangent line between the $`c`$ and $`aa`$ phases yields a lower energy in the intermediate strain regime, and thus a mixed phase of $`c`$ and $`aa`$ domains is expected.
The strain-induced phase transitions found using this approach compare well with the full DFT results previously reported for BaTiO<sub>3</sub>,Dieguez2004PRB PbTiO<sub>3</sub>,Bungaro2004PRB and SrTiO<sub>3</sub>.Antons2005PRB We look now in detail at this comparison for BaTiO<sub>3</sub>. Figure 3 shows the energy curves of the various phases as predicted by the KSV theory (right panel), to be compared with the full DFT results (left panel). The agreement between the two sets of results is very good, with the small differences present arising from two sources. First, the first-principles calculations in Ref. Dieguez2004PRB, were performed using the projector-augmented wave method,Blochl1994PRB while the first-principles calculations used to obtain the KSV coefficients in the present work were performed using ultrasoft pseudopotentials. Vanderbilt1990PRB Second, there are the intrinsic errors due to the use of a Taylor expansion described in the previous section, which are expected to grow as the strain and soft mode magnitudes increase.
Figure 4 shows the displacements of the atoms from their centrosymmetric perovskite positions as strain varies. Again, the agreement of the KSV results (right panel) with the full DFT results (left panel) is very good. In particular, the square root behavior predicted by the KSV theory is exhibited by the more exact DFT calculations. As the in-plane strain increases, we observe a second-order phase transition (c-r), and while the magnitude of the atomic displacements continues to diminish along , the displacements in the xy plane begin to grow. With increasing tensile strain, the displacements along vanish at the r-aa transition, while the displacements in the xy plane continue to grow smoothly. In this way, we see that the polarization vector continuously rotates in going from the $`c`$ phase through the $`r`$ phase to the $`aa`$ phase. A quantitative limitation of using a single misfit-strain-independent local mode in the KSV model is shown here in the form of an artificial constraint equating the $`\mathrm{\Delta }_y(\mathrm{O}_1)`$ and $`\mathrm{\Delta }_x(\mathrm{O}_3)`$ displacements. This constraint is removed when full DFT calculations are performed and the atoms are free to relax within the given space group, but the magnitude of these displacements is not very different from that obtained in the KSV theory.
Figure 5 shows the values of $`P_z`$ for the various compounds as a function of misfit strain. The square-root singularity at $`P_z`$=0 corresponds to an r-aa or c-p transition. The slope discontinuity visible at finite $`P_z`$ in some curves corresponds to the c-r transition of Fig. 2(a), while the termination of the curves at finite $`P_z`$ for CaTiO<sub>3</sub> and PbTiO<sub>3</sub> corresponds to the encounter with the tie line in Fig. 2(b). The general trend, of course, is a strong increase of $`P_z`$ with compressive strain, with considerable enhancement possible over the zero misfit strain value. Interestingly, according to this plot CaTiO<sub>3</sub> would have a large $`P_z`$ if the zone boundary distortions that are present in its actual ground-state crystal structure were suppressed. This result is supported by recent full DFT calculations.Nakhmanson2005
### III.2 Stress-strain phase diagrams
We now consider application of a nonzero normal stress $`\sigma `$. The stress-strain phase diagrams obtained for each of the eight perovskites are shown in Fig. 6.
All eight diagrams show a universal topology with straight-line phase boundaries meeting at a single crossing point. The perfect linearity of the boundaries is an artifact of the truncation of our energy expansion to fourth order, but the presence of a single crossing point is robust against the introduction of small higher-order terms. The crossing point strain $`\overline{\eta }_\times `$ and stress $`\sigma _\times `$ can be connected with the critical strain and stress in the isotropic cubic perovskite at which a structural instability first occurs as the pressure is reduced or made more negative. There are two main variants of the diagrams in Fig. 6, the first with four phases (c, r, p, and aa/a), and the second with three phases (c, p, aa/a and a mixed phase region). The varied behavior of the zero-stress diagrams of Fig. 2 can now be interpreted, in the context of Fig. 6, as reflecting whether the zero-stress axis lies above, or below, $`\sigma _\times `$.
More specifically, the coordinates of the crossing point can be expressed as functions of the total-energy coefficients as follows:
$`\overline{\eta }_\times `$ $`=`$ $`{\displaystyle \frac{B_\sigma CBC_\sigma }{B_{\overline{\eta }}C_\sigma B_\sigma C_{\overline{\eta }}}}={\displaystyle \frac{2\kappa }{B_{1xx}+2B_{1yy}}},`$ (33)
$`\sigma _\times `$ $`=`$ $`{\displaystyle \frac{BC_{\overline{\eta }}B_{\overline{\eta }}C}{B_{\overline{\eta }}C_\sigma B_\sigma C_{\overline{\eta }}}}=\overline{\eta }_\times (B_{11}+2B_{12}).`$ (34)
The strain and stress values at the crossing point are rather modest, with the single exception of PbZrO<sub>3</sub>. PbZrO<sub>3</sub> has the lowest soft-mode eigenvalue, and therefore the most negative value of $`\kappa `$ (Table 3). Its $`x`$-polarized soft-mode eigenvalue also changes more slowly on application of an $`\eta _1`$ strain, resulting in a less negative value of $`B_{1xx}`$. These values of $`\kappa `$ and $`B_{1xx}`$ result in $`\overline{\eta }_\times `$ and $`\sigma _\times `$ being about an order of magnitude more negative for PbZrO<sub>3</sub> than for the other seven compounds.
For external stresses below $`\sigma _\times `$, the behavior is similar in all eight compounds, showing a $`c`$-$`p`$-$`aa`$ ($`c`$-$`p`$-$`a`$ for BaZrO<sub>3</sub>) sequence of second-order phase transitions, with the elastic enthalpies of the phases behaving as in Fig. 2(b). In this fourth-order KSV theory, the phase boundaries are straight lines given by
$`\sigma _{c\text{-}p}`$ $`=`$ $`{\displaystyle \frac{C}{C_\sigma }}{\displaystyle \frac{C_{\overline{\eta }}}{C_\sigma }}\overline{\eta },`$ (35)
$`\sigma _{p\text{-}aa}`$ $`=`$ $`{\displaystyle \frac{B}{B_\sigma }}{\displaystyle \frac{B_{\overline{\eta }}}{B_\sigma }}\overline{\eta }.`$ (36)
Rewriting these expressions as functions of the fundamental coefficients of Table 3, we see that the value of $`B_{1yy}`$ plays a central role. For materials like the ones we are studying, for which $`B_{11}>0`$, $`B_{12}>0`$, and $`B_{1xx}<0`$, the slope of the $`c\text{-}p`$ transition line is found to be positive if
$$\frac{B_{12}}{B_{11}}>\frac{B_{1yy}}{B_{1xx}}.$$
(37)
For all eight perovskites this slope is indeed positive, since the left hand side of the inequality is positive, and $`B_{1yy}`$ is positive or only slightly negative but much smaller in magnitude than $`B_{1xx}`$ (the latter being the case for BaTiO<sub>3</sub> and PbTiO<sub>3</sub>). The slope of the $`p\text{-}aa`$ transition is positive if
$$\frac{B_{12}}{B_{11}}>\frac{1}{2}\left(1+\frac{B_{1xx}}{B_{1yy}}\right),$$
(38)
where the inequality is not satisfied for BaTiO<sub>3</sub> and PbTiO<sub>3</sub>, which have negative values for $`B_{1yy}`$. Therefore, for these two perovskites a transition from the $`aa`$ to the paraelectric $`p`$ phase is induced by applying a sufficiently high external tensile stress at fixed misfit strain, while for the others the transition would be from the $`p`$ to the $`aa`$ (or $`a`$, in the case of BaZrO<sub>3</sub>) phase.
For external stresses above $`\sigma _\times `$, two kinds of behaviors are found. Five of the perovskites (BaTiO<sub>3</sub>, SrTiO<sub>3</sub>, KNbO<sub>3</sub>, NaNbO<sub>3</sub>, and PbZrO<sub>3</sub>) show a $`c`$-$`r`$-$`aa`$ sequence of second-order phase transitions under these conditions, with the energies of the phases behaving as in Fig. 2(a). In this case, the phase boundaries are straight lines of the form
$`\sigma _{c\text{-}r}`$ $`=`$ $`{\displaystyle \frac{2BECF}{2B_\sigma EC_\sigma F}}`$ (39)
$`{\displaystyle \frac{2B_{\overline{\eta }}EC_{\overline{\eta }}F}{2B_\sigma EC_\sigma F}}\overline{\eta },`$
$`\sigma _{r\text{-}aa}`$ $`=`$ $`{\displaystyle \frac{2C(D+H/4)BF}{2C_\sigma (D+H/4)B_\sigma F}}`$ (40)
$`{\displaystyle \frac{2C_{\overline{\eta }}(D+H/4)B_{\overline{\eta }}F}{2C_\sigma (D+H/4)B_\sigma F}}\overline{\eta }.`$
For all five compounds, the slopes of both boundaries are positive.
The other behavior observed for external stress above $`\sigma _\times `$ is one in which $`c`$ and $`aa`$ (or $`a`$, in the case of BaZrO<sub>3</sub>) domains are expected. The energy curves behave either as shown in Fig. 2(c) (for CaTiO<sub>3</sub> and PbTiO<sub>3</sub>) or as shown in Fig. 2(d) (for BaZrO<sub>3</sub>). However, in the intermediate region, instead of a uniform phase, the system is expected to break into domains, as explained in the previous section for PbTiO<sub>3</sub> and CaTiO<sub>3</sub> at zero external stress.
## IV Discussion
In this section, we first consider in detail the several approximations that are responsible for the ease and simplicity with which we can generate stress-strain phase diagrams for perovskites. We then discuss and give examples of the applicability of the theory to realistic experimental studies of perovskite films and superlattices.
Within the KSV theory the thermodynamical potential is expanded as a Taylor series in strain and soft-mode amplitude, where the reference used is the perfect cubic perovskite of Fig. 1(a). The truncation at low order in the variables of the expansion means that the expansion decreases in accuracy for large distortions. As relevant misfit strains are generally rather small (less than 2%), this appears not to have significant implications. In addition, in the present form of the theory, the modes selected for the expansion allow only phases that involve five-atom unit cells to be considered. In particular, we do not take into account the possibility of cell-doubling oxygen octahedra rotations, which have been shown to be important in SrTiO<sub>3</sub>,Zhong1996PRB ; Vanderbilt1998F CaTiO<sub>3</sub>,Vanderbilt1998F and PbZrO<sub>3</sub>.Singh1997F In principle, such rotations could condense in the other compounds under high enough misfit strains, but for BaTiO<sub>3</sub> it has been shownDieguez2004PRB that this does not occur until one reaches experimentally irrelevant strains, and we expect that this will also be the case for most of these other compounds. Second, our calculations are done at zero temperature. Extending them to finite temperatures could in principle be done using an effective Hamiltonian method as we did in Ref. Dieguez2004PRB, for BaTiO<sub>3</sub>. However, this would involve designing for each perovskite an effective Hamiltonian along the lines described by Zhong, Vanderbilt, and Rabe,Zhong1994PRLandZhong1995PRB and is left for future work. Third, our model does not include the small effect of the zero-point motion of the ions (see Ref. Iniguez2002PRL, for a discussion). Finally, our theory relies on the LDA to compute the exchange and correlation terms in DFT. This introduces small systematic errors in the calculation, the most important of which is probably the error in the equilibrium lattice constant. However, such errors are well understood and well characterized in perovskites, and tend to be similar for different materials of this class, so that there is a tendency for cancellation of errors in relative quantities such as misfit strains (see, for example, Ref. KingSmith1994PRB, ).
The phenomenological Landau-Devonshire approach Pertsev1998PRL also requires approximations, which we review here for the purposes of comparison. Its starting thermodynamical potential is the bulk free energy expanded in polarization and stress, with linear temperature dependence in selected coefficients. Sixth order terms are needed as their importance increases at finite temperature.Iniguez2001 The reference used is the paraelectric cubic perovskite phase at the bulk critical temperature $`T_\text{c}`$, and the parameters are fit to reproduce experimental observations of the behavior near the bulk ferroelectric transition. For the epitaxial strain dependence, a Legendre transformation is then made to obtain the potential as a function of polarization and misfit strain. With parameters extrapolated to zero temperature, this can be compared to the potential in the present work using the linear relation between the polarization and the soft mode amplitude (31). Due to the way in which the parameters are fit, the Landau-Devonshire potential will give its most accurate results for small misfit strains and temperatures near the bulk $`T_\text{c}`$, while the first-principles potential will be more reliable for the zero-temperature misfit-strain phase diagram.
We now turn to the applicability of our theory to the prediction and understanding of properties of experimentally relevant systems. We first discuss the case of thin films, and then consider the case of strained-layer superlattices.
The theory presented here can be used directly to predict the structure and polarization of a single-domain perovskite-oxide thin film grown on a substrate with square-lattice symmetry. The effects of epitaxial strain will be most evident in films coherent with the substrate. In equilibrium, coherent epitaxial growth is possible up to a certain critical thickness, which depends upon the misfit between film and substrate materials as well as upon temperature and other growth conditions. Using low-temperature synthesis, coherence can be maintained far beyond the equilibrium critical thickness, as has been shown for BaTiO<sub>3</sub> grown on GdScO<sub>3</sub> ($`1.0`$% misfit strain).Choi2004S Under such coherent conditions, the full misfit should be used as the input for our theory. For example, in the case of BaTiO<sub>3</sub> on GdScO<sub>3</sub>, Fig. 6 shows a predicted enhancement of $`P_z`$ from 0.21 to 0.31 C/m<sup>2</sup> as a result of a 1.0% reduction of the in-plane lattice constant from the theoretical ground-state value of 7.448 bohr to 7.374 bohr. For thicker, partially-relaxed films, the strain should be taken to correspond to the in-plane lattice constant measured for the film Speck1994JAP ; this assumes that all the misfit dislocations responsible for the relaxation are located at the interface. These predictions are based on the assumption that the epitaxial strain strongly dominates other factors in determining the state of the film.
Similarly, for strained-layer superlattices, the structure and polarization of a perovskite-oxide layer in a superlattice will in general be significantly different from the corresponding bulk. The states of the component layers will largely be determined by the in-plane strain imposed by lattice matching to the other components and/or to the substrate, and can be obtained by referring to the theoretical phase diagrams at the relevant value of misfit strain. In addition, the normal component of the polarization in each layer, at the common in-plane strain, is an important consideration in determining the structure and polarization of the overall superlattice. If the normal polarization is discontinuous, electrostatic energy considerations will tend to polarize the low-polarization layer and depolarize the high-polarization layer to make the normal polarization uniform through the sublattice,Neaton2003 with accompanying changes in the normal strain of the layer. As an example, this analysis has proved useful in the interpretation of first-principles results and experiments on the structure of partially-relaxed SrTiO<sub>3</sub>/BaTiO<sub>3</sub> superlattices. Johnston2005 ; Rios2003JPCM
First-principles information about the effects of epitaxial strain on the structure and polarization of component layers also allows us to determine the relative stability of a given superlattice. It is well known that lattice matching of the components minimizes the elastic energy and thus increases the stability. In the case where one or more components have a nonzero normal polarization, the electrostatic energy of the superlattice can be minimized by โpolarization matching,โ where the components are selected to have the same normal polarization at the common in-plane strain. For example, according to our calculations, BaTiO<sub>3</sub> and NaNbO<sub>3</sub> have the same normal polarization of 0.34 C/m<sup>2</sup> at a common in-plane theoretical lattice constant of 7.336 bohr, corresponding to small compressive misfit strains of less that 0.2%, and thus this is a favorable combination with respect both to elastic and electrostatic energy. In general, however, it is not possible to minimize both elastic and electrostatic energy in this way, and a trade-off between the two is necessary to form the superlattice.
It should be kept in mind, however, that for both films and superlattices, the assumptions that the system is in a single domain and that the epitaxial strain strongly dominates other factors will not be valid in all cases. Phase diagrams including multiple-domain states have, for example, been discussed in Refs. Speck1994JAP, ; Pertsev2000PRL, ; Li2003APL, . Other influences that may be important include surface relaxation and reconstruction, atomic and electronic rearrangements at the interface, imperfectly compensated macroscopic electric fields, deviations from stoichiometry, and the presence of defects.
## V Summary
We have applied the first-principles total-energy parameterization of King-Smith and VanderbiltKingSmith1994PRB to study the effects of epitaxial strain and external stress on the structure and properties of perovskites. We report phase diagrams and polarizations for the same set of eight compounds as in Ref. KingSmith1994PRB, : BaTiO<sub>3</sub>, SrTiO<sub>3</sub>, CaTiO<sub>3</sub>, KNbO<sub>3</sub>, NaNbO<sub>3</sub>, PbTiO<sub>3</sub>, PbZrO<sub>3</sub>, and BaZrO<sub>3</sub>. An updated set of parameters, computed with a comparable first-principles method at higher precision, are provided in Table 3. The simple form of the parameterization is seen to be useful in reducing the computational effort for generating these phase diagrams relative to full first-principles calculations, and many features of the phase diagrams can be extracted and interpreted analytically.
We have discussed the use of these results in predicting the structure and polarization of epitaxial perovskite films and strained layer superlattices. Additional properties of interest, such as dielectric or piezoelectric constants can also be computed within this framework.
###### Acknowledgements.
We thank Javier Junquera for useful discussions. This work was supported by ONR Grants N0014-05-1-0054, N00014-00-1-0261, and N00014-01-1-0365, and DOE Grant DE-FG02-01ER45937.
|
warning/0506/math0506356.html
|
ar5iv
|
text
|
# The Hopf invariant of a Haefliger knot
## 1. Introduction
Haefliger discovered differentiably knotted spheres in codimension greater than three. In a particular case, he proved the isomorphism
$$\mathrm{\Omega }:C_{4k1}^{2k+1}๐,$$
where $`C_{4k1}^{2k+1}`$ denotes the group of differentiable isotopy classes of embeddings of $`S^{4k1}`$ in $`S^{6k}`$ ($`k1`$).
We deal with the case where $`k=1`$. We know that any such Haefliger knot $`F:S^3S^6`$ as above has a Seifert surface, that is, an embedding $`\stackrel{~}{F}:V^4S^6`$ of a compact oriented $`4`$-manifold whose restriction to the boundary $`V^4(=S^3)`$ coincides with $`F`$ (see ,\[8, p.95\] and \[24, Corollary 6.2\]). Furthermore, we can consider, associated to such a Seifert surface, the Hopf invariant $`H_{\stackrel{~}{F}}๐`$ of the map from $`S^3`$ into $`S^6F(S^3)S^2`$, determined by the outward normal field of $`F(S^3)\stackrel{~}{F}(V^4)`$. Then, by \[24, Corollary 6.5\], the Haefliger invariant $`\mathrm{\Omega }(F)`$ of $`F`$ is represented as
$$\mathrm{\Omega }(F)=\frac{1}{8}(\sigma (V^4)+H_{\stackrel{~}{F}}),$$
where $`\sigma (V^4)`$ denotes the signature of $`V^4`$.
Regarding this formula, we pose a natural question: for a given Haefliger knot, which integer can be realised as the Hopf invariant of a Seifert surface? In Corollaries 2.13 and 2.14, we give a complete answer to it; namely, we show that given a Haefliger knot and given an integer, we can choose a Seifert surface so that its Hopf invariant is equal to (or its signature is equal to) the given integer.
In ยง2.1, we will correct a sign error of our previous paper \[24, Lemma 5.2\]. With this correction, the sign should be changed in each term involving the square of the normal Euler class, appearing in \[24, Theorem 5.1, Corrollaries 6.2, 6.3(a) and 6.5\]. For details, see ยง2.1.
As a consequence, with the help of Ekholm and Szลฑcsโ formula for the Smale invariant $`\omega :Imm[S^3,๐^5]๐`$ (denoting by $`Imm[S^3,๐^5]`$ the group of regular homotopy classes of immersions of $`S^3`$ in $`๐^5`$), we show that for an embedding $`F:S^3๐^6`$ and an immersion $`f:S^3๐^5`$ with even Smale invariant, we can isotope $`F`$ so that its composition with the projection $`๐^6๐^5`$ becomes an immersion regularly homotopic to $`f`$ (Theorem 3.1). In particular, any immersion $`S^3๐^5`$ with even Smale invariant can be regularly homotoped to the projection of the unknot in $`C_3^3`$. In Corollary 3.4, we show that an immersion $`f:S^3๐^5`$ can be regularly homotoped to the projection of an embedding $`S^3๐^6`$ if and only if its Smale invariant $`\omega (f)`$ is even.
In ยง4, we discuss aspects of embeddings of punctured $`4`$-manifolds in $`S^6`$.
We work in the $`C^{\mathrm{}}`$-differentiable category; all manifolds and mappings are supposed to be differentiable of class $`C^{\mathrm{}}`$, unless otherwise explicitly stated. Throughout this paper, all $`C^{\mathrm{}}`$-mappings shall be in the so called nice dimensions (see ), where the two notions stable map and generic map are equivalent. We will use the term generic according to \[6, ยง2.1\]. For a map $`f:XY`$ from a manifold $`X`$ with non-empty boundary we often denote its restriction to the boundary by $`f(:=f|_X:XY`$).
We will suppose the spheres are oriented. If $`M`$ is an oriented manifold with non-empty boundary, then for the induced orientation of $`M`$ we adopt the outward vector first convention: we say an ordered basis of $`T_p(M)`$ ($`pM`$) is positively oriented if an outward vector followed by the basis is a positively oriented basis of $`T_pM`$. For a closed $`n`$-dimensional manifold $`M`$ we denote its punctured manifold by $`M_{}`$; i.e., $`M_{}:=MIntD^n`$.
We use the symbol โ$``$โ for a group isomorphism and โ$``$โ for a diffeomorphism between manifolds; the symbol โ$``$โ means a homotopy equivalence between two topological spaces. The homology and cohomology groups are supposed to be with integer coefficients unless otherwise explicitly noted.
## 2. Seifert surfaces for Haefliger knots and their Hopf invariants
Haefliger has proved in and \[10, ยง5.16\] that the group $`C_{4k1}^{2k+1}`$ of isotopy classes of embeddings $`S^{4k1}S^{6k}`$ is isomorphic to the integers $`๐`$ ($`k1`$). Since he has also given an explicit construction of an embedding representing a generator of $`C_{4k1}^{2k+1}`$, with respect to his generator we have the identification of $`C_{4k1}^{2k+1}`$ with $`๐`$, which we call the Haefliger invariant
$$\mathrm{\Omega }:C_{4k1}^{2k+1}๐.$$
For a Haefliger knot, we can consider its Seifert surface analogously as in the usual knot theory in codimension two.
###### Definition 2.1.
For an embedding $`F:S^{4k1}S^{6k}`$, a Seifert surface for $`F`$ is an embedding $`\stackrel{~}{F}:V^{4k}S^{6k}`$ of a compact connected oriented $`4k`$-manifold with boundary $`V^{4k}=S^{4k1}`$, whose restriction $`\stackrel{~}{F}:S^{4k1}S^{6k}`$ to the boundary coincides with $`F`$.
We will discuss the existence of such a Seifert surface in ยง2.2. Associated to a Seifert surface, we can consider the Hopf invariant as in .
###### Definition 2.2.
Let $`\stackrel{~}{F}:V^{4k}S^{6k}`$ be a Seifert surface for an embedding $`F:S^{4k1}S^{6k}`$. Then, the outward normal field of $`F(S^{4k1})\stackrel{~}{F}(V^{4k})`$ determines the map $`\nu _{\stackrel{~}{F}}:S^{4k1}S^{6k}F(S^{4k1})`$. We consider its homotopy class $`[\nu _{\stackrel{~}{F}}]`$ to be lying in $`\pi _{4k1}(S^{6k}F(S^{4k1}))=\pi _{4k1}(S^{2k})`$, via the homotopy equivalence $`p\psi :S^{6k}F(S^{4k1})\stackrel{}{}D^{4k}\times S^{2k}\stackrel{}{}S^{2k}`$, where $`\psi `$ is an orientation-preserving diffeomorphism (see \[20, Theorem 5.2\]) and $`p`$ is the projection. Thus, we define the Hopf invariant $`H_{\stackrel{~}{F}}`$ for $`\stackrel{~}{F}`$ to be the Hopf invariant of $`[\nu _{\stackrel{~}{F}}]`$.
### 2.1. A formula for the Haefliger invariant โ a correction to the paper โA geometric formula for Haefliger knotsโ \[Topology 43 (2004) 1425-1447\]
In our previous paper , we related the Hopf invariant for $`\stackrel{~}{F}`$ to the normal Euler class of $`\stackrel{~}{F}`$; however, there was an error about a sign. This subsection is devoted to correcting the sign error and reviewing a formula for the Haefliger invariant with the corrected sign (see also ).
The error occurs in the last sentence of the proof of Lemma 5.2 in \[24, p.1443\]. It reads โThus, we have $`H_{\stackrel{~}{F}}=[\stackrel{~}{F}(\mathrm{\Sigma }^{2k})].`$โ but should be โThus, the desired functional product is computed to be equal to $`[\stackrel{~}{F}(\mathrm{\Sigma }^{2k})]`$ and hence the Hopf invariant $`H_{\stackrel{~}{F}}`$ is equal to $`[\stackrel{~}{F}(\mathrm{\Sigma }^{2k})]H_{2k}(X)=๐`$.โ The reason is that the Hopf invariant is defined, in \[24, ยง2.2\], to be minus the functional cup product. Note that the similar argument can be found in the last part of the proof of Proposition 3.5 in \[24, p.1436\].
With this correction, each term involving the square of the normal Euler class, appearing in \[24, Theorem 5.1, Corrollaries 6.2, 6.3(a) and 6.5\], should change its sign, as follows.
First, Theorem 5.1 in \[24, p.1442\] should be written as:
###### Theorem 2.3 (\[24, Theorem 5.1\]).
Let $`\widehat{V}^{4k}`$ be a closed oriented $`4k`$-manifold and put $`V^{4k}:=\widehat{V}^{4k}IntD^{4k}`$. Let $`\stackrel{~}{F}:V^{4k}S^{6k}`$ be an embedding with normal Euler class $`e_{\stackrel{~}{F}}H^{2k}(V^{4k})=H^{2k}(\widehat{V}^{4k})`$. Then, the Hopf invariant $`H_{\stackrel{~}{F}}`$ (along the boundary) is equal to $`e_{\stackrel{~}{F}}e_{\stackrel{~}{F}}H^{4k}(\widehat{V}^{4k})=๐`$.
Putting $`M^{4k}:=S^{2k}\times S^{2k}IntD^{4k}`$, Corollary 6.2 in \[24, p.1444\] should be as:
###### Corollary 2.4 (\[24, Corollary 6.2\]).
Let $`\stackrel{~}{E}_{a,b}:M^{4k}S^{6k}`$ be an embedding with normal Euler class $`(2a,2b)H^{2k}(M^{4k})=H^{2k}(S^{2k}\times S^{2k})๐๐`$. Then $`E_{a,b}:=\stackrel{~}{E}_{a,b}|_{M^{4k}}:S^{4k1}S^{6k}`$ represents $`abC_{4k1}^{2k+1}=๐`$. In particular, $`E_{\pm 1,\pm 1}:S^{4k1}S^{6k}`$ represents the generator of $`C_{4k1}^{2k+1}`$.
Furthermore, Corollary 6.3(a) in \[24, p.1444\] should be as:
###### Corollary 2.5 (\[24, Corollary 6.3(a)\]).
(a) For an arbitrary embedding $`F:S^{4k1}S^{6k}`$, there exists an embedding $`\stackrel{~}{F}:V^{4k}S^{6k}`$ of a compact oriented $`4k`$-manifold $`V^{4k}`$ with $`V^{4k}=S^{4k1}`$ such that $`\stackrel{~}{F}|_{V^{4k}}=F`$. Furthermore,
$`\mathrm{\Omega }(F)`$ $`=`$ $`{\displaystyle \frac{1}{24}}(\overline{p}_k[\widehat{V}^{4k}]+3H_{\stackrel{~}{F}})`$
$`=`$ $`{\displaystyle \frac{1}{24}}(\overline{p}_k[\widehat{V}^{4k}]3e_{\stackrel{~}{F}}e_{\stackrel{~}{F}})`$
gives the isomorphism $`\mathrm{\Omega }:C_{4k1}^{2k+1}๐`$, where $`e_{\stackrel{~}{F}}H^{2k}(V^{4k})=H^{2k}(\widehat{V}^{4k})`$ is the normal Euler class for $`\stackrel{~}{F}`$ and $`e_{\stackrel{~}{F}}e_{\stackrel{~}{F}}H^{4k}(\widehat{V}^{4k})=๐`$ is the cup product.
Corollary 6.5 in \[24, p.1445\] in the case $`C_3^3`$ ($`k=1`$) should be as:
###### Corollary 2.6 (\[24, Corollary 6.5\], see also \[8, p.95\]).
Every embedding $`F:S^3S^6`$ extends to an embedding $`\stackrel{~}{F}:V^4S^6`$ of a compact oriented $`4`$-manifold $`V^4`$, and
$`\mathrm{\Omega }(F)`$ $`=`$ $`{\displaystyle \frac{1}{8}}(\sigma (V^4)+H_{\stackrel{~}{F}})`$
$`=`$ $`{\displaystyle \frac{1}{8}}(\sigma (V^4)e_{\stackrel{~}{F}}e_{\stackrel{~}{F}})`$
gives the isomorphism $`\mathrm{\Omega }:C_3^3๐.`$
### 2.2. Constructions of Seifert surfaces
From now on, we will confine ourselves to the case $`C_3^3`$ ($`k=1`$). In this subsection, we review some concrete constructions of Seifert surfaces for Haefliger knots, which have been used in \[8, p.95\] and \[24, Corollary 6.2\].
First, we write down the particular case of Corollary 2.4 in $`k=1`$.
###### Corollary 2.7.
Let $`\stackrel{~}{E}_{a,b}:(S^2\times S^2)_{}S^6`$ be an embedding with normal Euler class $`(2a,2b)H^2((S^2\times S^2)_{})=H^2(S^2\times S^2)๐๐`$. Then $`E_{a,b}:=\stackrel{~}{E}_{a,b}|_{(S^2\times S^2)_{}}:S^3S^6`$ represents $`abC_3^3=๐`$.
We see that for any (isotopy class of) Haefliger knot we can construct its Seifert surface by choosing a suitable pair $`(a,b)`$ of integers. In particular, if we put $`(a,b)=(1,1)`$ in Corollary 2.7, the embedding $`\stackrel{~}{E}_{1,1}:(S^2\times S^2)_{}S^6`$ is a Seifert surface for the generator of $`C_3^3=๐`$, of signature zero and with Hopf invariant $`8`$ (by Theorem 2.3).
A clue given for \[8, Exercise 3(ii) on p.95\] elicits another construction.
###### Proposition 2.8 (see \[8, p.95\]).
The punctured complex projective plane $`๐P_{}^{2}{}_{}{}^{}`$ can be embedded in $`S^6`$ with normal Euler class $`2k+1H^2(๐P_{}^{2}{}_{}{}^{})๐`$ $`(k๐)`$. If we denote by $`\stackrel{~}{E}_k:๐P_{}^{2}{}_{}{}^{}S^6`$ such an embedding with normal Euler class $`2k+1`$, then its Hopf invariant is equal to $`4k^24k1`$ and $`\stackrel{~}{E}_k:S^3S^6`$ represents $`k(k+1)/2C_3^3=๐`$.
###### Proof.
Consider the $`2`$-dimensional disc bundle $`\eta `$ over $`๐P_{}^{2}{}_{}{}^{}`$ with Euler class $`2k+1H^2(๐P_{}^{2}{}_{}{}^{};๐)=๐`$. Since the second Stiefel-Whitney class $`w_2(\eta )H^2(๐P_{}^{2}{}_{}{}^{};๐_2)=๐_2`$ of $`\eta `$ is congruent to $`2k+1`$ modulo $`2`$, we have $`w_2(T๐P_{}^{2}{}_{}{}^{}\eta )=w_2(T๐P_{}^{2}{}_{}{}^{})1+11=0H^2(๐P_{}^{2}{}_{}{}^{};๐_2)`$. Therefore, together with $`\pi _3(SO(6))=0`$, we see that $`T๐P_{}^{2}{}_{}{}^{}\eta `$ is trivial. Then, Hirschโs h-principle implies that $`๐P_{}^{2}{}_{}{}^{}`$ can be immersed in $`S^6`$ with normal bundle isomorphic to $`\eta `$. Let $`\stackrel{~}{F}:๐P_{}^{2}{}_{}{}^{}S^6`$ be such an immersion with normal Euler class $`2k+1`$. (Actually the above argument implies that the normal Euler class of such an immersion ought to be of the form $`2k+1H^2(๐P_{}^{2}{}_{}{}^{};๐)=๐`$).
Since $`๐P_{}^{2}{}_{}{}^{}`$ is the total space of a $`2`$-dimensional disc bundle over $`๐P^1S^2`$, the immersion $`\stackrel{~}{F}`$ is regularly homotopic to an embedding. Since a regular homotopy does not change the normal bundle, the first part of the proposition has been proved.
Let $`\stackrel{~}{E}_k:๐P_{}^{2}{}_{}{}^{}S^6`$ be an embedding with normal Euler class $`2k+1`$. Then, by Theorem 2.3, its Hopf invariant $`H_{\stackrel{~}{F}}`$ is equal to $`(2k+1)(2k+1)=4k^24k1H^4(๐P^2;๐)=๐`$, where we consider $`2k+1H^2(๐P_{}^{2}{}_{}{}^{};๐)`$ to be in $`H^2(๐P^2;๐)`$. Thus, by Corollary 2.6, we have $`\mathrm{\Omega }(\stackrel{~}{E}_k)=(14k^24k1)/8=k(k+1)/2`$. This completes the proof. โ
###### Remark 2.9.
In Proposition 2.8, if we use the complex projective plane $`\overline{๐P^2}`$ with the reversed orientation, we have a Seifert surface, of signature $`1`$ with the Hopf invariant $`4k^2+4k+1`$, for an embedding representing $`k(k+1)/2C_3^3=๐`$.
###### Remark 2.10.
Corollary 2.7 implies that the punctured $`S^2\times S^2`$ can be a Seifert surface for any (isotopy class of) Haefliger knot. On the other hand, Proposition 2.8 shows that the punctured complex projective plane $`๐P_{}^{2}{}_{}{}^{}`$ (resp. $`\overline{๐P^2}_{}`$) can be a Seifert surface only for non-negative classes (resp. for non-positive classes) of $`C_3^3=๐`$.
### 2.3. Seifert surfaces with given Hopf invariants
Regarding the formula in Corollary 2.6, it is natural to ask which integer can be realised as the Hopf invariant (or as the signature) of a Seifert surface for a given Haefliger knot. We will give a complete answer to it in Corollaries 2.13 and 2.14.
###### Proposition 2.11.
For the standard inclusion $`S^3S^6`$, there exists a Seifert surface whose Hopf invariant is equal to $`1`$.
###### Proof.
If we put $`k=0`$ in Proposition 2.8, then we have the embedding with Hopf invariant $`1`$, denoted by $`\stackrel{~}{P}:๐P_{}^2S^6`$, whose restriction to the boundary represents $`(11)/8=0C_3^3=๐`$ and hence is isotopic to the standard inclusion. Since a differentiable isotopy implies a differentiable ambient isotopy, the embedding $`\stackrel{~}{P}`$ followed by a suitable diffeomorphism on $`S^6`$ gives a desired Seifert surface for the standard inclusion. โ
###### Remark 2.12.
The similar argument for $`\overline{๐P^2}_{}`$ (putting $`k=0`$ in Remark 2.9) provides another Seifert surface $`\stackrel{~}{Q}:\overline{๐P^2}_{}S^6`$, of signature $`1`$ and with Hopf invariant $`1`$, for the standard inclusion $`S^3S^6`$.
Proposition 2.11 and Remark 2.12 imply that for a given Haefliger knot we can change its Seifert surface so as to have a desired Hopf invariant (in return for changing the signature), by taking the boundary connected sum with $`\stackrel{~}{P}`$ or with $`\stackrel{~}{Q}`$. Namely we have:
###### Corollary 2.13.
For an arbitrary embedding $`F:S^3S^6`$ and an arbitrary integer $`k๐`$, there exists a Seifert surface for $`F`$ whose Hopf invariant is equal to $`k`$.
###### Proof.
Take an arbitrary Seifert surface for $`F`$. Then, by taking the boundary connected sum with $`\stackrel{~}{P}`$ (resp. with $`\stackrel{~}{Q}`$), we obtain a new Seifert surface with signature which differs by $`1`$ (resp. by $`1`$) to the initial one, without changing the isotopy class of the embedding $`S^3S^6`$ on the boundary. Repeat this procedure until we have a Seifert surface with the Hopf invariant equal to $`k`$. Thus, composing it with a suitable ambient isotopy if necessary, we obtain a desired Seifert surface. โ
In view of the formula for the Haefliger invariant (Corollary 2.6), a very similar argument provides the following.
###### Corollary 2.14.
For an arbitrary embedding $`F:S^3S^6`$ and an arbitrary integer $`k๐`$, there exists a Seifert surface for $`F`$ with signature $`k`$.
Recall that the normal bundle of an embedding $`F:S^3S^6`$ is trivial and that the homotopy classes of its normal framings are classified by the elements of $`\pi _3(SO(3))๐`$. Furthermore, the homomorphism $`\pi _3(SO(3))\pi _3(S^2)`$ induced by the natural map $`SO(3)SO(3)/SO(2)=S^2`$ and the (usual) Hopf invariant $`\pi _3(S^2)๐`$ are both isomorphisms. Therefore, the following is just an interpretation of Corollary 2.13.
###### Corollary 2.15.
For an arbitrary embedding $`F:S^3S^6`$ and an arbitrary normal vector field to $`F(S^3)`$ in $`S^6`$, there exists a Seifert surface $`\stackrel{~}{F}:V^4S^6`$ for which the normal field along the boundary $`F(S^3)\stackrel{~}{F}(V^4)`$ is in accordance with the given normal vector field.
## 3. The projection of a Haefliger knot
The purpose of this section is to prove the following two Theorems 3.1 and 3.2. Let $`p:๐^6๐^5`$ be the projection, defined by dropping off the last coordinate: $`(x_0,\mathrm{},x_5,x_6)(x_0,\mathrm{},x_5)`$.
###### Theorem 3.1.
Let $`f:S^3๐^5`$ be an immersion with even Smale invariant and $`F:S^3๐^6`$ an embedding. Then, there exist an immersion $`f^{}:S^3๐^5`$ regularly homotopic to $`f`$ and an embedding $`F^{}:S^3๐^6`$ isotopic to $`F`$ such that $`f^{}=pF^{}`$.
###### Theorem 3.2.
If the projection $`pF`$ of an embedding $`F:S^3๐^6`$ is an immersion, then its Smale invariant $`\omega (pF)`$ is even.
As an easy corollary of Theorem 3.1, we have:
###### Corollary 3.3.
Any immersion $`f:S^3๐^5`$ with even Smale invariant can be regularly homotoped to the projection of the unknot in $`C_3^3`$.
Combining Theorem 3.1 and Theorem 3.2, we have the following.
###### Corollary 3.4.
An immersion $`f:S^3๐^5`$ can be regularly homotoped to the projection of an embedding $`S^3๐^6`$ if and only if its Smale invariant $`\omega (f)`$ is even.
###### Remark 3.5.
It has been shown in that if a closed $`n`$-manifold ($`n3`$) is non-orientable or odd-dimensional then there exists an immersion with normal crossings $`M^n๐^{2n1}`$ which can never be lifted to an embedding into $`๐^{2n}`$.
The proofs of Theorems 3.1 and 3.2 will depend on Ekholm and Szลฑcsโ geometric formula , which enables us to read off the Smale invariant of, hence the regular homotopy class of, an immersion of $`S^{4k1}`$ in $`๐^{4k+1}`$ ($`k1`$) in terms of its singular Seifert surface. We will first review their formula (only in our case, i.e., when $`k=1`$) in ยง3.1 and then give the proofs of Theorems 3.1 and 3.2 in ยง3.2.
### 3.1. Ekholm and Szลฑcsโ formula
We need to recall some definitions from \[6, ยง2.4\]. Let $`g:M^4๐^5`$ be a generic map of a compact oriented $`4`$-manifold. Then, the set $`\stackrel{~}{\mathrm{\Sigma }}(g)`$ of its singular points is a $`2`$-dimensional submanifold of $`M^4`$, in which finitely many isolated cusp points lie. Note that each cusp point has the naturally induced orientation \[22, Appendix A\].
###### Definition 3.6 (see \[6, Definition 2.10 and Remark 6.2\]).
Let $`g:M^4๐^5`$ be a generic map of an oriented $`4`$-manifold. Then let $`\xi (g)`$ denote the $`2`$-dimensional vector bundle over $`\stackrel{~}{\mathrm{\Sigma }}(g)`$, whose fibre over $`p\stackrel{~}{\mathrm{\Sigma }}(g)`$ is $`\xi (g)_p=T_{g(p)}๐^5/dg(T_pM^4)`$ and $`e[\xi (g)]`$ denote its Euler number (in the sense of \[6, Definition 2.10\]).
###### Remark 3.7 (\[6, Remark 6.4\]).
When a generic map $`g:M^4๐^5`$ of an oriented $`4`$-manifold $`M^4`$ is the composition of an immersion $`G:M^4๐^6`$ with the projection $`๐^6๐^5`$, we have a more convenient description of the Euler number $`e[\xi (g)]`$ in Definition 3.6. In such a situation, the homology class of $`\stackrel{~}{\mathrm{\Sigma }}(g)`$ in $`M^4`$ is dual to the normal Euler class of $`G`$, and the bundle $`\xi (g)`$ is isomorphic to the normal bundle of $`G`$ restricted to $`\stackrel{~}{\mathrm{\Sigma }}(g)`$, whose Euler class evaluated by the orientation class $`[\stackrel{~}{\mathrm{\Sigma }}(g)]`$ precisely equals $`e[\xi (g)]`$.
Ekholm and Szลฑcs \[6, Theorem 1.1 and Remark 3.1\] have given a formula for the Smale invariant
$$\omega :Imm[S^3,๐^5]๐,$$
which gives a group isomorphism between the group $`Imm[S^3,๐^5]`$ of regular homotopy classes of immersions of $`S^3`$ in $`๐^5`$ and the integers $`๐`$.
###### Theorem 3.8 (\[6, Theorem 1.1(a) and Remark 3.1\]).
Let $`f:S^3๐^5`$ be an immersion and $`\stackrel{~}{f}:V^4๐^5`$ be a singular Seifert surface for $`f`$, that is, a generic map with $`\stackrel{~}{f}=f`$ which has no singularity near the boundary. Then, we have
$`\omega (f)`$ $`=`$ $`{\displaystyle \frac{1}{2}}(3\sigma (V^4)+e[\xi (\stackrel{~}{f})])`$
$`=`$ $`{\displaystyle \frac{1}{2}}(3\sigma (V^4)+\mathrm{\#}\mathrm{\Sigma }^{1,1}(\stackrel{~}{f})),`$
where $`\mathrm{\#}\mathrm{\Sigma }^{1,1}(\stackrel{~}{f})`$ denotes the algebraic number of cusp points of $`\stackrel{~}{f}`$.
Note that this generalises the result of Hughes and Melvin since we can consider a usual non-singular Seifert surface for an embedding.
### 3.2. The proofs of Theorems 3.1 and 3.2
###### Proof of Theorem 3.1.
In what follows, we often consider $`F`$ to be an embedding in $`S^6`$ and use the same symbol for it: $`F:S^3S^6=๐^6\{\mathrm{}\}`$.
Since the Smale invariant $`\omega (f)`$ is even, we can take the pair $`(A,B)`$ of integers satisfying the following simultaneous linear equations:
$$\{\begin{array}{ccc}(AB)/8\hfill & =& \mathrm{\Omega }(F)\hfill \\ (3A+B)/2\hfill & =& \omega (f).\hfill \end{array}$$
According to Corollary 2.14, take a Seifert surface $`\stackrel{~}{F}:V^4S^6`$ for $`F`$ with $`\sigma (V^4)=A`$. Then, the Hopf invariant $`H_{\stackrel{~}{F}}`$ should be equal to $`B`$ since $`\mathrm{\Omega }(F)=(\sigma (V^4)+H_{\stackrel{~}{F}})/8=(AB)/8`$.
Consider the normal framing $`\nu =(\nu _1,\nu _2,\nu _3)`$ for $`F`$ such that the first vector field $`\nu _1`$ coincides with the normal field of $`F(S^3)\stackrel{~}{F}(V^4)`$. This is always possible since $`\pi _3(V_{3,1})\pi _3(SO(3))`$.
Then, by using the Compression Theorem , we can isotope $`F`$ to a compressible embedding $`F^{}`$ so that the composition $`pF^{}`$ is an immersion $`S^3๐^5`$. Here we mean by โcompressibleโ that the third normal vector field $`\nu _3`$ is parallel to $`/x_6`$. Furthermore, by using a suitable ambient isotopy, we can isotope $`\stackrel{~}{F}:V^4๐^6`$ to $`\stackrel{~}{F}^{}:V^4๐^6`$ with $`\stackrel{~}{F}^{}=F^{}`$, so that the composition $`p\stackrel{~}{F}^{}:V^4๐^5`$ has no singularity near its boundary. Since we can further assume that $`p\stackrel{~}{F}^{}`$ is generic, $`p\stackrel{~}{F}^{}:V^4๐^5`$ is a singular Seifert surface (see Theorem 3.8) for the immersion $`pF^{}:S^3๐^5`$.
Now, we will compute the Smale invariant $`\omega (pF^{})`$ by using Ekholm-Szลฑcsโ formula (Theorem 3.8):
$$\omega (pF^{})=\frac{1}{2}(3\sigma (V^4)+e[\xi (p\stackrel{~}{F}^{})]).$$
If we denote by $`e_{\stackrel{~}{F}^{}}`$ the normal Euler class of $`\stackrel{~}{F}^{}`$ and by $`e(\xi (p\stackrel{~}{F}^{}))`$ the Euler class of the bundle $`\xi (p\stackrel{~}{F}^{})`$ over the set $`\stackrel{~}{\mathrm{\Sigma }}(p\stackrel{~}{F}^{})`$ as in Definition 3.6, then, according to Remark 3.7 (see also \[6, Remark 6.4\]), the homology class of $`\stackrel{~}{\mathrm{\Sigma }}(p\stackrel{~}{F}^{})`$ in $`V^4`$ is dual to $`e_{\stackrel{~}{F}^{}}`$ and the bundle $`\xi (p\stackrel{~}{F}^{})`$ is isomorphic to the normal bundle of $`\stackrel{~}{F}^{}`$ restricted to $`\stackrel{~}{\mathrm{\Sigma }}(p\stackrel{~}{F}^{})`$. Therefore, we see that
$$e[\xi (p\stackrel{~}{F}^{})]=e(\xi (p\stackrel{~}{F}^{})),[\stackrel{~}{\mathrm{\Sigma }}(p\stackrel{~}{F}^{})]=e_{\stackrel{~}{F}^{}}e_{\stackrel{~}{F}^{}},[V^4,V^4],$$
which is further equal to minus the Hopf invariant $`H_{\stackrel{~}{F}}=B`$ by Theorem 2.3. Finally, we have
$$\omega (pF^{})=\frac{1}{2}(3\sigma (V^4)+e[\xi (p\stackrel{~}{F}^{})])=\frac{1}{2}(3A+B)=\omega (f).$$
This means that the immersion $`f^{}:=pF^{}`$ is regularly homotopic to the given immersion $`f`$. โ
###### Proof of Theorem 3.2.
Take a normal framing $`\nu =(\nu _1,\nu _2,\nu _3)`$ for $`F`$ such that the third vector field $`\nu _3`$ is in accordance with $`/x_6`$, which is not tangent to $`F(S^3)๐^6`$ since $`pF`$ is an immersion.
By Corollary 2.15, we can consider a Seifert surface $`\stackrel{~}{F}:V^4S^6`$ for which the normal field along the boundary $`F(S^3)\stackrel{~}{F}(V^4)`$ coincides with the first normal vector field $`\nu _1`$.
Then, the composition $`p\stackrel{~}{F}`$ of $`\stackrel{~}{F}`$ with the projection $`p`$ has no singularity near the boundary and can be considered to be a singular Seifert surface for $`pF`$.
By the same argument as in the proof of Theorem 3.1, the Smale invariant $`\omega (pF)`$ can be computed as:
$$\omega (pF)=\frac{1}{2}(3\sigma (V^4)+e_{\stackrel{~}{F}}e_{\stackrel{~}{F}}),$$
where $`e_{\stackrel{~}{F}}e_{\stackrel{~}{F}}H^4(V^4,V^4)=๐`$ is the square of the normal Euler class of $`\stackrel{~}{F}`$. Since $`\sigma (V^4)e_{\stackrel{~}{F}}e_{\stackrel{~}{F}}(mod8)`$, we have $`\omega (pF)2\sigma (V^4)(mod4)`$. Thus, $`\omega (pF)`$ is even. โ
###### Remark 3.9.
It seems that Theorem 3.2 can be deduced from the following argument. Denote by $`\delta (f)`$ the number of non-trivial components of the set of double points of an immersion $`f:S^3๐^5`$ (see ). Then, \[5, Proposition 7.5.2 and Corollary 8.3.2\] imply that $`\delta (f)(mod2)`$ determines the non-trivial homomorphism $`Imm[S^3,๐^5]๐_2`$, which consequently coincides with the Smale invariant $`\omega (f)(mod2)`$. According to Szลฑcs , if an immersion $`f:S^3๐^5`$ can be lifted to an embedding in $`S^6`$, then $`\delta (f)`$ and hence $`\omega (f)`$ must be even.
## 4. Punctured embeddings of $`4`$-manifolds in codimension $`2`$
It is natural to ask if a similar construction as in ยง2.2 is applicable to other oriented $`4`$-manifolds. We argue here the case of the Kummer surface $`K`$. The following is proved very similarly to Proposition 2.8.
###### Proposition 4.1.
The punctured Kummer surface $`K_{}`$ can be embedded in $`S^6`$ with normal Euler class $`((0,\mathrm{},0),(0,\mathrm{},0),(0,0),(0,0),(2a,2b))H^2(K_{})=H^2(K)=๐^8๐^8๐^2๐^2๐^2`$ (represented with respect to the decomposition $`E_8E_8\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$ of the intersection form of $`K`$) for any integers $`a,b๐`$. Such an embedding restricted to the boundary represents $`2+abC_3^3=๐`$.
###### Proof.
Since the Stiefel-Whitney class $`w_2(K)`$ is the reduction of the integral class $`\eta :=((0,\mathrm{},0),(0,\mathrm{},0),(0,0),(0,0),(2a,2b))H_2(V^4;๐)`$, by Hirschโs h-principle , there is an immersion of $`K_{}`$ in $`S^6`$ with normal Euler class $`\eta `$. Since the punctured Kummer surface $`K_{}`$ has a handlebody decomposition with a $`0`$-handle and several $`2`$-handles with even framings, this immersion is regularly homotopic to an embedding.
The latter part is just due to a computation of the Haefliger invariant:
$$\frac{1}{8}\left(\sigma (K_{})\left(\begin{array}{cc}2a& 2b\end{array}\right)\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)\left(\begin{array}{c}2a\\ 2b\end{array}\right)\right)=\frac{168ab}{8}=2+ab.$$
This completes the proof. โ
###### Remark 4.2.
In fact, the punctured Kummer surface can be embedded in $`S^5`$. Since such an embedding in $`S^5`$ has trivial normal bundle, its composition with the inclusion $`S^5S^6`$ restricted to the boundary represents $`2C_3^3=๐`$ (see \[10, Theorem 5.17\] and \[24, ยง7\]). Proposition 4.1, however, implies that for a suitable pair $`(a,b)`$ of integers, the punctured Kummer surface can be a Seifert surface for any (isotopy class of) Haefliger knot (see Remark 2.10).
###### Remark 4.3.
Combining Corollary 2.7, Propositions 2.8, 4.1 and Remark 2.9, we see that any indefinite symmetric unimodular form can be realised as the intersection form of a Seifert surface for a Haefliger knot.
We easily see that many other orientable 4-manifolds can be Seifert surfaces for Haefliger knots. For example, the punctured manifold $`V_{}^4`$ of a closed spin manifold $`V^4`$ can be embedded in $`S^6`$, since the double $`(V_{}^4\times [0,1])`$ of $`V_{}^4`$ is a closed spin $`4`$-manifold of signature zero and such a manifold can be embedded in $`S^6`$ (e.g., see and \[1, Theorem 2.5\]). We also see that a simply-connected punctured $`4`$-manifold can be embedded in $`S^6`$ (\[12, Corollary 4.2\]).
Note that if $`V^4`$ is a closed non-orientable manifold with non-zero third normal Stiefel-Whitney class $`\overline{w}_3(V^4)`$ then its punctured manifold $`V_{}^4`$ cannot be embedded in $`S^6`$. For if we have an embedding $`\stackrel{~}{F}:V_{}^4S^6`$, then since $`\stackrel{~}{F}:S^3S^6`$ is topologically unknotted , we can construct a topological embedding of the closed manifold $`V^4`$ in $`S^7`$; this contradicts $`\overline{w}_3(V^4)0`$ according to \[7, Theorem 1.2\].
## Acknowledgements
The author would like to convey his sincere thanks to Professor Osamu Saeki for his invaluable advice. He would also thank Professors Dennis Roseman, Takashi Nishimura and Yukio Matsumoto for constant encouragement and many helpful comments.
The author is partially supported by the Grant-in-Aid for JSPS Fellows.
|
warning/0506/quant-ph0506003.html
|
ar5iv
|
text
|
# Non-local thin films in Casimir force calculations
## I Introduction
The Casimir force between uncharged metallic plates Cas48 (see also reviews Mil94 ; Mos97 ; Kar99 ; Mil01 ; Bor01 ; Mil04 ) attracted considerable attention in the past years. The force was measured in a number of experiments with a high precision using different techniques and geometric configurations Lam97 ; Moh98 ; Roy99 ; Har00 ; Ede00 ; Cha01 ; Bres02 ; Dec03a ; Dec03b . On the other hand, the potential applications of the force in micro and nanomechanics is still largely unexplored. Actuation and nonlinear behavior of a mechanical oscillator with the Casimir force were demonstrated Cha01 and the importance of the force in adhesion and stiction has also been discussed Buk01 ; Joh02 ; Zha03 . Due to technological reasons thin coating layers or multilayered structures are often in use in micromechanical devices. The main question to be addressed in this paper is how important are the nonlocal effects when the film thickness is smaller than the mean free path of the electrons.
For the first time the problem of a thin metallic layer on top of another metal appeared in connection with the first atomic force microscope (AFM) experiments Moh98 ; Roy99 . In these experiments a relatively thick $`Al`$ layer was covered with $`Au/Pd`$ film of 20 nm Moh98 or 8 nm Roy99 thick to prevent aluminum oxidation. Because the film was thin enough to be transparent for the light with a characteristic frequency $`\omega _{ch}=c/2a`$, where $`a`$ is the distance between the bodies, it was concluded that $`Au/Pd`$ layer did not influence on the forcefootnote1 . In actual calculations Kli99 the thin $`Au/Pd`$ film was changed by vacuum. This approach was criticized Sve00a on the basis that according to the Lifshitz formula Lif56 ; LP9 the force depends on the dielectric function $`\epsilon \left(i\zeta \right)`$ at imaginary frequencies $`\omega =i\zeta `$. Kramers-Kronig relation shows that at $`\zeta _{ch}=c/2a`$ low real frequencies $`\omega c/2a`$ give significant contribution to $`\epsilon (i\zeta _{ch})`$. At low frequencies $`Au/Pd`$ film is not transparent and it should be taken into account. It was demonstrated that, indeed, even 8 nm thick film gave significant contribution to the force. The calculation in Ref. Kli00 supported this conclusion but the authors speculated that nonlocal effects due to small thickness of the film (smaller than the mean free path for electrons) allowed one to consider the film as transparent.
The question arose again in connection with a recent experiment Lis05 , where the force was measured between a plate and sphere covered with 10 nm or 200 nm $`Pd`$ film. For thin film the expected reduction of the force was clearly observed. It was indicated that spatial dispersion might be important for calculation of the force in the case of thin film. Also, Bostrรถm and Sernelius bostrom2000 pointed out to the need of detailed studies of nonlocal effects, while studying the retarded van der Waals force between thin metallic films within a local approximation.
There have been several works dealing with the problem of nonlocality in the Casimir force between half spaces. Katz katz was the first to point out the need of a quantitative study and in his work only a rough estimate of how spatial dispersion affected dispersive forces was given. Heindricks Heindricks was able to derive Lifshitz formula in an approximate way to include nonlocal effects. Similarly, Dubrava using a phenomenological approach described the Casimir attraction between thin films Dubrava . More recently, based on the formalism of nonlocal optics, the effects for thick metallic layers have been considered. Propagation of bulk plasmons Esq03 ; Esq05 ; Moch05 and electromagnetic response in the region of anomalous dispersion Esq04a were taken into account, showing that the spatial dispersion does not contribute significantly to the Casimir force. The method developed in Ref. Esq04a is very general and can be used for the analysis of all nonlocal effects including those arising in thin films.
## II Formalism
The Casimir force between two plates separated by a vacuum gap $`a`$ at a temperature $`T`$ is given by the Lifshitz formula Lif56 ; LP9 . The force is expressed via the reflection coefficients $`R_1`$ and $`R_2`$ of the plate 1 and 2, respectively, in the following way:
$$F_{pp}\left(a\right)=\frac{k_BT}{\pi }\underset{n=0}{\overset{\mathrm{}}{}}{}_{}{}^{}\underset{0}{\overset{\mathrm{}}{}}๐qqk_0\left[\left(R_{1s}^1R_{2s}^1\mathrm{exp}\left(2ak_0\right)1\right)^1+\left(R_{1p}^1R_{2p}^1\mathrm{exp}\left(2ak_0\right)1\right)^1\right],$$
(1)
where subscripts $`s`$ and $`p`$ denote the polarization states, $`๐ช`$ is the wave vector along the plates, $`q=\left|๐ช\right|`$, and $`k_0`$ is the normal component of the wave vector defined as
$$k_0=\sqrt{\zeta _n^2/c^2+q^2}.$$
(2)
In Eq. (1) the sum is calculated over the Matsubara frequencies
$$\zeta _n=\frac{2\pi k_BT}{\mathrm{}}n.$$
(3)
The reflection coefficients $`R_1`$ and $`R_2`$ are different for $`s`$ and $`p`$ polarizations and are functions of $`q`$ and imaginary frequencies $`\zeta _n`$. They comprise material properties of the plates and for this reason we start our analysis from the reflection coefficients.
### II.1 Local case
To set our notation, we first study briefly the known local case, when the optical response depends only on frequency. We start the analysis from a thin film of thickness $`h`$ on a substrate. It will be assumed here that the film is continuous.
For a film on an infinitely thick substrate the problem is rather simple. The Maxwell equations are solved with the boundary conditions which are the continuity of the tangential components of electric and magnetic fields on both boundaries of the film. The problem can be solved at real frequencies and then analytically continued to the imaginary axis. In the local limit the film and substrate are described by their local dielectric functions which will be denoted as $`\epsilon _1\left(\omega \right)`$ and $`\epsilon _2\left(\omega \right)`$. In general, the indexes marking the layers will increase from top to bottom of the plate. The dielectric function of vacuum will be taken as $`\epsilon _0\left(\omega \right)=1`$. The reflection coefficients in our case are well known in optics. At imaginary frequencies they are Zho95 :
$$R=\frac{r_{01}r_{21}\mathrm{exp}\left(2k_1h\right)}{1r_{01}r_{21}\mathrm{exp}\left(2k_1h\right)},$$
(4)
where $`r_{ml}`$ are the reflection coefficients from the boundary between media $`l`$ and $`m`$. These coefficients depend on the polarization, $`s`$ or $`p`$, and are defined as
$$r_{ml}^s=\frac{k_mk_l}{k_m+k_l},r_{ml}^p=\frac{\epsilon _lk_m\epsilon _mk_l}{\epsilon _lk_m+\epsilon _mk_l},$$
(5)
where $`k_m`$ is the normal component of the wave vector in the medium $`m`$:
$$k_m=\sqrt{\epsilon _m\left(i\zeta \right)\frac{\zeta ^2}{c^2}+q^2}.$$
(6)
It is easy to check that for $`h\mathrm{}`$ (thick film) $`Rr_{01}`$ and in the opposite limit $`h0`$ the reflection coefficient coincides with that for the substrate: $`Rr_{02}`$.
To understand the variation of the Casimir force with the film thickness, we first study the behavior of the reflection coefficients. For a qualitative analysis it will be assumed that a metal, film or substrate, can be described with the Drude dielectric function
$$\epsilon \left(i\zeta \right)=1+\frac{\omega _p^2}{\zeta \left(\zeta +\omega _\tau \right)},$$
(7)
where $`\omega _p`$ and $`\omega _\tau `$ are the Drude parameters which are different for each layer. For thin films, $`\omega _\tau `$ is a function of the film thickness. This dependence appears because, in addition to the internal scattering processes, for thin films scattering from the surfaces is important. These processes are independent of each other and the relaxation time in the Drude model is $`\omega _\tau =\omega _\tau ^{bulk}+\omega _\tau ^{surf}(h)`$. This effect becomes important when the thickness is smaller than the mean free path for electron. Dependence on $`h`$ of $`\omega _\tau ^{surf}`$ is explained by the Fuchs-Sondheimer theory Fuc38 ; Son52 . When $`h`$ is much smaller than the mean free path, this dependence is given by
$$\omega _\tau ^{surf}(h)=\frac{3}{8}(1p)\frac{v_F}{h},$$
(8)
where $`v_F`$ is the Fermi velocity and an electron has probability $`p`$ of being specularly reflected from the surface. As one can see from Eq. (8) only diffusely reflected electrons contribute to $`\omega _\tau ^{surf}(h)`$. Experimental results concerning the specularity are far from unique. Very different values of $`p`$ in the range $`0<p<1`$ were used to explain the experimental results Fis80 . In this paper we investigate the nonlocal effects for specular reflection of electrons on the surface and do not include in the consideration $`h`$-dependence of the relaxation frequency. But in any case our results are not very sensitive to the exact value of $`\omega _\tau `$.
Consider first the system consisting of $`SiO_2`$ substrate with $`\epsilon _2=4`$ and $`Au`$ film on top of it with the parameters $`\omega _p=9.0eV`$, $`\omega _\tau =0.035eV`$ Lam00 . It is convenient to introduce dimensionless variables and parameters as follows:
$$\mathrm{\Omega }=\frac{\zeta }{\omega _p},Q=\frac{cq}{\omega _p},\gamma =\frac{\omega _\tau }{\omega _p},H=\frac{\omega _ph}{c}.$$
(9)
The reflection coefficient for $`s`$-polarization as a function of the dimensionless frequency $`\mathrm{\Omega }`$ is shown in Fig. 1. The dashed curve corresponds to semi-infinite metal $`h\mathrm{}`$. It was calculated with $`Q=0.1`$. This value is taken for the characteristic wave number $`q1/2a`$ at $`a100nm`$. The solid lines marked as 1, 2, and 3 correspond to the dimensionless thickness $`H=0.3`$, $`1`$, and $`3`$, respectively. Note that $`H=1`$ gives the film thickness $`h`$ equal to the penetration depth $`\delta =c/\omega _p22nm`$ ($`Au`$). One can see that $`R_s`$ decreases fast with the thickness. When $`Q`$ increases the film also becomes more transparent for $`s`$-polarization. The other distinctive feature is that $`R_s`$ is going to zero in the limit $`\mathrm{\Omega }0`$. In this limit $`s`$-polarized field degenerates to pure magnetic field, which penetrate freely via the metallic film.
The reflection coefficient for $`p`$-polarization shows a different behavior as one can see in Fig. 2. The dashed line represents the thick film and the solid lines marked as 1 and 2 correspond to $`H=1`$ and $`0.1`$, respectively. Variation of $`R_p`$ with the film thickness is not very significant. The reason for this is the effective screening of the $`E_z`$ component even by a very thin metallic layer. An important conclusion can be drawn from this simple fact. The film thickness affects mostly the contribution of $`s`$-polarization, but the part of the force connected with $`p`$-polarization is changed weakly in the local case.
Consider now the effect of a thin film on top of a thick metallic layer. It will be assumed that both metals can be described by the Drude dielectric functions $`\epsilon _1\left(i\zeta \right)`$ and $`\epsilon _2\left(i\zeta \right)`$ which differ from each other only by the values of parameters $`\omega _{ip}`$ and $`\omega _{i\tau }`$ ($`i=1,2`$). It is clear that in dependence on the film thickness the reflection coefficients will be in between the lines describing metal 1 ($`h\mathrm{}`$) or metal 2 ($`h0`$). In Fig. 3 we present the case when the top layer is better reflector than the bottom one. The dotted line gives $`r_{02}`$ and the dashed line represents $`r_{01}`$. The results for the film with thickness $`H=1`$ and $`0.1`$ are marked as 1 and 2, respectively. In our calculations, the ratios $`\omega _{1p}/\omega _{2p}=2`$ and $`\omega _{1\tau }/\omega _{2\tau }=1`$ where used and dimensionless parameters (9) were defined relative to the parameters of the top layer 1. The relaxation frequencies, $`\omega _{i\tau }`$, influence mostly on low frequency behavior of $`R_s`$. They are not very important for the Casimir force because the main contribution in the force comes from the imaginary frequencies $`\mathrm{\Omega }c/2a\omega _p\gamma `$ where $`\omega _\tau `$ does not play significant role. The reflection coefficient, $`R_p`$, for $`p`$-polarization is shown in Fig. 4. The curves 1 and 2 correspond $`H=1`$ and $`H=0.1`$, respectively. Again one can conclude that the top layer is more important for $`s`$ than for $`p`$-polarization.
### II.2 Nonlocal case
For propagating photons the reflectivity of thin films in the nonlocal case has been analyzed in Ref. Jon69 . It was assumed that electrons are reflected specularly on both boundaries of the film. Let us consider first $`s`$-polarization. Similar to the case of a semi-infinite metal Kli68 the tangential component of the electric field is considered as even on each boundary:
$$E_y\left(mhz\right)=+E_y\left(mh+z\right),$$
(10)
where $`z`$ is the direction normal to the film surface, $`m`$ is an arbitrary integer, and the plane of incidence was chosen to be $`xz`$. The Maxwell equations and Eq. (10) demand for the magnetic field on the boundaries the following conditions:
$$H_x\left(mhz\right)=H_x\left(mh+z\right),H_z\left(mhz\right)=+H_z\left(mh+z\right).$$
(11)
Formally the conditions (10), (11) continue the film of finite thickness to the infinite layer. These conditions mean that the fields can be considered as periodic with period $`2h`$, and they can be expanded in a Fourier series.
In the nonlocal case the material is characterized by the impedance instead of local dielectric function. The impedance is defined as the ratio of tangential components of electric and magnetic fields just below the surface. For $`s`$ and $`p`$-polarizations the impedances of metallic film were found in Ref. Jon69 with the method which is direct generalization of the method used for semi-infinite layer Kli68 . The film has two surfaces and the impedances one can define on each of them:
$$Z_s=\frac{E_y}{H_x}|_{z=\delta ,h\delta },Z_p=\frac{E_x}{H_y}|_{z=\delta ,h\delta },$$
(12)
where $`\delta 0`$. It was noted Jon69 that instead of impedances (12) one can use a different couple for each polarization which can be easy calculated. These new impedances were introduced as the ratio of the fields even or odd relative to the film center $`z=h/2`$. Even or odd fields will be marked by the superscripts $`(1)`$ or $`(2)`$, respectively. The new impedances
$$Z_s^{(1,2)}=\frac{E_y^{(1,2)}}{H_x^{(1,2)}}|_{z=\delta },Z_p^{(1,2)}=\frac{E_x^{(1,2)}}{H_y^{(1,2)}}|_{z=\delta },$$
(13)
are the same on both boundaries of the film because of the symmetry conditions
$$E_{x,y}^{(1)}\left(\delta \right)=E_{x,y}^{(1)}\left(h\delta \right),E_{x,y}^{(2)}\left(\delta \right)=E_{x,y}^{(2)}\left(h\delta \right)$$
(14)
and similarly for the magnetic field.
Explicit expressions for these impedances were found in Ref. Jon69 :
$$Z_s^{(1,2)}=i\frac{2\omega }{ch}\underset{n=(odd,even)}{}\frac{1}{\frac{\omega ^2}{c^2}\epsilon _t(\omega ,k)\left(\frac{n\pi }{h}\right)^2q^2},$$
(15)
$$Z_p^{(1,2)}=i\frac{2\omega }{ch}\underset{n=(odd,even)}{}\frac{1}{k^2}\left[\frac{q^2}{\frac{\omega ^2}{c^2}\epsilon _l(\omega ,k)}+\frac{\left(\frac{n\pi }{h}\right)^2}{\frac{\omega ^2}{c^2}\epsilon _t(\omega ,k)\left(\frac{n\pi }{h}\right)^2q^2}\right].$$
(16)
where for even, $`(1)`$, or odd, $`(2)`$, fields the sum has to be calculated over $`n=2m+1`$ or $`n=2m`$, respectively. The transverse dielectric function $`\epsilon _t(\omega ,k)`$ contributes to $`Z_s`$. It describes the response of the material on the electric field transverse to the wave vector $`๐ค`$. In case of the $`p`$-polarization $`z`$-component of electric field creates a nonzero charge density in the metal producing the longitudinal field inside of metal. That is why $`Z_p`$ depends also on the longitudinal dielectric function $`\epsilon _l(\omega ,k)`$. In general, these functions are nonlocal, so they depend on both $`\omega `$ and $`k`$. The absolute value of the wave vector $`๐ค`$ in Eqs. (15), (16) is
$$k=\sqrt{\left(\frac{n\pi }{h}\right)^2+q^2}.$$
(17)
Let us consider now the reflection and transmission coefficients of the film on a substrate. Note that in Ref. Jon69 only a free standing film was considered. To find these coefficients one has to match the tangential components of the electric and magnetic fields outside and inside of the film. We assume for simplicity that the substrate can be described by a local dielectric function or equivalently by local impedances. This assumption is justified by the investigation of nonlocal effects at imaginary frequencies for semi-infinite metals Esq04a . It was demonstrated that in contrast with the real frequencies the nonlocal effect (anomalous skin effect) brings only minor influence on the reflection coefficients. Matching the electric field on both sides of the film for $`s`$-polarization one gets
$$\begin{array}{c}E_y^0\left(1+R_s\right)=E_y^{(1)}\left(\delta \right)+E_y^{(2)}\left(\delta \right),\\ E_y^0t_se^{ik_2h}=E_y^{(1)}\left(\delta \right)E_y^{(2)}\left(\delta \right),\end{array}$$
(18)
where $`E_y^0`$ is the incident field, $`t_s`$ is the transmission coefficient and the symmetry conditions (14) were taken into account. Similar equations are true for the magnetic field:
$$\begin{array}{c}H_x^0\left(1R_s\right)=H_x^{(1)}\left(\delta \right)+H_x^{(2)}\left(\delta \right),\\ H_x^0t_s\frac{k_2}{k_0}e^{ik_2h}=H_x^{(1)}\left(\delta \right)+H_x^{(2)}\left(\delta \right).\end{array}$$
(19)
Eqs. (18) and (19) can be solved for $`R_s`$ and $`t_s`$ using the impedance definition (13). As the result the reflection coefficient can be presented in the form:
$$R_s=\frac{\left(Z_{s1}^{(1)}Z_{s0}\right)\left(Z_{s1}^{(2)}+Z_{s2}\right)+\left(Z_{s1}^{(2)}Z_{s0}\right)\left(Z_{s1}^{(1)}+Z_{s2}\right)}{\left(Z_{s1}^{(1)}+Z_{s0}\right)\left(Z_{s1}^{(2)}+Z_{s2}\right)+\left(Z_{s1}^{(2)}+Z_{s0}\right)\left(Z_{s1}^{(1)}+Z_{s2}\right)}.$$
(20)
Here we introduced the following notations : $`Z_{s1}^{(1,2)}`$ are the nonlocal impedances of the film given by Eq. (13), $`Z_{s2}`$ is the local impedance of the substrate defined as
$$Z_{s2}=\frac{\omega }{ck_2},$$
(21)
and
$$Z_{s0}=\frac{\omega }{ck_0},$$
(22)
is the โimpedanceโ of the plane wave defined as the ratio of electric and magnetic fields in the wave. The formula (20) for $`R_s`$ cannot be presented in the same form (4) as in the local case. This is because we used the impedances (13) instead of that given by Eq. (12). As we will see both Eqs. (4) and (20) coincide in the local limit.
In the same way one can find the reflection coefficient for $`p`$ -polarization, $`R_p`$. In this case the equations similar to (18), (19) with the interchange $`xy`$ will be true, the impedance of the plane wave is defined as
$$Z_{p0}=\frac{ck_0}{\omega }=\frac{1}{Z_{s0}},$$
(23)
and the local impedance of the substrate is
$$Z_{p2}=\frac{ck_2}{\omega \epsilon _2\left(\omega \right)}.$$
(24)
The final expression for $`R_p`$ is
$$R_p=\frac{\left(Z_{p1}^{(1)}Z_{p0}\right)\left(Z_{p1}^{(2)}+Z_{p2}\right)+\left(Z_{p1}^{(2)}Z_{p0}\right)\left(Z_{p1}^{(1)}+Z_{p2}\right)}{\left(Z_{p1}^{(1)}+Z_{p0}\right)\left(Z_{p1}^{(2)}+Z_{p2}\right)+\left(Z_{p1}^{(2)}+Z_{p0}\right)\left(Z_{p1}^{(1)}+Z_{p2}\right)}.$$
(25)
It differs from Eq. (20) only by the general sign and the change $`sp`$.
If the substrate is changed by vacuum, $`Z_{\alpha 2}Z_{\alpha 0}`$ ($`\alpha =s,p`$), we reproduce the reflection coefficient found in Ref. Jon69 :
$$R_\alpha =\frac{1}{2}\left(r_\alpha ^{(1)}+r_\alpha ^{(2)}\right),\alpha =s,p,$$
(26)
where the โpartialโ reflection coefficients are connected with the impedances by the usual relations
$$r_s^{(1,2)}=\frac{Z_{s0}Z_s^{(1,2)}}{Z_{s0}+Z_s^{(1,2)}},r_p^{(1,2)}=\frac{Z_{p0}Z_p^{(1,2)}}{Z_{p0}+Z_p^{(1,2)}}.$$
(27)
In the local limit both the transverse $`\epsilon _t`$ and longitudinal $`\epsilon _l`$ dielectric functions coincide with the local function: $`\epsilon _t(\omega ,k)\epsilon _l(\omega ,k)\epsilon _1\left(\omega \right)`$. In this case the sums in Eqs. (15), (16) can be found explicitly. For example, for $`s`$-polarization one has
$$Z_{1s}^{(1),loc}=i\frac{\omega }{ck_1}\mathrm{tan}\frac{hk_1}{2},Z_{1s}^{(2),loc}=i\frac{\omega }{ck_1}\mathrm{cot}\frac{hk_1}{2}.$$
(28)
Substituting it in Eq. (20) one can check that the reflection coefficient for the local case given by Eq. (4) is reproduced.
All the equations above were written for real frequencies. Transition to imaginary frequencies, which are the main point of our interest, can be done by a simple analytic continuation. To get the nonlocal effects in the reflection coefficients, we have to fix the nonlocal dielectric functions. At imaginary frequencies in the Boltzmann approximation they are given by the relations Esq04a
$$\epsilon _l(\mathrm{\Omega },v)=1+\frac{f_l\left(v\right)}{\mathrm{\Omega }\left(\mathrm{\Omega }+\gamma \right)},f_l\left(v\right)=\frac{3}{v^2}\frac{v\mathrm{arctan}v}{v+\frac{\gamma }{\mathrm{\Omega }}\left(v\mathrm{arctan}v\right)},$$
(29)
$$\epsilon _t(\mathrm{\Omega },v)=1+\frac{f_t\left(v\right)}{\mathrm{\Omega }\left(\mathrm{\Omega }+\gamma \right)},f_t\left(v\right)=\frac{3}{2v^3}\left[v+\left(1+v^2\right)\mathrm{arctan}v\right],$$
(30)
$$v=\frac{v_F}{c}\frac{\sqrt{\left(\frac{n\pi }{H}\right)^2+Q^2}}{\mathrm{\Omega }+\gamma },$$
(31)
where $`v_F`$ is the Fermi velocity. The dimensionless variables (9) have been introduced in Eqs. (29)-(31) . In addition, we have neglected in Eqs. (29) -(30) the contribution due to the interband transitions.
The reflection coefficients in the nonlocal case were calculated numerically. In Fig. 5 the difference between local and nonlocal coefficients $`R_s`$ is shown for $`Au`$ film on top of $`SiO_2`$ substrate. The dashed curve corresponds to very thick film, $`H\mathrm{}`$. The solid lines marked as 1 and 2 are presented for $`H=1`$ and 0.1, respectively. As before, $`H=1`$ corresponds to the penetration depth of $`Au`$ ($`\delta =22`$ nm). The thick film clearly demonstrates the anomalous skin effect at $`\mathrm{\Omega }\gamma `$, although the magnitude of the effect is small as was already noted in Ref. Esq04a . Even this small effect decreases with the film thickness as the curves 1 and 2 show. The nonlocal effect increases with $`Q`$ but it is smaller than 1% even for $`Q=1`$. It should be noted that the Boltzmann approximation is good while $`\mathrm{\Omega }<1`$, but when $`\mathrm{\Omega }`$ approaching 1 the reflection coefficient itself becomes small and there is no sense to keep the nonlocal correction in this range. Similar result was found for the film on top of a metallic substrate. One can conclude that for $`s`$-polarization the nonlocal effect in the reflection coefficient is very small and can be neglected in calculation of the Casimir force.
The situation for $`p`$-polarization is shown in Fig. 6 for the film on top of metallic substrate. As in the local case the substrate was chosen to have the plasma frequency 2 times smaller than that for the film. The lower curve corresponds to $`Q=0.1`$ and $`H=1`$. The upper series of curves is given for $`Q=0.5`$. As one can see, the nonlocal effect manifests itself in a wider frequency range and does not disappear even for zero frequency. The latter is the result of Thomas-Fermi screening as was explained in Ref. Esq04a . The effect is still small but the nonlocal contribution in the Casimir force will be larger than that for $`s`$-polarization. This is because the nonlocal effect is the largest at frequencies which give the main contribution in the Casimir force.
## III Effects of Spatial Dispersion on the Casimir force
To quantify the effect of spatial dispersion on the Casimir force, we calculate the percent difference between the local case and nonlocal case ($`\mathrm{\Delta }\%=|(F_{local}F_{nonlocal})/F_{local}|`$), as a function of separation.
First we consider the case of free standing metallic films. The system is similar to that considered by Bostrรถm and Sernelius bostrom2000 . The percent difference $`\mathrm{\Delta }\%`$ as a function of separation is presented in Fig. 7, for three different thicknesses. The results for the thick film $`h=100nm`$ coincides with the results obtained for half spaces in our previous work Esq04a . As the thickness decreases the nonlocal effects become more relevant. Thin films have a more complicated nonlocal response than half spaces. For $`p`$-polarized waves, surface plasmons on each side of the film can interfere lopez82, creating standing waves that will increase the electromagnetic absorption of the field that will decrease the Casimir force. These resonance conditions are evident from Eq. (17) where $`k_z=n\pi /L`$.
The force is not affected significantly when the thin films are on substrates. In Figure 8 we have plotted the percent difference between two thin $`Au`$ films, each deposited on a dielectric substrate. Again, we assumed $`ฯต=4`$ for the dielectric, just as an illustrative example of the effect of substrate. The substrates reduce slightly the value of $`\mathrm{\Delta }\%`$ for both curves shown, with the obvious limit that when the substrate has the same dielectric function as the film, we recover the results for the force between half-spaces. This means that the effect of the substrate is to allow energy transfer out of the thin film into the substrate.
The difference between the local and nonlocal cases can be reduced in a system consisting of $`Au`$ half space and $`Au`$ coated substrate. Again, we took a dielectric ($`ฯต=4`$). The effect of spatial dispersion reduces significantly as compared to the cases treated in Figs. 7 and 8. This shows that the most important part of the spatial dispersion effect come from the thin films. If in current experiments the separation can go down to $`50nm`$, in the system shown in Fig. 9, the nonlocal correction is of the order of $`0.34\%`$.
The result holds for different substrates. This is shown in Table 1, where we presented the percent difference between the local and nonlocal forces for $`Au`$ film deposited on different substrates. All data are given for a separation of 50 nm. As before, $`\omega _{p1}/\omega _{p2}`$ is the ratio of the $`Au`$ plasma frequency to that of the metallic substrate, assuming the damping factor remains the same. The case $`\epsilon =1`$ corresponds to the free standing $`Au`$ thin film (no substrate).
## IV Conclusions
The role of thin metallic coatings in the calculation of Casimir forces has been studied taking into account spatial dispersion. The description of the nonlocal response of thin films is based on the Kliewer and Fuchs formalism that imposes a symmetrical behavior of the fields inside thin films. The study of the reflectivities shows that the main contribution to the nonlocal effect comes from $`p`$-polarized light that excites normal modes within the material. At very small separations, the effects can be appreciable but at best a percent difference of $`7\%`$ is found. However, for typical experimental setups and separations the percent difference between the local and nonlocal case is of the order of $`0.4\%`$, that can be regarded as negligible within current experimental precisions and the local description is good enough. The effect of thin films within a local approximation has been measured recently by Lisanti et al. Lis05 .
Along with the previous works on nonlocal effects between half-spaces Esq03 ; Esq04a ; Esq05 ; Moch05 , we can generally conclude that these effects will be difficult to detect at the current experimental precision. Our results indicate a decrease in the force due to spatial dispersion. However for half-spaces within a jellium model it has been shown Moch05 that the force can increase due to nonlocal effects because of decrease in the separation of the optical surfaces that might not coincide with the physical surface.
###### Acknowledgements.
Partial support from CONACyT project: 44306 and DGAPA-UNAM IN-101605. We thank W.L. Mochan and C. Villarreal for helpful discusions.
|
warning/0506/hep-ph0506167.html
|
ar5iv
|
text
|
# Width Effects on Near Threshold Decays of the Top Quark ๐โ{๐โข๐พโข๐พ,๐โข๐โข๐} and of Neutral Higgs Bosons
## I Introduction
The finite width of a particle is directly related to its instability. When its width is small with respect to its physical mass, finite width effects (FWE) are usually neglected except for decays in which a resonance can emerge when the particle appears as an intermediate state, or in decays that are kinematically allowed only very close to threshold and the particle is involved in either the initial or the final state. The former case is usually handled with the Breit-Wigner prescription, while the latter case, i.e., taking into account the FWE in processes occurring just around their kinematical threshold, needs special attention.
In this respect, there are two different methods proposed in the literature Altarelli:2000nt ; Mahlon:1994us ; Muta:1986is ; Calderon:2001qq . They were referred to by Altarelli, Conti and Lubicz Altarelli:2000nt as the decay-chain method (DCM) and the convolution method (CM).<sup>1</sup><sup>1</sup>1An alternative approach has been recently discussed by Kuksa Kuksa:2004cm , based on the uncertainty relation for the mass of the unstable particle. This method has a close analogy to the convolution method. In the first approach (i.e., the DCM), the dominant decay modes of the unstable final state particles are taken into account as subsequent decays to obtain the โtotalโ decay rate and then the branching ratio for the โsignalโ (i.e., with the unstable particle in the finite state) is calculated by taking the ratio of the โtotalโ decay rate to the multiplication of rates of the subsequent decay modes Altarelli:2000nt ; Mahlon:1994us . This method requires kinematical cuts in order to maintain the direct connection between the โsignalโ and the total number of events. That is, since the observed final state (with its subsequent decay products) could be produced through other channels, kinematical cuts are required to minimize this undesired background. Therefore, this method leads to physical quantities which depend on kinematical cuts and so it inherits some degree of experimental difficulties.
Alternatively, in the CM the instability of a final state particle is described instead by a Breit-Wigner-like density function whose central value and half-width are governed by the width and the physical invariant mass of the particle. In this way, the unstable particle produced can be seen effectively as a real physical particle, having an invariant mass which is controlled by its density function. Although this method does not require any kinematical cut, it doubles the number of phase space integrals, making it computationally more challenging.
In this paper we employ the CM to study FWE in the three-body flavor changing rare top decays $`tcWW`$ and $`tcZZ`$, by including the widths of $`W`$ and $`Z`$ bosons. These decay modes and other two and three-body rare flavor changing top decays mele , can provide a unique testing ground for the standard model (SM) Glashow-Iliopoulos-Maiani (GIM) mechanism and may give hints about - beyond the SM - flavor changing physics such as may occur in some variations of Two-Higgs Doublet Models (2HDMโs). FWE in these decay modes will be studied within the SM (in the case of $`tcWW`$) and in the context of the type III Two Higgs Doublet Model (in both $`tcWW`$ and $`tcZZ`$), which admits flavor changing neutral currents (FCNC) at the tree-level. The three-body top decays $`tcWW,bWZ,cZZ`$ have been considered before, without including FWE, in the SM Jenkins:1996zd ; Decker:1992wz ; Diaz-Cruz:1999ab , in 2HDMโs Diaz-Cruz:1999ab ; Bar-Shalom:1997tm ; Bar-Shalom:1997sj ; Li:zv ; DiazCruz:1999mq , in a generic formalism including scalar, vector or fermion exchanges 9707229 and in topcolor-assisted Technicolor model 0103081 . In addition, the top decays $`tbWh^0`$ and $`tbWA^0`$ have been analyzed in the context of a general 2HDM Iltan:2002am . Among the above decay modes, a simple threshold analysis shows that $`tcZZ`$ and $`tbWZ`$ are potentially the most sensitive to FWE. In particular, according to the recent CDF analysis based on the Tevatron RUN II data, the top mass is ($`1\sigma `$) recenttop1 : $`m_t=173.5_{2.6}^{+2.7}(stat)\pm 4.0(syst)`$<sup>2</sup><sup>2</sup>2Note that later D0 results from Tevatron RUN II, $`m_t=170.6\pm 4.2(stat)\pm 6.0(syst)`$ (see recenttop1 ), are based on less accumulated data and has larger statistical and systematic uncertainties. In fact, these later top mass measurements imply that for the stable Z-bosons case (i.e., without including FWE) the decay $`tcZZ`$ cannot occur if the top mass lies within its recent CDF and D0 $`1\sigma `$ limits. We, therefore, expect FWE to be substantial in this decay. Indeed, we find that FWE (due to the rather large $`๐ช(GeV)`$ Z-width) can give $`\mathrm{BR}(tcZZ)10^510^3`$ (as opposed to null in the stable case), within some range of the allowed parameter space of the type III 2HDM. Moreover, even for the decay $`tcWW`$, for which the central value of the top-quark mass (i.e., $`m_t=173.5`$) is about 10 GeV away from the kinematical threshold, we find that FWE from the unstable W-bosons can cause a several orders of magnitudes enhancement in the type III 2HDM with a light neutral Higgs of mass $`m_{h^0}\stackrel{<}{}2m_W`$, thus elevating the branching ratio from $`\mathrm{BR}(tcWW)10^910^8`$ to $`\mathrm{BR}(tcWW)10^410^3`$ in this case. Clearly, such large branching ratios would be accessible to the LHC and may even be detected at the Tevatron. A similar large enhancement due to FWE was found for the decay mode $`tbWZ`$ in both the CM Altarelli:2000nt and the DCM Altarelli:2000nt ; Mahlon:1994us . In particular, Altarelli:2000nt ; Mahlon:1994us have found that, in the SM, the FWE increase this decay width by orders of magnitude (with respect to the stable final state gauge bosons), giving $`\mathrm{BR}(tbWZ)2\times 10^6`$ for $`m_t176`$ GeV.
To demonstrate the potential importance of FWE in neutral Higgs decays, we also examine the three-body neutral Higgs decays $`h^0tbW`$ and $`A^0tbW`$, within the type III 2HDM, assuming that either $`h^0`$ (the lighter CP-even neutral Higgs) or $`A^0`$ (the CP-odd neutral Higgs) have masses around $`m_t+m_b+m_W`$ (i.e., close to the threshold). It is well known that, for a SM-like Higgs, the two-body decay modes to the heaviest fermions and to the gauge bosons are dominant, since its couplings to these particles are proportional to their masses. Three-body sub-threshold decays (e.g., to $`W^{}W`$ or $`Z^{}Z`$ pairs) can also have sizable BRโs despite the suppression factors involved Rizzo:1980gz . In the context of the minimal supersymmetric extension of the SM (MSSM) sub-threshold three-body decays of especially heavy Higgs bosons might also have a large branching ratio Djouadi:1995gv . In this paper we show that, including the top quark and the W boson width in the framework of the CM, the three-body Higgs decays $`h^0tbW`$ and $`A^0tbW`$ can be enhanced by about 3 orders of magnitudes in the type III 2HDM if they occur just around their kinematical threshold. For the case of $`A^0tbW`$, such an enhancement can push its BR to the level of tens of percents and may, therefore, become critical for experimental searches of $`A^0`$.
The paper is organized as follows: In Section II we describe the convolution method. In Section III we give a brief overview of the type III 2HDM. In section IV we examine the FWE in the top decays $`tcWW,cZZ`$ and in section V we study the FWE in the three-body Higgs decays $`h^0tbW`$ and $`A^0tbW`$. In Section VI we summarize our results.
## II The Convolution Method
Particles with large width imply a large uncertainty in its mass from the mass uncertainty relation Matthews:1958sc . The CM can be used to include such large width effects in decays involving unstable particles in the final state. Consider for example the main top decay $`tbW`$. Since the $`W`$ is unstable, we can define: $`\mathrm{\Gamma }(tbW)\mathrm{\Gamma }={\displaystyle \underset{i,j}{}}\mathrm{\Gamma }^0\left(tbf_i\overline{f}_j\right)`$, where the sum runs over all the $`W`$ decay modes. Furthermore, $`\mathrm{\Gamma }`$ can be decomposed into two parts corresponding to the transverse ($`\mathrm{\Gamma }_T`$) and longitudinal ($`\mathrm{\Gamma }_L`$) components of the intermediate $`W`$-boson (see e.g., Calderon:2001qq ):
$`\mathrm{\Gamma }`$ $`=`$ $`\mathrm{\Gamma }_T+\mathrm{\Gamma }_L,`$ (1)
where
$`\mathrm{\Gamma }_T`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle \underset{ij}{}}{\displaystyle _{(m_i+m_j)^2}^{(m_tm_b)^2}}๐p^2{\displaystyle \frac{\sqrt{p^2}\mathrm{\Gamma }^0\left(tbW(p^2)\right)\mathrm{\Gamma }^0\left(W(p^2)f_i\overline{f_j}\right)}{\left(p^2m_W^2\right)^2+\left(Im\mathrm{\Pi }_T(p^2)\right)^2}},`$ (2)
and $`\mathrm{\Gamma }_Lf(m_i,m_j)`$, with $`f0`$ as $`m_i,m_j0`$. Also, $`m_W`$ is the mass of the $`W`$ boson and $`Im\mathrm{\Pi }_T(p^2)`$ and $`Im\mathrm{\Pi }_L(p^2)`$ (appearing in $`\mathrm{\Gamma }_L`$) are the absorptive parts of the transverse and longitudinal vacuum polarization tensor (see e.g., Calderon:2001qq ; Atwood:nk ).
Using the Cutkotsky rule in the limit of massless fermion $`m_i,m_j0`$ ($`f_i,f_j`$, are the fermions exchanged in the W self energy diagram), one obtains
$`Im\mathrm{\Pi }_L(p^2)0`$ and:
$`Im\mathrm{\Pi }_T(p^2)=\sqrt{p^2}{\displaystyle \underset{i,j}{}}\mathrm{\Gamma }^0\left(W(p^2)f_i\overline{f}_j\right)={\displaystyle \frac{p^2}{m_W}}\mathrm{\Gamma }_W^0,`$ (3)
where $`\mathrm{\Gamma }_W^0`$ is the usual on-shell decay width of $`W`$ and $`\sqrt{p^2}m_i+m_j`$. Thus, in this limit $`\mathrm{\Gamma }`$ reduces to:
$`\mathrm{\Gamma }=\mathrm{\Gamma }_T`$ $`=`$ $`{\displaystyle _0^{\left(m_tm_b\right)^2}}๐p^2\rho (p^2,m_W,\mathrm{\Gamma }_W^0)\mathrm{\Gamma }^0\left(tbW(p^2)\right),`$ (4)
where $`\rho (p^2,m_W,\mathrm{\Gamma }_W^0)`$ is the โinvariant mass distribution functionโ, given by:
$`\rho (p^2,m_W,\mathrm{\Gamma }_W^0)`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle \frac{\frac{p^2}{m_W}\mathrm{\Gamma }_W^0}{\left(p^2m_W^2\right)^2+\left(\frac{p^2}{m_W}\mathrm{\Gamma }_W^0\right)^2}}.`$ (5)
Eqs. (4) and (5) describe the factorization of the production and the decay modes of the $`W`$ boson (in the limit of massless fermions). The case of a stable $`W`$ boson (i.e., $`\mathrm{\Gamma }_W^00`$) makes $`\rho \delta (p^2m_W^2)`$ which sets $`\mathrm{\Gamma }=\mathrm{\Gamma }^0(tbW)`$, where $`\mathrm{\Gamma }^0`$ is the width for an on-shell $`W`$ without FWE.
The above prescription can be generalized to the case of a generic three-body decay of the form $`abV_1V_2`$, where $`V_1`$ and $`V_2`$ are vector bosons:
$`\mathrm{\Gamma }(abV_1V_2)={\displaystyle _0^{\left(m_am_b\right)^2}}๐p_1^2{\displaystyle _0^{\left(m_am_b\sqrt{p_1^2}\right)^2}}๐p_2^2`$ $`\rho _1(p_1^2,m_{V_1},\mathrm{\Gamma }_{V_1}^0)\rho _2(p_2^2,m_{V_2},\mathrm{\Gamma }_{V_2}^0)`$ (6)
$`\times \mathrm{\Gamma }^0\left(abV_1(p_1^2)V_2(p_2^2)\right).`$
Furthermore, for consistency of the CM one needs also the following modifications:
1. The sum over polarization vectors of a gauge-boson with an invariant mass $`p^2`$ should be taken as:
$`{\displaystyle \underset{\lambda }{}}ฯต_\lambda ^\mu (p)ฯต_\lambda ^\nu (p)=g^{\mu \nu }+{\displaystyle \frac{p^\mu p^\nu }{p^2}}.`$ (7)
2. In calculating the โzerothโ width of the top-quark(Higgs) into the off-shell vector boson(s) \[i.e., $`\mathrm{\Gamma }^0\left(abV_1(p_1^2)V_2(p_2^2)\right)`$ in Eq. (6) or $`\mathrm{\Gamma }^0\left(ab(p_1)V(p_2^2)\right)`$ for $`=h^0`$ in Eq. (17)\], the tree-level propagator of the massive vector bosons should be modified as (in the unitary gauge):
$`{\displaystyle \frac{i}{p^2m_V^2+im_V\mathrm{\Gamma }_V^0}}\left[g^{\mu \nu }{\displaystyle \frac{p^\mu p^\nu }{m_V^2im_V\mathrm{\Gamma }_V^0}}\right].`$ (8)
This substitution is required since in the CM the invariant mass $`p^2`$ is allowed to vanish, as can be seen from the integration limits in Eqs. (4) and (6).
3. In order to restore gauge invariance in the $`R_\xi `$-gauge, the Feynman rules in which masses of such resonant intermediate particles appear should be modified to be functions of the corresponding invariant masses. Such a modification is, however, not necessary in the unitary gauge that we have used in all our calculations LopezCastro:1991nt .
## III The two Higgs doublet model of type III
One of the simplest extensions of the SM is obtained by enlarging the scalar sector with an additional $`SU(2)_L`$ doublet. In the most general case such a 2HDM gives rise to tree-level FCNC which are mediated by the physical Higgs bosons luke . To avoid such potentially dangerous FCNC, one usually imposes an ad-hoc discrete symmetry Glashow:1976nt that leads to the type I or type II 2HDM (see for example HHG and ourreview ). An alternative way for suppressing FCNC in a general 2HDM (i.e., without imposing discrete symmetries) was suggested by Cheng and Sher in Cheng:1987rs . In the Cheng and Sher Ansatz the arbitrary flavor changing couplings of the scalars to fermions are assumed to be proportional to the square root of masses of the fermions participating in the Higgs Yukawa vertex (see below).<sup>3</sup><sup>3</sup>3The Cheng and Sher Ansatz ensures the suppression of FCNC within the first two generations of quarks, as required by the experimental constraints on FCNC in meson transitions, see Atwood:1996vj .
Within the most general 2HDM one can always choose a basis where only one of the doublets acquires a vacuum expectation value (VEV): $`\mathrm{\Phi }_1=\left(0v/\sqrt{2}\right)^T\mathrm{and}\mathrm{\Phi }_2=0`$. A general 2HDM in this basis is often referred to as the type III 2HDM (or Model III) Atwood:1996vj ; Sher:1991km ; Atwood:1995ej . With this choice of basis, $`\mathrm{\Phi }_1`$ corresponds to the usual SM doublet and all the new flavor changing couplings are attributed to $`\mathrm{\Phi }_2`$. Note also that in this basis $`\mathrm{tan}\beta =v_1/v_2`$ has no physical meaning.<sup>4</sup><sup>4</sup>4โSwitching onโ $`\mathrm{tan}\beta `$ by allowing $`\mathrm{\Phi }_20`$ will not change any physical result.
As in any 2HDM, the physical Higgs sector of Model III consists of 3 neutral Higgs bosons (2 CP-even ones, $`h^0`$ and $`H^0`$, and one CP-odd state $`A^0`$) and a charged scalar with its conjugate $`H^\pm `$. The neutral bosons are given, in terms of the original SU(2) doublets, as:
$`h^0`$ $`=`$ $`\sqrt{2}\left[\left(Re\varphi _1^0v\right)\mathrm{sin}\alpha +Re\varphi _2^0\mathrm{cos}\alpha \right],`$
$`H^0`$ $`=`$ $`\sqrt{2}\left[\left(Re\varphi _1^0v\right)\mathrm{cos}\alpha +Re\varphi _2^0\mathrm{sin}\alpha \right],`$
$`A^0`$ $`=`$ $`\sqrt{2}Im\varphi _2^0.`$ (9)
The flavor changing part of the Yukawa Lagrangian in Model III is given by luke ; Atwood:1996vj :
$`_{Y,FC}=\xi _{ij}^U\overline{Q}_{iL}\stackrel{~}{\varphi }_2U_{jR}+\xi _{ij}^D\overline{Q}_{iL}\varphi _2D_{jR}+\mathrm{H}.\mathrm{c}.,`$ (10)
where $`\stackrel{~}{\varphi }_2=i\tau _2\varphi _2`$, $`Q`$ stands for the quark $`SU(2)_L`$ doublets, $`U(D)`$ for up-type (down-type) quark $`SU(2)_L`$ singlets and $`\xi ^U,\xi ^D`$ are $`3\times 3`$ non-diagonal matrices (in family space) that parametrize the strength of the FCNC vertices in the neutral Higgs sector. Adopting the Cheng and Sher Ansatz we set:<sup>5</sup><sup>5</sup>5Note that there is a factor of 1/2 difference between our definition for $`\xi _{ij}^{U,D}`$ in (Eq. 11) and the one used in Bar-Shalom:1997sj . This difference may be absorbed by redefining the arbitrary parameters $`\lambda _{ij}`$.
$`\xi _{ij}^{U,D}=\lambda _{ij}{\displaystyle \frac{\sqrt{m_im_j}}{v}},v=\left(\sqrt{2}G_F\right)^{1/2},`$ (11)
where for simplicity we assume the $`\lambda _{ij}`$โs to be real<sup>6</sup><sup>6</sup>6In this work we are not interested in CP-violating effects that may be driven by a possible phase contained in the $`\lambda _{ij}`$โs. and symmetric (i.e., $`\lambda _{ij}^{}=\lambda _{ji}`$) constants. For the Higgs-top-charm coupling we will take that $`\lambda _{tc}=\lambda _{ct}\lambda ๐ช(1)`$, which is compatible with all existing data, see Bar-Shalom:1997sj ; Atwood:1996vj for details.
Thus, for the top decays of our interest in this paper, the relevant terms in the Yukawa Lagrangian are Bar-Shalom:1997sj :
$`_{tc}`$ $`=`$ $`\lambda {\displaystyle \frac{\sqrt{m_cm_t}}{\sqrt{2}v}}f_{}\overline{c}t,`$
$`_{VV}`$ $`=`$ $`gm_WG_VS_{}g_{\mu \nu }V^\mu V^\nu ,`$ (12)
where $`=h^0`$ or $`H^0`$, $`V=W`$ or $`Z`$ and
$`f_{h^0;H^0}`$ $`=`$ $`\mathrm{cos}\alpha ;\mathrm{sin}\alpha ,`$
$`S_{h^0;H^0}`$ $`=`$ $`\mathrm{sin}\alpha ;\mathrm{cos}\alpha ,`$
$`G_{W;Z}`$ $`=`$ $`1;{\displaystyle \frac{m_Z^2}{m_W^2}}.`$ (13)
We will further need the $`q_iq_i`$ (with $`=h^0,A^0`$) and $`H^\pm tb`$ couplings Bar-Shalom:1997sj ; Atwood:1996vj :
$`_{q_iq_i}`$ $`=`$ $`{\displaystyle \frac{m_{q_i}}{v}}\overline{q}_i\left[h^0\left(\mathrm{sin}\alpha +{\displaystyle \frac{\lambda _{ii}}{\sqrt{2}}}\right)+iA^0{\displaystyle \frac{\lambda _{ii}}{\sqrt{2}}}\gamma _5\right]q_i,`$
$`_{H^{}t\overline{b}}`$ $`=`$ $`{\displaystyle \frac{1}{2v}}V_{tb}^{}H^{}\overline{b}[(\lambda _{bb}m_b\lambda _{tt}m_t)(\lambda _{bb}m_b+\lambda _{tt}m_t)\gamma _5]t.`$ (14)
## IV Finite width effects in the $`tcWW`$ and $`tcZZ`$ decays
In this section we will use the CM to evaluate the FWE in the top decays $`tcWW`$ and $`tcZZ`$. Kinematically, the naive threshold (i.e., not including FWE) for the decay $`tcZZ`$ is about 4 GeV away (i.e., larger) from the recent CDF $`1\sigma `$ limit (from Tevatron RUN II) on the top mass, $`m_t(1\sigma )180.2`$ GeV recenttop1 . Also, as will be shown below, even for $`tcWW`$ the available phase space can be (depending on the top mass) small enough for the FWE to become significant.
We will consider the decay $`tcWW`$ at the tree-level in both the SM and Model III, while $`tcZZ`$ will be analysed only within Model III, since in the SM this decay is doubly suppressed by both one-loop factors and non-diagonal Cabibbo-Kobayashi-Maskawa (CKM) elements and is, therefore, unobservably small.
In the SM, the tree-level decay $`tcWW`$ proceeds via $`td^{}W^+cW^{}W^+`$ ($`d=d,s`$ or $`b`$ quarks), with a BR of the order of $`๐ช(10^{14}10^{13})`$ (depending on the top-quark mass) if FWE are not taken into account Jenkins:1996zd . The dominant SM diagram is $`tb^{}W^+cW^{}W^+`$, since $`V_{tb}\times V_{cb}`$ is the largest out of the three possible products of CKM elements that enter this decay. In Model III there are two additional tree-level diagrams: $`tch^0cW^+W^{}`$ and $`tcH^0cW^+W^{}`$ Bar-Shalom:1997tm ; Bar-Shalom:1997sj . In this case, we will use the Breit-Wigner prescription for the propagators of $`=h^0`$ or $`H^0`$, i.e., $`(q^2m_{}^2+im_{}\mathrm{\Gamma }_{})^1`$, where $`\mathrm{\Gamma }_{}`$ is the total $``$ width calculated from the dominant $``$ decay modes: $`b\overline{b},t\overline{t},t\overline{c},ZZ,WW,WW^{},ZZ^{}`$.<sup>7</sup><sup>7</sup>7Note that in Model III the decay $`t\overline{c}`$ becomes important for $`\lambda _{tc}๐ช(1)`$.<sup>,</sup><sup>8</sup><sup>8</sup>8 Depending on the $``$ mass, only the kinematically allowed decays will be included in $`\mathrm{\Gamma }_{}`$.
Using the CM, the partial decay width for $`tcWW`$ in any given model $`M`$ can be written as \[see Eq. (6)\]:
$`\mathrm{\Gamma }_{\mathrm{conv}}^M(tcWW)=`$ $`{\displaystyle \frac{1}{512\pi ^3m_t^3}}{\displaystyle _0^{\left(m_tm_c\right)^2}}๐p_{W^+}^2\left[{\displaystyle \frac{p_{W^+}^2\mathrm{\Gamma }_W^0}{m_W\pi \left(\left(p_{W^+}^2m_W^2\right)^2+\left(\frac{p_{W^+}^2\mathrm{\Gamma }_W^0}{m_W}\right)^2\right)}}\right]`$ (15)
$`\times {\displaystyle _0^{\left(m_tm_c\sqrt{p_{W^+}^2}\right)^2}}dp_W^{}^2\left[{\displaystyle \frac{p_W^{}^2\mathrm{\Gamma }_W^0}{m_W\pi \left(\left(p_W^{}^2m_W^2\right)^2+\left(\frac{p_W^{}^2\mathrm{\Gamma }_W^0}{m_W}\right)^2\right)}}\right]`$
$`\times {\displaystyle _{\left(m_c+\sqrt{p_W^{}^2}\right)^2}^{\left(m_t\sqrt{p_{W^+}^2}\right)^2}}dx_1{\displaystyle _{x_{2,min}}^{x_{2,max}}}dx_2|_{\mathrm{conv}}^M(x_1,x_2,p_{W^+}^2,p_W^{}^2)|^2,`$
where the superscript $`M`$ stands for the model used for the calculation of the convoluted amplitude $`_{\mathrm{conv}}^M`$, and
$`x_{2,min}`$ $`=`$ $`(E_2+E_3)^2\left(\sqrt{E_2^2p_W^{}^2}+\sqrt{E_3^2p_{W^+}^2}\right)^2,`$
$`x_{2,max}`$ $`=`$ $`(E_2+E_3)^2\left(\sqrt{E_2^2p_W^{}^2}\sqrt{E_3^2p_{W^+}^2}\right)^2,`$
$`E_2`$ $`=`$ $`{\displaystyle \frac{x_1m_c^2+p_W^{}^2}{2\sqrt{x_1}}};E_3={\displaystyle \frac{x_1p_{W^+}^2+m_t^2}{2\sqrt{x_1}}}.`$ (16)
For the BR calculation, we approximate the total width of the top quark by its dominant decay $`tbW`$ which is computed at tree-level with the corresponding value of the top quark mass.
In Fig. 1 we plot the $`\mathrm{BR}(tcW^+W^{})`$ as a function of the top quark mass in the SM, with and without FWE. The case of stable $`W`$โs in the final state (i.e. without FWE) is obtained by taking the limit $`\rho (p_W^2,m_W^2,\mathrm{\Gamma }_W^0)\delta (p_W^2m_W^2)`$ \[see Eq. (5)\] which sets $`p_{W^\pm }^2=m_W^2`$ in the integrand of Eq. (15). The decay $`tcW^+W^{}`$ in the SM with stable $`W`$โs was calculated in Jenkins:1996zd and our result for this case agrees with hers. From Fig. 1 we see that for the CDF central value of the top mass, $`m_t=173.5`$ GeV, FWE can enhance the $`\mathrm{BR}(tcW^+W^{})`$ by about an order of magnitude, reaching $`210^{13}`$. For the lower $`1\sigma `$ CDF limit $`m_t167`$ GeV, the enhancement due to FWE is of about two orders of magnitudes. Unfortunately, even with such large FWE in the decay $`tcW^+W^{}`$, the BR in the SM is still too small to be measured - even at the LHC.
In Fig. 2 we show the $`\mathrm{BR}(tcW^+W^{})`$ in Model III with $`\lambda _{tc}=1`$, $`m_{H^0}=1`$ TeV and $`\alpha =\frac{\pi }{4}`$<sup>9</sup><sup>9</sup>9The dependence of $`\mathrm{BR}(tcW^+W^{})`$ and $`\mathrm{BR}(tcZZ)`$ on the Higgs mixing angle $`\alpha `$ in Model III can be found in Bar-Shalom:1997tm ; Bar-Shalom:1997sj . The maximum of these branching ratios with respect to $`\alpha `$ takes place at $`\alpha =\pi /8`$ (due to the dependency of the Higgs width on $`\alpha `$) and not at $`\alpha =\pi /4`$ which is used through out our analysis. (note that the SM tree-level contribution to $`tcWW`$, although included, is negligible in this case), as a function of $`m_t`$ with and without FWE, for several values of the light Higgs mass $`m_{h^0}=130,150,170,\mathrm{and}190`$ GeV, and as a function of $`m_{h^0}`$ with FWE, for the lower, upper and central CDF values of the top-quark mass $`m_t=166.9,173.5`$, and 180.2 GeV. As was found in Bar-Shalom:1997tm ; Bar-Shalom:1997sj , in Model III without FWE, the $`\mathrm{BR}(tcW^+W^{})`$ can at most reach the level of $`\mathrm{few}\times 10^5`$ if $`m_t`$ lies within its $`1\sigma `$ CDF limits and only if $`m_{h^0}m_t`$. On the other hand, when FWE are โturned onโ, a huge enhancement to the width arises within a large range of the Higgs mass. In particular, for $`100\mathrm{GeV}\stackrel{<}{}m_{h^0}\stackrel{<}{}165`$ GeV, we find $`\mathrm{BR}(tcW^+W^{})\stackrel{>}{}10^4`$, if $`167\mathrm{GeV}\stackrel{<}{}m_t\stackrel{<}{}180`$ GeV, in Model III when FWE are included. Note that, for the lower $`1\sigma `$ limit $`m_t167`$ GeV, i.e., close to the threshold for producing $`cWW`$, the FWE causes an up to six orders of magnitudes enhancement to the $`\mathrm{BR}(tcW^+W^{})`$ if, e.g., $`m_h130`$ GeV.
For the decay $`tcZZ`$ in Model III we use the analytical results of $`tcWW`$ with the replacements $`m_Wm_Z/\mathrm{cos}\theta _W`$ in the $`VV`$ vertex, $`p_W^{}p_{Z_1},p_W^+p_{Z_2}`$ in Eq. (15) and with an additional overall factor of 1/2 to take into account the symmetry factor for identical particles in the final state (i.e., $`Z`$ bosons). Fig. 3 shows the scaled branching ratio $`\mathrm{BR}(tcZZ)/\lambda ^2`$ ($`\lambda \lambda _{tc}`$) in Model III with $`m_{H^0}=1`$ TeV and $`\alpha =\pi /4`$ (see also footnote 9), as a function of $`m_t`$ with and without FWE, for $`m_{h^0}=130,150,170,\mathrm{and}190`$ GeV, and as a function of $`m_{h^0}`$ with FWE, for $`m_t=166.9,173.5`$, and 180.2 GeV. Note that the decay $`tcZZ`$ is fundamentally different from $`tcWW`$, since, unlike $`tcWW`$, this decay channel cannot occur for stable Z-bosons if $`m_t`$ lies within its $`1\sigma `$ limits. Thus, the inclusion of FWE in $`tcZZ`$ is crucial in this case. In particular, from Fig. 3 we see that a remarkably large $`\mathrm{BR}(tcZZ)10^510^3`$ is expected in Model III, if $`m_{h^0}`$ lies within $`90\mathrm{GeV}\stackrel{<}{}m_{h^0}\stackrel{<}{}170`$ GeV. Such a large BR will be accessible to the LHC and may even be detected at the Tevatron.
Finally we note that, following Altarelli:2000nt (who took $`m_b=m_B`$ for their calculation of $`tbWZ`$), we take $`m_c=m_D=1.87`$ GeV.
## V Finite width effects in the $`A^0(\overline{t}b+t\overline{b})W`$ and $`h^0(\overline{t}b+t\overline{b})W`$ decays
In this section we will examine FWE in three-body decays of neutral Higgs bosons in Model III. We will focus on the decay channels $`A^0\overline{t}bW^+`$ and $`h^0\overline{t}bW^+`$ which can have both theoretical and experimental advantages for Higgs searches and for investigating Higgs properties in the Higgs mass range $`200\mathrm{GeV}\stackrel{<}{}m_{h^0},m_{A^0}\stackrel{<}{}300`$ GeV.
The tree level diagrams contributing to these two decays in Model III are given in Fig. 5 (note that, for the $`A^0`$ decay, the diagram with an intermediate $`W`$-boson is missing, i.e., diagram (d), due to the absence of a tree-level $`A^0WW`$ coupling). A fomula analogous to Eq. (6) can be given for Higgs decays as
$`\mathrm{\Gamma }(b\overline{a}V)={\displaystyle _0^{\left(m_{}m_b\right)^2}}๐p_1^2{\displaystyle _0^{\left(m_{}m_b\sqrt{p_1^2}\right)^2}}๐p_2^2`$ $`\rho _1(p_1^2,m_t,\mathrm{\Gamma }_a^0)\rho _2(p_2^2,m_V,\mathrm{\Gamma }_V^0)`$ (17)
$`\times \mathrm{\Gamma }^0(b\overline{a}(p_1^2)V(p_2^2))),`$
where $`=h^0`$ or $`A^0`$ and $`a(b)`$ is the top(bottom) quark. Using the interaction terms in Section 3, we calculate the matrix element for each decay, where:
* The propagator of the intermediate $`W`$ is taken from Eq. (8).
* In the calculation of $`\mathrm{\Gamma }^0`$ in Eq. (17), the usual sum over the spins of the outgoing top-quark is modified to $`u(p_t)\overline{u}(p_t)=p_t\text{/}+\sqrt{p_t^2}`$ since, using the prescription of the CM, the final state top-quark is allowed to be off-shell.
* Throughout the following we assume that the Higgs mass spectrum respects $`m_{h^0}<m_{A^0}m_H^+,m_H^0`$, setting $`m_H^+=m_H^0=1`$ TeV. Thus, the contribution from the charged Higgs exchange, i.e., diagram (b) in Fig. 5, becomes negligible.
* The total width of $`A^0`$ is estimated from the decays $`A^0\tau \overline{\tau },b\overline{b},h^0Z,h^0Z^{},(t\overline{b}+\overline{t}b)W`$, and the total width of $`h^0`$ is estimated from the decays $`h^0\tau \overline{\tau },b\overline{b},W^+W^{},ZZ`$.
* We set all the relevant flavor diagonal $`\lambda `$โs of the Higgs Yukawa couplings in Eq. (14) to unity, i.e., $`\lambda _{qq}=1`$.
With the above assumptions, the remaining relevant input parameters (in Model III) for evaluating the branching ratios under consideration are $`m_{A^0}`$, $`m_{h^0}`$ and the Higgs mixing angle $`\alpha `$.
In Fig. 6 we depict the branching ratio of $`A^0(\overline{t}b+t\overline{b})W`$ as a function of $`m_{A^0}`$, for two values of the light Higgs mass $`m_{h^0}=170`$ and 230 GeV and for $`m_H^+=1`$ TeV, $`m_t=173.5`$ GeV and $`\alpha =\pi /4`$. We see that near threshold, i.e., $`m_{A^0}260`$ GeV, there is an enhancement of several orders of magnitude due to FWE, wherein the the branching ratio can reach $`\mathrm{BR}(A^0(\overline{t}b+t\overline{b})W)10^2`$. Away from threshold, the decay $`A^0(\overline{t}b+t\overline{b})W`$ is sensitive to the lightest neutral Higgs mass, $`m_{h^0}`$. In this case, the inclusion of FWE can increase the branching ratio by almost an order of magnitude, giving e.g. $`\mathrm{BR}(A^0(\overline{t}b+t\overline{b})W)\mathrm{few}\times 10^1`$ for $`m_{A^0}300`$ GeV and $`m_{h^0}=230`$ GeV. Thus, FWE in the three-body decay $`A^0(\overline{t}b+t\overline{b})W`$ can become very significant โ bringing its BR to the level of tens of percents and making it competitive with the $`A^0`$ two-body decays and, therefore, a viable experimental signature for studies of the properties of the Higgs sector.
Finally, let us consider the decay $`h^0(\overline{t}b+t\overline{b})W`$. In Fig. 7 we plot its branching ratio as a function of $`m_{h^0}`$ for the same input parameters (of Model III) as in Fig. 6. In this case, in spite of the large enhancement near threshold due to FWE, the $`\mathrm{BR}(h^0(\overline{t}b+t\overline{b})W)`$ remains rather small, i.e., at most of $`๐ช(10^5)`$, mainly due to the much larger $`h^0`$ total width caused by its tree-level decays to a pair of gauge-bosons $`h^0WW,ZZ`$.
## VI Summary
We have studied and emphasized the importance of FWE (finite width effects) in decays occurring just around their kinematical thresholds. For the inclusion of FWE we have adapted the so called CM (convolution method). In the CM, the unstable particle with 4-momentum $`p`$ is treated as a real physical particle with an invariant mass $`\sqrt{p^2}`$ and effectively weighted by a Breit-Wigner-like density function, which, becomes a Dirac-delta function in the limit that the particleโs total width approaches zero.
We first examined the FWE within the SM in the rare and flavor-changing tree-level top decay $`tcW^+W^{}`$ and then extended our analysis to FWE in the tree-level top decays $`tcW^+W^{}`$, $`tcZZ`$ and Higgs decays $`A^0,h^0t\overline{b}W`$ in a general two Higgs doublets model, the so called Model III, which gives rise to tree-level FCNC in the Higgs-fermion sector. In all these case we find that FWE can become substantial โ enhancing the branching ratios for the above decays by several orders of magnitudes near threshold.
Unfortunately, in the SM case, the top decay $`tcW^+W^{}`$ remains too small to be of any value in the upcoming high energy colliders, i.e., $`\mathrm{BR}^{\mathrm{SM}}(tcW^+W^{})10^{13}10^{12}`$, in spite of the large enhancement due to FWE. On the other hand, in Model III, the large enhancement due to FWE in all these three-body top and Higgs decays can make a difference with respect to experimental studies in the upcoming hadron colliders. In particular, the branching ratios for the top-decays $`tcW^+W^{}`$ and $`tcZZ`$ can reach the level of $`10^410^3`$ near threshold โ many orders of magnitudes larger than the corresponding branching ratio for the stable W and Z-bosons case (i.e., without FWE). For the $`tcZZ`$ decay, the inclusion of FWE is essential since such a large branching ratio arises even though the naive threshold for this decay is a few GeV away from the most recent $`1\sigma `$ upper limit on the top mass, $`m_t(1\sigma )180`$ GeV.
In the Higgs decays, FWE are more noticeable in the pseudo-scalar Higgs decay $`A^0(\overline{t}b+t\overline{b})W`$, elevating its branching ratio to the level of tens of percents, thus making this three-body decay channel dominant and competitive with its two-body decays and, therefore, extremely important for experimental studies.
Thus, our study shows that FWE is essential for a proper treatment of otherwise neglected finite widths of particles which emerge at the final state of decays or scattering processes occurring just around the threshold.
###### Acknowledgements.
I.T. would like to thank the HEP group members at Technion for their support and kind hospitality during his stay there. The work of G.E. was supported in part by the United States Department of Energy under Grant Contract No. DE-FG02-95ER40896 and by the Israel Science Foundation. The work of M.F. is supported in part by NSERC under grant number 0105354. I.T. also thanks Marc Sher for useful conversations.
|
warning/0506/cond-mat0506248.html
|
ar5iv
|
text
|
# Extension of Frรถhlichโs method to 4-fermion interactions
## I Introduction
The BCS theoryBardeen et al. (1957) proved extremely effective as a theory of superconductivity. The key idea of this theoryโthe attractive interaction which binds electrons into Cooper pairsโhas its essential origins in earlier conclusions drawn FrรถhlichFrรถhlich (1952), who performed a unitary transformation $`U`$ of the electron-phonon Hamiltonian $`H_{eph}`$. $`U`$ was adjusted so as to eliminate the electron-phonon interaction as far as possible and replace it by an effective interaction $`V_{eff}`$ between electrons dressed in the phonon field. $`V_{eff}`$ proved to be attractive for 1-electron energies close to $`\epsilon _F`$. A reduced form of $`V_{eff}`$ was subsequently used by Bardeen, Cooper and Schrieffer in their theoryBardeen et al. (1957).
It is worth emphasizing that Frรถhlichโs transformation is not strictly unitary, because the less significant terms of the resulting expansion of $`UH_{eph}U^1`$ were discarded. Correctness of the remaining terms, included into the Hamiltonian of a superconductor, was confirmed by the success of the BCS theory.
Unfortunately, BCS theory proved incapable of explaining superconductivity in type-II superconductors, heavy fermions and high-$`T_c`$ superconductors (HTSC). The search for an alternative theory of superconductivity proceeds in various directions and one of them exploits the idea of extending BCS theory by adding to the BCS Hamiltonian $`H_{BCS}`$ further interactions. RickayzenRickayzen (1964) suggested incorporation of a 4-fermion interaction, motivating such choice by analogies between theory of superconductivity and nuclear theory, where such interactions had been considered. This idea was also mentioned by VolovikVolovik (1992), but remained in the realm of theoretical concepts until the late nineties, when Maฤkowiak and one of authors (PT)Maฤkowiak and Tarasewicz (1996, 1997, 1998, 2000); Tarasewicz and Maฤkowiak (2000); Tarasewicz (2004a, b) (MT) proposed a Hamiltonian $`H_{MT}=H_{BCS}+W+V_{MT}`$, where
$$H_{BCS}=T+V_{BCS}=\underset{๐ค\sigma }{}\xi _๐คn_{๐ค\sigma }|\mathrm{\Lambda }|^1\underset{\mathrm{๐ค๐ค}^{}}{}G_{\mathrm{๐ค๐ค}^{}}a_{๐ค+}^{}a_๐ค^{}a_๐ค^{}a_{๐ค^{}+},$$
(1)
$$W=\underset{๐ค}{}\gamma _๐คn_{๐ค+}n_๐ค,$$
(2)
$$V_{MT}=|\mathrm{\Lambda }|^1\underset{๐ค,๐ค^{}}{}g_{\mathrm{๐ค๐ค}^{}}a_๐ค^{}a_{๐ค+}^{}a_๐ค^{}a_{๐ค+}^{}a_{๐ค^{}+}a_๐ค^{}a_{๐ค^{}+}a_๐ค^{},$$
(3)
$`|\mathrm{\Lambda }|`$ in Eqs. (1)โ(3) denotes the systemโs volume, whereas $`g_{\mathrm{๐ค๐ค}^{}}`$, $`G_{\mathrm{๐ค๐ค}^{}}`$ are bounded functions and $`\gamma _๐ค`$, $`g_{\mathrm{๐ค๐ค}^{}}`$, $`G_{\mathrm{๐ค๐ค}^{}}`$ are invariant under time reversal $`๐ค๐ค`$ or $`๐ค^{}๐ค^{}`$.
Considerations which led to the introduction of $`W`$ were founded on the analysis of the HTSC normal state. There are grounds to believe that this is not a normal Fermi liquid state. The interaction $`W`$, first added to $`H_{BCS}`$ by CzerwonkoCzerwonko (1994, 1995), guaranteed normal-state behaviour characteristic of the so-called statistical spin liquid, considered earlier by Spaลek and WรณjcikSpaลek and Wรณjcik (1986, 1988).
$`W`$ is a 2-electron interaction. This raises the question, whether it can be obtained by a reduction procedure (different than the BCS one) of the interaction derived by Frรถhlich. It can be easily verified that this is impossible. More precisely, for the unique possible reduction of momenta, the coupling vanishes, meaning that the nature of $`W`$ is not phononic.
Introduction of the 4-fermion interaction $`V_{MT}`$ was justified in Ref. Maฤkowiak and Tarasewicz, 1996 by its possible role as an attraction between pairs in HTSC, mediated by phonons or other quantaMaฤkowiak and Tarasewicz (2000). A conjecture put forward in Ref. Maฤkowiak and Tarasewicz, 1996 suggested also that $`V_{MT}`$ could be expected to arise as one of the higher-order terms of Frรถhlichโs expansion of $`UH_{eph}U^1`$. An alternative justification was given in Ref. Maฤkowiak and Tarasewicz, 2000, where $`V_{MT}`$ was viewed as a BCS-type interaction between quasi-particles of a free gas represented by $`W`$ written in the form $`_๐ค\gamma _๐คc_๐ค^{}c_๐ค`$, with $`c_๐ค=a_{๐ค+}a_๐ค`$. Both of these ideas essentially exploit the concept of phonon-type mediation of interactions. The significance of this mediation in HTSC was stressed by Wysokiลski Wysokiลski (1996).
The question of the form of higher-order terms of Frรถhlichโs expansion is interesting itself, not only as providing a possible explanation of the MT extension, but, first of all, because these terms could throw some light on further possible extensions of $`H_{BCS}`$ related to effects of electron-lattice interaction. Since Fermi-liquid theory will remain the foundation of our formalism, we shall focus our interest on effective electron interactions.
It is worth noting that the possible presence of fermion quadruples in superconductors and superfluids was considered in a number of papers. Schneider and KellerSchneider and Keller (1993) measured the various characteristics of some cuprates and Chevrel-phase superconductors, especially concentrating on the relation between the critical temperature and zero temperature condensate density. They noticed that the experimental data for e.g., $`\mathrm{YBa}_2\mathrm{Cu}_3\mathrm{O}_{6.602}`$ point to similarities with the behaviour of a dilute Bose gas. As a result they suggested Bose condensation of weakly interacting fermion pairs as a mechanism of transition from normal to superconducting state. Bunkov et al.Bunkov et al. (2000) pointed to presence of fermion quadruples in $`^3\mathrm{He}`$. Their work was devoted to the problem of influence of spatial disorder on the order parameter in superfluid $`^3\mathrm{He}`$. By resorting to the work of VolovikVolovik (1992), they suggested that unusual spectra of $`^3\mathrm{He}`$ in aerogel could be explained by a process in which impurities tend to destroy the anisotropic correlations of the order parameter, while correlations of higher symmetry can survive (e.g., four-particle correlations). Recently Schneider et al.Schneider et al. (2004) discovered of half-$`h/2e`$ magnetic flux in SQUIDs fabricated of bicrystalline $`\mathrm{YBa}_2\mathrm{Cu}_3\mathrm{O}_{7\delta }`$ films. This situation corresponds to the presence of fermion quadruples in the system. Based on this observation, Aligia et al.Aligia et al. (2005) investigated a model of interface between two superconductors, based on a one dimensional boson lattice model and proposed formation of quartets of electrons.
In Section II higher order terms of the expansion of Frรถhlichโs transformation $`UH_{eph}U^1`$ are discussed qualitatively in order to exhibit the emerging structure. Owing to the complexity of this procedure, only terms proportional to the third power of the electron-phonon coupling are found in Section III. The resulting extended Frรถhlich Hamiltonian is transformed in the next section by a second Frรถhlich-type transformation, which produces 4-electron terms. These are discussed in detail in Section V. In particular, a reduction similar to the BCS one is performed, which yields an interaction of the form $`V_{MT}`$ in Eq. (3). The expression for $`g_{\mathrm{๐ค๐ค}^{}}`$ is derived. Detailed analysis of this expression is performed in the next section, in particular, we showed, by applying some approximations, that this expression is negative i.e., $`V_{MT}`$ is attractive. The original Frรถhlichโs transformation and the resulting terms, up to 6th power of the coupling, are commented in Section VII. The final section contains a discussion, summary and open questions.
## II Higher order terms of the Frรถhlichโs transformation
Following FrรถhlichFrรถhlich (1952) (see Appendix A for details), let us consider the electron-phonon Hamiltonian:
$$H_{eph}=H_0+H_{int}=\underset{๐ค\sigma }{}\epsilon _๐คa_{๐ค\sigma }^{}a_{๐ค\sigma }+\underset{๐ฐ}{}\omega _๐ฐb_๐ฐ^{}b_๐ฐ+i\underset{๐ฐ}{}D_๐ฐ\left(b_๐ฐ\rho _๐ฐ^{}b_๐ฐ^{}\rho _๐ฐ\right),$$
(4)
where
$$\rho _๐ฐ=\underset{๐ค\sigma }{}a_{๐ค๐ฐ\sigma }^{}a_{๐ค\sigma },$$
(5)
and $`a_{๐ค\sigma }`$ ($`b_๐ค`$) are fermion (boson) operators. The coupling $`D_๐ฐ`$ will be assumed small and $`\mathrm{}1`$.
Since the interaction is spin-independent, the spin index will be suppressed. Summation over electron momenta will include summation over spins.
Frรถhlich performed a unitary transformation of $`H_{eph}`$ in order to eliminate (as far as possible) the interaction term. The transformed Hamiltonian is
$$H=\mathrm{e}^S^{}H_{\mathrm{e}\mathrm{ph}}\mathrm{e}^S=H_{\mathrm{e}\mathrm{ph}}[S,H_{\mathrm{e}\mathrm{ph}}]+\frac{1}{2}[S,[S,H_{\mathrm{e}\mathrm{ph}}]]+\mathrm{},$$
(6)
where
$$S=\underset{๐ช}{}S_๐ช=\underset{๐ช}{}\left(\gamma _๐ช^{}b_๐ช^{}\gamma _๐ชb_๐ช\right)=S^{},$$
(7)
$$\gamma _๐ช=\underset{๐ค}{}\varphi (๐ค,๐ช)a_๐ค^{}a_{๐ค๐ช},$$
(8)
and the unknown function $`\varphi (๐ค,๐ช):^3\times ^3^1`$ is adjusted to achieve the cancellation.
Subsequently, a term which is a combination of products, each with $`f`$ fermion operators and $`b`$ boson operators will be written as $`(f,b)`$. Clearly, $`f`$ will always be even. For example, $`H_0`$ consists of terms $`(2,0)`$ and $`(0,2)`$.
The rhs of Eq. (6) expresses in terms of commutators $`[(f_1,b_1),(f_2,b_2)]`$. One easily finds that
$$[(f_1,b_1),(f_2,b_2)]=[f_1,f_2]b_1b_2+f_2f_1[b_1,b_2]=[f_1,f_2]b_2b_1+f_1f_2[b_1,b_2].$$
(9)
The necessary commutators $`[f_1,f_2]`$, $`[b_1,b_2]`$ are given in Appendix C.
According to Eq. (6), the transformation can be performed, given commutators of the form occurring in Eq. (9) with the first argument equal $`S`$. The latter is a $`(2,1)`$ expression, hence
$$[S,(f,b)]=[(2,1),(f,b)]=(f,b+1)+(f+2,b1),$$
(10)
by virtue of Eqs. (9), (72). Clearly $`(f,b1)=0`$ for $`b=0`$.
Based on this ground, Frรถhlich obtained the transformed Hamiltonian (see Appendix A):
$$H_F=\underset{๐ค}{}\epsilon _๐คn_๐ค\frac{1}{2}\underset{\mathrm{๐ค๐ช๐ฐ}}{}\frac{D_๐ฐ^2\left(1+\mathrm{\Delta }(๐ค,๐ฐ)\right)\left(1\mathrm{\Delta }(๐ช,๐ฐ)\right)}{\epsilon _{๐ช๐ฐ}\epsilon _๐ช+\omega _๐ฐ}(a_๐ค^{}a_{๐ค๐ฐ}a_{๐ช๐ฐ}^{}a_๐ช+c.c.).$$
The second term represents an effective interaction between electrons dressed in the phonon field. If $`\epsilon _{๐ช๐ฐ}\epsilon _๐ช+\omega _๐ฐ>0`$, this interaction is attractive.
Omission of higher terms in Eq. (6) results in violation of unitarity. The question thus arises whether partial inclusion of these terms (first of all those of 3rd order in the coupling) could improve agreement between theory and experiment.
Following the general rule for the action of $`S`$ in consecutive orders, expressed by Eq. (10), one easily finds the form of subsequent terms:
$$\left\{\begin{array}{c}\hfill (2,0)\\ \hfill (0,2)\end{array}\right\}\stackrel{๐}{}(2,1)\stackrel{๐}{}\left\{\begin{array}{c}\hfill (4,0)\\ \hfill (2,2)\end{array}\right\}\stackrel{๐}{}(4,1)\stackrel{๐}{}\left\{\begin{array}{c}\hfill (6,0)\\ \hfill (4,2)\end{array}\right\}\stackrel{๐}{}(6,1)\stackrel{๐}{}\left\{\begin{array}{c}\hfill (8,0)\\ \hfill (6,2)\end{array}\right\}.$$
(11)
In each consecutive step one obtains terms proportional to the next power of the coupling.
From the view-point of Fermi liquid theory, the terms representing effective inter-electron interactions are most interesting. According to Eq. (11), new terms $`(6,0)`$ (proportional to $`D_๐ฐ^4`$), describing 3-electron interactions and $`(8,0)`$ (proportional to $`D_๐ฐ^6`$), representing 4-electron interactions, appear. Attractive interactions of this type leading to the formation of fermion triples and quadruples, could affect the behaviour of a superconductor. However, the total spin of an electron triple is nonzero, so such clusters are unstable, as they are not invariant under time inversion. So far, there has been no experimental evidence of such objects.
Most electrons in a superconductor below $`T_c`$ are paired, so the 4-electron interaction between Cooper pairs can be expected to prevail. Furthermore, quadruples with a total spin equal zero and appropriate 1-electron momenta are stable under time inversion. On the other hand, under Frรถhlichโs conditions for convergence of series (6), the effect of the terms $`(6,0)`$$`(8,0)`$ is weaker.
Evaluation of these terms is a complicated procedure. Before doing this, let us first examine the 3rd order corrections.
## III Third order of the transformation
Let us consider the effect of the first higher orders discarded by Frรถhlich, i.e., terms proportional to $`D_๐ฐ^3`$. Then the corrected Hamiltonian takes the form
$$H^{}=H_0\left([S,H_0]H_{int}\right)+\left(\frac{1}{2}[S,[S,H_0]][S,H_{int}]\right)\left(\frac{1}{6}[S,[S,[S,H_0]]]\frac{1}{2}[S,[S,H_{int}]]\right)+\mathrm{}.$$
(12)
The additional two terms in the last bracket are equal, explicitly,
$$\begin{array}{cc}& \frac{1}{2}[S,[S,H_{int}]]\frac{1}{6}[S,[S,[S,H_0]]]=\underset{\mathrm{๐ช๐ค}}{}A_{\mathrm{๐ค๐ช}}b_๐ช^{}n_๐ค\gamma _๐ช^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}b_๐ฐ^{}\{B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\}+c.c.:=H_e,\hfill \end{array}$$
(13)
where
$$2A_{\mathrm{๐ค๐ช}}=iD_๐ช\varphi ^{}(๐ค,๐ช)+iD_๐ช\varphi (๐ค+๐ช,๐ช)+\frac{1}{3}\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)|\varphi (๐ค,๐ช)|^2\frac{1}{3}\left(\epsilon _๐ค\epsilon _{๐ค+๐ช}+\omega _๐ช\right)|\varphi (๐ค+๐ช,๐ช)|^2+c.c.,$$
(14)
$$\begin{array}{cc}& 2B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}=iD_๐ช\left\{\varphi ^{}(๐ค^{},๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)\varphi ^{}(๐ค^{},๐ช)\varphi ^{}(๐ค,๐ฐ)+\varphi ^{}(๐ค,๐ฐ)\varphi (๐ค,๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)\varphi (๐ค๐ฐ,๐ช)\right\}\hfill \\ & +\frac{1}{3}\varphi ^{}(๐ค^{},๐ช)\left\{\varphi (๐ค,๐ช)\varphi ^{}(๐ค,๐ฐ)\left(\epsilon _๐ค\epsilon _{๐ค๐ช}+\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}\right)+\varphi (๐ค๐ฐ,๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)\left(\epsilon _{๐ค๐ฐ๐ช}\epsilon _{๐ค๐ฐ}\epsilon _{๐ค^{}๐ช}+\epsilon _๐ค^{}\right)\right\},\hfill \end{array}$$
(15)
$$\begin{array}{cc}\hfill 2C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}& =iD_๐ช\{\varphi (๐ค^{},๐ช)\varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)\varphi (๐ค^{},๐ช)\varphi ^{}(๐ค+๐ฐ,๐ฐ)\hfill \\ & \varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)\varphi ^{}(๐ค+๐ฐ,๐ช)+\varphi ^{}(๐ค+๐ฐ,๐ฐ)\varphi ^{}(๐ค,๐ช)\}\hfill \\ & +\frac{1}{3}\varphi (๐ค^{},๐ช)\{\varphi ^{}(๐ค+๐ฐ,๐ช)\varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)(\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}\epsilon _{๐ค+๐ฐ๐ช}+\epsilon _{๐ค+๐ฐ})\hfill \\ & +\varphi ^{}(๐ค,๐ช)\varphi ^{}(๐ค+๐ฐ,๐ฐ)(\epsilon _{๐ค๐ช}\epsilon _๐ค\epsilon _{๐ค^{}๐ช}+\epsilon _๐ค^{})\},\hfill \end{array}$$
(16)
It can be seen that a new phonon index $`๐ฐ`$ has appeared in Eq. (13). It results from the commutator of $`S`$ with the terms $`(4,0)`$, so the harmonic approximation has not been violated.
Substitution of the expression (13) into $`H^{}`$ yields
$$H^{}=H_a+H_b+H_c+H_d+H_e,$$
where
$$H_a=\frac{1}{2}\underset{๐ค}{}\epsilon _๐คn_๐ค+\frac{1}{2}\underset{๐ช}{}\omega _๐ชb_๐ช^{}b_๐ช+c.c.,$$
$$H_b=\underset{\mathrm{๐ช๐ค}}{}b_๐ช^{}b_๐ชn_๐ค\left\{iD_๐ช\left(\varphi (๐ค+๐ช,๐ช)\varphi (๐ค,๐ช)\right)\frac{1}{2}\left(\epsilon _๐ค\epsilon _{๐ค+๐ช}+\omega _๐ช\right)|\varphi (๐ค+๐ช,๐ช)|^2+\frac{1}{2}\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)|\varphi (๐ค,๐ช)|^2\right\}+c.c.,$$
(17)
$$H_c=\underset{\mathrm{๐ช๐ค}}{}b_๐ช\left(iD_๐ช+\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)\varphi (๐ค,๐ช)\right)a_๐ค^{}a_{๐ค๐ช}+c.c.,$$
(18)
$$H_d=\underset{\mathrm{๐ช๐ค๐ค}^{}}{}\left(iD_๐ช\varphi (๐ค,๐ช)+\frac{1}{2}(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช)\varphi (๐ค,๐ช)\varphi ^{}(๐ค^{},๐ช)\right)a_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+c.c..$$
(19)
$`H_b`$ is linear in $`b_๐ช^{}b_๐ช`$ but differs from $`H_a`$, $`H_c`$ is part of $`H_{int}`$ which had been excluded from the transformation to avoid convergence problems, $`H_d`$ is the 2-electron interaction obtained, whereas $`H_e`$ contains terms of the form $`(4,1)`$ which represent the obtained correction to Frรถhlichโs Hamiltonian.
The function $`\varphi (๐ค,๐ช)`$ remains in the form (61), which guarantees minimization of the contribution $`H_c`$.
## IV The second transformation
The structure of $`H^{}`$ bears similarities to that of $`H_{eph}`$. Both $`H^{}`$ and $`H_{eph}`$ contain terms describing the interaction of electrons with phonons: the counterpart of $`H_{int}`$ in $`H_{eph}`$ is $`H_e`$ in $`H^{}`$. To estimate the effect of $`H_e`$, let us repeatedly apply Frรถhlichโs method and perform a second unitary transformation adjusted to eliminate $`H_e`$ as far as possible. The form of $`H_e`$ suggests to take
$$S^{}=\underset{๐ฎ}{}\left(\eta _๐ฎ^{}b_๐ฎ^{}\eta _๐ฎb_๐ฎ+\xi _๐ฎ^{}b_๐ฎ^{}\xi _๐ฎb_๐ฎ+\zeta _๐ฎ^{}b_๐ฎ^{}\zeta _๐ฎb_๐ฎ\right),$$
where
$$\eta _๐ฎ^{}=\underset{\mathrm{๐ฅ๐ฆ}}{}\psi ^{}(๐ฅ,๐ฆ,๐ฎ)a_๐ฅ^{}a_๐ฅa_{๐ฆ๐ฎ}^{}a_๐ฆ,$$
(20)
$$\xi _๐ฎ^{}=\underset{\mathrm{๐ฅ๐ฆ๐ญ}}{}\chi ^{}(๐ฅ,๐ฆ,๐ญ,๐ฎ)a_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ญ}a_{๐ฆ๐ญ}^{}a_๐ฆ,$$
(21)
$$\zeta _๐ฎ^{}=\underset{\mathrm{๐ฅ๐ฆ๐ญ}}{}\phi ^{}(๐ฅ,๐ฆ,๐ญ,๐ฎ)a_๐ฅ^{}a_{๐ฅ๐ญ}a_{๐ฆ๐ญ}^{}a_{๐ฆ+๐ฎ}.$$
(22)
The explicit form of the functions $`\psi `$, $`\chi `$, $`\phi `$ will be found below.
The interaction $`H_e`$ is 3rd order in $`D_๐ช`$, therefore bearing in mind Frรถhlichโs approach, we shall restrict the expansion of $`\mathrm{exp}[S^{}]H^{}\mathrm{exp}[S^{}]=\widehat{H}`$ to terms which are 6th order in $`D_๐ช`$. These terms are
$$\widehat{H}=H^{}[S^{},H^{}]+\frac{1}{2}[S^{},[S^{},H_a]]+\mathrm{}.$$
(23)
To evaluate the rhs, one needs the commutators
$$\begin{array}{cc}\hfill [S^{},H_a]& =\underset{\mathrm{๐ฎ๐ค๐ฅ}}{}b_๐ฎ^{}\psi ^{}(๐ฅ,๐ค,๐ฎ)\left(\epsilon _๐ค\epsilon _{๐ค๐ฎ}\omega _๐ฎ\right)a_๐ฅ^{}a_๐ฅa_{๐ค๐ฎ}^{}a_๐ค\hfill \\ & +\underset{\mathrm{๐ฎ๐ค๐ฅ๐ญ}}{}b_๐ฎ^{}\chi ^{}(๐ฅ,๐ค,๐ญ,๐ฎ)\left(\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ฅ๐ญ}\epsilon _{๐ฅ๐ฎ}\omega _๐ฎ\right)a_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ญ}a_{๐ค๐ญ}^{}a_๐ค\hfill \\ & +\underset{\mathrm{๐ฎ๐ค๐ฅ๐ญ}}{}b_๐ฎ^{}\phi ^{}(๐ฅ,๐ค,๐ญ,๐ฎ)\left(\epsilon _{๐ค+๐ฎ}\epsilon _{๐ค๐ญ}+\epsilon _{๐ฅ๐ญ}\epsilon _๐ฅ\omega _๐ฎ\right)a_๐ฅ^{}a_{๐ฅ๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ฎ}+c.c.,\hfill \end{array}$$
$$[S^{},H_b]=\underset{\mathrm{๐ช๐ค}}{}b_๐ช^{}n_๐ค\left(\eta _๐ช^{}+\xi _๐ช^{}+\zeta _๐ช^{}\right)\left(D_{\mathrm{๐ค๐ช}}+D_{\mathrm{๐ค๐ช}}^{}\right)+c.c.,$$
where $`D_{\mathrm{๐ค๐ช}}`$ is defined by $`H_b=_๐ชb_๐ช^{}b_๐ชn_๐คD_{\mathrm{๐ค๐ช}}+c.c.`$ in Eq. (17),
$$\begin{array}{cc}& [S^{},H_c]=\underset{\mathrm{๐ช๐ค}}{}E_{๐ค,๐ช}a_๐ค^{}a_{๐ค๐ช}\left(\eta _๐ช^{}+\xi _๐ช^{}+\zeta _๐ช^{}\right)\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฅ}}{}E_{๐ค,๐ช}b_๐ช^{}b_๐ช\left\{\psi ^{}(๐ฅ,๐ค,๐ช)n_๐ฅ\left(n_{๐ค๐ช}n_๐ค\right)+\left(\psi ^{}(๐ค,๐ฅ,๐ช)\psi ^{}(๐ค๐ช,๐ฅ,๐ช)\right)a_๐ค^{}a_{๐ค๐ช}a_{๐ฅ๐ช}^{}a_๐ฅ\right\}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฅ๐ญ}}{}b_๐ช^{}b_๐ช\{(E_{๐ค,๐ช}\chi ^{}(๐ฅ,๐ค,๐ญ,๐ช)E_{๐ค๐ญ,๐ช}\chi ^{}(๐ฅ,๐ค๐ช,๐ญ,๐ช))a_{๐ฅ๐ช}^{}a_{๐ฅ๐ญ}a_{๐ค๐ญ}^{}a_{๐ค๐ช}\hfill \\ & +\left(E_{๐ค+๐ช๐ญ,๐ช}\chi ^{}(๐ค+๐ช,๐ฅ,๐ญ,๐ช)E_{๐ค,๐ช}\chi ^{}(๐ค,๐ฅ,๐ญ,๐ช)\right)a_๐ค^{}a_{๐ค๐ญ}a_{๐ฅ๐ญ}^{}a_๐ฅ\hfill \\ & +\left(E_{๐ค+๐ช+๐ญ,๐ช}\phi ^{}(๐ฅ,๐ค+๐ญ,๐ญ,๐ช)E_{๐ค,๐ช}\phi ^{}(๐ฅ,๐ค+๐ญ๐ช,๐ญ,๐ช)\right)a_๐ฅ^{}a_{๐ฅ๐ญ}a_๐ค^{}a_{๐ค+๐ญ}\hfill \\ & +(E_{๐ค,๐ช}\phi ^{}(๐ค+๐ญ,๐ฅ,๐ญ,๐ช)E_{๐ค+๐ญ,๐ช}\phi ^{}(๐ค+๐ญ๐ช,๐ฅ,๐ญ,๐ช))a_{๐ค+๐ญ}^{}a_{๐ค๐ช}a_{๐ฅ๐ญ}^{}a_{๐ฅ+๐ช}\}+c.c.,\hfill \end{array}$$
where $`E_{๐ค,๐ช}`$ is defined by $`H_c=_{\mathrm{๐ช๐ค}}b_๐ชa_๐ค^{}a_{๐ค๐ช}E_{๐ค,๐ช}+c.c.`$ in Eq. (18),
$$\begin{array}{cc}\hfill [S^{},H_d]& =\underset{\mathrm{๐ฎ๐ช๐ค๐ค}^{}}{}b_๐ฎ^{}(F_{๐ค,๐ค^{},๐ช}+F_{๐ค^{},๐ค,๐ช}^{})\{a_๐ค^{}a_{๐ค๐ช}[\eta _๐ฎ^{},a_{๐ค^{}๐ช}^{}a_๐ค^{}]+[\eta _๐ฎ^{},a_๐ค^{}a_{๐ค๐ช}]a_{๐ค^{}๐ช}^{}a_๐ค^{}\hfill \\ & +a_๐ค^{}a_{๐ค๐ช}[\xi _๐ฎ^{},a_{๐ค^{}๐ช}^{}a_๐ค^{}]+[\xi _๐ฎ^{},a_๐ค^{}a_{๐ค๐ช}]a_{๐ค^{}๐ช}^{}a_๐ค^{}+a_๐ค^{}a_{๐ค๐ช}[\zeta _๐ฎ^{},a_{๐ค^{}๐ช}^{}a_๐ค^{}]+[\zeta _๐ฎ^{},a_๐ค^{}a_{๐ค๐ช}]a_{๐ค^{}๐ช}^{}a_๐ค^{}\}+c.c.,\hfill \end{array}$$
where $`F_{๐ค,๐ค^{},๐ช}`$ is defined by $`H_d=_{\mathrm{๐ช๐ค๐ค}^{}}F_{๐ค,๐ค^{},๐ช}a_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+c.c.`$ in Eq. (19),
$$\begin{array}{cc}& [S^{},H_e]=\underset{\mathrm{๐ช๐ค}}{}A_{๐ค,๐ช}b_๐ช^{}b_๐ช\left\{[\eta _๐ช,n_๐ค\gamma _๐ช^{}]+[\xi _๐ช,n_๐ค\gamma _๐ช^{}]+[\zeta _๐ช,n_๐ค\gamma _๐ช^{}]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ฐ๐ค๐ค}^{}}{}B_{๐ค,๐ช,๐ฐ,๐ค^{}}b_๐ฐ^{}b_๐ฐ\left\{[\eta _๐ฐ,a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}]+[\xi _๐ฐ,a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}]+[\zeta _๐ฐ,a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ฐ๐ค๐ค}^{}}{}C_{๐ค,๐ช,๐ฐ,๐ค^{}}b_๐ฐ^{}b_๐ฐ\left\{[\eta _๐ฐ,a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}]+[\xi _๐ฐ,a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}]+[\zeta _๐ฐ,a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค}}{}A_{๐ค,๐ช}\left\{\eta _๐ชn_๐ค\gamma _๐ช^{}+\xi _๐ชn_๐ค\gamma _๐ช^{}+\zeta _๐ชn_๐ค\gamma _๐ช^{}\right\}\hfill \\ & \underset{\mathrm{๐ช๐ฐ๐ค๐ค}^{}}{}B_{๐ค,๐ช,๐ฐ,๐ค^{}}\left\{\eta _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\xi _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\zeta _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\right\}\hfill \\ & \underset{\mathrm{๐ช๐ฐ๐ค๐ค}^{}}{}C_{๐ค,๐ช,๐ฐ,๐ค^{}}\left\{\eta _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}+\xi _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}+\zeta _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\right\}+c.c.,\hfill \end{array}$$
where $`A_{๐ค,๐ช}`$$`B_{๐ค,๐ช,๐ฐ,๐ค^{}}`$$`C_{๐ค,๐ช,๐ฐ,๐ค^{}}`$ are given, respectively by Eqs. (14), (15) i (LABEL:C),
$$\begin{array}{cc}& [S^{},[S^{},H_a]]=\underset{\mathrm{๐ช๐ค๐ค}^{}}{}G_{๐ค,๐ค^{},๐ช}b_๐ช^{}b_๐ช\left\{[\eta _๐ช,a_๐ค^{}^{}a_๐ค^{}a_{๐ค๐ช}^{}a_๐ค]+[\xi _๐ช,a_๐ค^{}^{}a_๐ค^{}a_{๐ค๐ช}^{}a_๐ค]+[\zeta _๐ช,a_๐ค^{}^{}a_๐ค^{}a_{๐ค๐ช}^{}a_๐ค]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}H_{๐ค,๐ค^{},๐ญ,๐ช}b_๐ช^{}b_๐ช\left\{[\eta _๐ช,a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค]+[\xi _๐ช,a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค]+[\zeta _๐ช,a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}I_{๐ค,๐ค^{},๐ญ,๐ช}b_๐ช^{}b_๐ช\left\{[\eta _๐ช,a_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}]+[\xi _๐ช,a_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}]+[\zeta _๐ช,a_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}]\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค๐ค}^{}}{}G_{๐ค,๐ค^{},๐ช}\left\{\eta _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+\xi _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+\zeta _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}H_{๐ค,๐ค^{},๐ญ,๐ช}\left\{\eta _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+\xi _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+\zeta _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค\right\}\hfill \\ & \underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}I_{๐ค,๐ค^{},๐ญ,๐ช}\left\{\eta _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+\xi _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+\zeta _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}\right\}+c.c.,\hfill \end{array}$$
(24)
where
$$G_{๐ค,๐ค^{},๐ช}=\psi ^{}(๐ค^{},๐ค,๐ช)\left(\epsilon _๐ค\epsilon _{๐ค๐ช}\omega _๐ช\right),$$
(25)
$$H_{๐ค,๐ค^{},๐ญ,๐ช}=\chi ^{}(๐ค^{},๐ค,๐ญ,๐ช)\left(\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _{๐ค^{}๐ช}\omega _๐ช\right),$$
(26)
$$I_{๐ค,๐ค^{},๐ญ,๐ช}=\phi ^{}(๐ค^{},๐ค,๐ญ,๐ช)\left(\epsilon _{๐ค+๐ช}\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}\omega _๐ช\right).$$
(27)
The commutators in the terms containing phonon operators have not been evaluated, as we are interested first of all in expressions containing exclusively electron operators.
Given the transformed Hamiltonian $`\widehat{H}`$, we are now in position to minimize the effect of $`H_e`$ by imposing, similarly as Frรถhlich, the condition $`H_e[S^{},H_a]=0`$. This leads to equations for $`\psi `$, $`\chi `$, $`\phi `$, which determine these functions uniquely, viz,
$$\psi ^{}(๐ค^{},๐ค,๐ช)=\frac{\varphi ^{}(๐ค,๐ช)A_{๐ค^{},๐ช}}{\epsilon _๐ค\epsilon _{๐ค๐ช}\omega _๐ช}.$$
(28)
After substituting $`\varphi `$ and $`A`$ given by Eqs. (61), (14), one obtains
$$\psi ^{}(๐ค^{},๐ค,๐ช)=\frac{2iD_๐ช^3\left(1\mathrm{\Delta }(๐ค,๐ช)\right)}{3\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)^2}\left(\frac{1\mathrm{\Delta }(๐ค^{},๐ช)}{\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}+\omega _๐ช}\frac{1\mathrm{\Delta }(๐ค^{}+๐ช,๐ช)}{\epsilon _๐ค^{}\epsilon _{๐ค^{}+๐ช}+\omega _๐ช}\right).$$
(29)
Introduction of additional functions in order to preserve convergence is not necessary here, as $`\mathrm{\Delta }`$ already guarantees this property.
As for $`\chi `$ and $`\phi `$, additional functions preserving convergence are indispensable, viz,
$$\chi ^{}(๐ค^{},๐ค,๐ญ,๐ช)=\frac{B_{๐ค^{},๐ญ,๐ช,๐ค}}{\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _{๐ค^{}๐ช}\omega _๐ช}\left(1\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)\right),$$
(30)
where
$$\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)=\{\begin{array}{cc}1\text{,}\hfill & \text{if}\left|\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _{๐ค^{}๐ช}\omega _๐ช\right|<\stackrel{~}{\mathrm{\Gamma }}_๐ช\hfill \\ 0\text{,}\hfill & \text{if}\left|\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _{๐ค^{}๐ช}\omega _๐ช\right|\stackrel{~}{\mathrm{\Gamma }}_๐ช\hfill \end{array}.$$
(31)
Taking into account Eqs. (61), (15), one obtains
$$\begin{array}{cc}& \chi ^{}(๐ค^{},๐ค,๐ญ,๐ช)=\frac{iD_๐ญ^2D_๐ช}{6\left(\epsilon _๐ค\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _{๐ค^{}๐ช}\omega _๐ช\right)}(\frac{\left(1\mathrm{\Delta }(๐ค,๐ญ)\right)\left(1\mathrm{\Delta }(๐ค^{},๐ช)\right)}{\left(\epsilon _{๐ค๐ญ}\epsilon _๐ค+\omega _๐ญ\right)\left(\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}+\omega _๐ช\right)}(2+\mathrm{\Delta }(๐ค^{},๐ญ))\hfill \\ & \frac{\left(1\mathrm{\Delta }(๐ค,๐ญ)\right)\left(1\mathrm{\Delta }(๐ค^{}๐ญ,๐ช)\right)}{\left(\epsilon _{๐ค๐ญ}\epsilon _๐ค+\omega _๐ญ\right)\left(\epsilon _{๐ค^{}๐ญ๐ช}\epsilon _{๐ค^{}๐ญ}+\omega _๐ช\right)}\left(2+\mathrm{\Delta }(๐ค^{}๐ช,๐ญ)\right)+\frac{\left(1\mathrm{\Delta }(๐ค^{},๐ญ)\right)\left(1\mathrm{\Delta }(๐ค^{},๐ช)\right)}{\left(\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}+\omega _๐ญ\right)\left(\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}+\omega _๐ช\right)}\left(4\mathrm{\Delta }(๐ค,๐ญ)\right)\hfill \\ & \frac{\left(1\mathrm{\Delta }(๐ค^{}๐ช,๐ญ)\right)\left(1\mathrm{\Delta }(๐ค^{}๐ญ,๐ช)\right)}{\left(\epsilon _{๐ค^{}๐ช๐ญ}\epsilon _{๐ค^{}๐ช}+\omega _๐ญ\right)\left(\epsilon _{๐ค^{}๐ญ๐ช}\epsilon _{๐ค^{}๐ญ}+\omega _๐ช\right)}(4\mathrm{\Delta }(๐ค,๐ญ))\left)\right(1\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)),\hfill \end{array}$$
(32)
$$\phi ^{}(๐ค^{},๐ค,๐ญ,๐ช)=\frac{C_{๐ค,๐ญ,๐ช,๐ค^{}}}{\epsilon _{๐ค+๐ช}\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}\omega _๐ช}\left(1\widehat{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)\right),$$
(33)
where
$$\widehat{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)=\{\begin{array}{cc}1\text{,}\hfill & \text{if}\left|\epsilon _{๐ค+๐ช}\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}\omega _๐ช\right|<\widehat{\mathrm{\Gamma }}_๐ช\hfill \\ 0\text{,}\hfill & \text{if}\left|\epsilon _{๐ค+๐ช}\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}\omega _๐ช\right|\widehat{\mathrm{\Gamma }}_๐ช\hfill \end{array},$$
(34)
and substitution of $`\varphi `$ and $`C_{๐ค,๐ญ,๐ช,๐ค^{}}`$ from Eqs. (61), (LABEL:C) yields
$$\begin{array}{cc}& \phi ^{}(๐ค^{},๐ค,๐ญ,๐ช)=\frac{iD_๐ญ^2D_๐ช}{6\left(\epsilon _{๐ค+๐ช}\epsilon _{๐ค๐ญ}+\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}\omega _๐ช\right)}(\frac{\left(1\mathrm{\Delta }(๐ค^{},๐ญ)\right)\left(1\mathrm{\Delta }(๐ค+๐ช๐ญ,๐ช)\right)}{\left(\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}+\omega _๐ญ\right)\left(\epsilon _{๐ค๐ญ}\epsilon _{๐ค+๐ช๐ญ}+\omega _๐ช\right)}(2+\mathrm{\Delta }(๐ค+๐ช,๐ญ))\hfill \\ & \frac{\left(1\mathrm{\Delta }(๐ค^{},๐ญ)\right)\left(1\mathrm{\Delta }(๐ค+๐ช,๐ช)\right)}{\left(\epsilon _{๐ค^{}๐ญ}\epsilon _๐ค^{}+\omega _๐ญ\right)\left(\epsilon _๐ค\epsilon _{๐ค+๐ช}+\omega _๐ช\right)}\left(2+\mathrm{\Delta }(๐ค,๐ญ)\right)+\frac{\left(1\mathrm{\Delta }(๐ค+๐ช,๐ญ)\right)\left(1\mathrm{\Delta }(๐ค+๐ช๐ญ,๐ช)\right)}{\left(\epsilon _{๐ค+๐ช๐ญ}\epsilon _{๐ค+๐ช}+\omega _๐ญ\right)\left(\epsilon _{๐ค๐ญ}\epsilon _{๐ค+๐ช๐ญ}+\omega _๐ช\right)}\left(4\mathrm{\Delta }(๐ค^{},๐ญ)\right)\hfill \\ & \frac{\left(1\mathrm{\Delta }(๐ค,๐ญ)\right)\left(1\mathrm{\Delta }(๐ค+๐ช,๐ช)\right)}{\left(\epsilon _{๐ค๐ญ}\epsilon _๐ค+\omega _๐ญ\right)\left(\epsilon _๐ค\epsilon _{๐ค+๐ช}+\omega _๐ช\right)}(4\mathrm{\Delta }(๐ค^{},๐ญ))\left)\right(1\widehat{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)),\hfill \end{array}$$
(35)
Having established $`\psi `$, $`\chi `$, $`\phi `$, let us average $`\widehat{H}`$ over the phonon vacuum:
$$\begin{array}{cc}\hfill \widehat{H}_{\mathrm{av}}& =\frac{1}{2}\underset{๐ค}{}\epsilon _๐คn_๐ค+\underset{\mathrm{๐ช๐ค๐ค}^{}}{}F_{๐ค,๐ค^{},๐ช}a_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\underset{\mathrm{๐ช๐ค}}{}E_{๐ค,๐ช}a_๐ค^{}a_{๐ค๐ช}\left(\eta _๐ช^{}+\xi _๐ช^{}+\zeta _๐ช^{}\right)\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ค}^{}}{}\left(A_{๐ค^{},๐ช}\varphi ^{}(๐ค,๐ช)\frac{1}{2}G_{๐ค,๐ค^{},๐ช}\right)\left\{\eta _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+\xi _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+\zeta _๐ชn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค\right\}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}\left(B_{๐ค^{},๐ญ,๐ช,๐ค}\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}\right)\left\{\eta _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+\xi _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+\zeta _๐ชa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค\right\}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ค}^{}๐ญ}{}\left(C_{๐ค,๐ญ,๐ช,๐ค^{}}\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}\right)\left\{\eta _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+\xi _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+\zeta _๐ชa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}\right\}+c.c.,\hfill \end{array}$$
(36)
$`\widehat{H}_{\mathrm{av}}`$ contains the free-electron term, the 2-electron Frรถhlich interaction and terms representing 3-electron and 4-electron interactions. The 3-electron terms were generated by the second transformation of $`H_c`$. The source of these terms is thus the non-transformed part of the original interaction, discarded by BCS. If higher-order corrections to BCS theory are of interest, all terms arising from that part, i.e., the 3-electron ones, can be therefore neglected. The same conclusion was drawn above on the grounds of unstability.
## V 4-fermion interactions
The Hamiltonian $`\widehat{H}_{\mathrm{av}}`$ contains several terms representing 4-fermion interactions. Using Eqs. (25), (28), (30), (33), one finds
$$A_{๐ค^{},๐ช}\varphi ^{}(๐ค,๐ช)\frac{1}{2}G_{๐ค,๐ค^{},๐ช}=\frac{1}{2}G_{๐ค,๐ค^{},๐ช}=\frac{1}{2}\varphi ^{}(๐ค,๐ช)A_{๐ค^{},๐ช}.$$
Additionally, taking into account Eqs. (26), (27), we get
$$B_{๐ค^{},๐ญ,๐ช,๐ค}\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}=\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}=\frac{1}{2}B_{๐ค^{},๐ญ,๐ช,๐ญ}\left(1\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)\right),$$
$$C_{๐ค,๐ญ,๐ช,๐ค^{}}\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}=\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}=\frac{1}{2}C_{๐ค,๐ญ,๐ช,๐ค^{}}\left(1\widehat{\mathrm{\Delta }}(๐ค,๐ค^{},๐ญ,๐ช)\right).$$
In terms of $`G_{๐ค,๐ค^{},๐ช}`$$`H_{๐ค,๐ค^{},๐ญ,๐ช}`$$`I_{๐ค,๐ค^{},๐ญ,๐ช}`$ the 4-fermion interactions present in $`\widehat{H}_{\mathrm{av}}`$ express as
$$H_4^1=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ}}{}\frac{1}{2}G_{๐ค,๐ค^{},๐ช}\psi (๐ฅ,๐ฆ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ช}n_๐ฅn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+c.c.,$$
(37)
$$H_4^2=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ}}{}\frac{1}{2}G_{๐ค,๐ค^{},๐ช}\chi (๐ฅ,๐ฆ,๐ญ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ญ}a_{๐ฅ๐ญ}^{}a_{๐ฅ๐ช}n_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+c.c.,$$
(38)
$$H_4^3=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ}}{}\frac{1}{2}G_{๐ค,๐ค^{},๐ช}\phi (๐ฅ,๐ฆ,๐ญ,๐ช)a_{๐ฆ+๐ช}^{}a_{๐ฆ๐ญ}a_{๐ฅ๐ญ}^{}a_๐ฅn_๐ค^{}a_{๐ค๐ช}^{}a_๐ค+c.c.,$$
(39)
$$H_4^4=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ}}{}\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}\psi (๐ฅ,๐ฆ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ช}n_๐ฅa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+c.c.,$$
(40)
$$H_4^5=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ๐ฐ}}{}\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}\chi (๐ฅ,๐ฆ,๐ฐ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ฐ}a_{๐ฅ๐ฐ}^{}a_{๐ฅ๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+c.c.,$$
(41)
$$H_4^6=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ๐ฐ}}{}\frac{1}{2}H_{๐ค,๐ค^{},๐ญ,๐ช}\phi (๐ฅ,๐ฆ,๐ฐ,๐ช)a_{๐ฆ+๐ช}^{}a_{๐ฆ๐ฐ}a_{๐ฅ๐ฐ}^{}a_๐ฅa_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_๐ค+c.c.,$$
(42)
$$H_4^7=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ}}{}\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}\psi (๐ฅ,๐ฆ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ช}n_๐ฅa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+c.c.,$$
(43)
$$H_4^8=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ๐ฐ}}{}\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}\chi (๐ฅ,๐ฆ,๐ฐ,๐ช)a_๐ฆ^{}a_{๐ฆ๐ฐ}a_{๐ฅ๐ฐ}^{}a_{๐ฅ๐ช}a_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+c.c.,$$
(44)
$$H_4^9=\underset{\mathrm{๐ช๐ค๐ค}^{}\mathrm{๐ฅ๐ฆ๐ญ๐ฐ}}{}\frac{1}{2}I_{๐ค,๐ค^{},๐ญ,๐ช}\phi (๐ฅ,๐ฆ,๐ฐ,๐ช)a_{๐ฆ+๐ช}^{}a_{๐ฆ๐ฐ}a_{๐ฅ๐ฐ}^{}a_๐ฅa_๐ค^{}^{}a_{๐ค^{}๐ญ}a_{๐ค๐ญ}^{}a_{๐ค+๐ช}+c.c..$$
(45)
Five of these interactions, viz, $`H_4^1`$, $`H_4^2`$, $`H_4^3`$, $`H_4^4`$, $`H_4^7`$, contain the operator $`n_๐ค`$, therefore they are not reducible to the 4-fermion MT potential (3). This potential is particularly interesting not only for its possible relevance to the physics of superconductors, but also because the thermodynamics of the Hamiltonian $`H_{BCS}+W+V_{MT}`$ is exactly solvable (Brankov et al.Brankov et al. (2001)).
Let us consider the 8-fold product of operators in $`H_4^5`$ with explicit spin indices:
$$a_{๐ฆ\sigma ^{\prime \prime }}^{}a_{๐ฆ๐ฐ\sigma ^{\prime \prime }}a_{๐ฅ๐ฐ\sigma ^{\prime \prime \prime }}^{}a_{๐ฅ๐ช\sigma ^{\prime \prime \prime }}a_{๐ค^{}๐ช\sigma }^{}a_{๐ค^{}๐ญ\sigma }a_{๐ค๐ญ\sigma ^{}}^{}a_{๐ค\sigma ^{}}.$$
With respect to 1-fermion momenta these are nine possible reductions to the form (3), and each of them allows two or four possibilities related to spin reduction. The values assumed by each momentum index are collected in Table 1. In four cases $`๐ช=0`$ and the coupling vanishes, since $`\chi (๐ฅ,๐ฆ,๐ฐ,0)=0`$, for all $`๐ฅ`$, $`๐ฆ`$, $`๐ฐ`$. As a consequence, $`H_4^5`$ assumes the reduced form
$$\begin{array}{cc}\hfill H_{4(red)}^5& =\underset{๐ค๐ฆ}{}\{2H_{๐ค,2๐ค+๐ฆ,๐ค+๐ฆ,2๐ค+2๐ฆ}\chi (2๐ฆ+๐ค,๐ฆ,๐ค+๐ฆ,2๐ค+2๐ฆ)\hfill \\ & H_{๐ค,๐ฆ,๐ค+๐ฆ,2๐ฆ}\left(\chi (2๐ฆ+๐ค,๐ฆ,๐ค+๐ฆ,2๐ฆ)+\chi (2๐ฆ๐ค,๐ฆ,๐ฆ๐ค,2๐ฆ)\right)\hfill \\ & \chi (๐ค,๐ฆ,๐ค+๐ฆ,2๐ค)(H_{๐ค,2๐ค+๐ฆ,๐ค+๐ฆ,2๐ค}+H_{๐ค,2๐ค๐ฆ,๐ค๐ฆ,2๐ค})\}a_๐ฆ^{}a_{๐ฆ+}^{}a_๐ฆ^{}a_{๐ฆ+}^{}a_{๐ค+}a_๐คa_{๐ค+}a_๐ค+c.c..\hfill \end{array}$$
(46)
Terms with $`๐ค=๐ฆ`$ have been excluded, similarly as in $`V_{BCS}`$. Their contribution is accounted for by a shift of 1-fermion energies. The coefficients on the rhs result after performing summation over spins.
This procedure has been also applied to $`H_4^6`$, $`H_4^8`$ $`H_4^9`$. The corresponding values of 1-fermion momenta are given in Tables 2, 3, 4. The additional remark $`๐ค๐ค2๐ฆ`$ (or similar), indicates that a translation of one momentum index is necessary after reduction.
After reduction, $`H_4^6`$, $`H_4^8`$, $`H_4^9`$ take the following forms:
$$\begin{array}{cc}& H_{4(red)}^6=\underset{๐ค๐ฆ}{}\{2H_{๐ค,2๐ค+๐ฆ,๐ค+๐ฆ,2๐ค+2๐ฆ}\phi (๐ค,2๐ค๐ฆ,๐ค๐ฆ,2๐ค+2๐ฆ)\hfill \\ & H_{๐ค,๐ฆ,๐ค+๐ฆ,2๐ฆ}\left(\phi (๐ค,๐ฆ,๐ค๐ฆ,2๐ฆ)+\phi (๐ค,๐ฆ,๐ฆ๐ค,2๐ฆ)\right)\hfill \\ & \phi (๐ค,๐ฆ2๐ค,๐ค+๐ฆ,2๐ค)(H_{๐ค,2๐ค+๐ฆ,๐ค+๐ฆ,2๐ค}+H_{๐ค,2๐ค๐ฆ,๐ค๐ฆ,2๐ค})\}a_๐ฆ^{}a_{๐ฆ+}^{}a_๐ฆ^{}a_{๐ฆ+}^{}a_{๐ค+}a_๐คa_{๐ค+}a_๐ค+c.c.,\hfill \end{array}$$
(47)
$$\begin{array}{cc}\hfill H_{4(red)}^8& =\underset{๐ค๐ฆ}{}\{2I_{๐ค2๐ฆ,๐ฆ,๐ค๐ฆ,2๐ค+2๐ฆ}\chi (๐ค+2๐ฆ,๐ฆ,๐ค+๐ฆ,2๐ค+2๐ฆ)\hfill \\ & I_{๐ค2๐ฆ,๐ฆ,๐ค๐ฆ,2๐ฆ}\left(\chi (๐ค+2๐ฆ,๐ฆ,๐ค+๐ฆ,2๐ฆ)+\chi (๐ค+2๐ฆ,๐ฆ,๐ฆ๐ค,2๐ฆ)\right)\hfill \\ & \chi (๐ค,๐ฆ,๐ค+๐ฆ,2๐ค)(I_{๐ค,๐ฆ,๐ค+๐ฆ,2๐ค}+I_{๐ค,๐ฆ,๐ค๐ฆ,2๐ค})\}a_๐ฆ^{}a_{๐ฆ+}^{}a_๐ฆ^{}a_{๐ฆ+}^{}a_{๐ค+}a_๐คa_{๐ค+}a_๐ค+c.c.,\hfill \end{array}$$
(48)
$$\begin{array}{cc}& H_{4(red)}^9=\underset{๐ค๐ฆ}{}\{2I_{๐ค2๐ฆ,๐ฆ,๐ค๐ฆ,2๐ค+2๐ฆ}\phi (๐ค,2๐ค๐ฆ,๐ค๐ฆ,2๐ค+2๐ฆ)\hfill \\ & I_{๐ค2๐ฆ,๐ฆ,๐ค๐ฆ,2๐ฆ}\left(\phi (๐ค,๐ฆ,๐ค๐ฆ,2๐ฆ)+\phi (๐ค,๐ฆ,๐ค๐ฆ,2๐ฆ)\right)\hfill \\ & \phi (๐ค,2๐ค+๐ฆ,๐ค+๐ฆ,2๐ค)(I_{๐ค,๐ฆ,๐ค+๐ฆ,2๐ค}+I_{๐ค,๐ฆ,๐ค๐ฆ,2๐ค})\}\times a_๐ฆ^{}a_{๐ฆ+}^{}a_๐ฆ^{}a_{๐ฆ+}^{}a_{๐ค+}a_๐คa_{๐ค+}a_๐ค+c.c..\hfill \end{array}$$
(49)
Collecting all terms in Eqs. (46)โ(49), one obtains a $`V_{MT}`$ interaction as in Eq. (3) with
$$g_{\mathrm{๐ฆ๐ค}}=4\omega _{2๐ค+2๐ฆ}\mathrm{\Lambda }_{\mathrm{๐ค๐ฆ}}\mathrm{\Lambda }_{\mathrm{๐ฆ๐ค}}^{}+2\mathrm{\Theta }_{\mathrm{๐ค๐ฆ}}+2\mathrm{\Theta }_{\mathrm{๐ฆ๐ค}}^{},$$
(50)
where
$$\mathrm{\Lambda }_{\mathrm{๐ค๐ฆ}}=\chi ^{}(2๐ค+๐ฆ,๐ค,๐ค+๐ฆ,2๐ค+2๐ฆ)+\phi ^{}(๐ฆ,๐ค2๐ฆ,๐ค๐ฆ,2๐ค+2๐ฆ),$$
(51)
$$\begin{array}{cc}& \mathrm{\Theta }_{\mathrm{๐ค๐ฆ}}=\omega _๐ฆ\left(\chi ^{}(๐ฆ,๐ค,๐ค+๐ฆ,2๐ฆ)+\phi ^{}(๐ฆ,๐ค2๐ฆ,๐ค๐ฆ,2๐ฆ)\right)\hfill \\ & \times \left(\chi (2๐ฆ+๐ค,๐ฆ,๐ค+๐ฆ,2๐ฆ)+\chi (2๐ฆ๐ค,๐ฆ,๐ฆ๐ค,2๐ฆ)+\phi (๐ค,๐ฆ,๐ค๐ฆ,2๐ฆ)+\phi (๐ค,๐ฆ,๐ค๐ฆ,2๐ฆ)\right).\hfill \end{array}$$
(52)
## VI 4-fermion interaction coupling
The most important question about the 4-fermion interaction is positive- or negative-valuedness of $`g_{\mathrm{๐ฆ๐ค}}`$. In particular, it would be of interest to determine the domain of $`g_{\mathrm{๐ฆ๐ค}}`$ in momentum space where it is attractive. Unfortunately, the form of $`g_{\mathrm{๐ฆ๐ค}}`$ given by Eq. (50) is extremely complicated, so this problem cannot be resolved in general.
First, let us specify the quantities occurring in Eq. (50). We assume that $`\epsilon _๐ค=ak^2`$, $`\omega _๐ค=bk`$, $`\stackrel{~}{\mathrm{\Gamma }}_๐ค=ck`$, $`\mathrm{\Gamma }_๐ค=dk`$ and $`\widehat{\mathrm{\Gamma }}_๐ค=ek`$, where $`a`$, $`b`$, $`c`$, $`d`$, $`e`$ are real, positive constants.
Following DavydovDavydov (1976), we apply some approximations in order to estimate $`g_{\mathrm{๐ฆ๐ค}}`$. We assume that the most significant contribution to the 4-fermion interaction (analogously as in the BCS theory) comes from 1-electron momenta $`๐ฆ`$, $`๐ค`$ which satisfy the condition $`mk`$.
The case of exact equality is, of course, excluded (in accord with the restriction on summation in Eqs. (46) โ (49)), so we use $`g_{\mathrm{๐ค๐ค}}`$ only as an abbreviation for $`g_{\mathrm{๐ฆ๐ค}}`$ under our approximation $`mk`$.
Thus, the interaction coupling $`g_{\mathrm{๐ค๐ค}}`$ is given by
$$g_{\mathrm{๐ค๐ค}}=4\omega _{4๐ค}\mathrm{\Lambda }_{\mathrm{๐ค๐ค}}\mathrm{\Lambda }_{\mathrm{๐ค๐ค}}^{}+2\mathrm{\Theta }_{\mathrm{๐ค๐ค}}+2\mathrm{\Theta }_{\mathrm{๐ค๐ค}}^{}.$$
(53)
We take into account only vectors with almost compatible directions, otherwise $`g_{\mathrm{๐ค๐ค}}`$ vanishes.
We can thus rewrite Eq. (53) in the form:
$$g_{\mathrm{๐ค๐ค}}=16\omega _๐ค|\mathrm{\Lambda }_{\mathrm{๐ค๐ค}}|^2+4\mathrm{}\mathrm{\Theta }_{\mathrm{๐ค๐ค}},$$
(54)
where the first term is always nonpositive and $`\mathrm{}`$ means real part. The second term requires detailed analysis. Detailed calculations are performed in Appendix B. We have found $`g_{\mathrm{๐ค๐ค}}`$ in all cases for different values of the constants $`a`$, $`b`$, $`c`$, $`d`$, $`e`$. Most interesting is the case $`b>c`$, $`b>e`$ and $`b>d`$, because the essential part of initial Hamiltonian is transformed under these conditions. Under a further approximation, we find the form of $`g_{\mathrm{๐ค๐ค}}`$ for $`kk_F`$, where $`k_F`$ is Fermi momentum. Additionally we put $`c=d=e`$ and $`\epsilon _{๐ค_F}\omega _{๐ค_F}`$ (this is at least true for metals). Under these assumptions
$$\mathrm{\Theta }_{๐ค_F๐ค_F}\frac{D_{๐ค_F}^6}{4\omega _{๐ค_F}^5},\mathrm{\Lambda }_{๐ค_F๐ค_F}\frac{iD_{๐ค_F}^3}{12\omega _{๐ค_F}^2\epsilon _{๐ค_F}},$$
which implies
$$g_{๐ค_F๐ค_F}\frac{D_{๐ค_F}^6}{\omega _{๐ค_F}^5}<0.$$
Now we can compare the magnitudes of the coupling constants of 4-fermion interaction and 2-fermion interaction (under the same assumptions). It is shown in Appendix A that $`G_{๐ค_F๐ค_F}=D_{๐ค_F}^2/\omega _{๐ค_F}`$. Thus the interaction coupling for 4- and 2-fermion (BCS) interactions fulfils the relation $`g_{๐ค_F๐ค_F}=G_{๐ค_F๐ค_F}^3/\omega _{๐ค_F}^2`$. As a consequence, for a strong pairing, there is a significant contribution from 4-fermion interactions.
Moreover, the 4-fermion interaction coupling $`g_{\mathrm{๐ค๐ค}}`$ is also negative in most cases without imposing the approximations $`c=d=e`$ and $`kk_F`$. This is shown in Appendix B.
## VII 4-fermion interactions in Frรถhlichโs expansion
As demonstrated in Section III, 4-fermion interaction terms appear in sixth order of the expansion of Frรถhlichโs original transformation. These terms are derived in Appendix C. The various resulting 4-fermion terms considerably outnumber those obtained by applying a second transformation in Section V. Again a reduction procedure to $`V_{MT}`$ is possible for all 4-fermion expressions (76), which do not contain the particle number operator and possess three phonon indices. However, the resulting couplings are also complicated functions and, therefore, will not be examined in detail.
The disadvantage of this method is appearance of 3-fermion interactions, which in the double transformation method gave a small contribution. Here additional arguments must be used to discard these terms.
## VIII Concluding remarks
We have extended Frรถhlichโs transformation of $`H_{\mathrm{e}\mathrm{ph}}`$ to higher order terms. This has been done by performing a second transform of the first terms discarded by Frรถhlich. The resulting interactions are of 3- and 4-fermion type. The 3-fermion terms can be expected to be inessential in superconductivity because of their instability under time inversion. The 4-fermion terms are, in general, reducible to $`V_{\mathrm{MT}}`$, a BCS-type 4-fermion interaction. The resulting 4-fermion coupling is extremely complicated, but under reasonable approximations it is negative-valued. Moreover, for 1-fermion momenta in the neighbourhood of Fermi momentum, it has the simple form $`D_{๐ค_F}^6/\omega _{๐ค_F}^5=G_{๐ค_F๐ค_F}^3/\omega _{๐ค_F}^2`$, where $`G_{๐ค_F๐ค_F}`$ is the 2-fermion coupling. This fact implies that 4-fermion interactions are significant for systems with strong pairing and allows to estimate (relative to 2-fermion coupling and phonon energy at Fermi momentum) their magnitude.
As in BCS theory, where the relation between the gap parameter $`\mathrm{\Delta }`$ and coupling constant $`G`$ of the BCS interaction allows to estimate the magnitude of $`G`$, it can be expected that detailed thermodynamics of $`H_{MT}`$ will provide a relation between $`|g_{\mathrm{๐ฆ๐ค}}|`$ and other parameters of the theory, thereby allowing to estimate the magnitude of $`g_{\mathrm{๐ฆ๐ค}}`$. Another emerging question is Cooperโs problemCooper (1956) for a bound quadruple in the presence of $`V_{\mathrm{MT}}`$ or $`V_{BCS}+V_{MT}`$. Kamei and Miyake deal with this question in a recent workKamei and Miyake (2005).
The double transformation has unveiled the structure of 3-, 4-, 5-fermion interactions. Since they are proportional to 4th, 6th, 8th power of $`D_๐ฐ`$, they are relatively weak, so inclusion of these terms, apart from 4-fermion ones, appears unjustified at present, although Schneider et al.Schneider et al. (2004) suggest existence of both quadruples and sextets in some HTSC.
The higher-order expansion terms of the transformed $`H_{\mathrm{e}\mathrm{ph}}`$, including quadruple, sextet etc. interactions can be expected to reveal themselves in materials with extremely strong pairing correlations between spin $`1/2`$ fermions, since the presence of strongly correlated pairs implies at least some kind of weak interaction between them.
On the other hand, our extension of Frรถhlichโs transformation shows that if phonon mediation exists in a superconductor, the 4-fermion and 6-fermion interactions are always present as supplementary to the BCS one. Unfortunately, the Hamiltonian $`H_{\mathrm{BCS}}+V_{\mathrm{MT}}`$, although exactly solvable, leads to intricate mean-field equations.
Our results obtained can be generalized in many respects. First of all, the process of averaging over phonon vacuum could be replaced by averaging over the phonon equilibrium state, which could be justified at higher temperatures. This would already lead to an additional 1-fermion term in Frรถhlichโs Hamiltonian and modification of 2-fermion and 3-fermion terms in our method. Another extension would result by going beyond the harmonic approximation and including all products of phonon operators. Method of BogoliubovBogoliubov (1958) could be applied to โdangerous termsโ (divergent), omitted in Frรถhlichโs method. These questions will be dealt with in further investigations.
## Appendix A The Frรถhlichโs transformation
Let us recall Frรถhlichโs methodFrรถhlich (1952), using in some details the more elegant approach due to DavydovDavydov (1976).
Consider the electron-phonon Hamiltonian:
$$H_{\mathrm{e}\mathrm{ph}}=H_0+H_{\mathrm{int}}=\underset{๐ค\sigma }{}\epsilon _๐คa_{๐ค\sigma }^{}a_{๐ค\sigma }+\underset{๐ฐ}{}\omega _๐ฐb_๐ฐ^{}b_๐ฐ+i\underset{๐ฐ}{}D_๐ฐ\left(b_๐ฐ\rho _๐ฐ^{}b_๐ฐ^{}\rho _๐ฐ\right),$$
where
$$\rho _๐ฐ=\underset{๐ค\sigma }{}a_{๐ค๐ฐ\sigma }^{}a_{๐ค\sigma },$$
and $`a_{๐ค\sigma }`$ ($`b_๐ค`$) are fermion (boson) operators. The coupling $`D_๐ฐ`$ will be assumed small and $`\mathrm{}1`$.
The form of the interaction term of $`H_{\mathrm{e}\mathrm{ph}}`$ arises under a number of assumptions: the ions of the lattice move collectively, the coupling depends only on $`๐ฐ`$ and electrons interact only with longitudinal phonons for which $`\omega _๐ฐ=ws`$, $`s`$ denoting the velocity of sound. Our interest is focused on the behaviour of electrons, therefore variations of the phonon spectrum will be accounted for only through $`s`$.
Frรถhlich performed a unitary transformation of $`H_{\mathrm{e}\mathrm{ph}}`$ in order to eliminate (as far as possible) the interaction term. The transformed Hamiltonian is
$$H=\mathrm{e}^S^{}H_{\mathrm{e}\mathrm{ph}}\mathrm{e}^S=H_{\mathrm{e}\mathrm{ph}}[S,H_{\mathrm{e}\mathrm{ph}}]+\frac{1}{2}[S,[S,H_{\mathrm{e}\mathrm{ph}}]]+\mathrm{},$$
(55)
where
$$S=\underset{๐ช}{}S_๐ช=\underset{๐ช}{}\left(\gamma _๐ช^{}b_๐ช^{}\gamma _๐ชb_๐ช\right)=S^{},\gamma _๐ช=\underset{๐ค}{}\varphi (๐ค,๐ช)a_๐ค^{}a_{๐ค๐ช},$$
and the unknown function $`\varphi (๐ค,๐ช):^3\times ^3^1`$ is adjusted to achieve the cancellation.
Collecting terms of the same order in the coupling $`D_๐ฐ`$, one obtains Frรถhlichโs expansion:
$$H=H_0\left([S,H_0]H_{\mathrm{int}}\right)+\left(\frac{1}{2}[S,[S,H_0]][S,H_{\mathrm{int}}]\right)+\mathrm{}.$$
(56)
Subsequently, a term which is a combination of products, each with $`f`$ fermion operators and $`b`$ boson operators will be written as $`(f,b)`$. Clearly, $`f`$ will always be even. For example, $`H_0`$ consists of terms $`(2,0)`$ and $`(0,2)`$.
The rhs of Eq. (56) expresses in terms of commutators $`[(f_1,b_1),(f_2,b_2)]`$. One easily finds that
$$[(f_1,b_1),(f_2,b_2)]=[f_1,f_2]b_1b_2+f_2f_1[b_1,b_2]=[f_1,f_2]b_2b_1+f_1f_2[b_1,b_2].$$
(57)
The necessary commutators $`[f_1,f_2]`$, $`[b_1,b_2]`$ are given in Appendix C.
According to Eq. (55), the transformation can be performed, given commutators of the form occurring in Eq. (57) with the first argument equal $`S`$. The latter is a $`(2,1)`$ expression, hence
$$[S,(f,b)]=[(2,1),(f,b)]=(f,b+1)+(f+2,b1),$$
by virtue of Eqs. (57), (72). Clearly $`(f,b1)=0`$ for $`b=0`$.
One finds
$$[S,H_0]=[(2,1),(2,0)]+[(2,1),(0,2)]=(2,1),$$
so the term arising from $`H_0`$ in first order, has the same form as the one arising from $`H_{\mathrm{int}}`$ in zeroth order. These terms are used to eliminate the interaction in $`H_{\mathrm{e}\mathrm{ph}}`$.
Similarly, for $`H_{\mathrm{int}}`$
$$[S,H_{\mathrm{int}}]=[(2,1),(2,1)]=(2,2)+(4,0).$$
Frรถhlich additionally assumed that introduction of experimentally measured velocity of sound $`s`$ allows to discard all terms containing $`b_๐ฐ^2`$, $`(b_๐ฐ^{})^2`$, $`b_๐ฐb_๐ฏ`$, $`b_๐ฐ^{}b_๐ฏ`$ if $`๐ฐ๐ฏ`$ etc. In other words, all $`(f,b)`$ terms for $`b3`$ and part of $`(f,2)`$ terms are neglected. As a consequence, the expansion (55) up to 2nd order reduces toFrรถhlich (1952):
$$[S,H_0]=\underset{\mathrm{๐ค๐ช}}{}b_๐ช\left(\epsilon _๐ค\epsilon _{๐ค๐ช}\omega _๐ช\right)\varphi (๐ค,๐ช)a_๐ค^{}a_{๐ค๐ช}+c.c.,$$
(58)
$$[S,H_{\mathrm{int}}]=\underset{๐ช}{}iD_๐ช\gamma _๐ช\rho _๐ช+\underset{\mathrm{๐ค๐ช}}{}iD_๐ชb_๐ช^{}b_๐ช\varphi (๐ค,๐ช)\left(n_๐คn_{๐ค๐ช}\right)+c.c.,$$
(59)
$$[S,[S,H_0]]=\underset{\mathrm{๐ค๐ช}}{}\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)b_๐ช^{}b_๐ช|\varphi (๐ค,๐ช)|^2\left(n_๐คn_{๐ค๐ช}\right)+\underset{\mathrm{๐ค๐ช}}{}\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)\varphi (๐ค,๐ช)a_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}+c.c..$$
(60)
$`\varphi `$ is now adjusted so as to minimize the contribution of terms $`(2,1)`$. This is achieved by the choiceFrรถhlich (1952)
$$\varphi (๐ค,๐ช)=\frac{iD_๐ช}{\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช}\left(1\mathrm{\Delta }(๐ค,๐ช)\right),$$
(61)
where
$$\mathrm{\Delta }(๐ค,๐ช)=\{\begin{array}{cc}1\text{,}\hfill & \text{if}\left|\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right|<\mathrm{\Gamma }_๐ช\hfill \\ 0\text{,}\hfill & \text{if}\left|\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right|\mathrm{\Gamma }_๐ช\hfill \end{array}.$$
(62)
$`\mathrm{\Delta }(๐ค,๐ช)`$ is introduced to avoid divergence of the series (55) and $`\mathrm{\Gamma }_๐ช`$ is positive energy choosing for convergence.
Eqs. (58)โ(61) yield
$$\begin{array}{cc}\hfill H& =\underset{๐ค}{}\epsilon _๐คa_๐ค^{}a_๐ค+\underset{๐ฐ}{}\omega _๐ฐb_๐ฐ^{}b_๐ฐ+\underset{\mathrm{๐ฐ๐ค}}{}b_๐ฐ^{}b_๐ฐ(n_๐คn_{๐ค๐ฐ})(\frac{1}{2}(\epsilon _{๐ค๐ฐ}\epsilon _๐ค+\omega _๐ฐ)|\varphi (๐ค,๐ฐ)|^2iD_๐ฐ\varphi (๐ค,๐ฐ)+c.c.)\hfill \\ & +i\underset{\mathrm{๐ค๐ฐ}}{}D_๐ฐ(b_๐ฐa_๐ค^{}a_{๐ค๐ฐ}b_๐ฐ^{}a_{๐ค๐ฐ}^{}a_๐ค)\mathrm{\Delta }(๐ค,๐ฐ)\frac{1}{2}\underset{\mathrm{๐ค๐ช๐ฐ}}{}\frac{D_๐ฐ^2\left(1+\mathrm{\Delta }(๐ค,๐ฐ)\right)\left(1\mathrm{\Delta }(๐ช,๐ฐ)\right)}{\epsilon _{๐ช๐ฐ}\epsilon _๐ช+\omega _๐ฐ}(a_๐ค^{}a_{๐ค๐ฐ}a_{๐ช๐ฐ}^{}a_๐ช+c.c.).\hfill \end{array}$$
(63)
Discarding the non-transformed part of the initial interaction (4th sum on the rhs) and taking the average in phonon vacuum, one obtains the Frรถhlichโs Hamiltonian:
$$H_\mathrm{F}=\underset{๐ค}{}\epsilon _๐คn_๐ค\frac{1}{2}\underset{\mathrm{๐ค๐ช๐ฐ}}{}\frac{D_๐ฐ^2\left(1+\mathrm{\Delta }(๐ค,๐ฐ)\right)\left(1\mathrm{\Delta }(๐ช,๐ฐ)\right)}{\epsilon _{๐ช๐ฐ}\epsilon _๐ช+\omega _๐ฐ}(a_๐ค^{}a_{๐ค๐ฐ}a_{๐ช๐ฐ}^{}a_๐ช+c.c.).$$
(64)
The second term represents an effective interaction between electrons dressed in the phonon field. If $`\epsilon _{๐ช๐ฐ}\epsilon _๐ช+\omega _๐ฐ>0`$, this interaction is attractive.
## Appendix B Calculation of $`g_{\mathrm{๐ค๐ค}}`$
Our objective is to find the explicit form of the following expression:
$$g_{\mathrm{๐ค๐ค}}=16\omega _๐ค|\mathrm{\Lambda }_{\mathrm{๐ค๐ค}}|^2+4\mathrm{}\mathrm{\Theta }_{\mathrm{๐ค๐ค}}.$$
(65)
Taking into account Eq. (52), we have
$$\begin{array}{cc}\hfill \mathrm{\Theta }_{\mathrm{๐ค๐ค}}& =\omega _๐ค\left(\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)+\phi ^{}(๐ค,๐ค,0,2๐ค)\right)\hfill \\ & \times \left(\chi (3๐ค,๐ค,2๐ค,2๐ค)+\chi (๐ค,๐ค,0,2๐ค)+\phi (๐ค,๐ค,0,2๐ค)+\phi (๐ค,๐ค,2๐ค,2๐ค)\right)\hfill \\ & =\omega _๐ค\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)\left(\chi (3๐ค,๐ค,2๐ค,2๐ค)+\phi (๐ค,๐ค,2๐ค,2๐ค)\right),\hfill \end{array}$$
(66)
because $`\phi ^{}(๐ค,๐ค,0,2๐ค)=\chi (๐ค,๐ค,0,2๐ค)=\phi (๐ค,๐ค,0,2๐ค)=0`$, as can be seen from the explicit form e.g., of the first of these functions (see Eq. (33)):
$$\begin{array}{cc}\hfill \phi ^{}(๐ค,๐ค,0,2๐ค)& =\frac{iD_0^2D_{2๐ค}}{6\omega _๐ค^2\omega _0}\left(1\widehat{\mathrm{\Delta }}(๐ค,๐ค,0,2๐ค)\right)\left(1\mathrm{\Delta }(๐ค,2๐ค)\right)\hfill \\ & \times \{(1\mathrm{\Delta }(๐ค,0))(4+\mathrm{\Delta }(๐ค,0))+(1\mathrm{\Delta }(๐ค,0))(4\mathrm{\Delta }(๐ค,0))\}=0,\hfill \end{array}$$
because $`\mathrm{\Delta }(๐ค,0)=\mathrm{\Delta }(๐ค,0)=1`$, for all $`๐ค`$ (see Eq. (62)). Similarly for $`\chi (๐ค,๐ค,0,2๐ค)`$ and $`\phi (๐ค,๐ค,0,2๐ค)`$.
Similarly, from Eq. (51), we get
$$\mathrm{\Lambda }_{\mathrm{๐ค๐ค}}=\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)+\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค).$$
(67)
On the grounds of Eq. (30), we obtain
$$\begin{array}{cc}& \chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{12\omega _k}(1\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค,2๐ค,2๐ค))\{\frac{3\left(1\mathrm{\Delta }(๐ค,2๐ค)\right)}{\omega _๐ค^2}\hfill \\ & \frac{\left(1\mathrm{\Delta }(๐ค,2๐ค)\right)\left(1\mathrm{\Delta }(๐ค,2๐ค)\right)}{2\omega _๐ค\left(4\epsilon _๐ค+\omega _๐ค\right)}(2+\mathrm{\Delta }(๐ค,2๐ค))\frac{\left(1\mathrm{\Delta }(๐ค,2๐ค)\right)}{2\left(4\epsilon _๐ค+\omega _๐ค\right)^2}(4\mathrm{\Delta }(๐ค,2๐ค))\}.\hfill \end{array}$$
(68)
The functions $`\stackrel{~}{\mathrm{\Delta }}`$ and $`\mathrm{\Delta }`$ are equal $`0`$ or $`1`$, depending on their argument (see Eqs. (31), (62)), so we obtain several conditions. The most important one is
$$\stackrel{~}{\mathrm{\Delta }}(๐ค,๐ค,2๐ค,2๐ค)=0\omega _๐ค>\stackrel{~}{\mathrm{\Gamma }}_๐คb>c,$$
(69)
and $`\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=0`$ otherwise. Other conditions are not so strong, because they do not destroy all of parts of Eq. (68). We have
$$\mathrm{\Delta }(๐ค,2๐ค)=0\omega _๐ค>\mathrm{\Gamma }_๐คb>d,$$
(70)
$$\mathrm{\Delta }(๐ค,2๐ค)=0|4\epsilon _๐ค+\omega _๐ค|>\mathrm{\Gamma }_๐คk>\frac{db}{4a}.$$
(71)
Taking into account Eqs. (69), (70) and (71) we get
* If $`bc`$, then $`\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>c`$ and $`b>d`$, then
$$\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3\epsilon _๐ค\left(5\omega _๐ค+12\epsilon _๐ค\right)}{3\omega _๐ค^3\left(\omega _๐ค+4\epsilon _๐ค\right)^2}.$$
* If $`b>c`$, $`bd`$ and $`k(0,\frac{db}{4a}]`$, then $`\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>c`$, $`bd`$ and $`k(\frac{db}{4a},\mathrm{})`$, then
$$\chi ^{}(๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{8\omega _๐ค\left(\omega _๐ค+4\epsilon _๐ค\right)^2}.$$
Proceeding similarly, we get the form of other functions occurring in Eqs. (66) and (67). For $`\chi (3๐ค,๐ค,2๐ค,2๐ค)`$, we have:
* If $`bc`$, then $`\chi (3๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>c`$, $`b>d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\mathrm{})`$, then
$$\chi (3๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3\epsilon _๐ค\left(5\omega _๐ค12\epsilon _๐ค\right)}{3\omega _๐ค^3\left(\omega _๐ค4\epsilon _๐ค\right)^2}.$$
* If $`b>c`$, $`b>d`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then
$$\chi (3๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{4\omega _๐ค^3}.$$
* If $`b>c`$, $`bd`$ and $`k[0,\frac{b+d}{4a}]`$, then $`\chi (3๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>c`$, $`bd`$ and $`k(\frac{b+d}{4a},\mathrm{})`$, then
$$\chi (3๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{8\omega _๐ค\left(\omega _๐ค4\epsilon _๐ค\right)^2}.$$
For $`\phi (๐ค,๐ค,2๐ค,2๐ค)`$:
* If $`be`$, then $`\phi (๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>e`$, $`b>d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\mathrm{})`$, then
$$\phi (๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3\epsilon _๐ค\left(\omega _๐ค+12\epsilon _๐ค\right)}{3\omega _๐ค^3\left(\omega _๐ค^216\epsilon _๐ค^2\right)}.$$
* If $`b>e`$, $`b>d`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then
$$\phi (๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{4\omega _๐ค^3}.$$
* If $`b>e`$, $`bd`$ and $`k[0,\frac{b+d}{4a}]`$, then $`\phi (๐ค,๐ค,2๐ค,2๐ค)=0`$.
* If $`b>e`$, $`bd`$ and $`k[\frac{b+d}{4a},\mathrm{}]`$, then
$$\phi (๐ค,๐ค,2๐ค,2๐ค)=\frac{i\sqrt{2}D_๐ค^3}{8\omega _๐ค\left(\omega _๐ค^216\epsilon _๐ค^2\right)}.$$
For $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)`$:
* If $`bc`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=0`$.
* If $`b>c`$, $`b<d/3`$ and $`k(0,\frac{db}{2a}]`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=0`$.
* If $`b>c`$, $`b<d/3`$ and $`k(\frac{db}{2a},\frac{d+b}{2a})`$, then
$$\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3}{16\omega _๐ค\left(\omega _๐ค+2\epsilon _๐ค\right)\left(\omega _๐ค+4\epsilon _๐ค\right)}:=\chi ^{(1)}.$$
* If $`b>c`$, $`b<d/3`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then
$$\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\frac{3iD_๐ค^3\epsilon _๐ค}{4\left(\omega _๐ค^216\epsilon _๐ค^2\right)\left(\omega _๐ค^24\epsilon _๐ค^2\right)}:=\chi ^{(2)}.$$
* If $`b>c`$, $`d/3bd`$ and $`k(0,\frac{db}{2a}]`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=0`$.
* If $`b>c`$, $`d/3bd`$ and $`k(\frac{db}{2a},\frac{d+b}{2a}]`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\chi ^{(1)}`$.
* If $`b>c`$, $`d/3bd`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\chi ^{(2)}`$.
* If $`b>c`$, $`d<b3d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{2a},\mathrm{})`$, then
$$\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\epsilon _๐ค\left(7\omega _๐ค^216\epsilon _๐ค^2\right)}{6\omega _๐ค^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)\left(\omega _๐ค^24\epsilon _๐ค^2\right)}:=\chi ^{(3)}.$$
* If $`b>c`$, $`d<b3d`$ and $`k[\frac{bd}{4a},\frac{bd}{2a})`$, then
$$\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\left(3\omega _๐ค^2+40\epsilon _๐ค^2+22\omega _๐ค\epsilon _๐ค\right)}{48\omega _๐ค^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)\left(\omega _๐ค+4\epsilon _๐ค\right)}:=\chi ^{(4)}.$$
* If $`b>c`$, $`d<b3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then
$$\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\left(3\omega _๐ค+4\epsilon _๐ค\right)}{24\omega _๐ค^2\left(\omega _๐ค+2\epsilon _๐ค\right)\left(\omega _๐ค+4\epsilon _๐ค\right)}:=\chi ^{(5)}.$$
* If $`b>c`$, $`b>3d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\frac{bd}{2a})(\frac{b+d}{2a},\mathrm{})`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\chi ^{(3)}`$.
* If $`b>c`$, $`b>3d`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\chi ^{(4)}`$.
* If $`b>c`$, $`b>3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then $`\chi ^{}(3๐ค,๐ค,2๐ค,4๐ค)=\chi ^{(5)}`$.
For $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)`$:
* If $`be`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=0`$.
* If $`b>e`$, $`b<d/3`$ and $`k(0,\frac{db}{2a}]`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=0`$.
* If $`b>e`$, $`b<d/3`$ and $`k(\frac{db}{2a},\frac{d+b}{2a}]`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3}{16\omega _๐ค\left(\omega _๐ค+2\epsilon _๐ค\right)\left(\omega _๐ค4\epsilon _๐ค\right)}:=\phi ^{(1)}.$$
* If $`b>e`$, $`b<d/3`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\epsilon _๐ค}{4\left(\omega _๐ค^216\epsilon _๐ค^2\right)\left(\omega _๐ค^24\epsilon _๐ค^2\right)}:=\phi ^{(2)}.$$
* If $`b>e`$, $`d/3bd`$ and $`k(0,\frac{d+b}{4a}]`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=0`$.
* If $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{4a},\frac{d+b}{2a}]`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\phi ^{(1)}`$.
* If $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\phi ^{(2)}`$.
* If $`b>e`$, $`d<b3d`$ and $`k(0,\frac{bd}{4a})[\frac{b+d}{2a},\mathrm{})`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\epsilon _๐ค\left(\omega _๐ค^2+16\epsilon _๐ค^2\right)}{6\omega _๐ค^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)\left(\omega _๐ค^216\epsilon _๐ค^2\right)}:=\phi ^{(3)}.$$
* If $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{4a},\frac{bd}{2a})`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\left(3\omega _๐ค^2+14\epsilon _๐ค\omega _๐ค+40\epsilon _๐ค^2\right)}{48\omega _๐ค^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)\left(\omega _๐ค+4\epsilon _๐ค\right)}:=\phi ^{(4)}.$$
* If $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{4a}]`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3}{16\omega _๐ค^2\left(\omega _๐ค+2\epsilon _๐ค\right)}.$$
* If $`b>e`$, $`d<b3d`$ and $`k(\frac{b+d}{4a},\frac{b+d}{2a})`$, then
$$\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\frac{iD_๐ค^3\left(3\omega _๐ค4\epsilon _๐ค\right)}{24\omega _๐ค^2\left(\omega _๐ค+2\epsilon _๐ค\right)\left(\omega _๐ค4\epsilon _๐ค\right)}:=\phi ^{(5)}.$$
* If $`b>e`$, $`3d<b`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\frac{bd}{2a})(\frac{b+d}{2a},\mathrm{})`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\phi ^{(3)}`$.
* If $`b>e`$, $`3d<b`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\phi ^{(4)}`$.
* If $`b>e`$, $`3d<b`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then $`\phi ^{}(๐ค,3๐ค,2๐ค,4๐ค)=\phi ^{(5)}`$.
On the basis of this results and Eq. (65), we get the 4-fermion interaction coupling $`g_{\mathrm{๐ค๐ค}}`$:
* If $`bc`$ and $`be`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`bc`$, $`b>e`$, $`b<d/3`$ and $`k(0,\frac{db}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`bc`$, $`b>e`$, $`b<d/3`$ and $`k(\frac{db}{2a},\frac{d+b}{2a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6}{16\omega _๐ค\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค4\epsilon _๐ค\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(1)}<0.$$
* If $`bc`$, $`b>e`$, $`b<d/3`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\omega _๐ค\epsilon _๐ค^2}{\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(2)}<0.$$
* If $`bc`$, $`b>e`$, $`d/3bd`$ and $`k(0,\frac{d+b}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`bc`$, $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{4a},\frac{d+b}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(1)}`$.
* If $`bc`$, $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(2)}`$.
* If $`bc`$, $`b>e`$, $`d<b3d`$ and $`k(0,\frac{bd}{4a})[\frac{b+d}{2a},\mathrm{})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{4D_๐ค^6\epsilon _๐ค^2\left(\omega _๐ค^2+16\epsilon _๐ค^2\right)^2}{9\omega _๐ค^3\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(3)}<0.$$
* If $`bc`$, $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{4a},\frac{bd}{2a})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(3\omega _๐ค^2+14\epsilon _๐ค\omega _๐ค+40\epsilon _๐ค^2\right)^2}{144\omega _๐ค^3\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(4)}<0.$$
* If $`bc`$, $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{4a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6}{16\omega _๐ค^3\left(\omega _๐ค+2\epsilon _๐ค\right)^2}<0.$$
* If $`bc`$, $`b>e`$, $`d<b3d`$ and $`k(\frac{b+d}{4a},\frac{b+d}{2a})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(3\omega _๐ค4\epsilon _๐ค\right)^2}{36\omega _๐ค^3\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค4\epsilon _๐ค\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(5)}<0.$$
* If $`bc`$, $`b>e`$, $`3d<b`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\frac{bd}{2a})(\frac{b+d}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(3)}`$.
* If $`bc`$, $`b>e`$, $`3d<b`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(4)}`$.
* If $`bc`$, $`b>e`$, $`3d<b`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(5)}`$.
* If $`b>c`$, $`be`$, $`b<d/3`$ and $`k(0,\frac{b+d}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`b>c`$, $`be`$, $`b<d/3`$ and $`k(\frac{b+d}{4a},\frac{db}{2a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6}{8\omega _๐ค\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}<0.$$
* If $`b>c`$, $`be`$, $`b<d/3`$ and $`k(\frac{db}{2a},\frac{b+d}{2a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{3D_๐ค^6\left(\omega _๐ค^2+8\epsilon _๐ค^2\right)}{16\omega _๐ค\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(6)}<0.$$
* If $`b>c`$, $`be`$, $`b<d/3`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(\omega _๐ค^4+64\omega _๐ค^2\epsilon _๐ค^2+16\epsilon _๐ค^4\right)}{8\omega _๐ค\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(7)}<0.$$
* If $`b>c`$, $`be`$, $`d/3bd`$ and $`k(0,\frac{db}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`b>c`$, $`be`$, $`d/3bd`$ and $`k(\frac{db}{2a},\frac{d+b}{4a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6}{16\omega _๐ค\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(8)}<0.$$
* If $`b>c`$, $`be`$, $`d/3bd`$ and $`k(\frac{d+b}{4a},\frac{d+b}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(6)}`$.
* If $`b>c`$, $`be`$, $`d/3bd`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(7)}`$.
* If $`b>c`$, $`be`$, $`d<b3d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{2a},\mathrm{})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{4D_๐ค^6\epsilon _๐ค^2\left(\omega _๐ค^6+464\omega _๐ค^4\epsilon _๐ค^22848\omega _๐ค^2\epsilon _๐ค^4+4608\epsilon _๐ค^6\right)}{9\omega _๐ค^5\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(9)}.$$
We see that $`g_{\mathrm{๐ค๐ค}}^{(9)}<0`$ iff
$$\omega _๐ค^6+464\omega _๐ค^4\epsilon _๐ค^22848\omega _๐ค^2\epsilon _๐ค^4+4608\epsilon _๐ค^6>0.$$
Taking into account $`\omega _๐ค=bk`$ and $`\epsilon _๐ค=ak^2`$, we obtain a 3rd order algebraic inequality
$$4608l^32848xl^2+464x^2lx^3>0,$$
for $`l=a^2k^2`$, $`x=b^2`$. This can be easily solved<sup>1</sup><sup>1</sup>1this and following inequalities are solved by *Mathematica* packageโthere exist only one real root $`l_0^{(9)}0,002x`$, so
$$g_{\mathrm{๐ค๐ค}}^{(9)}<0\text{iff}k>k_0^{(9)}0.047b/a<b/4a.$$
.
* If $`b>c`$, $`be`$, $`d<b3d`$ and $`k[\frac{bd}{4a},\frac{bd}{2a})`$, then
$$\begin{array}{cc}\hfill g_{\mathrm{๐ค๐ค}}& =\frac{D_๐ค^6}{144\omega _๐ค^5\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)^2}\hfill \\ & \times (9\omega _๐ค^6+348\omega _๐ค^5\epsilon _๐ค+1396\omega _๐ค^4\epsilon _๐ค^22080\omega _๐ค^3\epsilon _๐ค^37616\omega _๐ค^2\epsilon _๐ค^4+7680\omega _๐ค\epsilon _๐ค^5+18432\epsilon _๐ค^6):=g_{\mathrm{๐ค๐ค}}^{(10)}.\hfill \end{array}$$
It turns out that $`g_{\mathrm{๐ค๐ค}}^{(10)}<0`$ for all $`k>0`$, because ($`l=ak`$) the equation
$$18432l^6+7680bl^57616b^2l^42080b^3l^3+1396b^4l^2+348b^5l+9b^6=0$$
has only two real roots and both are negative: $`l_1^{(10)}0.208b`$ and $`l_1^{(10)}=0.029b`$.
* If $`b>c`$, $`be`$, $`d<b3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{4a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(9\omega _๐ค^4+144\omega _๐ค^3\epsilon _๐ค+784\omega _๐ค^2\epsilon _๐ค^2+1632\omega _๐ค\epsilon _๐ค^3+1152\epsilon _๐ค^4\right)}{36\omega _๐ค^5\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)^2}<0.$$
* If $`b>c`$, $`be`$, $`d<b3d`$ and $`k(\frac{b+d}{4a},\frac{b+d}{2a}]`$, then
$$\begin{array}{cc}\hfill g_{\mathrm{๐ค๐ค}}& =\frac{D_๐ค^6}{36\omega _๐ค^5\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2}\hfill \\ & \times (9\omega _๐ค^648\omega _๐ค^5\epsilon _๐ค832\omega _๐ค^4\epsilon _๐ค^22944\omega _๐ค^3\epsilon _๐ค^3+1664\omega _๐ค^2\epsilon _๐ค^4+18432\omega _๐ค\epsilon _๐ค^5+18432\epsilon _๐ค^6):=g_{\mathrm{๐ค๐ค}}^{(11)}.\hfill \end{array}$$
$`g_{\mathrm{๐ค๐ค}}^{(11)}<0`$ iff $`k(0,0.073b/a)(0.412b/a,\mathrm{})`$.
* If $`b>c`$, $`be`$, $`b>3d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\frac{bd}{2a})(\frac{b+d}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(9)}`$.
* If $`b>c`$, $`be`$, $`b>3d`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(10)}`$.
* If $`b>c`$, $`be`$, $`b>3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(11)}`$.
* If $`b>c`$, $`b>e`$, $`b<d/3`$ and $`k(0,\frac{b+d}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`b>c`$, $`b>e`$, $`b<d/3`$ and $`k(\frac{b+d}{4a},\frac{db}{2a})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6}{4\left(\omega _๐ค+4\epsilon _๐ค\right)^3\left(\omega _๐ค4\epsilon _๐ค\right)^2}<0.$$
* If $`b>c`$, $`b>e`$, $`b<d/3`$ and $`k[\frac{db}{2a},\frac{b+d}{2a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(\omega _๐ค^2+4\omega _๐ค\epsilon _๐ค+2\epsilon _๐ค^2\right)}{2\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2\left(\omega _๐ค+2\epsilon _๐ค\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)}:=g_{\mathrm{๐ค๐ค}}^{(12)}<0.$$
* If $`b>c`$, $`b>e`$, $`b<d/3`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(\omega _๐ค^44\omega _๐ค^2\epsilon _๐ค^2+16\omega _๐ค\epsilon _๐ค^3+16\epsilon _๐ค^4\right)}{4\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)}:=g_{\mathrm{๐ค๐ค}}^{(13)}.$$
$`g_{\mathrm{๐ค๐ค}}^{(13)}<0`$ for all $`k>0`$.
* If $`b>c`$, $`b>e`$, $`d/3bd`$ and $`k(0,\frac{db}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=0`$.
* If $`b>c`$, $`b>e`$, $`d/3bd`$ and $`k(\frac{db}{2a},\frac{d+b}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(8)}`$.
* If $`b>c`$, $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{4a},\frac{d+b}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(12)}`$.
* If $`b>c`$, $`b>e`$, $`d/3bd`$ and $`k(\frac{d+b}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(13)}`$.
* If $`b>c`$, $`b>e`$, $`d<b3d`$ and $`k(0,\frac{bd}{4a})`$, then
$$\begin{array}{cc}\hfill g_{\mathrm{๐ค๐ค}}& =\frac{32D_๐ค^6\epsilon _๐ค^2}{9\omega _๐ค^5\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)}\hfill \\ & \times (3\omega _๐ค^720\omega _๐ค^6\epsilon _๐ค+84\omega _๐ค^5\epsilon _๐ค^2+400\omega _๐ค^4\epsilon _๐ค^3568\omega _๐ค^3\epsilon _๐ค^42400\omega _๐ค^2\epsilon _๐ค^5+1152\omega _๐ค\epsilon _๐ค^6+4608\epsilon _๐ค^7):=g_{\mathrm{๐ค๐ค}}^{(14)}.\hfill \end{array}$$
$`g_{\mathrm{๐ค๐ค}}^{(14)}<0`$ iff $`k(0.268b/a,\mathrm{})`$.
* If $`b>c`$, $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{4a},\frac{bd}{2a})`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{16D_๐ค^6\epsilon _๐ค\left(60\omega _๐ค^5+225\omega _๐ค^4\epsilon _๐ค120\omega _๐ค^3\epsilon _๐ค^2752\omega _๐ค^2\epsilon _๐ค^3+960\omega _๐ค\epsilon _๐ค^4+2304\epsilon _๐ค^5\right)}{9\omega _๐ค^5\left(\omega _๐ค^24\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)^2}:=g_{\mathrm{๐ค๐ค}}^{(15)}.$$
$`g_{\mathrm{๐ค๐ค}}^{(15)}<0`$ for all $`k>0`$.
* If $`b>c`$, $`b>e`$, $`d<b3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{4a}]`$, then
$$g_{\mathrm{๐ค๐ค}}=\frac{D_๐ค^6\left(81\omega _๐ค^4+1320\omega _๐ค^3\epsilon _๐ค+6544\omega _๐ค^2\epsilon _๐ค^2+13056\omega _๐ค\epsilon _๐ค^3+9216\epsilon _๐ค^4\right)}{9\omega _๐ค^5\left(\omega _๐ค+4\epsilon _๐ค\right)^2\left(\omega _๐ค+2\epsilon _๐ค\right)^2}<0.$$
* If $`b>c`$, $`b>e`$, $`d<b3d`$ and $`k(\frac{b+d}{4a},\frac{b+d}{2a}]`$, then
$$\begin{array}{cc}\hfill g_{\mathrm{๐ค๐ค}}& =\frac{D_๐ค^6}{9\omega _๐ค^5\left(\omega _๐ค^216\epsilon _๐ค^2\right)^2\left(\omega _๐ค+4\epsilon _๐ค\right)\left(\omega _๐ค+2\epsilon _๐ค\right)^2}\hfill \\ & \times (9\omega _๐ค^7+36\omega _๐ค^6\epsilon _๐ค336\omega _๐ค^5\epsilon _๐ค^22560\omega _๐ค^4\epsilon _๐ค^33264\omega _๐ค^3\epsilon _๐ค^414592\omega _๐ค^2\epsilon _๐ค^5+46080\omega _๐ค\epsilon _๐ค^6+36864\epsilon _๐ค^7):=g_{\mathrm{๐ค๐ค}}^{(16)}.\hfill \end{array}$$
$`g_{\mathrm{๐ค๐ค}}^{(16)}<0`$ iff $`k(0,0.142b/a)(0.345b/a,\mathrm{})`$.
* If $`b>c`$, $`b>e`$, $`d<b3d`$ and $`k(\frac{b+d}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(14)}`$.
* If $`b>c`$, $`b>e`$, $`b>3d`$ and $`k(0,\frac{bd}{4a})(\frac{b+d}{4a},\frac{bd}{2a})(\frac{b+d}{2a},\mathrm{})`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(14)}`$.
* If $`b>c`$, $`b>e`$, $`b>3d`$ and $`k[\frac{bd}{4a},\frac{b+d}{4a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(15)}`$.
* If $`b>c`$, $`b>e`$, $`b>3d`$ and $`k[\frac{bd}{2a},\frac{b+d}{2a}]`$, then $`g_{\mathrm{๐ค๐ค}}=g_{\mathrm{๐ค๐ค}}^{(16)}`$.
## Appendix C Higher order terms of Frรถhlichโs transformation
Evaluation of successive commutators of the expansion (55) is a time-consuming task. Below the final results are given for the relevant orders.
Since $`S=BB^{}`$ in Eq. (7), all expansion terms have the form $`g+g^{}`$. We focus on the first summand.
The following formulae for boson operators
$$[b_1,b_2b_3]=[b_1,b_2]b_3+b_2[b_1,b_3],$$
$$[b_1b_2,b_3b_4]=b_1b_3[b_2,b_4]+b_1[b_2,b_3]b_4+b_3[b_1,b_4]b_3+[b_1,b_3]b_4b_2,$$
and fermion operators
$$[f_1,f_2f_3]=\{f_1,f_2\}f_3f_2\{f_1,f_3\},$$
$$[f_1f_2,f_3f_4]=\{f_1,f_3\}f_4f_2f_3\{f_1,f_4\}f_2+f_1\{f_2,f_3\}f_4f_1f_3\{f_2,f_4\},$$
will be used. In particular, for fermion creation and annihilation operators,
$$[a_k^{}a_l,a_q^{}a_r]=\delta _{lq}a_k^{}a_r\delta _{kr}a_q^{}a_l.$$
(72)
### C.1 Fourth order
To calculate 4th order terms, let us rewrite the 3rd order expression (13) as a sum of $`H_{\mathrm{int}}^{(3)}=\frac{1}{2}[S,[S,H_{\mathrm{int}}]]`$ and $`H_0^{(3)}=\frac{1}{6}[S,[S,[S,H_0]]]`$:
$$H_{\mathrm{int},0}^{(3)}=\underset{\mathrm{๐ช๐ค}}{}A_{\mathrm{๐ค๐ช}}^{int,0}b_๐ช^{}n_๐ค\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}b_๐ฐ^{}\left\{B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{int,0}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{int,0}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\right\}+c.c.,$$
with
$$A_{\mathrm{๐ค๐ช}}^{\mathrm{int}}=\frac{1}{2}iD_๐ช\left(\varphi ^{}(๐ค,๐ช)+\varphi (๐ค+๐ช,๐ช)\right)+c.c.,$$
$$A_{\mathrm{๐ค๐ช}}^0=\frac{1}{6}\left(\epsilon _{๐ค๐ช}\epsilon _๐ค+\omega _๐ช\right)|\varphi (๐ค,๐ช)|^2\frac{1}{6}\left(\epsilon _๐ค\epsilon _{๐ค+๐ช}+\omega _๐ช\right)|\varphi (๐ค+๐ช,๐ช)|^2+c.c.,$$
$$B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{\mathrm{int}}=\frac{1}{2}iD_๐ช\left(\varphi ^{}(๐ค^{},๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)\varphi ^{}(๐ค^{},๐ช)\varphi ^{}(๐ค,๐ฐ)+\varphi ^{}(๐ค,๐ฐ)\varphi (๐ค,๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)\varphi (๐ค๐ฐ,๐ช)\right),$$
$$\begin{array}{cc}\hfill B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^0& =\frac{1}{6}\varphi ^{}(๐ค^{},๐ช)\{\varphi (๐ค,๐ช)\varphi ^{}(๐ค,๐ฐ)(\epsilon _๐ค\epsilon _{๐ค๐ช}+\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{})\hfill \\ & +\varphi (๐ค๐ฐ,๐ช)\varphi ^{}(๐ค๐ช,๐ฐ)(\epsilon _{๐ค๐ฐ๐ช}\epsilon _{๐ค๐ฐ}\epsilon _{๐ค^{}๐ช}+\epsilon _๐ค^{})\},\hfill \end{array}$$
$$\begin{array}{cc}\hfill C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{\mathrm{int}}& =\frac{1}{2}iD_๐ช(\varphi (๐ค^{},๐ช)\varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)\varphi (๐ค^{},๐ช)\varphi ^{}(๐ค+๐ฐ,๐ฐ)\hfill \\ & \varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)\varphi ^{}(๐ค+๐ฐ,๐ช)+\varphi ^{}(๐ค+๐ฐ,๐ฐ)\varphi ^{}(๐ค,๐ช)),\hfill \end{array}$$
$$\begin{array}{cc}\hfill C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^0& =\frac{1}{6}\varphi (๐ค^{},๐ช)\{\varphi ^{}(๐ค+๐ฐ,๐ช)\varphi ^{}(๐ค+๐ฐ๐ช,๐ฐ)(\epsilon _{๐ค^{}๐ช}\epsilon _๐ค^{}\epsilon _{๐ค+๐ฐ๐ช}+\epsilon _{๐ค+๐ฐ})\hfill \\ & +\varphi ^{}(๐ค,๐ช)\varphi ^{}(๐ค+๐ฐ,๐ฐ)(\epsilon _{๐ค๐ช}\epsilon _๐ค\epsilon _{๐ค^{}๐ช}+\epsilon _๐ค^{})\}.\hfill \end{array}$$
The 4th order term in Eq. (6) equals
$$H^{(4)}=\frac{1}{6}[S,[S,[S,H_{\mathrm{int}}]]]+\frac{1}{24}[S,[S,[S,[S,H_0]]]]=\frac{1}{3}[S,H_{\mathrm{int}}^{(3)}]\frac{1}{4}[S,H_0^{(3)}]:=H_{\mathrm{int}}^{(4)}+H_0^{(4)}.$$
Explicitly,
$$\begin{array}{cc}& H_{\mathrm{int},0}^{(4)}=\underset{\mathrm{๐ค๐ช}}{}A_{\mathrm{int},0}^{(4)}\gamma _๐ชn_๐ค\gamma _๐ช^{}+\underset{\mathrm{๐ค๐ช๐ค}^{}๐ฐ}{}B_{\mathrm{int},0}^{(4)}\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\underset{\mathrm{๐ค๐ช๐ค}^{}๐ฐ}{}C_{\mathrm{int},0}^{(4)}\gamma _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\hfill \\ & +\underset{\mathrm{๐ค๐ช๐ค}^{}}{}D_{\mathrm{int},0}^{(4)}b_๐ช^{}b_๐ชn_๐คn_๐ค^{}+\underset{\mathrm{๐ค๐ช}}{}E_{\mathrm{int},0}^{(4)}b_๐ช^{}b_๐ชa_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}+\underset{\mathrm{๐ค๐ช๐ค}^{}๐ฐ}{}F_{\mathrm{int},0}^{(4)}b_๐ฐ^{}b_๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ฐ}+\hfill \\ & +\underset{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}{}G_{\mathrm{int},0}^{(4)}b_๐ฐ^{}b_๐ฐa_{๐ค^{}+๐ฐ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}+\underset{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}{}I_{\mathrm{int},0}^{(4)}b_๐ฐ^{}b_๐ฐa_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+c.c.,\hfill \end{array}$$
(73)
where
$$A_{\mathrm{int}}^{(4)}=\frac{1}{3}A_{\mathrm{๐ค๐ช}}^{\mathrm{int}},A_0^{(4)}=\frac{1}{4}A_{\mathrm{๐ค๐ช}}^0,B_{\mathrm{int}}^{(4)}=\frac{1}{3}B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{\mathrm{int}},B_0^{(4)}=\frac{1}{4}B_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^0.$$
Other coefficients in Eq. (73) arise according to the same scheme, viz,
$$C_{\mathrm{int}}^{(4)}=\frac{1}{3}C_{\mathrm{๐ค๐ช๐ฐ๐ค}^{}}^{\mathrm{int}},D_{\mathrm{int}}^{(4)}=\frac{1}{3}A_{\mathrm{๐ค๐ช}}^{\mathrm{int}}\left(|\varphi (๐ค^{},๐ช)|^2|\varphi (๐ค^{}+๐ช,๐ช)|^2\right),$$
$$E_{\mathrm{int}}^{(4)}=\frac{1}{3}\left(A_{๐ค๐ช,๐ช}^{\mathrm{int}}A_{\mathrm{๐ค๐ช}}^{\mathrm{int}}\right)\varphi (๐ค,๐ช),F_{\mathrm{int}}^{(4)}=\frac{1}{3}\left(B_{๐ค,๐ช,๐ฐ,๐ค^{}๐ฐ}^{\mathrm{int}}\varphi (๐ค^{}๐ช,๐ฐ)B_{๐ค,๐ช,๐ฐ,๐ค^{}}^{\mathrm{int}}\varphi (๐ค^{},๐ฐ)\right),$$
$$G_{\mathrm{int}}^{(4)}=\frac{1}{3}\left(C_{๐ค,๐ช,๐ฐ,๐ค^{}}^{\mathrm{int}}\varphi (๐ค^{}+๐ฐ,๐ฐ)C_{๐ค,๐ช,๐ฐ,๐ค^{}+๐ฐ}^{\mathrm{int}}\varphi (๐ค^{}+๐ฐ๐ช,๐ฐ)\right),$$
$$I_{\mathrm{int}}^{(4)}=\frac{1}{3}\left(B_{๐ค,๐ช,๐ฐ,๐ค^{}}^{\mathrm{int}}\varphi (๐ค,๐ฐ)B_{๐ค+๐ฐ,๐ช,๐ฐ,๐ค^{}}^{\mathrm{int}}\varphi (๐ค+๐ฐ๐ช,๐ฐ)+C_{๐ค^{}๐ฐ,๐ช,๐ฐ,๐ค}^{\mathrm{int}}\varphi (๐ค^{}๐ช,๐ฐ)C_{๐ค^{},๐ช,๐ฐ,๐ค}^{\mathrm{int}}\varphi (๐ค^{}+๐ฐ,๐ฐ)\right).$$
and similarly for $`C_0^{(4)}`$, $`D_0^{(4)}`$, etc.
### C.2 Fifth order
Proceeding analogously as with 4th order, one obtains
$$H^{(5)}=\frac{1}{24}[S,[S,[S,[S,H_{\mathrm{int}}]]]]\frac{1}{120}[S,[S,[S,[S,[S,H_0]]]]]=\frac{1}{4}[S,H_{\mathrm{int}}^{(4)}]\frac{1}{5}[S,H_0^{(4)}]:=H_{\mathrm{int}}^{(5)}+H_0^{(5)},$$
(74)
$$\begin{array}{cc}& H_{\mathrm{int},0}^{(5)}=\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}A_{int,0}^{(5)}b_๐ฐ^{}\gamma _๐ชn_๐คa_{๐ค^{}๐ฐ}^{}a_{๐ค^{}+๐ช}+\underset{\mathrm{๐ช๐ค๐ฐ}}{}B_{\mathrm{int},0}^{(5)}b_๐ฐ^{}\gamma _๐ชa_๐ค^{}a_{๐ค๐ฐ}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}C_{\mathrm{int},0}^{(5)}b_๐ฐ^{}a_{๐ค^{}๐ฐ}a_{๐ค^{}๐ช}n_๐ค\gamma _๐ช^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}D_{\mathrm{int},0}^{(5)}b_๐ฎ^{}\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}+๐ฎ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}E_{\mathrm{int},0}^{(5)}b_๐ฎ^{}\gamma _๐ฐa_{๐ค๐ฐ๐ฎ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}F_{\mathrm{int},0}^{(5)}b_๐ฎ^{}a_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ฐ}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}G_{\mathrm{int},0}^{(5)}b_๐ฎ\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ฎ}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}I_{\mathrm{int},0}^{(5)}b_๐ฎ\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช๐ฎ}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}J_{\mathrm{int},0}^{(5)}b_๐ฎa_{๐ฅ+๐ฐ}^{}a_{๐ฅ๐ฎ}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}K_{\mathrm{int},0}^{(5)}b_๐ฎ^{}\gamma _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช๐ฎ}^{}a_{๐ค+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}L_{\mathrm{int},0}^{(5)}b_๐ฎ^{}\gamma _๐ฐa_{๐ค^{}๐ฎ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}M_{\mathrm{int},0}^{(5)}b_๐ฎ^{}a_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ช}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค^{}+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}N_{\mathrm{int},0}^{(5)}b_๐ฎ\gamma _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ๐ฎ}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}O_{\mathrm{int},0}^{(5)}b_๐ฎ\gamma _๐ฐa_{๐ค^{}+๐ฎ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}P_{\mathrm{int},0}^{(5)}b_๐ฎa_{๐ฅ+๐ฐ}^{}a_{๐ฅ๐ฎ}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ค}^{}}{}R_{\mathrm{int},0}^{(5)}b_๐ช^{}n_๐คn_๐ค^{}\gamma _๐ช^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค}}{}S_{\mathrm{int},0}^{(5)}b_๐ช^{}a_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค}}{}S_{\mathrm{int},0}^{(5)}b_๐ช\gamma _๐ชa_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}T_{\mathrm{int},0}^{(5)}b_๐ฐ^{}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ฐ}\gamma _๐ฐ^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}U_{\mathrm{int},0}^{(5)}b_๐ฐ^{}a_{๐ค^{}+๐ฐ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\gamma _๐ฐ^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}W_{\mathrm{int},0}^{(5)}b_๐ฐ^{}a_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\gamma _๐ฐ^{}+c.c.,\hfill \end{array}$$
where
$$A_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(A_{\mathrm{int},0}^{(4)}+A_{\mathrm{int},0}^{(4)}\right)\times \left(\varphi ^{}(๐ค^{},๐ฐ)\varphi ^{}(๐ค^{}+๐ช,๐ช)\varphi ^{}(๐ค^{}+๐ช,๐ฐ)\varphi ^{}(๐ค^{}+๐ช๐ฐ,๐ช)\right),$$
$$B_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(A_{\mathrm{int},0}^{(4)}+A_{\mathrm{int},0}^{(4)}A_{\mathrm{int},0}^{(4)๐ค๐ค๐ฐ}A_{\mathrm{int},0}^{(4)๐ค๐ค๐ฐ}\right)\varphi ^{}(๐ค,๐ฐ),$$
$$C_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(A_{\mathrm{int},0}^{(4)}+A_{\mathrm{int},0}^{(4)}\right)\left(\varphi ^{}(๐ค^{},๐ฐ)\varphi ^{}(๐ค^{},๐ช)\varphi ^{}(๐ค^{}๐ช,๐ฐ)\varphi ^{}(๐ค^{}๐ฐ,๐ช)\right),$$
$$D_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(B_{\mathrm{int},0}^{(4)๐ค^{}๐ค^{}+๐ฎ}\varphi ^{}(๐ค^{}+๐ฎ๐ช,๐ฎ)B_{\mathrm{int},0}^{(4)}\varphi ^{}(๐ค^{}+๐ฎ,๐ฎ)\right),$$
$$E_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(B_{\mathrm{int},0}^{(4)}\varphi ^{}(๐ค๐ฐ,๐ฎ)B_{\mathrm{int},0}^{(4)๐ค๐ค๐ฎ}\varphi ^{}(๐ค๐ช,๐ฎ)\right),$$
$$F_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}B_{\mathrm{int},0}^{(4)}\left(\varphi ^{}(๐ฅ,๐ฎ)\varphi (๐ฅ,๐ฐ)\varphi ^{}(๐ฅ๐ฐ,๐ฎ)\varphi (๐ฅ๐ฎ,๐ฐ)\right),$$
$$G_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(B_{\mathrm{int},0}^{(4)๐ค^{}๐ค^{}๐ฎ}\varphi (๐ค^{}๐ช,๐ฎ)B_{\mathrm{int},0}^{(4)}\varphi (๐ค^{},๐ฎ)\right),$$
$$I_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(B_{\mathrm{int},0}^{(4)๐ค๐ค๐ฎ}\varphi (๐ค๐ฐ,๐ฎ)B_{\mathrm{int},0}^{(4)}\varphi (๐ค๐ช,๐ฎ)\right),$$
$$J_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}B_{\mathrm{int},0}^{(4)}\left(\varphi (๐ฅ+๐ฐ,๐ฎ)\varphi (๐ฅ๐ฎ+๐ฐ,๐ฐ)\varphi (๐ฅ,๐ฎ)\varphi (๐ฅ+๐ฐ,๐ฐ)\right),$$
$$K_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(C_{\mathrm{int},0}^{(4)}\varphi ^{}(๐ค๐ช,๐ฎ)C_{\mathrm{int},0}^{(4)๐ค๐ค๐ฎ}\varphi ^{}(๐ค+๐ฐ,๐ฎ)\right),$$
$$L_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(C_{\mathrm{int},0}^{(4)}\varphi ^{}(๐ค^{},๐ฎ)C_{\mathrm{int},0}^{(4)๐ค^{}๐ค^{}๐ฎ}\varphi ^{}(๐ค^{}๐ช,๐ฎ)\right),$$
$$M_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}C_{\mathrm{int},0}^{(4)}\left(\varphi ^{}(๐ฅ,๐ฎ)\varphi (๐ฅ,๐ฐ)\varphi ^{}(๐ฅ๐ฐ,๐ฎ)\varphi (๐ฅ๐ฎ,๐ฐ)\right),$$
$$N_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(C_{\mathrm{int},0}^{(4)๐ค๐ค๐ฎ}\varphi (๐ค๐ช,๐ฎ)C_{\mathrm{int},0}^{(4)}\varphi (๐ค+๐ฐ,๐ฎ)\right),$$
$$O_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(C_{\mathrm{int},0}^{(4)}\varphi (๐ค^{}+๐ฎ,๐ฎ)C_{\mathrm{int},0}^{(4)๐ค^{}๐ค^{}+๐ฎ}\varphi (๐ค^{}+๐ฎ๐ช,๐ฎ)\right),$$
$$P_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}C_{\mathrm{int},0}^{(4)}\left(\varphi (๐ฅ+๐ฐ,๐ฎ)\varphi (๐ฅ๐ฎ+๐ฐ,๐ฐ)\varphi (๐ฅ,๐ฎ)\varphi (๐ฅ+๐ฐ,๐ฐ)\right),$$
$$R_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(D_{\mathrm{int},0}^{(4)}+D_{\mathrm{int},0}^{(4)}\right),S_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}E_{\mathrm{int},0}^{(4)},T_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(F_{\mathrm{int},0}^{(4)}+F_{\mathrm{int},0}^{(4)๐ค๐ค^{}}\right),$$
$$U_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(G_{\mathrm{int},0}^{(4)}+G_{\mathrm{int},0}^{(4)๐ค๐ค^{}}\right),W_{\mathrm{int},0}^{(5)}=\{\frac{1}{4},\frac{1}{5}\}\left(I_{\mathrm{int},0}^{(4)}+I_{\mathrm{int},0}^{(4)๐ค๐ค^{}}\right).$$
$`1/4`$ in the curly brackets refering to int terms and $`1/5`$ to $`0`$ terms.
### C.3 Sixth order
$$H^{(6)}=\frac{1}{120}[S,[S,[S,[S,[S,H_{\mathrm{int}}]]]]]+\frac{1}{720}[S,[S,[S,[S,[S,[S,H_0]]]]]]=\frac{1}{5}[S,H_{\mathrm{int}}^{(5)}]\frac{1}{6}[S,H_0^{(5)}]:=H_{\mathrm{int}}^{(6)}+H_0^{(6)},$$
(75)
$$\begin{array}{cc}& H_{\mathrm{int},0}^{(6)}=\{\frac{1}{5},\frac{1}{6}\}\{\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}A_{\mathrm{int},0}^{(5)}\gamma _๐ฐ\gamma _๐ชn_๐คa_{๐ค^{}๐ฐ}^{}a_{๐ค^{}+๐ช}+\underset{\mathrm{๐ช๐ค๐ฐ}}{}B_{\mathrm{int},0}^{(5)}\gamma _๐ฐ\gamma _๐ชa_๐ค^{}a_{๐ค๐ฐ}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}C_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค^{}๐ฐ}a_{๐ค^{}๐ช}n_๐ค\gamma _๐ช^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}D_{\mathrm{int},0}^{(5)}\gamma _๐ฎ\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}+๐ฎ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}E_{\mathrm{int},0}^{(5)}\gamma _๐ฎ\gamma _๐ฐa_{๐ค๐ฐ๐ฎ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}F_{\mathrm{int},0}^{(5)}\gamma _๐ฎa_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ฐ}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}G_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ฎ}\gamma _๐ฎ^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}I_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช๐ฎ}a_{๐ค^{}๐ช}^{}a_๐ค^{}\gamma _๐ฎ^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}J_{\mathrm{int},0}^{(5)}a_{๐ฅ+๐ฐ}^{}a_{๐ฅ๐ฎ}a_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\gamma _๐ฎ^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}K_{\mathrm{int},0}^{(5)}\gamma _๐ฎ\gamma _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช๐ฎ}^{}a_{๐ค+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}L_{\mathrm{int},0}^{(5)}\gamma _๐ฎ\gamma _๐ฐa_{๐ค^{}๐ฎ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}M_{\mathrm{int},0}^{(5)}\gamma _๐ฎa_{๐ฅ๐ฎ}^{}a_{๐ฅ๐ช}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค^{}+๐ฐ}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}N_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ๐ฎ}\gamma _๐ฎ^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}๐ฎ}{}O_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค^{}+๐ฎ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\gamma _๐ฎ^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}\mathrm{๐ฎ๐ฅ}}{}P_{\mathrm{int},0}^{(5)}a_{๐ฅ+๐ฐ}^{}a_{๐ฅ๐ฎ}a_๐ค^{}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\gamma _๐ฎ^{}+\underset{\mathrm{๐ช๐ค๐ค}^{}}{}R_{\mathrm{int},0}^{(5)}\gamma _๐ชn_๐คn_๐ค^{}\gamma _๐ช^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค}}{}S_{\mathrm{int},0}^{(5)}\gamma _๐ชa_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค}}{}S_{\mathrm{int},0}^{(5)}\gamma _๐ชa_๐ค^{}a_{๐ค๐ช}\gamma _๐ช^{}\gamma _๐ช^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}T_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค๐ฐ}^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_{๐ค^{}๐ฐ}\gamma _๐ฐ^{}\hfill \\ & +\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}U_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_{๐ค^{}+๐ฐ}^{}a_{๐ค^{}๐ช}a_{๐ค๐ช}^{}a_{๐ค+๐ฐ}\gamma _๐ฐ^{}+\underset{\mathrm{๐ช๐ค๐ฐ๐ค}^{}}{}W_{\mathrm{int},0}^{(5)}\gamma _๐ฐa_๐ค^{}a_{๐ค๐ช}a_{๐ค^{}๐ช}^{}a_๐ค^{}\gamma _๐ฐ^{}+b^{}b\left\{\right\}+c.c.\}.\hfill \end{array}$$
(76)
The last term denoting all $`(6,2)`$ expressions which are irrelevant after averaging over phonon vacuum. The coefficients are the same as in the 5th order terms, except for the factors $`1/5`$$`1/6`$.
###### Acknowledgements.
The authors would like to thank prof. J. Maฤkowiak for valuable comments and suggestions.
|
warning/0506/astro-ph0506076.html
|
ar5iv
|
text
|
# Dynamics of assisted quintessence
## I Introduction
An attractive hypothesis for the fundamental nature of dark energy is quintessence โ a scalar field evolving in a non-zero potential energy quint . Such modelling has proven highly successful in implementing inflation models in the early Universe. It has however met with much less success in the present Universe, primarily due to the difficulty of obtaining equations of state close enough to the cosmological constant value $`w=1`$ to satisfy observational bounds obsw . One strategy is to ensure that the quintessence energy density starts at such a low value that the field only begins evolving close to the present, but such tuning is hardly more satisfactory than a cosmological constant. The alternative is for the field to evolve significantly, ideally exploiting a โtrackerโ behaviour rendering the late-time evolution almost independent of initial conditions. Unfortunately however this is viable only for very particular potentials: amongst monotonic potentials, exponentials do not give acceleration and power-laws can give negative enough $`w`$ only for exponents well below 2. Among the potentials regarded as giving satisfactory phenomenology, many in fact feature a minimum tuned to match the observed value of the cosmological constant.
In this paper we investigate whether this situation might be alleviated by allowing the quintessence to arise from several fields, which we assume to have the same potential energies. While our interest is primarily phenomenological, we note that such situations may arise from higher-dimensional theories; in fact there are many dynamical modulus fields in string theory corresponding to the size of compactified dimensions. In the context of early Universe inflation, such a set of fields has been shown to give the phenomenon of assisted inflation LMS , whereby they may collectively drive inflation even if each individual field has too steep a potential to do so on its own, i.e. yielding an effective equation of state closer to $`w=1`$. It is therefore worth considering the possibility of assisted quintessence behaviour, in order to see whether it may be better able to match observations. Assisted quintessence is particularly attractive in the context of tracking models, as each field will separately converge onto the tracking solution making it entirely natural that they all play a dynamical role.
In this paper we will investigate some aspects of assisted quintessence, mostly restricting ourselves to the simplest case where each field has the same potential and always assuming there are no interactions between fields. The most closely-related paper is that of Blais and Polarski BP , who analyzed several multi-field quintessence models both analytically and numerically, though with a different focus directed mainly at attempting to realize models where the acceleration is a transient phenomenon. We will primarily study the exponential and inverse power-law cases in detail, in the former case considering different exponents for the two potentials as already extensively analyzed by Coley and van den Hoogen CvdH . In this paper we will also provide an algorithm for relating assisted quintessence to an equivalent single-field model under more general circumstances.
Our scenario is distinct from two types of scenario already extensively investigated in the literature. One is the double exponential potential models of Refs. BCN ; double , where there was only a single field (Ref. BCN briefly mentioned a multiple-field case but not with uncoupled fields). Another is the two-field models of dark energy which have received some attention recently as a way of crossing the cosmological constant boundary ($`w=1`$) to give rise to a phantom behaviour at late times Feng ; GPZZ ; crossing . Unlike our case, at least one of those fields must have a negative kinetic energy.
## II Assisted quintessence
If one accepts the possibility of multiple fields, particularly sharing the same form of potential $`V(\varphi _i)`$, then the idea of assisted quintessence emerges very naturally provided the potentials have tracker solutions for at least some values of the fields. Tracker solutions arise when the fields are initially sub-dominant as compared to a perfect fluid, the field contribution to the Friedmann equation then being neglected to give equations of the form
$`H^2`$ $``$ $`{\displaystyle \frac{8\pi }{3m_{\mathrm{Pl}}^2}}\rho _\mathrm{f},`$ (1)
$`\ddot{\varphi }_i`$ $`=`$ $`3H\dot{\varphi }_i{\displaystyle \frac{\mathrm{d}V}{\mathrm{d}\varphi _i}},`$ (2)
where $`H`$ is the Hubble rate, $`m_{\mathrm{Pl}}`$ is the Planck mass and a dot denotes the derivative with respect to a cosmic time $`t`$. Here the fluid density $`\rho _\mathrm{f}`$ might for instance be broken up into matter and radiation components $`\rho _\mathrm{m}`$ and $`\rho _\mathrm{r}`$. In this set-up, the scalar fields are completely unaware of each otherโs existence (their normal channel of communication being via the Friedmann equation), and hence separately evolve onto the tracker solution in response to the fluid. At sufficiently-late times one would therefore have all the $`\varphi _i`$ equal to each other, and hence of equal potential energy.
For an individual field $`\varphi `$ with potential $`V(\varphi )`$, whether or not there is tracking behaviour can be determined from the value of the function
$$\mathrm{\Gamma }\frac{VV^{\prime \prime }}{V^2},$$
(3)
where a prime represents the derivative in terms of $`\varphi `$. Solutions converge to a tracker provided that it satisfies $`\mathrm{\Gamma }>1(2\gamma )/(4+2\gamma )`$ where the equation of state is $`p_\mathrm{f}=(\gamma 1)\rho _\mathrm{f}`$, the interesting case however being $`\mathrm{\Gamma }>1`$ which is required for the field energy density to grow relative to the fluid allowing eventual domination SWZ .
The convergence of different fields to the same tracking solution does not in itself amount to assisted quintessence, as the fields are not generating any gravitational effect on the background evolution. The convergence does however set up the initial conditions for such a behaviour. What is mainly of interest is what happens once the energy density of the fields, all evolving together, is no longer subdominant, and that is the situation addressed in the rest of this paper. We will therefore be considering the full Friedmann equation
$$H^2=\frac{8\pi }{3m_{\mathrm{Pl}}^2}\left(\rho _\mathrm{f}+\underset{i}{}\rho _{\varphi _i}\right),$$
(4)
where $`\rho _{\varphi _i}V_i(\varphi _i)+\dot{\varphi }_i^2/2`$ and we assume spatial flatness throughout.
## III Exponential potentials
We consider two fields $`\varphi _1`$ and $`\varphi _2`$ each with a separate exponential potential
$`V(\varphi _1,\varphi _2)`$ $`=`$ $`Ae^{\lambda _1\kappa \varphi _1}+Be^{\lambda _2\kappa \varphi _2}`$
$``$ $`V_1(\varphi _1)+V_2(\varphi _2),`$
where $`\kappa ^28\pi /m_{\mathrm{Pl}}^2`$. For generality, we will allow the potentials to have different slopes.
### III.1 Assisted quintessence solutions
For the case where no matter is present, this system is exactly the original assisted inflation scenario of Liddle et al. LMS , where the multiple fields evolve to give dynamics matching a single-field model with
$$\frac{1}{\lambda _{\mathrm{eff}}^2}=\frac{1}{\lambda _1^2}+\frac{1}{\lambda _2^2}.$$
(6)
For a single-field potential $`V(\varphi )=V_0e^{\lambda \kappa \varphi }`$ the scale factor evolves as $`at^p`$, where $`p=2/\lambda ^2`$. Then the multi-field case Eq. (6) corresponds to an effective power-law index given by $`p_{\mathrm{eff}}2/\lambda _{\mathrm{eff}}^2=p_i`$ LMS . The expansion rate is therefore more rapid the more fields there are. A particularly comprehensive analysis of multiple fields in exponential potentials has been given by Collinucci et al. CNV .
As it happens, this assisted behaviour continues to be valid in the presence of a perfect fluid with equation of state $`p=(\gamma 1)\rho `$. This was first noted by Coley and van den Hoogen CvdH , who also allowed for the possibility of spatial curvature. This is because the method used to relate multi-field dynamics to an equivalent single-field dynamics in Ref. LMS is a property of the scalar field sector alone, being valid in the presence of any other matter sources.
### III.2 Critical points and stability
The critical points for this system were classified by Coley and van den Hoogen CvdH , and we will reiterate their results only briefly before embarking on some numerical analysis of the evolution towards them. In Ref. CvdH , spatial curvature was included as a degree of freedom and the fluid assumed to have $`\gamma >1`$, whereas we assume spatial flatness and $`\gamma `$ in the wider range $`0`$ to $`2`$.
To study the critical point structure and stability of the system, it is convenient to introduce the following dimensionless quantities CLW
$$x_i\frac{\kappa \dot{\varphi }_i}{\sqrt{6}H},y_i\frac{\kappa \sqrt{V_i}}{\sqrt{3}H},i=1,2.$$
(7)
Then we obtain
$`{\displaystyle \frac{\mathrm{d}x_i}{\mathrm{d}N}}`$ $`=`$ $`3x_i+\lambda _i\sqrt{{\displaystyle \frac{3}{2}}}y_i^2+{\displaystyle \frac{3}{2}}x_i\left[2x_1^2+2x_2^2+\gamma (1x_1^2y_1^2x_2^2y_2^2)\right]i=1,2,`$ (8)
$`{\displaystyle \frac{\mathrm{d}y_i}{\mathrm{d}N}}`$ $`=`$ $`\lambda _i\sqrt{{\displaystyle \frac{3}{2}}}x_iy_i+{\displaystyle \frac{3}{2}}y_i\left[2x_1^2+2x_2^2+\gamma (1x_1^2y_1^2x_2^2y_2^2)\right]i=1,2,`$ (9)
$`{\displaystyle \frac{1}{H}}{\displaystyle \frac{\mathrm{d}H}{\mathrm{d}N}}`$ $`=`$ $`{\displaystyle \frac{3}{2}}\left[2x_1^2+2x_2^2+\gamma (1x_1^2y_1^2x_2^2y_2^2)\right],`$ (10)
where $`N\mathrm{ln}a`$, together with the constraint
$$x_1^2+y_1^2+x_2^2+y_2^2+\frac{\kappa ^2\rho _\mathrm{f}}{3H^2}=1.$$
(11)
We define the density parameters $`\mathrm{\Omega }_{\varphi _i}`$ and the equation of state $`w_{\varphi _i}`$ for scalar fields, as
$$\mathrm{\Omega }_{\varphi _i}=x_i^2+y_i^2,w_{\varphi _i}=\frac{x_i^2y_i^2}{x_i^2+y_i^2},i=1,2.$$
(12)
We note that the density parameter for matter is given by $`\mathrm{\Omega }_m\kappa ^2\rho _\mathrm{f}/3H^2=1\mathrm{\Omega }_{\varphi _1}\mathrm{\Omega }_{\varphi _2}`$ from Eq. (11).
These equations are in fact valid for any uncoupled potentials, with $`\lambda _i`$ defined by
$$\lambda _i\frac{1}{\kappa V_i}\frac{\mathrm{d}V_i}{\mathrm{d}\varphi _i}.$$
(13)
Our case of exponential potentials corresponds to $`\lambda _i`$ both constant. It is straightforward to extend our analysis to the case of a dynamically changing $`\lambda `$ as studied in single-field models in Ref. MP .
The classification of critical points and their stability is given in Table LABEL:table:bw, and agrees with results in Ref. CvdH .<sup>1</sup><sup>1</sup>1We note that a full analysis has also been carried out for the case where one of the fields is of phantom type (opposite sign of the kinetic energy) GPZZ , where the late-time behaviour is domination by the phantom field. There are eight types of critical point, seven being discrete points and one a circular locus of critical points. They are readily compared to the five types of critical point found for the single-field system in Ref. CLW . Our cases 1, 2 with $`\theta =0`$ and $`\pi `$, 3 and 4 correspond to the five points in the single-field case for the field $`\varphi _1`$, with $`\varphi _2`$ playing no role. Our cases 1, 2 with $`\theta =\pi /2`$ and $`3\pi /2`$, 5 and 6 are the same solutions for the $`\varphi _2`$ field. Finally, cases 7 and 8 are critical points in which both fields play a role, and which have no direct analogue with the single-field case.
In the single-field case, the stable late-time attractor is either scalar-field dominated (case 3 or 5) or a scaling solution (case 4 or 6) depending on the relative values of $`\lambda `$ and $`\gamma `$ CLW . Once a second field is added, the new degrees of freedom always render those solutions unstable. The late-time attractors instead become either the assisted scalar-field dominated solution case 7 (for $`\lambda _{\mathrm{eff}}^2<3\gamma `$) or the assisted scaling solution case 8 (for $`\lambda _{\mathrm{eff}}^2>3\gamma `$).
To compare our results with Ref. CvdH , we note that under their assumptions the curvature always dominates the fluid at late times, and behaves like a $`\gamma =2/3`$ fluid. Accordingly, they always find solutions with non-zero fluid density to be unstable to eventual curvature domination, whereas our assumption of spatial flatness renders them stable. Otherwise, the classification and stability shown in our Table matches their Tables I and III. Additionally they correctly describe the spatially-flat case in their text.
### III.3 Multi-field phenomenology
For the exponential potential, the assisted quintessence phenomenon does indeed exist, with the extra fields being equivalent to a flatter single-field potential. Unfortunately this result does not seem particularly useful phenomenologically, as the scaling solutions do not give acceleration as required by observations, while the scalar-field dominated solutions would long ago have made the matter density negligible.
There is however one scenario that might be of interest, which is to imagine there are a large number of exponential potentials with different initial conditions. As the Universe evolves, more and more fields would join the assisted quintessence attractor, reducing $`\lambda _{\mathrm{eff}}`$. Eventually, this could switch the attractor from the scaling regime $`\lambda _{\mathrm{eff}}^2>3\gamma `$ into the regime of late-time scalar field dominance $`\lambda _{\mathrm{eff}}^2<3\gamma `$ CvdH .
Figure 1 shows a two-field example of such evolution, with the single-field scaling solution becoming unstable as the second field becomes important, switching the evolution into late-time scalar field domination. From Table I the first scaling regime corresponds to case 4 ($`x_1=y_1=\sqrt{3/8}`$, $`x_2=y_2=0`$) with no acceleration, whereas the final stable attractor is case 7 ($`x_1=0.365`$, $`y_1=0.560`$, $`x_2=0.406`$, $`y_2=0.623`$) with acceleration. We checked that the values of the fixed points agree very well with numerical results.
However even if the initial conditions were fine-tuned to bring this transition into the recent past, it is hard to see how the equation of state $`w`$ could reach a sufficiently-negative value to be observationally viable, current observations indicating roughly $`w\gamma 1<0.8`$ obsw . In fact, if we compare the two-field scenarios with a single-field scenario in which an accelerated expansion occurs at late times ($`\lambda <\sqrt{2}`$), we find that typically the two-field models give a larger present value of $`w`$ (here $`w`$ is the equation of state of scalar fields including the contribution of both $`\varphi _1`$ and $`\varphi _2`$). Figure 2 compares a two-field model with $`\lambda _1=3`$ and $`\lambda _2=1`$ against a single-field model with $`\lambda =1`$, in both cases with the field with slope 1 starting with a low value and coming to dominate only at the present epoch (the โthawingโ regime as according to Ref. CL ). We see that the equation of state of the thawing field alone is indeed closer to $`1`$ at the present (identified as when the matter density is 0.3 of critical) than in the single-field case, but the combined $`w`$ of the two fields is larger.
## IV Inverse power-law potentials
We now consider the inverse-power law case $`V(\varphi )=V_0\varphi ^\beta `$, where we will take the same exponent and normalization for each field. Here one might hope that the assisted phenomenon would give rise to an effective $`\beta _{\mathrm{eff}}`$ which is smaller than the individual $`\beta `$, so that observationally-viable models can be achieved in steeper potentials than the single-field case. Unfortunately that turns out not to be true, as we now see.
These potentials are favoured because they exhibit tracker solutions, and it is therefore legitimate to suppose that after some early time we can take $`\varphi _1=\varphi _2=\mathrm{}\varphi `$. The Friedmann and scalar wave equations become
$`H^2`$ $`=`$ $`{\displaystyle \frac{8\pi }{3m_{\mathrm{Pl}}^2}}\left(\rho _\mathrm{f}+nV_0\varphi ^\beta +{\displaystyle \frac{n}{2}}\dot{\varphi }^2\right);`$ (14)
$`\ddot{\varphi }`$ $`=`$ $`3H\dot{\varphi }+\beta V_0\varphi ^{\beta 1},`$ (15)
where $`n`$ is the number of fields. This is not yet equivalent to a single-field model, but can be made so (loosely following the method of Ref. LMS ) by the redefinitions
$$\chi =\sqrt{n}\varphi ;W_0=\sqrt{n^\beta }nV_0,$$
(16)
which then gives the equations of a single-field model with potential $`W(\chi )=W_0\chi ^\beta `$.
This rescaling indicates that the assisted behaviour renormalizes the amplitude of the potential in this case, but does not renormalize its exponent. In such models the amplitude has to be adjusted in order to give the present-day value of the matter density $`\mathrm{\Omega }_\mathrm{m}`$, and having done that the multi-field system then has identical dynamics to the single-field model, in particular predicting the same present-day value of the equation of state $`w`$.
## V Assisted quintessence dynamics
We end with a more general construction for analyzing assisted quintessence, extending the analysis of the previous section to the case where the potential is arbitrary, but the same for all fields. Again we assume tracking behaviour so that we can take $`\varphi _i=\varphi `$, giving
$`H^2`$ $`=`$ $`{\displaystyle \frac{8\pi }{3m_{\mathrm{Pl}}^2}}\left[\rho _\mathrm{f}+nV(\varphi )+{\displaystyle \frac{n}{2}}\dot{\varphi }^2\right],`$ (17)
$`\ddot{\varphi }`$ $`=`$ $`3H\dot{\varphi }{\displaystyle \frac{\mathrm{d}V}{\mathrm{d}\varphi }}.`$ (18)
To find an equivalent single-field system, we first note that the kinetic term in the Friedmann equation forces the correspondence
$$\chi =\sqrt{n}\varphi .$$
(19)
The equations can then be transformed into single-field form with potential
$$W(\chi )=nV(\chi /\sqrt{n}),$$
(20)
which obviously has the desired effect in the Friedmann equation, but which also renders the fluid equation into single-field form. This formula therefore represents an algorithm for finding a single-field potential $`W`$ which will generate the same evolution as multiple fields evolving together in the potential $`V`$.
For simplicity, we henceforth consider the two-field case, though the generalization is straightforward. The correspondence then is
$$\chi =\sqrt{2}\varphi ;W(\chi )=2V(\chi /\sqrt{2}).$$
(21)
This allows us to ask what condition would have to be satisfied in order to have no assisted behaviour, i.e. for $`W`$ and $`V`$ have the same functional form apart from an overall constant. This happens for potentials obeying the condition
$$V(\chi )=2CV(\chi /\sqrt{2}),$$
(22)
for all $`\chi `$, where $`C`$ is a constant. The general solution to this equation is
$$V(\chi )\chi ^\beta \times f(\chi ),$$
(23)
where $`f(\chi )`$ is any function periodic in $`\mathrm{ln}\chi `$ with period $`\mathrm{ln}\sqrt{2}`$ (i.e. a Fourier series with this periodicity). While this can be any of an infinite class of potentials, having such a periodicity is clearly artificial. The only interesting case therefore is $`f(\chi )`$ equals a constant, giving the power-law potential. This proves that the inverse power-law potentials are the unique monotonic potentials which do not exhibit assisted quintessence behaviour. In all other cases, the equivalent single-field potential has a different functional form. In the exponential case this algorithm correctly reproduces Eq. (6) for the case $`\lambda _1=\lambda _2`$.
As a final point, we note that under the correspondence Eq. (20), the tracking parameter $`\mathrm{\Gamma }_WWW^{\prime \prime }/W^2`$ is equal to $`\mathrm{\Gamma }_V=VV^{\prime \prime }/V^2`$ (the primes here being derivatives wrt the arguments $`\chi `$ and $`\varphi `$ respectively). As one would expect, the tracking conditions on the multi-field model and its single-field dynamical equivalent are the same.
## VI Conclusions
We have studied various aspects of assisted quintessence dynamics. Such dynamics arises naturally if there are several fields with the same potential, provided the potential exhibits tracking behaviour for at least some stage of its early evolution.
Our most powerful result is Eq. (20), which provides a general algorithm for finding a single-field model which mimics the dynamics of a multi-field assisted quintessence model. Applied to inverse power-law models, it shows that they are the unique (monotonic) potentials for which there is no assisted behaviour, the collection of fields behaving as a single field in the same potential (up to overall normalization). All other potentials will exhibit assisted behaviour, the exponential potential being an explicit example.
It should be possible to extend our analysis to the case of more general dark energy models in which the Lagrangian includes non-canonical kinematic terms, such as the tachyon, k-essence and ghost condensate. For theories whose Lagrangian $`p`$ is a function of the field $`\varphi `$ and $`X(\varphi )^2/2`$, the existence of scaling solutions restricts the form of Lagrangian to be $`p=Xg(Xe^{\lambda \varphi })`$, where $`g`$ is an arbitrary function and $`\lambda `$ is a constant PT . It would be certainly of interest to investigate whether the assisted behaviour we found for the canonical scalar field with an exponential potential persists in such general dark energy models.
It is interesting that the multi-field system can be analyzed so simply. Regrettably, however, we have not uncovered any scenarios where the assisted quintessence phenomenon appears to improve the situation with regard to the observations. In fact our results show the contrary; in potentially the most interesting scenario of the inverse power-law it turns out that there is no assisted quintessence effect.
###### Acknowledgements.
S.A.K. was supported by the Korean government, A.R.L. by PPARC, and S.T. by JSPS (No. 30318802).
|
warning/0506/astro-ph0506114.html
|
ar5iv
|
text
|
# Ultra-low Amplitude Cepheids in the Large Magellanic Cloud
## 1. Introduction
The MACHO, EROS and OGLE data of the Large Magellan Cloud (LMC) have already been searched for variables, and in particular classical Cepheid variables (e.g., Beaulieu et al. (1995), Welch et al. (1995), Udalski et al. (1999), Kanbur et al. (2003)). The lowest reported amplitudes in these analyses have been around 0.01 mag. Our purpose for redoing such an analysis is to detect as many Cepheids as possible, with particular emphasis on those with very small pulsation amplitudes. Such low amplitude stars are expected from an evolutionary point of view, either just ramping up their amplitudes after entering the instability strip (IS), or decaying after exiting the strip. More details will be given in the discussion. Our analysis finds that the MACHO and OGLE data are accurate enough to go down to several milli-magnitude pulsations.
We have examined the 636 MACHO stars of Field 77 in a parallelogram in the HR diagram defined by $`14<V<16`$ and $`17.64<V+16.39(VR_\mathrm{c})<24.03`$ This region was chosen by visual inspection to include the instability strip and colors 0.15 blueward and redward. It contains a mixture of non-oscillatory giants of spectral type F, and variable stars such as Cepheids, W Vir stars, and ellipsoidal variables (binaries). This region was converted from Johnson $`V`$ and Cousins $`R_\mathrm{c}`$ into MACHO blue magnitude (M<sub>B</sub>) and red magnitude (M<sub>R</sub>) using the transforms given in Alcock et al. (1999).
Fourier analysis is known to be very good at detecting periodicity in datasets even in the presence of large noise. We have performed a Fourier analysis of the MACHO $`M_R`$ and $`M_B`$ datasets of the 636 objects with MUFRAN (multi-frequency analysis, Kollรกth 1990) in the frequency range 0.02 โ 0.98 d<sup>-1</sup>.
We first reduced the set of objects to those in which there are coincidences among the 8 largest Fourier peaks in MACHO $`M_R`$ and $`M_B`$. For the usual, large amplitude Cepheids the peaks are extremely sharp and these Cepheids are thus readily identified. Interestingly, there are a number of objects for which the peaks are not very pronounced, but nevertheless there are coincidences among the highest Fourier peaks. Each of these cases has had to be examined individually to ascertain that the detected variability is not spurious. Independently, we have used the Phase Dispersion Minimization routine PDM in IRAF (Stellingwerf 1978) to confirm the detected common MUFRAN frequency.
How can we be sure that the detected variability is real? One of the tests already mentioned is to compare the amplitude spectra of the red and the blue MACHO lightcurves which are to a large degree independent of each other. However, since both datasets were obtained in common observing conditions and at the same sampling points in time, they could potentially have a common spurious periodicity. Therefore, as a completely independent test, we have performed a Fourier analysis of the $`I`$ lightcurves obtained from OGLE-II and OGLE-III observations.
One possible source of a spurious common signal in both the MACHO and OGLE data could be a nearby large-amplitude Cepheid. In order to eliminate this possibility, we have checked the online MACHO variable star catalog and the OGLE-II catalog of variable stars in the LMC (Zebrun et al. 2001) for any variable within 8<sup>โฒโฒ</sup> of the low amplitude Cepheids. In no case did we find a nearby Cepheid of similar period.
To summarize then, an object with marginal periodicity had to show a prominent Fourier peak in OGLE $`I`$ as well as in MACHO red amplitude spectra and, in addition, it had to have a peak in MACHO blue at the same frequency. One final requirement was that the phase, as well as the period, of all three lightcurves was the same.
## 2. Results
In Fig. 1 we present an example of the comparison of the $`M_\mathrm{R}`$ and $`M_\mathrm{B}`$ data of the MACHO star 77.7430.18 and $`I`$ data of the corresponding OGLE star SC4 323401. The Fourier amplitude spectra of both the MACHO red and OGLE data show a very sharp peak near $`f_\mathrm{o}`$ = 0.30194 d<sup>-1</sup> (3.3121 d). The MACHO blue data appear to be much noisier, and by themselves would be rejected as just noise, but there is a peak at the same frequency. Note that the second sharp peak in the Fourier spectra, located at $`1f_\mathrm{o}`$, is the result of aliasing because of a large, but very sharp window peak at 1 d.
The right panels show the data folded with the common period of 3.3121 d. This common period has been determined by concatenating the MACHO red and the OGLE $`I`$ data and adjusting the zero-point of the OGLE data (on a yearly basis because of the sizeable yearly shifts in the OGLE data.) A comparison of the panels shows that the 3 datasets are in phase and that the variability is real. This is further confirmed by least-squares fitting of single frequency sinusoids to the data: the fits are displayed as thick solid lines. Thus, in all respects, this star appears as a very clear case of an ultra low amplitude (ULA) Cepheid.
Fig. 2 gives a second example of the comparison of the $`M_\mathrm{R}`$ and $`M_\mathrm{B}`$ data of the MACHO star 77.7428.36 and of the corresponding OGLE star SC4 296029. The Fourier amplitude spectra of both the MACHO red and OGLE $`I`$ data show a common, relatively sharp peak at $`f_\mathrm{o}`$ = 0.4495 d<sup>-1</sup> (2.224 d). The MACHO blue amplitude spectra appear noisier, although the common peak is the second largest in the blue spectrum. The diagrams on the right show the folded lightcurves together with the single frequency (sinusoidal) fit. Note that each of the 3 datasets on its own merit would be considered weak or marginal, but the occurrence of a common peak in all three, and the phase correlation between the oscillatory parts provides strong evidence that the periodic variability is real.
Following the above procedures we have identified the Cepheids in MACHO Field 77. Our 10 lowest amplitude Cepheids are displayed in the Table. The identification of periodic variability is solid for all objects. As far as the quoted amplitudes are concerned, they should be considered upper limits because the observational noise adds power to the peak. We estimate that the actual amplitudes could be 10โ20% lower in the noisiest cases.
In Fig. 3 we present our results in an amplitude histogram. Among the 144 variables that we have identified as Cepheids we find 14 that have amplitudes $`<0.05`$ mag in $`M_\mathrm{R}`$, of which 7 have an amplitude $`<0.006`$ mag: we call the latter group of stars ULA Cepheids. It is possible that out of caution we have discarded some additional ULA Cepheids because the signal to noise was too small.
Fig. 4a exhibits the periodโluminosity (PL) relation. To guide the eye we have added the slanted line, which is approximately parallel to the fundamental mode blue edge. It is defined by $`W`$ = $`\alpha \mathrm{Log}P`$ \+ $`W_\mathrm{o}`$, with $`\alpha `$ = โ3.3, $`W_\mathrm{o}`$ = โ16.01, where $`W=M_\mathrm{R}4(M_\mathrm{B}M_\mathrm{R})`$ is a reddening corrected magnitude (e.g. Alcock et al. 1995).
Fig. 4b gives the periodโamplitude relation for all the Cepheids. The crosses represent the Cepheids that lie below the slanted line of Fig. 4a.
On the grounds that a time related to pulsation, e.g., the amplitude growth time $`\tau _{gr}`$, is much shorter than the stellar evolution time $`\tau _{evol}`$, one might expect a star to always achieve its full limit cycle pulsation amplitude $`A_{LC}`$, i.e., the amplitude that a standard hydrodynamics code would compute. If that were indeed the case then the detection of any Cepheids with very small amplitudes upon entering or exiting the IS would be very unlikely, because at the boundaries of the IS the limit cycle amplitude sets in, respectively decays, with a vertical slope ($`dA_{LC}/dt=\mathrm{}`$; cf. Fig. 5.) An amplitude histogram would therefore be expected to be devoid of small amplitude Cepheids.
However, we have identified a surprisingly large number ($`5\%`$) of ULA Cepheids ($`<0.006`$mag). These stars seem to fall preferentially near the edges of the Cepheid IS in a PโL diagram. (Ideally, a colorโmagnitude plot should also show this feature, but it has too much scatter because color is uncertain by at least $`\pm `$0.05 mag due to reddening variations (Keller & Wood 2002) while the whole width of the IS is less than 0.2 mag.) It turns out that a recent theoretical development predicted the existence of such ULA Cepheids (Buchler & Kollรกth 2002), and provided the incentive for this search.
At the edges of the IS the amplitude growth rate of a star actually vanishes, so that $`\tau _{gr}`$ can be much longer than $`\tau _{evol}`$ in the immediate vicinity of the IS. As a consequence, as the star enters the IS, the amplitude does not immediately achieve its full pulsation amplitude. Rather it stays at a very low amplitude for a time of the order of $`\sqrt{\tau _{evol}\times \tau _{gr}}`$, but then rapidly grows to full amplitude. Therefore very few Cepheids should be detectable during that rapid amplitude growth, and one would expect to find a gap in the amplitude distribution between ULA Cepheids and the (usual) full limit cycle amplitudes. On their way out of the IS, for the same reason the amplitude should not decay to zero right at the edge of the IS, but the stars should linger in a state of small amplitude. This behavior is sketched in Fig. 5, on the left for a star entering the IS, and on the right for a star leaving the IS. Stellar evolution calculations further show that Cepheids cross the IS in both directions. ULA Cepheids should therefore be found in the vicinity of both the blue and the red edges of the IS. (Note that a similar behavior is predicted for RR Lyrae.)
The existence of ULA Cepheids in the vicinities of the blue and red edges is thus fully compatible with theoretical predictions (when they are interpreted as radial pulsators). A more quantitative comparison of the Cepheid amplitude distribution should be made with the construction of a synthetic HR diagram that would combine stellar evolutionary tracks with the results of nonlinear hydrodynamic simulations along the lines of Szabรณ, Kollรกth & Buchler (2004).
A second theoretical development is the prediction by Buchler et al. (1997) of the self-excitation of radial overtone modes in Cepheids and RR Lyrae that are predominantly surface modes. They dubbed these modes โstrange modesโ because of their similarity to the strange modes found by Wood (1976) in high luminosity stars (see also Saio, Wheeler & Cox, 1984). This brings us to a discussion of the two egregious stars, namely 77.6940.14 and 77.7548.11. They appear as numbers 9 and 10 in the Table and are indicated by triangles in Figs. 4aโb.
First we have ascertained that they are not contaminated by nearby objects. But that still leaves the possibility of binarity or rotation of a star with spots. Using bolometric corrections and $`T_{\mathrm{ef}\mathrm{f}}`$ from Bessell & Germany (1999), a distance modulus 18.55 and $`A_v`$= 0.18 (Keller & Wood 2002; Marconi & Clementini 2005), we find for star 77.6940.14, $`L`$=2510 $`L_{}`$, $`M`$=5 $`M_{}`$, $`T_{\mathrm{ef}\mathrm{f}}`$=6200 K and $`R`$= 44$`R_{}`$; and for star 77.7548.11: $`L`$ =6730 $`L_{}`$, $`M`$= 6.5 $`M_{}`$, $`T_{\mathrm{ef}\mathrm{f}}`$ = 5400 K, $`R`$=94 $`R_{}`$. Given the observed periods $`P`$=1.32 d and $`P`$= 6.03 d, respectively, the corresponding orbital radius of any (light) companion would be $`R_{orb}`$=9 $`R_{}`$ and $`R_{orb}`$= 26 $`R_{}`$, or 0.14 and 0.28 times the stellar radius. Under the above assumptions, the small orbital radii required definitively rule out the binary hypothesis. Similarly, rotation can also be eliminated as an explanation because the rotation velocity derived from the observed period and stellar radius is greater than the rotational breakup velocity (by factors of 11.4 and 6.8, respectively). We note, however, that the above arguments fail if these two stars are foreground stars of much lower luminosity than assumed. Unfortunately, these stars are in a region of the LMC HR-diagram where foreground contamination is large (the number of LMC and foreground stars are comparable - see Alcock et al. 2000). Radial velocity measurements would be required to confirm LMC membership. However, the fact that they fall this close to the Cepheid PL relation suggests that they are probably LMC members.
Assuming these two stars are in the LMC, pulsation in strange modes is the likely explanation for their variability. The periods of the two stars are indeed a factor of 5โ6 lower than the periods of F Cepheid periods of equal apparent magnitude, and in agreement with the expected periods of Strange Cepheids (Buchler & Kollath, 2001). The amplitudes, $``$0.03 mag, of these stars are perhaps a little larger than the theoretical estimates, but they are within their uncertainty. A search for Strange Cepheids in the remaining MACHO LMC fields would be desirable as it would strengthen the status of this novel type of Cepheids.
One of us (JRB) gratefully acknowledges the hospitality of Mount Stromlo Observatory. We wish to thank Zoltรกn Kollรกth for providing us with the MUFRAN code. This work has been supported by NSF (AST-0307281, OISE-0417772) at the University of Florida. This paper utilizes public domain data obtained by the MACHO Project, jointly funded by the US DoE through the LLNL at the University of California (W-7405-Eng-48), by the NSF through the UC Center for Particle Astrophysics (AST-8809616), and by the Mount Stromlo and Siding Spring Observatory, part of the ANU. Support for OGLE was provided by Polish MNII (2P03D02124), NSF (AST-0204908) and NASA (NAG5-12212).
|
warning/0506/quant-ph0506102.html
|
ar5iv
|
text
|
# Classical-interference analog of quantum fluctuations for bound-state soliton pairs
## Abstract
Quantum photon-number fluctuation and correlation of bound soliton pairs in mode-locked fiber lasers are studied based on the complex Ginzburg-Landau equation model. We find that, depending on their phase difference, the total photon-number noise of the bound soliton pair can be larger or smaller than that of a single soliton and the two solitons in the soliton pairs are with positive or negative photon-number correlation, correspondingly. It is predicted for the first time that out-of-phase soliton pairs can exhibit less noises due to negative correlation.
Optical solitons, Nonlinear guided waves
Quantum solitons have attracted a great deal of research interest in the contexts of nonlinear quantum optics, condensed-matter physics, and quantum information science due to their remarkable nonclassical properties. In particular, quantum solitons in optical fibers largely resemble their classical counterparts, but with additional quantum fluctuations around the mean fields. It has been possible to achieve squeezing through quantum solitons in optical fibers, Carter87 ; Drummond87 ; Lai89a ; Lai89b and they may also serve as a new platform for quantum information applications Silberhorn01 ; Silberhorn02 ; Konig02 .
Quantum solitons are macroscopic optical wave packets which offer a testbed for quantum optics and quantum field theories. For the quantum nonlinear Schrรถdinger equation (NLSE), exact soliton states can be constructed as combinations of eigenstates of the Hamiltonian of the one-dimensional Bose gas with $`\delta `$-like (contact) interaction through the Bethe ansatz method Lai89b . In the large photon number limit, which corresponds to the usual optical solitons generated by lasers, the many photon wave function of the quantum soliton is well approximated by a single-photon wave function (the Hartree approximation) Lai89a . Linearization around such a soliton Haus90 ; Lai93 successfully explains experimental observations of quantum fluctuations for temporal fiber solitons, provided that optical loss and higher-order effects are negligible Rosenbluh ; Bergman91 ; Friberg ; Krylov99 ; Spalter .
It is well known that the force between adjacent solitons in the NLSE model is attractive or repulsive, depending on the phase difference between them Agrawal95 . Stationary bound soliton states in this conservative model do not exist. Formation of effectively stable double-, triple-, and multi-soliton bound states was predicted in models based on the complex Ginzburg-Landau equation (CGLE) Malomed91 ; Akhmediev96 ; Soto-Crespo03 , and observed experimentally in various passively mode-locked fiber lasers Tang01 ; Seong02 ; Grelu03 . The separation between the solitons in these bound states are โquantizedโ, taking a set of discrete values. The amplitude noise in triplet bound states generated by a stretched-pulse ytterbium-doped double-clad fiber laser was observed to be reduced compared to the single soliton pulse Ortac04 . It is an issue of straightforward interest to study the noise of these bound solitons, and to understand why the mode-locked fiber lasers operate more stably in the bound-state regime.
The passively mode-locked fiber lasers are quite accurately described by the cubic-quintic CGLE. In a normalized form, the equation is
$`iU_z+(D/2)U_{tt}+|U|^2U`$ $`=`$ $`i\delta U+iฯต|U|^2U+i\beta U_{tt}`$ (1)
$`+`$ $`i\mu |U|^4U\nu |U|^4U,`$
where $`U`$ is the local amplitude of the electromagnetic wave, $`z`$ is the propagation distance, $`t`$ is the retarded time, and $`D=+1`$ and $`1`$ correspond, respectively, to the anomalous and normal dispersion. Besides the group-velocity dispersion (GVD) and the Kerr effect, which are accounted for by conservative terms on the left-hand side of Eq. (1), the model also includes the quintic correction to the Kerr nonlinearity, through the coefficient $`\nu `$, and non-conservative terms. The coefficients $`\delta `$, $`ฯต`$, $`\mu `$, and $`\beta `$ account for the linear, cubic, and quintic loss or gain, and spectral filtering, respectively.
In the CGLE model, with suitable parameters degenerate bound-state soliton pairs are known to exist through the balance between the gain and loss, in the form Malomed91 ; Akhmediev96 , $`U(z,t)=U_0(z,t+\rho )e^{i\theta /2}+U_0(z,t\rho )e^{i\theta /2}`$, where $`U_0`$ is a single soliton solution, and $`\rho `$ and $`\theta `$ are the separation and phase difference between the solitons. In this Letter, we focus on the consideration of three fundamentally different cases, corresponding to the bound states with the same separation and amplitude, and $`\theta =0`$, $`\pi /2`$, and $`\pi `$ (the in-phase, orthogonal, and out-of-phase pair), respectively.
We compute the quantum fluctuations of these soliton pairs by dint of a numerically implemented back-propagation method Lai95 , which may be summarized as follows. First of all, we replace the classical function $`U(z,t)`$ in Eq. (1) by the quantum-field operator variable, $`\widehat{U}(z,t)`$, which satisfies the equal-coordinate Bosonic commutation relations. Next, the equation is linearized around the classical solution through the substitution of $`\widehat{U}(z,t)=U_0(z,t)+\widehat{u}(z,t)`$, assuming large photon numbers in the solitons. Then, a zero-mean additional noise operator, $`\widehat{n}(z,t)`$, is introduced to make the quantum perturbation fields in the linearized equation satisfy the Bosonic communication relations (see Ref. \[References\] for more details). By imposing suitable correlation functions for the noise operator, the minimum quantum noise in the considered nonconservative model is introduced. Therefore the results presented here represent a lower limit required by the fundamental principles of quantum mechanics.
Figure 1 shows the photon-number correlation parameter for the two solitons in the bound soliton pair, which is defined as
$$C_{12}=\frac{:\mathrm{\Delta }\widehat{N}_1\mathrm{\Delta }\widehat{N}_2:}{\sqrt{\mathrm{\Delta }\widehat{N}_1^2\mathrm{\Delta }\widehat{N}_2^2}}.$$
Here, the colons stand for the normal ordering of the operators and $`\mathrm{\Delta }\widehat{N}_{1,2}`$ are perturbations of the photon-number operators for the two solitons, which are numbered (1,2) according to their position in the time domain. Initially, the two solitons are assumed to be uncorrelated, with fluctuations around each soliton obeying the coherent-state statistics. For the in-phase pair, the photon-number correlation between the solitons gradually increases to positive values and eventually saturates around $`C_{12}=0.36`$. But for the out-of-phase pair, $`C_{12}`$ gradually decreases to negative values and then saturates too. In between, the correlation parameter for the case of $`\theta =\pi /2`$ remains close to zero as long as the computation is run. For the former two cases, the saturation of the photon-number correlation parameter is due to the nonconservative effects in the CGLE model.
To further demonstrate the behavior difference of the photon-number correlation for soliton pairs with different relative phases, in Fig. 2 we display the time-domain photon-number correlation patterns for them. The plotted correlation coefficients, $`\eta _{ij}`$, are defined through the normally-ordered covariance,
$$\eta _{ij}\frac{:\mathrm{\Delta }\widehat{n}_i\mathrm{\Delta }\widehat{n}_j:}{\sqrt{\mathrm{\Delta }\widehat{n}_i^2\mathrm{\Delta }\widehat{n}_j^2}},$$
(2)
where $`\mathrm{\Delta }\widehat{n}_j`$ is the photon-number fluctuation in the $`j`$-th time slot $`\mathrm{\Delta }t_j`$,
$$\mathrm{\Delta }\widehat{n}_j=_{\mathrm{\Delta }t_j}๐t[U_0(z,t)\widehat{u}^{}(z,t)+U_0^{}(z,t)\widehat{u}(z,t)].$$
Here the integral is taken over the given time slot, with the same time-division length $`\mathrm{\Delta }t`$. Clearly, in Fig. 2 (A) one can see that there is a strong *positive*-correlation band connecting the quantum correlation patterns of the bound solitons when they are in phase, $`\theta =0`$. In Fig. 2 (C) there exists a *negative*-correlation pattern between two solitons for the out-of-phase case, $`\theta =\pi `$. Moreover, for the case of $`\theta =\pi /2`$, in Fig. 2 (B), the correlation patterns of bound solitons are almost isolated. In classical physics, in-phase and out-of-phase fields will lead respectively to the constructive and destructive interference. Here we observe a similar effect for the quantum noises. What is more important, in Fig. 3 we compute the total photon number noise of the bound soliton pair and compare it to the case of a single soliton (these results are amenable to straightforward experimental verification). As one may expect, the photon-number noise of the in-phase pair is larger than that for the single soliton, which may be explained as the fluctuation enhancement due to constructive interference. On the other hand, the noise is reduced for the case of out-of-phase pair as the result of destructive interference. The orthogonal soliton pair with $`\theta =\pi /2`$ may be viewed, in the first approximation, as independent two single solitons, which explains why it features almost the same noise level as the single soliton, even though small oscillation of the noise level originated from the residual interaction between the two solitons can still be seen.
In conclusion, we have presented theoretical results on the photon-number correlation and total photon-number noise for bound-state soliton pairs in the model of complex cubic-quintic Ginzburg-Landau equation. The cases of the in-phase, orthogonal, and out-of-phase soliton pairs have been considered in detail. We conclude that the interference of the quantum fluctuations in the soliton pair is constructive or destructive depending on the *classical* relative phase of the solitons. An important consequence of the results is that the operation regime of the fiber laser should be more stable when it is based on the *out-of-phase* soliton pairs.
|
warning/0506/math0506236.html
|
ar5iv
|
text
|
# Convergence of Quantum Cohomology by Quantum Lefschetz
## 1. Introduction
Quantum cohomology is a deformation of the ring structure of the ordinary cohomology. The structure constants of quantum cohomology are formal power series whose coefficients consist of Gromov-Witten invariants. We do not know a priori whether or not the structure constants are convergent. In this paper, we discuss the compatibility of quantum Lefschetz principle and the convergence of quantum cohomology.
There are several cases where the convergence is trivial. If $`c_1(X)>0`$, the small quantum cohomology of $`X`$ is defined over the polynomial ring by the degree constraints. If $`c_1(X)<0`$, even the big quantum cohomology is defined over the polynomial ring for the same reason. Hence, the problem is the intermediate case, i.e. when there exist two curves $`C_1,C_2`$ in $`X`$ such that $`c_1(X),[C_1]0`$ and $`c_1(X),[C_2]0`$. The main theorem in this paper is the following.
###### Theorem 1.1.
Let $`X`$ be a smooth projective variety and $``$ be a nef line bundle on $`X`$. If the big quantum cohomology $`QH^{}(X)`$ of $`X`$ has convergent structure constants, then the twisted quantum cohomology $`QH_{S^1}^{}(X,)`$ by $``$ also has convergent structure constants. (This holds true when $``$ is replaced by a sum of nef line bundles. )
Coates-Giventalโs quantum Lefschetz theorem gives the twisted quantum cohomology $`QH_{S^1}^{}(X,)`$ in terms of $`QH^{}(X)`$. Here, $`QH_{S^1}^{}(X,)`$ is a cohomology theory closely related to the quantum cohomology of an intersection $`YX`$ with respect to the line bundle $``$. More precisely, $`QH_{S^1}^{}(X,)`$ gives us the information on the structure constants of $`QH^{}(Y)`$ with respect to the cohomology classes coming from the ambient space $`X`$. Therefore, if the convergence of $`QH^{}(X)`$ is known, we can know the convergence of $`QH^{}(Y)`$ partially. The main tool in the proof is a ring of formal power series with certain estimates for coefficients.
In the second half of the paper, we give a description of mirror symmetry for a not necessarily nef toric variety. In , the author calculated the quantum cohomology $`D`$-module of a toric variety $`X`$. The method there was to embed $`X`$ into another Fano toric variety $`X^{}`$ as a complete intersection and to use quantum Lefschetz theorem together with a mirror theorem for a Fano toric variety . We will recast the consequences of in terms of the following oscillatory integral, which was introduced as a mirror of a toric variety in :
$$_\mathrm{\Gamma }(q_1,\mathrm{},q_r,\mathrm{})=_{\mathrm{\Gamma }_qY_q}e^{_{i=1}^{r+N}๐_i/\mathrm{}}\omega _q,Y_q=\left\{(๐_i)_{i=1}^{r+N}(^{})^{r+N};_{i=1}^{r+N}๐_i^{m_{ia}}=q_a\right\}.$$
Here, $`\omega _q`$ is a holomorphic volume form on $`Y_q`$ and $`\mathrm{\Gamma }_q`$ is a non-compact cycle. These oscillatory integrals define the following mirror $`D`$-module $`M_{\mathrm{mir}}`$ (denoted by $`FH_0`$ in the main text):
$$M_{\mathrm{mir}}:=q_1,\mathrm{},q_r,\mathrm{}_1,\mathrm{},\mathrm{}_r,\mathrm{}/I_{\mathrm{poly}},_a=q_a/q_a.$$
Here, $`I_{\mathrm{poly}}`$ is a left ideal consisting of polynomial differential operators annihilating $`_\mathrm{\Gamma }(q,\mathrm{})`$. Mirror symmetry for a Fano toric variety $`X`$ states that the mirror $`D`$-module $`M_{\mathrm{mir}}`$ is isomorphic to the big quantum cohomology $`D`$-module $`QDM^{}(X)`$ restricted to $`H^2(X)`$. Here, $`QDM^{}(X)`$ is a $`D`$-module over the total cohomology ring $`H^{}(X)`$ which is defined by $`QH^{}(X)`$. For a general toric variety $`X`$, we obtain the following description:
###### Theorem 1.2 (see Theorem 5.6 for details).
Let $`\widehat{M}_{\mathrm{mir}}`$ be a $`D`$-module on the formal germ $`(^r,0)`$ obtained as the completion of $`M_{\mathrm{mir}}`$ with respect to its natural $`q`$-adic topology (denoted by $`FH_{S^1}^{}`$ in the main text). There exists a formal embedding $`๐ข๐ช๐:(^r,0)(H^{}(X),0)`$ such that we have an isomorphism of $`D`$-modules:
$$\mathrm{\Phi }_{๐ข๐ช๐}:๐ข๐ช๐^{}(QDM^{}(X))\widehat{M}_{\mathrm{mir}}.$$
Here, $`QDM^{}(X)`$ is the big quantum $`D`$-module of the toric variety $`X`$. If $`X`$ is Fano, the image of $`๐ข๐ช๐`$ coincides with the linear subspace $`H^2(X)H^{}(X)`$.
Because of the completion in the above description, it is not clear if $`QH^{}(X)`$ is convergent. Our main theorem 1.1 is not directly applicable to $`X`$ because $`X`$ is a complete intersection in $`X^{}`$ with respect to a sum of not necessarily nef line bundles. Using techniques similar to the proof of Theorem 1.1, however, we show the following:
###### Theorem 1.3 (Theorem 5.7, Corollary 5.12).
The big quantum cohomology of a smooth projective toric variety is convergent and generically semisimple. The embedding $`๐ข๐ช๐`$ in the above theorem is complex analytic.
Note that the isomorphism $`\mathrm{\Phi }_{๐ข๐ช๐}`$ is not convergent unless $`X`$ is nef (Proposition 5.13). The asymptotic expansion of the oscillatory integral $`_\mathrm{\Gamma }(q,\mathrm{})`$ in $`\mathrm{}`$ is shown to give a formal solution to $`QDM^{}(X)`$ for special choices of cycles $`\mathrm{\Gamma }`$ (Corollary 6.9). We also prove the $`R`$-conjecture for equivariant quantum cohomology of toric varieties (Theorem 6.10). Here, the $`R`$-conjecture implies the Virasoro constraints by Giventalโs theory .
Our result on the semisimplicity is also a successful test for Bayer and Maninโs modified Dubrovinโs conjecture . The modified Dubrovinโs conjecture claims that $`(p,p)`$-part of quantum cohomology of a projective variety $`X`$ is generically semisimple if its bounded derived category $`D_{\mathrm{coh}}^b(X)`$ of coherent sheaves admits a full exceptional collection. In fact, Kawamata recently showed that toric varieties have full exceptional collections .
For the application of the main theorem 1.1, we need to know the convergence of big quantum cohomology of ambient spaces. We prove that if $`H^{}(X)`$ is generated by $`H^2(X)`$ and if the small quantum cohomology of $`X`$ has convergent structure constants, so does the big quantum cohomology (Corollary 5.9). In particular, the big quantum cohomology of a Fano variety with $`H^2`$-generated cohomology always has convergent structure constants. In a subsequent paper , we will also prove that the big quantum cohomology (and higher genus potential $`_g`$ also) is convergent for a projective manifold which admits Hamiltonian torus action with only isolated fixed points and isolated one dimensional orbits.
We should remark that in this paper, we only consider the even part of (quantum) cohomology. For example, $`H^{}(X)`$ always means $`H^{\mathrm{even}}(X)`$.
The paper is organized as follows. In section 2 and 3, we review the quantum $`D`$-modules and the quantum Lefschetz theorem by Coates and Givental. In section 4, we prove the main theorem. In section 5, we discuss the mirror symmetry for a non-nef toric variety. In section 6, we prove the $`R`$-conjecture for any toric variety.
Acknowledgments. Thanks are due to Professor Martin Guest, Professor Hiraku Nakajima and Kazushi Ueda for valuable discussions. Part of this paper was written while the author stayed at Mathematical Sciences Research Institute. He thanks MSRI for excellent working conditions. He is also grateful to anonymous referee for thier helpful comments. This research is supported by Grant-in-Aid for JSPS Fellows and Scientific Research 15-5482.
## 2. Quantum $`D`$-modules
In this section, we introduce quantum $`D`$-modules twisted by the equivariant Euler class following . Let $`X`$ be a smooth projective variety and $``$ be a line bundle over $`X`$. Let $`\overline{M}_{0,n}(X,๐
)`$ be the moduli space of genus 0, degree $`๐
`$ stable maps to $`X`$ with $`n`$ marked points, where $`๐
H_2(X,)`$. We have the following diagram:
$$\begin{array}{ccc}\overline{M}_{0,n+1}(X,๐
)& \stackrel{e_{n+1}}{}& X\\ \pi _{n+1}& & \\ \overline{M}_{0,n}(X,๐
)\end{array}$$
where $`e_i`$ is the evaluation map and $`\pi _i`$ is the forgetful map. We introduce the fiber-wise $`S^1`$ action on $``$ by scalar multiplication. Let $`\lambda `$ be a generator of the $`S^1`$ equivariant cohomology of a point. Define the twisted correlator
$$\alpha _1,\mathrm{},\alpha _n_{S^1,๐
}^{}=_{[\overline{M}_{0,n}(X,๐
)]^{\mathrm{virt}}}\underset{i=1}{\overset{n}{}}e_i^{}(\alpha _i)\mathrm{Euler}_{S^1}(R^{}\pi _{n+1}^{}{}_{}{}^{}e_{n+1}^{}),$$
where $`\alpha _1,\mathrm{},\alpha _nH^{}(X,)`$ and $`[\overline{M}_{0,n}(X,๐
)]^{\mathrm{virt}}`$ is the virtual fundamental class. The right hand side is in $`[\lambda ,\lambda ^1]`$. Let $`\{p_0,\mathrm{},p_s\}`$ be a basis of $`H^{}(X,)`$. We assume that $`p_0`$ is a unit and that $`p_1,\mathrm{},p_r`$ form a nef integral basis of $`H^2(X,)`$ ($`rs`$). Let $`t_0,\mathrm{},t_s`$ be linear coordinates dual to the basis $`p_0,\mathrm{},p_s`$. We write $`q_a:=\mathrm{exp}(t_a)`$ and $`q^๐
:=q_1^{p_1,๐
}q_2^{p_2,๐
}\mathrm{}q_r^{p_r,๐
}=\mathrm{exp}(_{a=1}^rp_a,๐
t_a)`$ for $`๐
H_2(X,)`$. Define a twisted pairing $`,_{S^1}^{}`$ by
$$\alpha ,\beta _{S^1}^{}=_X\alpha \beta \mathrm{Euler}_{S^1}().$$
The twisted quantum product $`_{}`$ is defined by the formula:
$`\alpha _{}\beta ,\gamma _{S^1}^{}`$ $`={\displaystyle \underset{๐
\mathrm{\Lambda }}{}}{\displaystyle \underset{n0}{}}{\displaystyle \frac{1}{n!}}\alpha ,\beta ,\gamma ,(_{j=0}^st_jp_j)^n_{S^1,๐
}^{}`$
$`={\displaystyle \underset{๐
\mathrm{\Lambda }}{}}{\displaystyle \underset{n0}{}}{\displaystyle \frac{1}{n!}}\alpha ,\beta ,\gamma ,(_{j=r+1}^st_jp_j)^n_{S^1,๐
}^{}q^๐
,`$
where $`\alpha ,\beta ,\gamma H^{}(X,)`$ and $`\mathrm{\Lambda }H_2(X,)`$ is a semigroup generated by effective curves. The product $`_{}`$ is extended linearly over $`[\lambda ,\lambda ^1][[t_0,q_1,\mathrm{},q_r,t_{r+1},\mathrm{},t_s]]`$ and
$$QH_{S^1}^{}(X,):=(H^{}(X)[\lambda ,\lambda ^1][[t_0,q_1,\mathrm{},q_r,t_{r+1},\mathrm{},t_s]],_{})$$
becomes an associative and commutative ring. We define $`\mathrm{deg}t_j=2\mathrm{deg}p_j`$ for $`j=0`$ or $`j>r`$, $`\mathrm{deg}q^๐
=2c_1(X)c_1(),๐
`$ and $`\mathrm{deg}\lambda =2`$. Then $`QH_{S^1}^{}(X,)`$ becomes a graded ring. For simplicity, we also use the following notation:
$$x=(x_0,x_1,\mathrm{},x_r,x_{r+1},\mathrm{},x_s)=(t_0,q_1,\mathrm{},q_r,t_{r+1},\mathrm{},t_s),\tau =\underset{j=0}{\overset{s}{}}t_jp_j.$$
When $`c_1()`$ is nef, the product $`_{}`$ can be defined over $`[\lambda ][[x]]`$ (see ). Therefore in this case, we can also consider the non-equivariant ($`\lambda =0`$) version $`QH^{}(X,)=(H^{}(X)[[x]],_{})`$.
The usual (non-twisted) quantum cohomology $`QH^{}(X)=(H^{}(X)[[x]],_X)`$ is defined by removing all the Euler classes in the definition. Let $`Y`$ be a smooth intersection in $`X`$ with respect to a transverse section $`s\mathrm{\Gamma }(X,)`$. By the main theorem of , if $`c_1()`$ is nef, we have at $`\lambda =0`$,
$$\alpha _1,\mathrm{},\alpha _n_๐
^{}=i^{}\alpha _1,\mathrm{},i^{}\alpha _n_๐
^Y$$
for the inclusion $`i:YX`$ and the correlator $`\mathrm{}_๐
^Y`$ for $`Y`$. Therefore, from $`_{}`$, we can read the structure constants of $`QH^{}(Y)`$ with respect to the classes coming from the ambient space.
We can endow a $`D`$-module structure on the quantum cohomology. Let $``$ denote $`_{}`$ or $`_X`$. The dual Givental connection $`^{\mathrm{}}`$ is defined by
$$_j^{\mathrm{}}=\mathrm{}\frac{}{t_j}+p_j(0js),$$
where $`\mathrm{}`$ is a formal variable of degree two. This connection is regular singular along $`q_1=\mathrm{}=q_r=0`$ and is known to be flat. It defines the non-twisted or twisted quantum $`D`$-modules:
$$QDM^{}(X)=(H^{}(X)[\mathrm{}][[x]],^{\mathrm{}}),QDM_{S^1}^{}(X,)=(H^{}(X)[\mathrm{},\lambda ,\lambda ^1][[x]],^{\mathrm{}})$$
When $`c_1(L)`$ is nef, we can also consider the non-equivariant version $`QDM^{}(X,)`$. It is easy to see that quantum $`D`$-module is generated by $`1`$ over the Heisenberg algebra $`[\mathrm{}][[x]][_0^{\mathrm{}},\mathrm{},_s^{\mathrm{}}]`$. There exists a unique fundamental solution $`L(\tau ,\mathrm{})`$ for the flat connection $`^{\mathrm{}}`$ such that
(1)
$$\mathrm{}dL(\tau ,\mathrm{})=L(\tau ,\mathrm{})^{\mathrm{}},$$
$$L(\tau ,\mathrm{})=e^{\tau /\mathrm{}}T(x,\mathrm{}),T\mathrm{End}(H^{}(X))[\mathrm{}^1][[x]],T|_{q=0}=\mathrm{id}.$$
where $`\tau H^{}(X)`$ is considered as an operator acting on $`H^{}(X)`$ by the cup product. This $`L(\tau ,\mathrm{})`$ satisfies the following unitarity :
(2)
$$L(\tau ,\mathrm{})\alpha ,L(\tau ,\mathrm{})\beta =\alpha ,\beta $$
where $`,`$ denotes the Poincarรฉ pairing $`,^X`$ of $`X`$ in case of $`QDM^{}(X)`$ and the twisted Poincarรฉ pairing $`,_{S^1}^{}`$ in case of $`QDM_{S^1}^{}(X,)`$. This $`L(\tau ,\mathrm{})`$ also defines the $`J`$-function of quantum cohomology by
$$J(\tau ,\mathrm{}):=L(\tau ,\mathrm{})1,1H^{}(X)\text{ is a unit.}$$
The $`J`$-function is a realization of a generator $`1`$ of the quantum $`D`$-module as a function.
## 3. Coates-Giventalโs Quantum Lefschetz
Coates and Giventalโs quantum Lefschetz theorem describes the relationship between the two quantum cohomologies $`QH^{}(X)`$ and $`QH_{S^1}^{}(X,)`$. It was described in terms of symplectic transformations of Lagrangian cones in the infinite dimensional space $`H^{}(X)[[\mathrm{},\mathrm{}^1]]`$. In this paper, we describe it in terms of gauge transformations by translating the language of cones into that of quantum $`D`$-modules.
### 3.1. Symplectic formalism
First we will review the infinite dimensional symplectic formalism in briefly. Consider the following general multiplicative characteristic class for a complex vector bundle $`E`$:
$$๐(E)=\mathrm{exp}\left(\underset{k=0}{\overset{\mathrm{}}{}}s_k\mathrm{ch}_k(E)\right).$$
Here, $`๐=(s_0,s_1,s_2,\mathrm{})`$ are infinite number of arbitrary parameters. Let $``$ be the following infinite dimensional vector space:
$$:=H^{}(X)[\mathrm{},\mathrm{}^1][[๐]]=_+_{},$$
where
$$_+=H^{}(X)[\mathrm{}][[๐]],_{}=\mathrm{}^1H^{}(X)[\mathrm{}^1][[๐]].$$
For any holomorphic vector bundle $`E`$ on $`X`$, define a symplectic form $`\mathrm{\Omega }_๐`$ on $``$ by
$$\mathrm{\Omega }_๐(f(\mathrm{}),g(\mathrm{}))=\mathrm{Res}_{\mathrm{}=0}d\mathrm{}_Xf(\mathrm{})g(\mathrm{})๐(E),f,g.$$
Then, $`_+`$ and $`_{}`$ become Lagrangian with respect to this symplectic form $`\mathrm{\Omega }_๐`$ and give a polarization of $``$. The fundamental solution $`L(\tau ,\mathrm{})`$ of the quantum $`D`$-module $`QDM^{}(X)`$ (see (1)) defines a Lagrangian subspace $`๐_\tau `$ of $`(,\mathrm{\Omega }_0)`$ for each $`\tau `$.
(3)
$$๐_\tau :=L(\tau ,\mathrm{})(_+)$$
Here, we assume the convergence of $`L(\tau ,\mathrm{})`$ for the sake of simplicity. For a rigorous argument, we need to introduce a Novikov ring $`\mathrm{\Lambda }_{\mathrm{nov}}`$ and replace $``$ with the module $`H^{}(X,\mathrm{\Lambda }_{\mathrm{nov}}[[๐]])\{\mathrm{},\mathrm{}^1\}`$ of convergent power series in $`\mathrm{}`$ with respect to the adic topology (as explained in ). These semi-infinite subspaces sweep a germ of Lagrangian cone $`๐_0`$ in $`(,\mathrm{\Omega }_0)`$.
(4)
$$๐_0:=\underset{\tau H^{}(X)}{}\mathrm{}๐_\tau $$
As explained in , the tangent space of $`๐_0`$ at any point in $`\mathrm{}๐_\tau `$ equals the Lagrangian subspace $`๐_\tau `$. In other words, $`๐_0`$ has a remarkable property that it is ruled by $`\mathrm{}`$ times its tangent spaces. Let $`J(\tau ,\mathrm{})=L(\tau ,\mathrm{})1`$ be the $`J`$-function of $`QH^{}(X)`$. Then the vectors $`\mathrm{}J(\tau ,\mathrm{})`$ parametrized by $`\tau H^{}(X)`$ lie on the cone $`๐_0`$. The derivatives $`\{\mathrm{}_{t_j}J(\tau ,\mathrm{})\}_j`$ form a basis of the tangent space $`๐_\tau `$ of $`๐_0`$ over $`[\mathrm{}][[๐]]`$. Thus, the $`J`$-function recovers the whole Lagrangian cone $`๐_0`$ by (4).
Similarly, the twisted theory by the characteristic class $`๐`$ and a vector bundle $`E`$ defines a Lagrangian cone $`๐_๐(,\mathrm{\Omega }_๐)`$. Coates and Givental proved that two cones $`๐_0`$ and $`๐_๐`$ are related by a linear symplectic transformation.
###### Theorem 3.1 (, Corollary 4).
The linear symplectic transformation
$$๐(E)^{1/2}\mathrm{exp}\left(\underset{l,k0}{}s_{2k+l1}\frac{B_{2k}}{(2k)!}\mathrm{ch}_l(E)\mathrm{}^{2k1}\right):(,\mathrm{\Omega }_0)(,\mathrm{\Omega }_๐)$$
sends the Lagrangian cone $`๐_0`$ to $`๐_๐ฌ`$. Here, $`B_{2k}`$ is the Bernoulli number defined by $`x/(1e^x)=x/2+_{k0}B_{2k}x^{2k}/(2k)!`$ and $`s_1=0`$.
In this paper, we only consider a twist by the equivariant Euler class and the case where $`E`$ is a line bundle $``$. In this case, values of the parameters $`s_i`$ are set as follows:
(5)
$$s_0=\mathrm{log}\lambda ,s_k=(1)^{k1}\frac{(k1)!}{\lambda ^k}k>0.$$
Using this substitution, $`๐()`$ equals the equivariant Euler class $`c_1()+\lambda `$, where $`S^1`$ acts on $``$ by scalar multiplication on each fiber and $`\lambda `$ is a generator of $`H_{S^1}^{}(\mathrm{pt})`$. Coates and Givental introduced the following hypergeometric modification of the $`J`$-function:
$$I_{}(\tau ,\mathrm{}):=\underset{๐
\mathrm{\Lambda }}{}\frac{_{k=\mathrm{}}^{\rho ,๐
}(\rho +k\mathrm{}+\lambda )}{_{k=\mathrm{}}^0(\rho +k\mathrm{}+\lambda )}J_๐
(t,\mathrm{})q^๐
,\rho :=c_1(),$$
where $`J(\tau ,\mathrm{})=_{๐
\mathrm{\Lambda }}J_๐
(t,\mathrm{})q^๐
`$ is the $`J`$-function of $`QH^{}(X)`$.
###### Theorem 3.2 (\[4, Theorem 2\]).
The vectors $`\mathrm{}I_{}(\tau ^{},\mathrm{})`$ parametrized by $`\tau ^{}H^{}(X)`$ lie on the cone $`๐_\lambda `$ defined by the twisted quantum cohomology $`QH_{S^1}^{}(X,)`$. The derivatives $`\{\mathrm{}_{t_j}I_{}(\tau ^{},\mathrm{})\}_j`$ span the tangent spaces to $`๐_\lambda `$ and reconstruct the whole cone $`๐_\lambda `$.
### 3.2. Symplectic transformation as a gauge transformation
Following \[17, section 5\], we will interpret the linear symplectic transformation in Theorem 3.1 as a gauge transformation of quantum $`D`$-modules. The quantum $`D`$-module $`QDM^{}(X)=(H^{}(X)[\mathrm{}][[x]],^{\mathrm{}})`$ is a vector bundle on a formal neighborhood $`\mathrm{Spec}[[x_0,\mathrm{},x_s]]`$ with fiber $`H^{}(X)`$ endowed with a flat connection $`^{\mathrm{}}`$ with parameter $`\mathrm{}`$. This vector bundle has a canonical trivialization by definition. This canonical trivialization together with a canonical origin $`\tau =0`$ of the cohomology fixes a choice of a fundamental solution (1) and in turn defines the Lagrangian subspace (3) and the Lagrangian cone (4). Suppose that one changes the trivialization by a gauge transformation $`g(x,\mathrm{})\mathrm{End}(H^{}(X))[\mathrm{}][[x]]`$ such that
$$g_0(\mathrm{}):=g|_{q=0}=A_0\mathrm{exp}(a_1\mathrm{}+a_3\mathrm{}^3+\mathrm{}),A_0,a_iH^{}(X).$$
Then the dual Givental connection $`^{\mathrm{}}`$ changes as
$$g^{}_j^{\mathrm{}}=g^1_j^{\mathrm{}}g=\mathrm{}\frac{}{t_j}+g^1(p_j)g+\mathrm{}g^1\frac{g}{t_j}$$
Suppose also that one takes another point $`\tau =c`$ as an origin. Then the fundamental solution changes as
(6)
$$L^{g,c}(\tau ,\mathrm{})=e^{c/\mathrm{}}g_0(\mathrm{})^1L(\tau ,\mathrm{})g(x,\mathrm{}).$$
This new fundamental solution $`L^{g,c}(\tau ,\mathrm{})`$ is uniquely determined by a differential equation $`\mathrm{}dL^{g,c}(\tau ,\mathrm{})=L^{g,c}(\tau ,\mathrm{})(g^{}^{\mathrm{}})`$ and a shifted initial condition:
$$L^{g,c}(\tau ,\mathrm{})=e^{(\tau c)/\mathrm{}}T^{g,c}(x,\mathrm{}),T^{g,c}|_{q=0}=\mathrm{id}.$$
By the same formulas (3), (4) as before, $`L^{g,c}(\tau ,\mathrm{})`$ defines the following Lagrangian cone:
$$e^{c/\mathrm{}}g_0(\mathrm{})^1๐_0=A_0^1\mathrm{exp}\left(\frac{c}{\mathrm{}}+a_1\mathrm{}+a_3\mathrm{}^3+\mathrm{}\right)๐_0.$$
Therefore, $`e^{c/\mathrm{}}g_0(\mathrm{})^1`$ is identified with the symplectic transformation in Theorem 3.1. In case of equivariant Euler class, Theorem 3.1 is restated as follows:
###### Proposition 3.3.
The dual Givental connection of the twisted quantum $`D`$-module $`QDM_{S^1}^{}(X,)`$ is obtained from that of $`QDM^{}(X)`$ by a gauge transformation $`^{\mathrm{}}g^{}^{\mathrm{}}=g^1^{\mathrm{}}g`$, $`g(x,\mathrm{})\sqrt{\lambda }\mathrm{End}(H^{}(X))[\mathrm{}][\lambda ,\lambda ^1]][[x]]`$ and a coordinate change $`\tau \widehat{\tau }=\widehat{\tau }(\tau )`$ such that
(7)
$$g|_{q=0}=(\lambda +\rho )^{1/2}\mathrm{exp}\left(\underset{k1,l0}{}(1)^l\frac{(2k+l2)!}{\lambda ^{2k+l1}}\frac{B_{2k}}{(2k)!}\frac{\rho ^l}{l!}\mathrm{}^{2k1}\right).$$
(8)
$$\widehat{\tau }(\tau =c)=0,\text{where }c=\rho \mathrm{log}\lambda +\underset{l=2}{\overset{dimX}{}}(1)^l\frac{(l2)!}{\lambda ^{l1}}\frac{\rho ^l}{l!}.$$
The hypergeometric modification in Theorem 3.2 can be considered as an intermediate step to find a gauge transformation $`g`$ and a coordinate change $`\widehat{\tau }=\widehat{\tau }(\tau )`$ in the above proposition. If one changes a trivialization of $`QDM^{}(X)`$ by a gauge transformation $`g`$ satisfying (7) and also shifts the origin by a coordinate change satisfying (8), the set of column vectors $`L^{g,c}(\tau ,\mathrm{})p_j`$ of the new fundamental solution gives a basis of tangent spaces to the cone $`๐_\lambda `$ defined by the twisted quantum cohomology. Conversely, a matrix formed by the basis $`\{\mathrm{}_{t_j}I_{}(\tau ^{},\mathrm{})\}_j`$ of tangent spaces to the cone $`๐_\lambda `$ (given in Theorem 3.2) satisfies the following initial condition:
(9)
$$\stackrel{~}{L}(\tau ^{},\mathrm{}):=\left[\begin{array}{ccc}|& & |\\ \mathrm{}_{t_0}I_{}(\tau ^{},\mathrm{})& \mathrm{}& \mathrm{}_{t_s}I_{}(\tau ^{},\mathrm{})\\ |& & |\end{array}\right]=e^{\tau ^{}/\mathrm{}}\stackrel{~}{T}(x,\mathrm{}),\stackrel{~}{T}(x,\mathrm{})|_{q=0}=\mathrm{id}.$$
From this it follows that there exists an intermediate gauge transformation $`g_1(\tau ,\mathrm{})`$ satisfying (7) and a coordinate change $`\tau ^{}=\tau ^{}(\tau )`$ satisfying (8) such that
$$L^{g_1,c}(\tau ,\mathrm{})=\stackrel{~}{L}(\tau ^{},\mathrm{}),$$
where the left hand side is given in (6). Let $`\stackrel{~}{}^{\mathrm{}}`$ be the connection in this new gauge and coordinates:
(10)
$$\stackrel{~}{}^{\mathrm{}}:=\mathrm{}d+\underset{j=0}{\overset{s}{}}\mathrm{\Omega }_jdt_j,\mathrm{\Omega }_j(x,\mathrm{}):=\stackrel{~}{L}(\tau ,\mathrm{})^1\mathrm{}_{t_j}\stackrel{~}{L}(\tau ,\mathrm{}).$$
Here, the connection $`\stackrel{~}{}^{\mathrm{}}`$ at a point $`\tau ^{}(\tau )`$ is equal to $`g_1^{}^{\mathrm{}}`$ at $`\tau `$. By abuse of notation, we use $`\tau `$ instead of $`\tau ^{}`$ as an argument of this new connection $`\stackrel{~}{}^{\mathrm{}}`$ and solution $`\stackrel{~}{L}`$.
###### Proposition 3.4.
(i) The connection matrix $`\mathrm{\Omega }_j`$ is in the ring $`\mathrm{End}(H^{}(X))[\mathrm{}][\lambda ,\lambda ^1]][[x]]`$. If $`c_1()`$ is nef, this is also in the ring $`\mathrm{End}(H^{}(X))[\mathrm{},\lambda ][[x]]`$.
(ii) The pairing $`\alpha ,\beta `$ defined by
$$\alpha ,\beta :=\stackrel{~}{L}(\tau ,\mathrm{})\alpha ,\stackrel{~}{L}(\tau ,\mathrm{})\beta _{S^1}^{},\alpha ,\beta H^{}(X)$$
takes values in $`[\mathrm{}][\lambda ,\lambda ^1]][[x]]`$. If $`c_1()`$ is nef, this takes values in $`[\mathrm{},\lambda ][[x]]`$.
###### Proof.
(i) A tangent space to the cone $`๐_\lambda `$ is a vector space over $`[\mathrm{}][\lambda ,\lambda ^1]]`$. The former part follows from this and the argument in using a ruling property (4) of the cone $`๐_\lambda `$. If $`c_1()`$ is nef, the hypergeometric modification is of the form
$$I_{}(\tau ,\mathrm{})=\underset{๐
\mathrm{\Lambda }}{}\underset{k=1}{\overset{\rho ,๐
}{}}(\rho +\lambda +k\mathrm{})J_๐
(t,\mathrm{})q^๐
$$
and does not contain negative powers of $`\lambda `$. The latter part follows from this.
(ii) This follows from that $`\stackrel{~}{L}(\tau ,\mathrm{})\alpha `$ is a tangent vector of $`๐_\lambda `$ and that $`๐_\lambda `$ is Lagrangian with respect to the symplectic form $`\mathrm{Res}_{\mathrm{}=0}f(\mathrm{}),g(\mathrm{})_{S^1}^{}d\mathrm{}`$. โ
The dual Givental connection of $`QDM_{S^1}^{}(X,)`$ can be obtained from $`\stackrel{~}{}^{\mathrm{}}`$ by a further gauge transformation by $`g_2\mathrm{End}(H^{}(X))[\mathrm{}][\lambda ,\lambda ^1]][[x]]`$ and a coordinate change $`x\widehat{x}`$. The gauge transformation $`g_2`$ must satisfy
(11)
$$g_2|_{q=0}=\mathrm{id},g_2^{}\stackrel{~}{}^{\mathrm{}}=\mathrm{}d+\underset{j=0}{\overset{s}{}}\widehat{\mathrm{\Omega }}_jdt_j,\widehat{\mathrm{\Omega }}_j\text{ does not depend on }\mathrm{}.$$
The new coordinates $`\widehat{x}=(\widehat{t}_0,\widehat{q}_1=e^{\widehat{t}_1},\mathrm{},\widehat{q}_r=e^{\widehat{t}_r},\widehat{t}_{r+1},\mathrm{},\widehat{t}_s)`$ are of the form
(12)
$$\widehat{t}_0=t_0+F_0(x,\lambda ),\mathrm{log}\widehat{q}_a=\mathrm{log}q_a+F_a(x,\lambda )(1ar),$$
$$\widehat{t}_j=t_j+F_j(x,\lambda ),(r+1js)$$
for some $`F_j(x,\lambda )[\lambda ,\lambda ^1]][[x]]`$ satisfying $`F_j(x,\lambda )|_{q=0}=0`$. When written in the new coordinate system $`(\widehat{t}_0,\mathrm{},\widehat{t}_s)`$, the connection matrix $`\widehat{\mathrm{\Omega }}_{\widehat{j}}`$ must satisfy
(13)
$$\widehat{\mathrm{\Omega }}_{\widehat{j}}(1)=p_j,\widehat{\mathrm{\Omega }}_{\widehat{j}}:=\underset{i=0}{\overset{s}{}}\frac{t_i}{\widehat{t}_j}\widehat{\mathrm{\Omega }}_i$$
since $`\widehat{\mathrm{\Omega }}_{\widehat{j}}`$ is identified with the quantum multiplication by $`p_j`$ in $`QH_{S^1}^{}(X,)`$. Note that a gauge transformation and a coordinate change satisfying (11), (12) do not change the cone $`๐_\lambda `$. As shown in , this gauge transformation $`g_2`$ can be obtained as the positive part of the Birkhoff factorization of the fundamental solution:
$$g_2=\stackrel{~}{L}_+,\stackrel{~}{L}(\tau ,\mathrm{})=\stackrel{~}{L}_{}(\tau ,\mathrm{})\stackrel{~}{L}_+(x,\mathrm{}),$$
where $`\stackrel{~}{L}_{}=\mathrm{id}+O(\mathrm{}^1)`$ and $`\stackrel{~}{L}_+`$ is regular at $`\mathrm{}=0`$. As shown in , Theorem 4.6 and 4.8, $`g_2`$ and $`\widehat{x}`$ can be uniquely determined by the conditions (11), (13).
Summarizing, we can find the gauge transformation $`g`$ in Proposition 3.3 as a composite $`g_1g_2`$ of two gauge transformations. First gauge transformation $`g_1`$ corresponds to the hypergeometric modification $`\stackrel{~}{L}(\tau ,\mathrm{})`$ of the fundamental solution and the second one $`g_2`$ can be obtained by the Birkhoff factorization of $`\stackrel{~}{L}(\tau ,\mathrm{})`$. In the next section, we study the analytic property of the connection $`\mathrm{\Omega }_j`$ and the second gauge transformation $`g_2`$ in detail.
## 4. Proof of the main theorem
### 4.1. Formal power series with estimates
In this subsection, we give an estimate for the connection matrix $`\mathrm{\Omega }_j`$ which is obtained after the gauge transformation $`g_1`$. We will show that matrix elements of $`\mathrm{\Omega }_j`$ are not necessarily convergent power series, but their coefficients satisfy certain estimates. We assume that $`c_1()=\rho `$ is nef and that the ambient quantum cohomology $`QH^{}(X)`$ converges. We use the notation in Section 3.
Let $`L(\tau ,\mathrm{})`$ be the fundamental solution (1) of $`QDM^{}(X)`$.
$$L(\tau ,\mathrm{})=e^{\tau /\mathrm{}}T(x,\mathrm{})=e^{\tau /\mathrm{}}\underset{๐,n0}{}T_{๐,n}x^๐\mathrm{}^n,$$
where $`๐=(m_1,\mathrm{},m_s)`$ is a multi-index in $`_0^s`$. ($`T(x,\mathrm{})`$ does not depend on $`x_0=t_0`$.)
###### Lemma 4.1.
There exist positive constants $`C_1,C_2>1`$ such that
$$T_{๐,n}C_1C_2^{|๐|+n}\frac{1}{n!},$$
where $``$ is the operator norm and $`|๐ฆ|=_{j=1}^sm_j`$.
###### Proof.
The function $`T(x,\mathrm{})`$ is known to satisfy the following homogeneity:
$$2\mathrm{}\frac{T}{\mathrm{}}+\underset{a=1}{\overset{r}{}}(\mathrm{deg}q_a)\frac{T}{q_a}+\underset{j=r+1}{\overset{s}{}}(\mathrm{deg}t_j)t_j\frac{T}{t_j}+[\mu ,T]=0,$$
where $`\mu `$ is a constant matrix defined by $`\mu (p_j)=(\mathrm{deg}p_j)p_j`$ and $`\mathrm{deg}q_a`$ is the degree defined in non-twisted theory. By using $`\mathrm{}L/t_j=L(p_j_X)`$, we can rewrite the above equation as
$$2\mathrm{}\frac{T}{\mathrm{}}+\underset{a=1}{\overset{r}{}}(\mathrm{deg}q_a)\frac{1}{\mathrm{}}(T(p_a_X)p_aT)+\underset{j=r+1}{\overset{s}{}}(\mathrm{deg}t_j)\frac{t_j}{\mathrm{}}T(p_j_X)+[\mu ,T]=0$$
Set $`T_n(x)=_๐T_{๐,n}x^๐`$. By this equation, we have
$$(2n\mathrm{ad}(\mu ))T_n(x)=T_{n1}(x)A(x)c_1(X)T_{n1}(x),$$
where $`A(x)=_{a=1}^r(\mathrm{deg}q_a)(p_a_X)+_{j=r+1}^st_j(\mathrm{deg}t_j)(p_j_X)`$. Therefore, we have
$$T_n=\frac{1}{2n}\left(1\frac{\mathrm{ad}(\mu )}{2n}\right)^1(T_{n1}Ac_1(X)T_{n1}).$$
By the assumption that $`QH^{}(X)`$ is convergent, there exist a neighborhood $`U`$ of $`x=0`$ and a constant $`C>0`$ such that
$$\underset{xU}{sup}A(x)C,\underset{n1}{sup}\left(1\frac{\mathrm{ad}(\mu )}{2n}\right)^1C,c_1(X)C.$$
Since $`T_0(x)=\mathrm{id}`$, we can see that $`sup_{xU}T_n(x)C^{2n}/n!`$. The lemma follows from this estimate. โ
For a multi-index $`๐=(m_1,\mathrm{},m_s)`$, we write $`\rho ,๐=_{a=1}^r\xi _am_a`$, where $`\rho =c_1()=_{a=1}^r\xi _ap_a`$. Note that $`\rho ,๐0`$ for $`๐_0^s`$ because $`\rho `$ is nef. Let $`\stackrel{~}{L}(\tau ,\mathrm{})`$ be the fundamental solution given by hypergeometric modification (9).
$$\stackrel{~}{L}(\tau ,\mathrm{})=e^{\tau /\mathrm{}}\stackrel{~}{T}(x,\mathrm{},\lambda )=e^{\tau /\mathrm{}}\underset{๐0,n}{}\stackrel{~}{T}_{๐,n}(\lambda )x^๐\mathrm{}^n$$
###### Lemma 4.2.
There exists a positive constant $`C_3(\lambda )`$ depending continuously on $`\lambda `$ such that
$$\stackrel{~}{T}_{๐,n}(\lambda )C_1C_3(\lambda )^{|๐|+|n|}\{\begin{array}{cc}1/n!\hfill & n0,\hfill \\ |n|!\hfill & n0.\hfill \end{array}$$
Here we set $`0!=1`$. Moreover, $`\stackrel{~}{T}_{๐ฆ,n}=0`$ for $`n>\rho ,๐ฆ`$.
###### Proof.
Because $`\rho `$ is nef, we have
(14)
$$\underset{n}{}\stackrel{~}{T}_{๐,n}(\lambda )\mathrm{}^n=\underset{k=1}{\overset{\rho ,๐}{}}(\rho +\lambda +k\mathrm{})\underset{n0}{}T_{๐,n}\mathrm{}^n.$$
Therefore, by Lemma 4.1, we have
$`\stackrel{~}{T}_{๐,n}(\lambda )`$ $`={\displaystyle \underset{0l\rho ,๐}{}}{\displaystyle \underset{1k_1<k_2<\mathrm{}<k_l\rho ,๐}{}}(\rho +\lambda )^{\rho ,๐l}k_1k_2\mathrm{}k_lT_{๐,n+l}`$
$`{\displaystyle \underset{0l\rho ,๐}{}}(\rho +\lambda )^{\rho ,๐l}\left({\displaystyle \genfrac{}{}{0pt}{}{\rho ,๐}{l}}\right)(\rho ,๐l+1)\mathrm{}\rho ,๐C_1C_2^{|๐|+n+l}{\displaystyle \frac{1}{(n+l)!}}`$
$`C_1C_2^{|๐|+n}2^{\rho ,๐}{\displaystyle \underset{0l\rho ,๐}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{\rho ,๐}{l}}\right)(\rho +\lambda )^{\rho ,๐l}C_2^l{\displaystyle \frac{l!}{(n+l)!}}`$
$`C_1C_2^{|๐|+|n|}2^{\rho ,๐}(\rho +\lambda +2C_2)^{\rho ,๐}\{\begin{array}{cc}1/n!\hfill & n0,\hfill \\ |n|!\hfill & n0.\hfill \end{array}`$
Thus, we obtain the estimate. The latter part follows from (14). โ
Let $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ be the set of formal power series of the form $`_{๐0,n}A_{๐,n}(\lambda )x^๐\mathrm{}^n`$ in $`(\lambda )[\mathrm{},\mathrm{}^1][[x]]`$ satisfying the following conditions:
(i) $`A_{๐,n}=0`$ for $`n>\rho ,๐`$.
(ii) There exist positive continuous functions $`B(\lambda ),C(\lambda )`$ defined on the complement of a finite subset of $``$ such that
$$|A_{๐,n}(\lambda )|B(\lambda )C(\lambda )^{|๐|+|n|}\{\begin{array}{cc}1/n!\hfill & n0,\hfill \\ |n|!\hfill & n0.\hfill \end{array}$$
The above lemma says that each matrix element of $`\stackrel{~}{T}(x,\mathrm{},\lambda )`$ is contained in $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$.
###### Lemma 4.3.
The set $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ is a subring of $`(\lambda )[\mathrm{},\mathrm{}^1][[x]]`$.
###### Proof.
We must check that $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ is closed under product. Let $`_{๐0,n}A_{๐,n}x^๐\mathrm{}^n`$, $`_{๐0,n}B_{๐,n}x^๐\mathrm{}^n`$ be in $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$. By taking the maximum if necessary, we can assume that these two elements are estimated by the same functions $`B(\lambda ),C(\lambda )>1`$. Set $`C_{๐,n}=_{๐_1+๐_2=๐}_{n_1+n_2=n}A_{๐_1,n_1}B_{๐_2,n_2}`$. When $`n0`$, we have
$`|C_{๐,n}|`$ $`B(\lambda )^2C(\lambda )^{|๐|+|n|}\left({\displaystyle \underset{\begin{array}{c}n_1+n_2=n,\\ n_i0\end{array}}{}}{\displaystyle \frac{1}{n_1!n_2!}}+2{\displaystyle \underset{i=1}{\overset{\rho ,๐}{}}}C(\lambda )^{2i}{\displaystyle \frac{i!}{(n+i)!}}\right){\displaystyle \underset{๐_1+๐_2=๐}{}}1`$
$`B(\lambda )^2C(\lambda )^{|๐|+|n|}{\displaystyle \frac{1}{n!}}\left(2^n+2{\displaystyle \frac{C(\lambda )^{2\rho ,๐+2}}{C(\lambda )^21}}\right)C_{}^{}{}_{}{}^{|๐|}`$
for some $`C^{}>0`$. When $`n0`$, we have
$`|C_{๐,n}|`$ $`B(\lambda )^2C(\lambda )^{|๐|+|n|}\left({\displaystyle \underset{\begin{array}{c}n_1+n_2=n,\\ n_i0\end{array}}{}}|n_1|!|n_2|!+2{\displaystyle \underset{i=1}{\overset{\rho ,๐|n|}{}}}C(\lambda )^{2i}{\displaystyle \frac{(i+|n|)!}{i!}}\right){\displaystyle \underset{๐_1+๐_2=๐}{}}1`$
$`B(\lambda )^2C(\lambda )^{|๐|+|n|}|n|!\left(\rho ,๐+1+2(1+C(\lambda )^2)^{\rho ,๐}\right)C_{}^{}{}_{}{}^{|๐|}`$
for some $`C^{}>0`$. Therefore, we have the desired estimate for the product. โ
Let $`\mathrm{\Omega }_j(x,\mathrm{},\lambda )`$ be the connection matrix of $`\stackrel{~}{}^{\mathrm{}}`$ given in (10). Let $`,`$ be the bilinear form defined in Proposition 3.4 (ii).
###### Lemma 4.4.
The elements of the form $`\alpha ,\beta ,\alpha ,\mathrm{\Omega }_j\beta `$ are contained in both $`[\mathrm{},\lambda ][[x]]`$ and $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ for $`\alpha ,\beta H^{}(X)`$.
###### Proof.
By Proposition 3.4, we can see that $`\alpha ,\beta `$ and $`\alpha ,\mathrm{\Omega }_j\beta `$ are in $`[\mathrm{},\lambda ][[x]]`$. Since we have
$$\alpha ,\beta =\stackrel{~}{T}(x,\mathrm{},\lambda )\alpha ,\stackrel{~}{T}(x,\mathrm{},\lambda )\beta _{S^1}^{},$$
we can see that $`\alpha ,\beta `$ is in $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ by Lemma 4.2. On the other hand, we have
$`\alpha ,\mathrm{\Omega }_j\beta `$ $`=\stackrel{~}{L}(\tau ,\mathrm{})\alpha ,\mathrm{}{\displaystyle \frac{}{t_j}}\stackrel{~}{L}(\tau ,\mathrm{})\beta _{S^1}^{}`$
$`=\stackrel{~}{T}(x,\mathrm{},\lambda )\alpha ,_j^{\mathrm{cl}}\stackrel{~}{T}(x,\mathrm{},\lambda )\beta _{S^1}^{},`$
where $`_j^{\mathrm{cl}}=\mathrm{}_{t_j}+p_j`$. Therefore, $`\alpha ,\mathrm{\Omega }_j\beta `$ is also in $`๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )`$ by Lemma 4.2. โ
Define another ring $`๐ช_\lambda ^{\mathrm{}}(\rho )`$ by
$$๐ช_\lambda ^{\mathrm{}}(\rho )=๐ช_\lambda ^{\mathrm{},\mathrm{}^1}(\rho )(\lambda )[\mathrm{}][[x]].$$
By the above lemma, $`\alpha ,\beta `$, $`\alpha ,\mathrm{\Omega }_j\beta `$ are contained in $`๐ช_\lambda ^{\mathrm{}}(\rho )`$.
###### Lemma 4.5.
The ring $`๐ช_\lambda ^{\mathrm{}}(\rho )`$ is a local ring.
###### Proof.
Let $`_{๐0}_{n=0}^{\rho ,๐}A_{๐,n}(\lambda )x^๐\mathrm{}^n`$ be in $`๐ช_\lambda ^{\mathrm{}}(\rho )`$. We will show that if $`A_{0,0}(\lambda )=1`$, this element is invertible in $`๐ช_\lambda ^{\mathrm{}}(\rho )`$. We can assume that $`|A_{๐,n}|B(\lambda )C(\lambda )^{|๐|+n}n!`$ for some positive functions $`B(\lambda ),C(\lambda )>1`$. Set $`1+_{๐>0,n0}B_{๐,n}x^๐\mathrm{}^n=(1+_{๐>0,n0}A_{๐,n}x^๐\mathrm{}^n)^1`$. Then we have
$`|B_{๐,n}|`$ $`=\left|{\displaystyle \underset{l1}{}}(1)^l{\displaystyle \underset{\begin{array}{c}๐=๐_1+\mathrm{}+๐_l\\ ๐_i>0\end{array}}{}}{\displaystyle \underset{\begin{array}{c}n=n_1+\mathrm{}+n_l\\ 0n_i\rho ,๐_i\end{array}}{}}A_{๐_1,n_1}\mathrm{}A_{๐_l,n_l}\right|`$
$`{\displaystyle \underset{l1}{}}{\displaystyle \underset{\begin{array}{c}๐=๐_1+\mathrm{}+๐_l\\ ๐_i>0\end{array}}{}}{\displaystyle \underset{\begin{array}{c}n=n_1+\mathrm{}+n_l\\ 0n_i\rho ,๐_i\end{array}}{}}B(\lambda )^lC(\lambda )^{|๐|+n}n_1!\mathrm{}n_l!`$
$`B(\lambda )^{|๐|}C(\lambda )^{|๐|+n}n!{\displaystyle \underset{l1}{}}{\displaystyle \underset{\begin{array}{c}๐=๐_1+\mathrm{}+๐_l\\ ๐_i>0\end{array}}{}}\left({\displaystyle \genfrac{}{}{0pt}{}{n+l1}{n}}\right)`$
$`B(\lambda )^{|๐|}C(\lambda )^{|๐|+n}2^{n+|๐|1}n!{\displaystyle \underset{l1}{}}{\displaystyle \underset{\begin{array}{c}๐=๐_1+\mathrm{}+๐_l\\ ๐_i>0\end{array}}{}}1,`$
where we used $`l|๐|`$. The summation factor is of exponential order in $`|๐|`$ because
$$1+\underset{l1}{}\underset{\begin{array}{c}๐=๐_1+\mathrm{}+๐_l,\\ ๐_i>0\end{array}}{}y^{|๐|}=\frac{1}{1_{๐>0}y^{|๐|}}=\frac{1}{1(1/(1y)^s1)}$$
is analytic around $`y=0`$. โ
###### Proposition 4.6.
Each entry of the connection matrix $`\mathrm{\Omega }_j(x,\mathrm{},\lambda )`$ is contained in the ring $`๐ช_\lambda ^{\mathrm{}}(\rho )`$.
###### Proof.
By Lemma 4.4, $`\eta _{kl}=p_k,p_l`$ and $`p_k,\mathrm{\Omega }_jp_l`$ belong to $`๐ช_\lambda ^{\mathrm{}}(\rho )`$. It is easy to see that $`lim_{q0}\eta _{kl}=p_k,p_l_{S^1}^{}`$. Because the matrix $`(p_k,p_l_{S^1}^{})`$ is invertible in $`\mathrm{Mat}(s+1,(\lambda ))`$ and $`๐ช_\lambda ^{\mathrm{}}(\rho )`$ is a local ring, the matrix $`(\eta _{kl})`$ is invertible in $`\mathrm{Mat}(s+1,๐ช_\lambda ^{\mathrm{}}(\rho ))`$. Therefore, matrix elements $`\mathrm{\Omega }_{j;kl}=_i\eta ^{ki}p_i,\mathrm{\Omega }_jp_l`$ are in $`๐ช_\lambda ^{\mathrm{}}(\rho )`$. โ
### 4.2. Gauge fixing
In this subsection, we will find a gauge transformation $`g_2`$ which changes the connection matrices into $`\mathrm{}`$-independent ones.
Let $`๐ช_\lambda ^{\mathrm{}}`$ be the set of formal power series $`_{๐,n0}A_{๐,n}(\lambda )x^๐\mathrm{}^n`$ in $`(\lambda )[\mathrm{}][[x]]`$ satisfying the following conditions:
(i) $`A_{๐,n}=0`$ if $`|๐|_0=0,n>0`$,
(ii) There exists positive continuous functions $`B(\lambda )`$ and $`C(\lambda )`$ defined on the complement of a finite subset of $``$ such that
$$|A_{๐,n}(\lambda )|B(\lambda )C(\lambda )^{|๐|+n}|๐|_0^n,$$
where $`|๐|_0=_{a=1}^rm_a(|๐|)`$. It is easy to see that $`๐ช_\lambda ^{\mathrm{}}(\rho )`$ is contained in $`๐ช_\lambda ^{\mathrm{}}`$.
###### Lemma 4.7.
$`๐ช_\lambda ^{\mathrm{}}`$ is a local ring.
We omit the proof because it is similar to Lemma 4.5.
###### Proposition 4.8.
There exists a unique gauge transformation $`g_2`$ with entries in $`๐ช_\lambda ^{\mathrm{}}`$ such that $`g_2|_{q=0}=\mathrm{id}`$ and that the new connection matrix $`\widehat{\mathrm{\Omega }}_j`$ of $`g_2^{}\stackrel{~}{}^{\mathrm{}}=\mathrm{}d+_{j=0}^s\widehat{\mathrm{\Omega }}_jdt_j`$ is $`\mathrm{}`$-independent. Moreover, $`\widehat{\mathrm{\Omega }}_j`$ is convergent.
This proposition is considered to be a general gauge fixing lemma. It is applicable to any flat connection of the form $`\mathrm{}d+\mathrm{\Omega }`$ which is defined over $`๐ช_\lambda ^{\mathrm{}}`$, regular singular along $`q_1q_2\mathrm{}q_r=0`$ and whose residue matrices at $`q=0`$ are nilpotent. In , Theorem 4.6, we showed the existence and the uniqueness of $`g_2`$ in $`\mathrm{End}(H^{}(X))[\mathrm{},\lambda ][[x]]`$ by using a formal Birkhoff factorization. Here, we will show that $`g_2`$ also belongs to $`\mathrm{End}(H^{}(X))๐ช_\lambda ^{\mathrm{}}`$. This gauge fixing can be considered as a procedure of renormalization. A divergent connection can be renormalized by $`g_2`$ to yield a finite (convergent) result.
###### Proof.
Once we establish the existence of $`g_2`$ in $`\mathrm{End}(H^{}(X))๐ช_\lambda ^{\mathrm{}}`$, we can see that $`\widehat{\mathrm{\Omega }}_j`$ is convergent because it is contained in $`\mathrm{End}(H^{}(X))๐ช_\lambda ^{\mathrm{}}`$ and $`\mathrm{}`$-independent at the same time. Thus, it suffices to solve for $`g_2`$ in $`\mathrm{End}(H^{}(X))๐ช_\lambda ^{\mathrm{}}`$. Set $`\mathrm{\Omega }_j=_{๐
,n0}\mathrm{\Omega }_{j;๐
,n}(t_{r+1},\mathrm{},t_s,\lambda )q^๐
\mathrm{}^n`$. Because $`\mathrm{\Omega }_j`$ is defined over $`๐ช_\lambda ^{\mathrm{}}`$, there exist a neighborhood $`U^{sr}`$ of $`0`$ and continuous functions $`B(\lambda ),C(\lambda )>0`$ such that $`\mathrm{\Omega }_{j;๐
,n}(t,\lambda )B(\lambda )C(\lambda )^{|๐
|+n}|๐
|^n`$ for $`(t_{r+1},\mathrm{},t_s)U`$ and $`0js`$. From now on, we omit $`\lambda `$ and $`(t_{r+1},\mathrm{},t_s)`$ in the notation, but $`(t_{r+1},\mathrm{},t_s)`$ is always assumed to be in $`U`$. We set $`g_2=_{๐
,n0}G_{๐
,n}q^๐
\mathrm{}^n,\widehat{\mathrm{\Omega }}_j=_{๐
,n0}\widehat{\mathrm{\Omega }}_{j;๐
}q^๐
`$. By expanding the relation
$$g_2\widehat{\mathrm{\Omega }}_a=\mathrm{\Omega }_ag_2+\mathrm{}q_a\frac{g_2}{q_a},1ar,$$
we obtain the following equations:
(15) $`d_aG_{๐
,n1}+\mathrm{ad}(p_a)G_{๐
,n}`$ $`={\displaystyle \underset{\begin{array}{c}๐
_1+๐
_2=๐
,\\ ๐
_1>0,๐
_2>0\end{array}}{}}G_{๐
_2,n}\widehat{\mathrm{\Omega }}_{a;๐
_1}{\displaystyle \underset{\begin{array}{c}๐
_1+๐
_2=๐
,\\ n=n_1+n_2\\ ๐
_1>0\end{array}}{}}\mathrm{\Omega }_{a;๐
_1,n_1}G_{๐
_2,n_2}(n1),`$
(16) $`\mathrm{ad}(p_a)G_{๐
,0}`$ $`=\widehat{\mathrm{\Omega }}_{a;๐
}+{\displaystyle \underset{\begin{array}{c}๐
_1+๐
_2=๐
,\\ ๐
_1>0,๐
_2>0\end{array}}{}}G_{๐
_2,0}\widehat{\mathrm{\Omega }}_{a;๐
_1}{\displaystyle \underset{\begin{array}{c}๐
_1+๐
_2=๐
,\\ ๐
_1>0\end{array}}{}}\mathrm{\Omega }_{a;๐
_1,0}G_{๐
_2,0},`$
where we used $`G_{0,n}=\delta _{0,n}`$, $`\widehat{\mathrm{\Omega }}_{a;0}=p_a`$ and $`\mathrm{\Omega }_{a;0,n}=\delta _{0,n}p_a`$. Note that $`\mathrm{\Omega }_{a;๐
,n}`$ is known and $`G_{๐
,n}`$ and $`\widehat{\mathrm{\Omega }}_{a;๐
}`$ are unknown. Assume by induction that we know $`G_{๐
^{},n}`$ and $`\widehat{\mathrm{\Omega }}_{a;๐
^{}}`$ for all $`๐
^{}`$ with $`|๐
^{}|<\overline{d}`$. For a multi-index $`๐
`$ with $`|๐
|=\overline{d}`$, we first solve for $`G_{๐
,n}`$ for all $`n0`$ by using (15), and then we solve for $`\widehat{\mathrm{\Omega }}_{a;๐
}`$ by using (16).
More precisely, we must solve for $`G_{๐
,n},\widehat{\mathrm{\Omega }}_{a;๐
}`$ with estimates. Introduce the following notation:
$$g_{d,n}=\frac{1}{d^n}\underset{|๐
|=d}{}G_{๐
,n},\omega _{a;d,n}=\frac{1}{d^n}\underset{|๐
|=d}{}\mathrm{\Omega }_{a;๐
,n},\widehat{\omega }_{a;d}=\underset{|๐
|=d}{}\widehat{\mathrm{\Omega }}_{a;๐
},$$
$$H_{a;๐
,n}=\text{the right hand side of (}\text{15}\text{)},h_{d,n}=\underset{1ar}{\mathrm{max}}\frac{1}{d^n}\underset{|๐
|=d}{}H_{a;๐
,n}(n1).$$
There exist positive $`A_1,B_1`$ such that $`\omega _{a;d,n}A_1B_1^{d+n}`$. Assume by induction that
(17)
$$g_{d,n}\frac{B_2^d}{(d+1)^M}B_3^n,\widehat{\omega }_{a;d}A_2\frac{B_2^d}{(d+1)^M}(1ar)$$
hold for all $`d<\overline{d}`$. We take $`A_2`$ so that $`\widehat{\omega }_{a;0}=p_aA_2`$ holds. Then (17) is valid for $`d=0`$ because $`g_{0,n}=\delta _{0,n}`$. We will specify $`B_2,B_3,M`$ later. Take a $`๐
`$ with $`|๐
|=\overline{d}`$. Let $`a(๐
)`$ be an index such that $`d_{a(๐
)}=\mathrm{max}\{d_1,\mathrm{},d_r\}`$. Let $`C`$ be a constant satisfying $`\mathrm{ad}(p_a)C`$ for $`1ar`$. By (15), we have
$`G_{๐
,n}`$ $`={\displaystyle \frac{1}{d_{a(๐
)}}}(H_{a(๐
);๐
,n+1}\mathrm{ad}(p_{a(๐
)})G_{๐
,n+1})`$
$`={\displaystyle \frac{1}{d_{a(๐
)}}}H_{a(๐
);๐
,n+1}{\displaystyle \frac{\mathrm{ad}(p_{a(๐
)})}{d_{a(๐
)}^2}}H_{a(๐
);๐
,n+2}+\mathrm{}+{\displaystyle \frac{\mathrm{ad}(p_{a(๐
)})^{2N}}{d_{a(๐
)}^{2N+1}}}H_{a(๐
);๐
,n+2N+1},`$
where $`N=dim_{}X`$ and we used $`\mathrm{ad}(p_{a(๐
)})^{2N+1}=0`$. By using $`(|๐
|/d_{a(๐
)})r`$, we have
(18)
$$g_{\overline{d},n}rh_{\overline{d},n+1}+r^2Ch_{\overline{d},n+2}+\mathrm{}+r^{2N+1}C^{2N}h_{\overline{d},n+2N+1}.$$
Also we have
(19) $`h_{\overline{d},n}`$ $`\underset{1ar}{\mathrm{max}}\left({\displaystyle \underset{\begin{array}{c}|๐
_1|+|๐
_2|=\overline{d},\\ ๐
_1>0,๐
_2>0\end{array}}{}}{\displaystyle \frac{1}{\overline{d}^n}}G_{๐
_2,n}\widehat{\mathrm{\Omega }}_{a;๐
_1}+{\displaystyle \underset{\begin{array}{c}|๐
_1|+|๐
_2|=\overline{d},\\ n_1+n_2=n,๐
_1>0\end{array}}{}}{\displaystyle \frac{1}{\overline{d}^{n_1}}}\mathrm{\Omega }_{a;๐
_1,n_1}{\displaystyle \frac{1}{\overline{d}^{n_2}}}G_{๐
_2,n_2}\right)`$
$`\underset{1ar}{\mathrm{max}}\left({\displaystyle \underset{d_1=1}{\overset{\overline{d}1}{}}}g_{\overline{d}d_1,n}\widehat{\omega }_{a;d_1}+{\displaystyle \underset{d_1=1}{\overset{\overline{d}}{}}}{\displaystyle \underset{k=0}{\overset{n}{}}}\omega _{a;d_1,k}g_{\overline{d}d_1,nk}\right).`$
By using (16), we have
(20) $`\widehat{\omega }_{a;\overline{d}}`$ $`C{\displaystyle \underset{|๐
|=\overline{d}}{}}G_{๐
,0}+{\displaystyle \underset{\begin{array}{c}|๐
_1|+|๐
_2|=\overline{d},\\ ๐
_1>0,๐
_2>0\end{array}}{}}G_{๐
_2,0}\widehat{\mathrm{\Omega }}_{a;๐
_1}+{\displaystyle \underset{\begin{array}{c}|๐
_1|+|๐
_2|=\overline{d},\\ ๐
_1>0\end{array}}{}}\mathrm{\Omega }_{a;๐
_1,0}G_{๐
_2,0}`$
$`Cg_{\overline{d},0}+{\displaystyle \underset{d_1=1}{\overset{\overline{d}1}{}}}g_{\overline{d}d_1,0}\widehat{\omega }_{a;d_1}+{\displaystyle \underset{d_1=1}{\overset{\overline{d}}{}}}\omega _{a;d_1,0}g_{\overline{d}d_1,0}.`$
By (19) and the assumption (17), we obtain
(21) $`h_{\overline{d},n}`$ $`A_2B_2^{\overline{d}}B_3^n{\displaystyle \underset{i=1}{\overset{\overline{d}1}{}}}{\displaystyle \frac{1}{(\overline{d}i+1)^M(i+1)^M}}+A_1B_2^{\overline{d}}B_3^n{\displaystyle \underset{i=1}{\overset{\overline{d}}{}}}{\displaystyle \underset{k=0}{\overset{n}{}}}\left({\displaystyle \frac{B_1}{B_2}}\right)^i\left({\displaystyle \frac{B_1}{B_3}}\right)^k{\displaystyle \frac{1}{(\overline{d}i+1)^M}}`$
$`\left(A_2\epsilon _1(M)+{\displaystyle \frac{A_1}{1B_1/B_3}}\epsilon _2(B_2,M)\right){\displaystyle \frac{B_2^{\overline{d}}}{(\overline{d}+1)^M}}B_3^n,`$
where we set
$$\epsilon _1(M)=\underset{\overline{d}1}{sup}\underset{i=1}{\overset{\overline{d}1}{}}\frac{(\overline{d}+1)^M}{(\overline{d}i+1)^M(i+1)^M},\epsilon _2(B_2,M)=\underset{\overline{d}1}{sup}\underset{i=1}{\overset{\overline{d}}{}}\left(\frac{B_1}{B_2}\right)^i\frac{(\overline{d}+1)^M}{(\overline{d}i+1)^M}.$$
By (18) and (21), we obtain
(22)
$$g_{\overline{d},n}\epsilon _3(B_2,B_3,M)\frac{B_2^{\overline{d}}}{(\overline{d}+1)^M}B_3^n,$$
where we set
$$\epsilon _3(B_2,B_3,M)=\left(A_2\epsilon _1(M)+\frac{A_1}{1B_1/B_3}\epsilon _2(B_2,M)\right)rB_3\frac{(rB_3C)^{2N+1}1}{rB_3C1}.$$
By (22) and (20), we have
(23)
$$\widehat{\omega }_{a;\overline{d}}\epsilon _3(B_2,B_3,M)(C+A_2\epsilon _1(M)+A_1\epsilon _2(B_2,M))\frac{B_2^{\overline{d}}}{(\overline{d}+1)^M}.$$
In order to complete the induction step, we need to specify the parameters $`B_2,B_3,M`$. First we set $`B_3=2B_1`$. To choose $`B_2`$ and $`M`$, we need the following lemma:
###### Lemma 4.9.
$`lim_M\mathrm{}\epsilon _1(M)=0,lim_{B_2\mathrm{}}\epsilon _2(B_2,M)=0`$.
The proof will be given in the Appendix. For sufficiently large $`M`$, we have
$$A_2\epsilon _1(M)rB_3\frac{(rB_3C)^{2N+1}1}{rB_3C1}<\mathrm{min}\{\frac{1}{2},\frac{A_2}{3C}\}\text{ and }\epsilon _1(M)<\frac{1}{3}.$$
Next, for sufficiently large $`B_2`$, we have
$$\frac{A_1}{1B_1/B_3}\epsilon _2(B_2,M)rB_3\frac{(rB_3C)^{2N+1}1}{rB_3C1}<\frac{1}{2}\text{ and }A_1\epsilon _2(B_2,M)<\frac{A_2}{3}.$$
Now, it is easy to check that $`\epsilon _3(B_2,B_3,M)<1`$ and $`\epsilon _3(B_2,B_3,M)(C+A_2\epsilon _1(M)+A_1\epsilon _2(B_2,M))<A_2`$. Therefore, by (22) and (23), we complete the induction step. โ
###### Proof of Theorem 1.1.
As explained at the end of the section 3, in order to obtain the structure constants of $`QH_{S^1}^{}(X,)`$ from $`g_2^{}\stackrel{~}{}^{\mathrm{}}`$, it suffices to find a new coordinate system $`\widehat{x}=\{\widehat{t}_0,\widehat{q}_1=\mathrm{exp}(\widehat{t}_1),\mathrm{},\widehat{q}_r=\mathrm{exp}(\widehat{t}_r),\widehat{t}_{r+1},\mathrm{},\widehat{t}_s\}`$ such that the connection matrix $`\widehat{\mathrm{\Omega }}_{\widehat{j}}`$ defined by $`g_2^{}\stackrel{~}{}^{\mathrm{}}=\mathrm{}d+_{j=0}^s\widehat{\mathrm{\Omega }}_jdt_j=\mathrm{}d+_{j=0}\widehat{\mathrm{\Omega }}_{\widehat{j}}d\widehat{t}_j`$ satisfies $`\widehat{\mathrm{\Omega }}_{\widehat{j}}1=p_j`$. Then $`\widehat{\mathrm{\Omega }}_{\widehat{j}}`$ gives the twisted quantum product $`p_j_{}`$. Because $`\widehat{\mathrm{\Omega }}_j`$ is already convergent, new coordinates $`\widehat{x}`$ also become convergent functions in $`x`$. โ
## 5. Mirror symmetry for non-nef toric varieties
In this section, we will study mirror symmetry and the quantum cohomology of a not necessarily nef toric variety. In , we showed that the quantum $`D`$-module of $`QDM^{}(X)`$ of a toric variety $`X`$ can be reconstructed from the equivariant Floer cohomology $`FH_{S^1}^{}`$ by a generalized mirror transformation. There, the quantum $`D`$-module $`QDM^{}(X)`$ was described in terms of hypergeometric series ($`I`$-function). Here, we describe $`QDM^{}(X)`$ in terms of the mirror oscillatory integral introduced by Givental . The oscillatory integral satisfies (a generalized version of) the Mellin system of hypergeometric differential equations. We will show that the equivariant Floer cohomology $`FH_{S^1}^{}`$ is isomorphic to the $`q`$-adic completion of the Mellin system. Then, results in can be restated as $`QDM^{}(X)`$ restricted to some non-linear subspace of $`H^{}(X)`$ is isomorphic to the completion of the Mellin system. Using a method similar to section 4, we will show the convergence of the quantum cohomology of toric varieties. By using mirror symmetry, we will also show the semisimplicity.
### 5.1. Mirrors and the Mellin system
Let $`X`$ be a smooth projective toric variety defined by a fan $`\mathrm{\Sigma }^N`$. Take a nef integral basis $`\{p_1,\mathrm{},p_r\}`$ of $`H^2(X,)`$. Let $`D_1,\mathrm{},D_{r+N}`$ be all the torus invariant prime divisors and $`w_1,\mathrm{},w_{r+N}H^2(X,)`$ be their Poincarรฉ duals. We write $`w_i=_{a=1}^rm_{ia}p_a`$. We can recover $`X`$ from the data $`w_i`$ and Kรคhler class $`\eta _X`$ in $`H^2(X,)`$. Set $`=\{I\{1,\mathrm{},N+r\}|\eta _X_{iI}_{>0}w_i\}`$. Then we have
$$X=_{}^{r+N}/(^{})^r,_{}^{r+N}:=^{r+N}\underset{I}{}\{(z_1,\mathrm{},z_{r+N})|z_i=0\text{ for }iI\}.$$
Here, $`(^{})^r`$ acts on $`^{r+N}`$ as $`(z_1,\mathrm{},z_{r+N})(t^{w_1}z_1,\mathrm{},t^{w_{r+N}}z_{r+N})`$, where $`t^{w_i}=_{a=1}^rt_a^{m_{ia}}`$. The divisor $`D_i`$ corresponds to $`\{z_i=0\}`$. Let $`๐_i`$ denotes a coordinate of the mirror corresponding to the class $`w_i`$. Let $`\pi :Y=(^{})^{r+N}(^{})^r`$ be a family of algebraic tori defined by $`\pi (๐_1,\mathrm{},๐_{r+N})=(q_1,\mathrm{},q_r)`$, $`q_a=_{i=1}^{r+N}๐_i^{m_{ia}}`$. Define a function $`F(๐)`$ on $`Y`$ as $`F(๐)=๐_1+\mathrm{}+๐_{r+N}`$. We write $`Y_q=\pi ^1(q)`$ and $`F_q=F|_{Y_q}`$ for $`q(^{})^r`$. The mirror oscillatory integral is given by
$$_\mathrm{\Gamma }(q,\mathrm{})=_{\mathrm{\Gamma }_qY_q}e^{F_q/\mathrm{}}\omega _q,$$
where $`\mathrm{\Gamma }_q`$ is a non-compact real $`N`$-cycle in $`Y_q`$ such that the integral converges. The holomorphic volume form $`\omega _q`$ is given below. Take coordinates $`(s_1,\mathrm{},s_N)`$ of fibers $`Y_q`$ of the form $`s_b=_{i=1}^{r+N}๐_i^{m_{ib}^{}}`$, where the $`(r+N)\times N`$ matrix $`(m_{ib}^{}):^{r+N}^N`$ gives a splitting of the exact sequence
(24)
$$\begin{array}{ccccccccc}0& & ^N& \stackrel{\mathrm{ker}(m_{ia})}{}& ^{r+N}& \stackrel{(m_{ia})}{}& ^rH^2(X,)& & 0.\end{array}$$
By this splitting, we can write $`๐_i=_{a=1}^rq_a^{l_{ai}}_{b=1}^Ns_b^{x_{ib}}`$. Here, $`\stackrel{}{x}_i=(x_{i1},\mathrm{},x_{iN})^N`$ gives the primitive generator of the $`i`$-th one dimensional cone of the fan $`\mathrm{\Sigma }`$. The matrix $`(l_{ai}):^r^{r+N}`$ also gives a splitting of the exact sequence (24). We can choose this splitting so that $`l_{ai}0`$ because $`\{p_1,\mathrm{},p_r\}`$ is a nef basis. We define $`\omega _q`$ by
$$\omega _q=\frac{d\mathrm{log}๐_1\mathrm{}d\mathrm{log}๐_{r+N}}{d\mathrm{log}q_1\mathrm{}d\mathrm{log}q_r}=d\mathrm{log}s_1\mathrm{}d\mathrm{log}s_N|_{Y_q}.$$
This is independent of a choice of the splitting.
###### Proposition 5.1.
The oscillatory integral $`_\mathrm{\Gamma }(q,\mathrm{})`$ satisfies $`๐ซ_๐_\mathrm{\Gamma }(q,\mathrm{})=0`$ for all $`๐^r`$, where $`๐ซ_๐`$ is a differential operator defined by
$$๐ซ_๐
=q^๐
\underset{w_i,๐
<0}{}\underset{k=0}{\overset{w_i,๐
1}{}}\left(\underset{a=1}{\overset{r}{}}m_{ia}\mathrm{}_ak\mathrm{}\right)\underset{w_i,๐
>0}{}\underset{k=0}{\overset{w_i,๐
1}{}}\left(\underset{a=1}{\overset{r}{}}m_{ia}\mathrm{}_ak\mathrm{}\right),$$
where $`_a=q_a/q_a`$.
This proposition may be well-known, but we include a proof for completeness.
###### Proof.
First we have
$`{\displaystyle _\mathrm{\Gamma }}\mathrm{}{\displaystyle \frac{e^{F_q/\mathrm{}}}{\mathrm{log}๐_i}}\omega _q`$ $`={\displaystyle _\mathrm{\Gamma }}\left({\displaystyle \underset{a=1}{\overset{r}{}}}m_{ia}\mathrm{}_ae^{F_q/\mathrm{}}+{\displaystyle \underset{b=1}{\overset{N}{}}}m_{ib}^{}\mathrm{}{\displaystyle \frac{e^{F_q/\mathrm{}}}{\mathrm{log}s_b}}\right){\displaystyle \underset{b}{}}d\mathrm{log}s_b`$
$`=\left({\displaystyle \underset{a=1}{\overset{r}{}}}m_{ia}\mathrm{}_a\right){\displaystyle _\mathrm{\Gamma }}e^{F_q/\mathrm{}}{\displaystyle \underset{b}{}}d\mathrm{log}s_b+{\displaystyle \underset{c=1}{\overset{N}{}}}m_{ic}^{}{\displaystyle _\mathrm{\Gamma }}\mathrm{}d\left(e^{F_q/\mathrm{}}{\displaystyle \underset{bc}{}}d\mathrm{log}s_b\right)`$
$`=\left({\displaystyle \underset{a=1}{\overset{r}{}}}m_{ia}\mathrm{}_a\right){\displaystyle _\mathrm{\Gamma }}e^{F_q/\mathrm{}}\omega _q.`$
By using this, we have
$`๐ซ_๐
(q,\mathrm{})`$ $`={\displaystyle _\mathrm{\Gamma }}\left(q^๐
{\displaystyle \underset{w_i,๐
<0}{}}{\displaystyle \underset{k=0}{\overset{w_i,๐
1}{}}}(\mathrm{}๐_i{\displaystyle \frac{}{๐_i}}k\mathrm{}){\displaystyle \underset{w_i,๐
>0}{}}{\displaystyle \underset{k=0}{\overset{w_i,๐
1}{}}}(\mathrm{}๐_i{\displaystyle \frac{}{๐_i}}k\mathrm{})\right)e^{F_q/\mathrm{}}\omega _q`$
$`={\displaystyle _\mathrm{\Gamma }}\left(q^๐
{\displaystyle \underset{w_i,๐
<0}{}}๐_i^{w_i,๐
}{\displaystyle \underset{w_i,๐
>0}{}}๐_i^{w_i,๐
}\right)e^{F_q/\mathrm{}}\omega _q=0,`$
where we used $`q^๐
=_i๐_i^{w_i,๐
}`$. โ
We call the system of hypergeometric differential equations $`\{๐ซ_๐
(q,\mathrm{})=0|๐
^r\}`$ the Mellin system. Let $`M_{\mathrm{Mell}}`$ be the $`D`$-module corresponding to the Mellin system
$$M_{\mathrm{Mell}}=q^\pm ,\mathrm{},\mathrm{}/I_{\mathrm{Mell}},I_{\mathrm{Mell}}=\underset{๐
^r}{}q^\pm ,\mathrm{},\mathrm{}๐ซ_๐
,$$
where $`q^\pm `$ and $`\mathrm{}`$ are shorthand for $`q_1^\pm ,\mathrm{},q_r^\pm `$ and $`\mathrm{}q_1/q_1,\mathrm{},\mathrm{}q_r/q_r`$. The oscillatory integral $`_\mathrm{\Gamma }(q,\mathrm{})`$ gives a solution of $`M_{\mathrm{Mell}}`$ for each non-compact cycle $`\mathrm{\Gamma }`$.
### 5.2. $`S^1`$-equivariant Floer cohomology
We review the algebraic construction of the equivariant Floer cohomology $`FH_{S^1}^{}`$ for a toric variety $`X`$ briefly (see for detail). For each $`๐
^r`$, we put $`H_{S^1}^{}(L_๐
^{\mathrm{}})=[P_1,\mathrm{},P_r,\mathrm{}]`$. This is an algebra over $`H_{S^1}^{}(\mathrm{pt})=[\mathrm{}]`$. When $`w_i,๐
w_i,๐
^{}`$ for all $`i`$, we define a push-forward map $`i_{๐
,๐
^{}}:H_{S^1}^{}(L_๐
^{\mathrm{}})H_{S^1}^{}(L_๐
^{}^{\mathrm{}})`$ by $`i_{๐
,๐
^{}}(\alpha )=\alpha _{i=1}^N_{k=w_i,๐
^{}}^{w_i,๐
1}W_{i,k}`$, where $`W_{i,k}=_{a=1}^rm_{ia}P_ak\mathrm{}`$. Then, we have an inductive system $`(H_{S^1}^{}(L_๐
^{\mathrm{}}),i_{๐
,๐
^{}})`$. Let $`H_{S^1}^{\mathrm{}/2}=\underset{}{\mathrm{lim}}_๐
H_{S^1}^{}(L_๐
^{\mathrm{}})`$ be its direct limit. We define an $`H_{S^1}^{}(\mathrm{pt})`$-algebra homomorphism $`Q^๐
=_{a=1}^rQ_a^{d_a}:H_{S^1}^{}(L_๐
^{}^{\mathrm{}})H_{S^1}^{}(L_{๐
^{}+๐
}^{\mathrm{}})`$ by $`Q^๐
(P_a)=P_ad_a\mathrm{}`$. This is compatible with the direct limit and we have a module homomorphism $`Q^๐
:H_{S^1}^{\mathrm{}/2}H_{S^1}^{\mathrm{}/2}`$. We can check that the multiplication by $`P_a`$ and the action of $`Q_b`$ on $`H_{S^1}^{\mathrm{}/2}`$ satisfies the commutation relation $`[P_a,Q_b]=\mathrm{}\delta _{ab}Q_b`$. Hence, $`H_{S^1}^{\mathrm{}/2}`$ has a $`D`$-module structure when we regard $`P_a`$ as a differential operator $`\mathrm{}Q_a(/Q_a)`$. Let $`\mathrm{\Delta }_๐
H_{S^1}^{\mathrm{}/2}`$ be the image of $`1H_{S^1}^{}(L_๐
^{\mathrm{}})`$. We also write $`\mathrm{\Delta }=\mathrm{\Delta }_0`$. Let $`FH_0`$ be the submodule $`P_1,\mathrm{},P_r,Q_1,\mathrm{},Q_r,\mathrm{}\mathrm{\Delta }`$ of $`H_{S^1}^{\mathrm{}/2}`$. This $`FH_0`$ has a natural $`Q`$-adic topology and we define $`FH_{S^1}^{}`$ as the completion of $`FH_0`$:
$$FH_{S^1}^{}=\widehat{FH_0}=\underset{}{\mathrm{lim}}_nFH_0/๐ช^nFH_0,$$
where $`๐ช=_aQ_a[Q,\mathrm{}]`$. This becomes a module over $`[\mathrm{}][[Q_1,\mathrm{},Q_r]]P_1,\mathrm{},P_r`$.
###### Proposition 5.2.
The $`D`$-module $`H_{S^1}^{\mathrm{}/2}`$ is generated by $`\mathrm{\Delta }`$ as a $`Q^\pm ,P,\mathrm{}`$-module and all the relations are generated by $`\stackrel{~}{๐ซ}_๐=Q^๐_{w_i,๐<0}_{k=0}^{w_i,๐1}W_{i,k}_{w_i,๐>0}_{k=0}^{w_i,๐1}W_{i,k}`$ for $`๐^r`$. Therefore, $`H_{S^1}^{\mathrm{}/2}`$ is isomorphic to $`M_{\mathrm{Mell}}`$.
###### Proof.
By the relation $`\mathrm{\Delta }_๐
=Q^๐
\mathrm{\Delta }`$, $`H_{S^1}^{\mathrm{}/2}`$ is generated by $`\mathrm{\Delta }`$. It is easy to check that $`\stackrel{~}{๐ซ}_๐
\mathrm{\Delta }=0`$. Assume that $`f(P,Q,\mathrm{})\mathrm{\Delta }=0`$. We set $`f(P,Q,\mathrm{})=_if_i(P,\mathrm{})Q^{๐
_i}`$. There exists $`๐
`$ such that $`w_j,๐
_i+๐
>0`$ for all $`i,j`$. Then we have
(25) $`Q^๐
f(P,Q,\mathrm{})`$ $`={\displaystyle \underset{i}{}}f_i(P_ap_a,๐
\mathrm{},\mathrm{})Q^{๐
_i+๐
}`$
$`={\displaystyle \underset{i}{}}f_i(P_ap_a,๐
\mathrm{},\mathrm{})\left(\stackrel{~}{๐ซ}_{๐
_i+๐
}{\displaystyle \underset{j}{}}{\displaystyle \underset{k=0}{\overset{w_j,๐
_i+๐
1}{}}}W_{j,k}\right).`$
When applying this to $`\mathrm{\Delta }`$, we have $`_if_i(P_ap_a,๐
\mathrm{},\mathrm{})_j_{k=0}^{w_j,๐
_i+๐
1}W_{j,k}\mathrm{\Delta }=0`$. Because the canonical map $`H_{S^1}^{}(L_0^{\mathrm{}})H_{S^1}^{\mathrm{}}`$ is injective, we have
$$\underset{i}{}f_i(P_ap_a,๐
\mathrm{},\mathrm{})\underset{j}{}\underset{k=0}{\overset{w_j,๐
_i+๐
1}{}}W_{j,k}=0.$$
Therefore, by (25), we have $`f(P,Q,\mathrm{})=_if_i(P,\mathrm{})Q^๐
\stackrel{~}{๐ซ}_{๐
_i+๐
}`$. โ
###### Corollary 5.3.
Under the correspondence $`q_aQ_a`$, $`\mathrm{}_aP_a`$, we have
$$FH_0q,\mathrm{},\mathrm{}/I_{\mathrm{poly}},I_{\mathrm{poly}}=I_{\mathrm{Mell}}q,\mathrm{},\mathrm{}.$$
Hereafter, we identify $`Q_a`$ and $`P_a`$ with $`q_a`$ and $`\mathrm{}_a`$. From this, we can describe $`FH_{S^1}^{}`$ as follows.
###### Proposition 5.4.
$$FH_{S^1}^{}=\widehat{FH}_0[\mathrm{}][[q]]\mathrm{}/\overline{I}_{\mathrm{poly}},$$
where $`\overline{I}_{\mathrm{poly}}[\mathrm{}][[q]]\mathrm{}`$ is the closure of $`I_{\mathrm{poly}}`$ with respect to the $`q`$-adic topology.
###### Proof.
In , section 4.4, it is proved that $`FH_{S^1}^{}`$ is generated by $`\mathrm{\Delta }`$ over $`[\mathrm{}][[Q]]P`$. Therefore we have a surjection $`[\mathrm{}][[Q]]PFH_{S^1}^{}`$. Assume $`f(P,Q,\mathrm{})=_{๐
0}f_๐
(P,\mathrm{})Q^๐
[\mathrm{}][[Q]]P`$ satisfies $`f(P,Q,\mathrm{})\mathrm{\Delta }=0`$. Set $`f_n=_{|๐
|n}f_๐
(P,\mathrm{})Q^๐
`$. Then,
$$f_n\mathrm{\Delta }=\underset{|๐
|>n}{}f_๐
(P,\mathrm{})Q^๐
\mathrm{\Delta }$$
belongs to $`FH_0๐ช^{n+1}\widehat{FH}_0=๐ช^{n+1}FH_0`$. Therefore, there exists $`g_n`$ in $`๐ช^{n+1}Q,P,\mathrm{}`$ such that $`f_n\mathrm{\Delta }=g_n\mathrm{\Delta }`$. Because $`f_ng_nI_{\mathrm{poly}}`$ and $`f_ng_nf`$ as $`n\mathrm{}`$, we have $`f\overline{I}_{\mathrm{poly}}`$. โ
Later, we will see that $`\overline{I}_{\mathrm{poly}}`$ does not necessarily coincide with $`[\mathrm{}][[q]]\mathrm{}I_{\mathrm{poly}}`$. In such a case, we need to add non-algebraic differential equations. We set $`๐ช_{\mathrm{small}}^{\mathrm{}}=๐ช_\lambda ^{\mathrm{}}[\mathrm{}][[q]]`$ ($`๐ช_\lambda ^{\mathrm{}}`$ was introduced in section 4):
$$๐ช_{\mathrm{small}}^{\mathrm{}}=\{\underset{๐
,n0}{}A_{๐
,n}q^๐
\mathrm{}^n[\mathrm{}][[q]]|\begin{array}{c}A_{0,n}=0\text{ for }n>0,\hfill \\ |A_{๐
,n}|BC^{|๐
|}|๐
|^n(B,C>0).\hfill \end{array}\}$$
Let $`\stackrel{~}{FH}_{S^1}^{}`$ be the following submodule of $`FH_{S^1}^{}`$:
$$\stackrel{~}{FH}_{S^1}^{}=๐ช_{\mathrm{small}}^{\mathrm{}}P_1,\mathrm{},P_r\mathrm{\Delta }FH_{S^1}^{}.$$
Then we have $`\stackrel{~}{FH}_{S^1}^{}๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}/\overline{I}_{\mathrm{poly}}`$, where $`\overline{I}_{\mathrm{poly}}`$ is the closure of $`I_{\mathrm{poly}}`$ in $`๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}`$ with respect to the $`q`$-adic topology. Then we have a surjection
(26)
$$M_{\mathrm{Mell}}_{[\mathrm{},q^\pm ]}๐ช_{\mathrm{small}}^{\mathrm{}}[q^1]=\left(๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}/๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}}\right)[q^1]\stackrel{~}{FH}_{S^1}^{}[q^1]$$
###### Proposition 5.5.
Let $`\{T_i(y_1,\mathrm{},y_r)\}_{i=0}^s`$ be homogeneous polynomials such that $`\{T_i(p_1,\mathrm{},p_r)\}_{i=0}^s`$ forms a basis of $`H^{}(X)`$. Then, $`\{T_i(P)\mathrm{\Delta }\}_{i=1}^s`$ forms a free basis of $`\stackrel{~}{FH}_{S^1}^{}`$ as an $`๐ช_{\mathrm{small}}^{\mathrm{}}`$-module. Moreover, we have $`FH_{S^1}^{}=\stackrel{~}{FH}_{S^1}^{}_{๐ช_{\mathrm{small}}^{\mathrm{}}}[\mathrm{}][[q]]`$.
###### Proof.
In , section 4.4, we showed that $`\{T_i(P)\mathrm{\Delta }\}_{i=0}^s`$ forms a free basis of $`FH_{S^1}^{}`$ as a $`[\mathrm{}][[q]]`$-module. Here, we will show this is also a basis of $`\stackrel{~}{FH}_{S^1}^{}`$ as a $`๐ช_{\mathrm{small}}^{\mathrm{}}`$-module. It suffices to show that the connection matrix $`\mathrm{\Omega }_a=(\mathrm{\Omega }_{a;ij})`$ defined by
$$P_aT_j(P)\mathrm{\Delta }=\underset{i=0}{\overset{s}{}}\mathrm{\Omega }_{a;ij}(q,\mathrm{})T_i(P)\mathrm{\Delta }$$
has entries in $`๐ช_{\mathrm{small}}^{\mathrm{}}`$. We use a method similar to the proof in section 4.1. By , section 4.3, there exists a map $`\mathrm{\Xi }:FH_{S^1}^{}H^{}(X)[\mathrm{},\mathrm{}^1][[q]]`$ and a pairing $`(,):FH_{}^{S^1}\times FH_{S^1}^{}[\mathrm{}][[q]]`$ such that
(27)
$$\mathrm{\Xi }(P_a\alpha )=(\mathrm{}_a+p_a)\mathrm{\Xi }(\alpha ),(\overline{\alpha },\beta )=_X\overline{\mathrm{\Xi }(\alpha )}\mathrm{\Xi }(\beta ),$$
for $`\alpha ,\beta FH_{S^1}^{}`$. Here, the equivariant Floer homology $`FH_{}^{S^1}`$ is the Poincarรฉ dual theory of $`FH_{S^1}^{}`$ with a $`[[q]]`$-module isomorphism $`\overline{}:FH_{S^1}^{}FH_{}^{S^1}`$ such that $`\overline{\mathrm{}\alpha }=\mathrm{}\overline{\alpha }`$. The operator $`\overline{}`$ also acts on $`H^{}(X)[\mathrm{},\mathrm{}^1][[q]]`$ by changing the sign of $`\mathrm{}`$. The function $`\stackrel{~}{I}(q,\mathrm{})=\mathrm{\Xi }(\mathrm{\Delta })`$ is written as
(28)
$$\stackrel{~}{I}(q,\mathrm{})=\underset{๐
0}{}q^๐
\underset{i=1}{\overset{r+N}{}}\frac{_{k=1+w_i,๐
}^{\mathrm{}}(w_i+k\mathrm{})}{_{k=1}^{\mathrm{}}(w_i+k\mathrm{})}.$$
Here, $`I(q,\mathrm{})=e^{p\mathrm{log}q/\mathrm{}}\stackrel{~}{I}(q,\mathrm{})`$ is frequently referred to as $`I`$-function. We write $`\stackrel{~}{I}(q,\mathrm{})=_{๐
0}I_๐
(\mathrm{})q^๐
`$. Take $`C_1>0`$ such that $`sup_{|\mathrm{}|=1,k1,i}(w_i,1+w_i/(k\mathrm{}),(1+w_i/(k\mathrm{}))^1)C_1`$. Then we have for $`|\mathrm{}|=1`$
$`I_๐
(\mathrm{})`$ $`{\displaystyle \frac{_{w_i,๐
<0}|w_i,๐
|!}{_{w_i,๐
>0}w_i,๐
!}}C_1^{_{i=1}^{r+N}|w_i,๐
|}`$
$`C_2^{|๐
|}\{\begin{array}{cc}1/c_1(X),๐
!\hfill & \text{ if }c_1(X),๐
0,\hfill \\ |c_1(X),๐
|!\hfill & \text{ if }c_1(X),๐
0.\hfill \end{array}`$
for some $`C_2>0`$. Here, we used $`c_1(X)=_{i=1}^{r+N}w_i`$. Let $`๐ช_{\mathrm{deg}}^{\mathrm{},\mathrm{}^1}`$ be the set of power series $`_๐
A_๐
(\mathrm{})q^๐
`$ in $`[\mathrm{},\mathrm{}^1][[q]]`$ satisfying
$$\underset{|\mathrm{}|=1}{sup}|A_๐
(\mathrm{})|BC^๐
\{\begin{array}{cc}1/c_1(X),๐
!\hfill & \text{ if }c_1(X),๐
0,\hfill \\ |c_1(X),๐
|!\hfill & \text{ if }c_1(X),๐
0,\hfill \end{array}$$
for some $`B,C>0`$. ($`B`$, $`C`$ depend on each element.) As in section 4.1, we can prove that $`๐ช_{\mathrm{deg}}^{\mathrm{},\mathrm{}^1}`$ is a subring of $`[\mathrm{},\mathrm{}^1][[q]]`$. Each component of $`\stackrel{~}{I}(q,\mathrm{})`$ belongs to the ring $`๐ช_{\mathrm{deg}}^{\mathrm{},\mathrm{}^1}`$. Set $`\eta _{kj}=(\overline{T_k(P)\mathrm{\Delta }},T_j(P)\mathrm{\Delta })`$. By using (27), we have
$`\eta _{kj}`$ $`={\displaystyle _X}T_k(\mathrm{}+p)\stackrel{~}{I}(q,\mathrm{})T_j(\mathrm{}+p)\stackrel{~}{I}(q,\mathrm{})`$
$`{\displaystyle \underset{i=0}{\overset{s}{}}}\eta _{ki}\mathrm{\Omega }_{a;ij}`$ $`=(\overline{T_k(P)\mathrm{\Delta }},P_aT_j(P)\mathrm{\Delta })`$
$`={\displaystyle _X}T_k(\mathrm{}+p)\stackrel{~}{I}(q,\mathrm{})(\mathrm{}_a+p_a)T_j(\mathrm{}+p)\stackrel{~}{I}(q,\mathrm{}).`$
Therefore, $`\eta _{kj}`$ and $`_i\eta _{ki}\mathrm{\Omega }_{a;ij}`$ are contained in $`๐ช_{\mathrm{deg}}^{\mathrm{},\mathrm{}^1}[\mathrm{}][[q]]`$. Moreover, these elements are homogeneous with respect to the grading $`\mathrm{deg}q^๐
=2c_1(X),๐
`$ and $`\mathrm{deg}\mathrm{}=2`$. The homogeneity means that the coefficient of $`q^๐
\mathrm{}^n`$ is non-vanishing only when $`c_1(X),๐
+n`$ equals a given constant. Therefore, by Cauchyโs residue theorem, we can see that $`\eta _{kj}`$ and $`_i\eta _{ki}\mathrm{\Omega }_{a;ij}`$ are also contained in $`๐ช_{\mathrm{small}}^{\mathrm{}}`$. Because $`lim_{q0}\eta _{kj}=_XT_k(p)T_j(p)`$ is an invertible matrix and $`๐ช_{\mathrm{small}}^{\mathrm{}}`$ is a local ring, we have $`\mathrm{\Omega }_{a;ij}๐ช_{\mathrm{small}}^{\mathrm{}}`$. โ
### 5.3. Generalized mirror transformation
We can describe how to reconstruct the quantum $`D`$-module $`QDM^{}(X)`$ from $`FH_{S^1}^{}`$ as follows:
###### Theorem 5.6 (, Theorem 4.9, 5.4).
There exists a formal embedding $`๐ข๐ช๐:(^r,0)(^{s+1},0)`$ and an isomorphism of $`D`$-modules $`\mathrm{\Phi }_{๐ข๐ช๐}:๐ข๐ช๐^{}(QDM(X))FH_{S^1}^{}`$. The map $`๐ข๐ช๐`$ is given by equations of the form
$$\widehat{t}_0=F_0(q),\widehat{q}_1=q_1\mathrm{exp}(F_1(q)),\mathrm{},\widehat{q}_r=q_r\mathrm{exp}(F_r(q)),\widehat{t}_{r+1}=F_{r+1}(q),\mathrm{},\widehat{t}_s=F_s(q)$$
for some $`F_i(q)[[q]]`$, $`F_i(0)=0`$ and $`\mathrm{\Phi }_{๐ข๐ช๐}|_{q=0}`$ is determined by the canonical isomorphism $`H^{}(X)[\mathrm{}]FH_{S^1}^{}/_{a=1}^rq_aFH_{S^1}^{}`$. Moreover, we can reconstruct $`QDM^{}(X)`$ from $`FH_{S^1}^{}`$ by the following steps:
(i) Take a free basis $`\{T_0,\mathrm{},T_s\}`$ of $`FH_{S^1}^{}`$ as a $`[\mathrm{}][[q]]`$-module as in Proposition 5.5 and calculate a connection matrix $`\mathrm{\Omega }_a`$ defined by $`P_aT_j=_{i=0}^s\mathrm{\Omega }_{a;ij}T_i`$.
(ii) Find a gauge transformation $`g`$ such that $`g|_{q=0}=\mathrm{id}`$ and that the new connection matrix $`\widehat{\mathrm{\Omega }}_a`$ is $`\mathrm{}`$-independent, where $`\widehat{\mathrm{\Omega }}_a=g^1\mathrm{\Omega }_ag+g^1\mathrm{}_ag`$.
(iii) Solve for matrix-valued functions $`\widehat{\mathrm{\Omega }}_j(t_0,q,t_{r+1},\mathrm{},t_s)`$ $`(0js)`$ from $`\widehat{\mathrm{\Omega }}_a(q)`$ $`(1ar)`$, where $`\widehat{\mathrm{\Omega }}_0=\mathrm{id}`$. This procedure will be reviewed in Proposition 5.8.
(iv) Find a new coordinate system $`(\widehat{t}_0,\widehat{q}_1,\mathrm{},\widehat{q}_r,\widehat{t}_{r+1},\mathrm{},\widehat{t}_s)`$ of the form $`\widehat{t}_0=t_0+F_0(q)`$, $`\widehat{q}_a=q_a\mathrm{exp}(F_a(q))`$ $`(1ar)`$, $`\widehat{t}_j=t_j+F_j(q)`$ $`(r+1js)`$ such that $`\widehat{\mathrm{\Omega }}_{\widehat{j}}=_{i=0}^s(t_i/\widehat{t}_j)\widehat{\mathrm{\Omega }}_i`$ satisfies $`\widehat{\mathrm{\Omega }}_{\widehat{j}}(1)=p_j`$.
We study these four steps from an analytic point of view. By Proposition 5.5, the connection matrix $`\mathrm{\Omega }_a(q)`$ in step (i) has its matrix elements in $`๐ช_{\mathrm{small}}^{\mathrm{}}`$. Then by Proposition 4.8, the connection matrix $`\widehat{\mathrm{\Omega }}_a`$ in step (ii) becomes a convergent function of $`q`$. In Proposition 5.8, we will see that the reconstruction step (iii) preserves the convergence. The last step (iv) can be done in the convergent category, therefore $`QH^{}(X)`$ has convergent structure constants. Summarizing,
###### Theorem 5.7.
The quantum $`D`$-module $`QDM^{}(X)`$ of a toric variety is defined over convergent power series. The embedding $`๐ข๐ช๐:(^r,0)(^{s+1},0)`$ in Theorem 5.6 is complex analytic and $`\stackrel{~}{FH}_{S^1}^{}๐ข๐ช๐^{}(QDM_{\mathrm{an}}^{}(X))_{๐ช_{\mathrm{an}}^{\mathrm{}}}๐ช_{\mathrm{small}}^{\mathrm{}}`$, where $`QDM_{\mathrm{an}}^{}(X)=(H^{}(X)๐ช_{\mathrm{an}}^{\mathrm{}},^{\mathrm{}})`$ and $`๐ช_{\mathrm{an}}^{\mathrm{}}=\{f(\mathrm{},x)\{\mathrm{},x\}|f|_{q=0}\text{ is constant.}\}`$.
In order to state the compatibility of reconstruction with convergence, we consider the following general situation. Consider a flat connection $`^{\mathrm{}}=\mathrm{}d+_{a=1}^r\mathrm{\Omega }_a(q)dq_a/q_a`$ of the bundle $`H^{}(X)\times UU`$ regular singular along $`q_1\mathrm{}q_r=0`$. Here, $`U`$ is a neighborhood of $`0`$ in $`^r`$ and $`\mathrm{\Omega }_a(q)`$ is an $`\mathrm{}`$-independent holomorphic function on $`U`$.
###### Proposition 5.8.
Assume that $`H^{}(X)`$ is generated by $`1`$ under the action of residue matrices $`p_a:=\mathrm{\Omega }_a(0)`$. For a set of vectors $`v_1,\mathrm{},v_l`$ in $`H^{}(X)`$, $`^{\mathrm{}}`$ can be uniquely extended to a flat connection $`\stackrel{~}{}^{\mathrm{}}=\mathrm{}d+_{a=1}^r\mathrm{\Omega }_a(q,t)dq_a/q_a+_{j=1}^l\mathrm{\Phi }_j(q,t)dt_j`$ defined on a neighborhood $`U^{}`$ of $`0`$ in $`^{r+l}`$ under the condition that $`\mathrm{\Phi }_j(q,t)1=v_j`$.
###### Proof.
In , Theorem 4.9, we proved that $`\stackrel{~}{}^{\mathrm{}}`$ can be reconstructed uniquely as a formal connection. Here, we prove that this is convergent. Because we can extend $`^{\mathrm{}}`$ to the $`t_j`$-direction independently for different $`t_j`$โs, we can assume that $`l=1`$. Flatness of the connection $`\stackrel{~}{}^{\mathrm{}}`$ implies $`_t\mathrm{\Omega }_a=q_a_{q_a}\mathrm{\Phi }`$ and $`[\mathrm{\Omega }_a,\mathrm{\Phi }]=0`$. By taking a smaller $`U`$ if necessary, we can assume that $`H^{}(X)`$ is generated by $`1`$ under the action of $`\mathrm{\Omega }_a(q)`$ for each $`qU`$. Then, we have a surjection $`[\mathrm{\Omega }_1(q),\mathrm{},\mathrm{\Omega }_r(q)]H^{}(X)`$ and this determines a product $`_q`$ on $`H^{}(X)`$ for each $`qU`$. On the locus $`\{t=0\}`$, $`\mathrm{\Phi }(q,0)`$ must be a multiplication $`(v_1_q)`$ by $`v_1`$ because it commutes with $`\mathrm{\Omega }_a(q)`$. Therefore, $`\mathrm{\Phi }(q,0)`$ becomes holomorphic on $`U`$. On the locus $`\{q=0\}`$, it is easy to see that $`\mathrm{\Omega }_a(0,t)=p_a`$ and $`\mathrm{\Phi }(0,t)=v_1`$. Expand $`\mathrm{\Phi }`$ and $`\mathrm{\Omega }_a`$ as $`\mathrm{\Phi }=_{๐
,m0}\mathrm{\Phi }_{๐
,m}q^๐
t^m`$ and $`\mathrm{\Omega }_a=_{๐
,m0}\mathrm{\Omega }_{a;๐
,m}q^๐
t^m`$. Note that $`\mathrm{\Phi }_{0,m}=\delta _{0m}(v_1)`$ and $`\mathrm{\Omega }_{a;0,m}=\delta _{0m}p_a`$. Then by $`_t\mathrm{\Omega }_a=q_a_{q_a}\mathrm{\Phi }`$, we have
(29)
$$\mathrm{\Omega }_{a;๐
,m+1}=\frac{d_a}{m+1}\mathrm{\Phi }_{๐
,m}.$$
Let $`\{T_0,\mathrm{},T_s\}`$ be a basis of $`H^{}(X)`$. By the assumption, we can write
$$T_i=\underset{l,a_1,\mathrm{},a_l}{}A_{a_1,\mathrm{},a_l}^{(i)}p_{a_1}\mathrm{}p_{a_l}(1)$$
for some $`A_{a_1,\mathrm{},a_l}^i`$. Then by $`\mathrm{\Phi }_{๐
,m+1}(1)=0`$, we have
(30) $`\mathrm{\Phi }_{๐
,m+1}T_i`$ $`={\displaystyle \underset{l,a_1,\mathrm{},a_l}{}}A_{a_1,\mathrm{},a_l}^{(i)}{\displaystyle \underset{k=1}{\overset{l}{}}}p_{a_1}\mathrm{}[\mathrm{\Phi }_{๐
,m+1},p_{a_k}]\mathrm{}p_{a_l}(1)`$
$`={\displaystyle \underset{l,a_1,\mathrm{},a_l}{}}A_{a_1,\mathrm{},a_l}^{(i)}{\displaystyle \underset{k=1}{\overset{l}{}}}p_{a_1}\mathrm{}\left({\displaystyle \underset{๐
_1>0}{}}{\displaystyle \underset{j=0}{\overset{m+1}{}}}[\mathrm{\Omega }_{a_k;๐
_1,j},\mathrm{\Phi }_{๐
๐
_1,m+1j}]\right)\mathrm{}p_{a_l}(1)`$
where we used $`[\mathrm{\Phi },\mathrm{\Omega }_{a_k}]=0`$. Assume by induction that we know $`\mathrm{\Omega }_{a;๐
^{},m}`$ and $`\mathrm{\Phi }_{๐
^{},m}`$ for all $`|๐
^{}|<\overline{d}`$ and all $`m0`$. Set $`\omega _{d,m}=\mathrm{max}_a_{|๐
|=d}\mathrm{\Omega }_{a;๐
,m}`$ and $`\varphi _{d,m}=_{|๐
|=d}\mathrm{\Phi }_{๐
,m}`$. Assume also that there exist constants $`A,C,M>1`$ such that
(31)
$$\omega _{d,m}A\frac{C^{d+m}}{(d+1)^M}\frac{d^m}{m!},\varphi _{d,m}A\frac{C^{d+m}}{(d+1)^M}\frac{d^m}{m!}$$
holds for all $`d<\overline{d}`$ and all $`m0`$. We must choose $`A,C,M`$ so that this estimate is valid for $`m=0`$ and all $`d0`$. Take $`๐
`$ such that $`|๐
|=\overline{d}`$. First we can solve for $`\mathrm{\Omega }_{a;๐
,1}`$ by (29). Then we solve for $`\mathrm{\Phi }_{๐
,1}`$ by (30). Next we solve for $`\mathrm{\Omega }_{a;๐
,2}`$ by (29) and repeat this process. Assume that the estimate (31) holds for $`d=\overline{d}`$ and up to $`m`$ ($`m0`$). We have by (29),
$$\omega _{\overline{d},m+1}\frac{\overline{d}}{m+1}\varphi _{\overline{d},m}A\frac{C^{\overline{d}+m}}{(\overline{d}+1)^M}\frac{\overline{d}^{m+1}}{(m+1)!}.$$
Next we have by (30),
(32) $`\varphi _{\overline{d},m+1}`$ $`B_1{\displaystyle \underset{0<i<\overline{d}}{}}{\displaystyle \underset{j=0}{\overset{m+1}{}}}\omega _{i,j}\varphi _{\overline{d}i,m+1j}+B_1\omega _{\overline{d},m+1}(v_1)`$
$`B_1A^2{\displaystyle \frac{C^{\overline{d}+m+1}}{(\overline{d}+1)^M}}{\displaystyle \frac{\overline{d}^{m+1}}{(m+1)!}}{\displaystyle \underset{i=1}{\overset{\overline{d}1}{}}}{\displaystyle \frac{(\overline{d}+1)^M}{(i+1)^M(\overline{d}i+1)^M}}+B_2A{\displaystyle \frac{C^{\overline{d}+m}}{(\overline{d}+1)^M}}{\displaystyle \frac{\overline{d}^{m+1}}{(m+1)!}}`$
$`A{\displaystyle \frac{C^{\overline{d}+m+1}}{(\overline{d}+1)^M}}{\displaystyle \frac{\overline{d}^{m+1}}{(m+1)!}}\{B_1A\epsilon _1(M)+B_2/C\}`$
where $`B_1,B_2>0`$ are constants determined only by $`(H^{}(X),)`$ and $`\epsilon _1(M)`$ is a function defined in the proof of Proposition 4.8. To complete the induction, we need to choose $`C`$ and $`M`$ carefully. Because $`lim_M\mathrm{}\epsilon _1(M)=0`$, for sufficiently large $`M`$, we have $`B_1A\epsilon _1(M)1/2`$. Next we choose $`C`$ sufficiently large so that the estimate (31) is valid for $`m=0`$ and all $`d0`$ and that $`C2B_2`$. Then by (32), we have the desired estimate for $`\varphi _{\overline{d},m+1}`$. After we obtain the estimate (31) for all $`d,m0`$, we can see that $`\mathrm{\Omega }_a(q,t)`$ and $`\mathrm{\Phi }(q,t)`$ are convergent because
$$\underset{d0}{}\underset{m0}{}C^{d+m}\frac{d^m}{m!}x^dy^m=\frac{1}{1Cx\mathrm{exp}(Cy)}$$
is holomorphic around $`(x,y)=(0,0)`$. โ
When we apply this proposition to the dual Givental connections, we have
###### Corollary 5.9.
Let $`X`$ be a smooth projective variety. If $`H^{}(X)`$ is generated by $`H^2(X)`$ and if the small quantum cohomology of $`X`$ has convergent structure constants, so does the big quantum cohomology $`QH^{}(X)`$.
### 5.4. Characteristic variety and semisimplicity
We study the characteristic variety of the Mellin system and proves the semisimplicity. For a differential operator $`f(q,\mathrm{},\mathrm{})`$, we define its principal symbol as $`\sigma (f)=f(q,๐,0)`$. Here, $`๐_1,\mathrm{},๐_r`$ are conjugate variables. We also define
$`\sigma (M_{\mathrm{Mell}})`$ $`=[q^\pm ,๐]/\sigma (I_{\mathrm{Mell}}),`$
$`\sigma (\stackrel{~}{FH}_{S^1}^{})`$ $`=\{q\}[๐]_\sigma \stackrel{~}{FH}_{S^1}^{}\{q\}[๐]/\sigma (\overline{I}_{\mathrm{poly}}).`$
The characteristic varieties of $`\mathrm{Ch}(M_{\mathrm{Mell}})`$ and $`\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})`$ are defined as analytic spectra of $`\sigma (M_{\mathrm{Mell}})`$ and $`\sigma (\stackrel{~}{FH}_{S^1}^{})`$ respectively. They are (germs of) analytic subvarieties of $`(^{})^r\times ^r=T^{}(^{})^r`$. By the surjection (26), we have an embedding
$$\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})|_{q_1\mathrm{}q_r0}\mathrm{Ch}(M_{\mathrm{Mell}})$$
The characteristic variety of $`M_{\mathrm{Mell}}`$ is the zero set of $`\sigma (๐ซ_๐
)`$ for all $`๐
`$. Because there exists $`๐
`$ such that $`w_i,๐
>0`$ for all $`i`$, we can see from $`\sigma (๐ซ_๐
)=0`$ that $`_{b=1}^rm_{ib}๐_b0`$ if $`q(^{})^r`$. (Here, we need the compactness of $`X`$.) Thus,
$$\mathrm{Ch}(M_{\mathrm{Mell}})=\left\{(q,๐)(^{})^r\times ^r\right|\underset{b=1}{\overset{r}{}}m_{ib}๐_b0,q_a=\underset{i=1}{\overset{r+N}{}}\left(\underset{b=1}{\overset{r}{}}m_{ib}๐_b\right)^{m_{ia}}\}.$$
Let $`\mathrm{Crit}(F_q)`$ be the set of critical points of $`F_q`$ in $`Y_q`$. This forms a family over $`(^{})^r`$. For $`x\mathrm{Crit}(F_q)`$, the cotangent vector $`d_xF`$ in $`T_x^{}Y`$ equals $`\pi ^{}(_{a=1}^r๐_ad\mathrm{log}q_a)`$ for some $`๐_a`$. Hence we have a map $`dF:_{q(^{})^r}\mathrm{Crit}(F_q)T^{}(^{})^r`$, $`x_{a=1}^r๐_ad\mathrm{log}q_a`$.
###### Lemma 5.10.
$`dF:_{q(^{})^r}\mathrm{Crit}(F_q)\mathrm{Ch}(M_{\mathrm{Mell}})T^{}(^{})^r`$.
###### Proof.
By $`dF=\pi ^{}(_{a=1}^r๐_ad\mathrm{log}q_a)`$ on $`_q\mathrm{Crit}(F_q)`$, we see that $`๐_i=_{a=1}^rm_{ia}๐_a`$ and so $`(q,๐)\mathrm{Ch}(M_{\mathrm{Mell}})`$. The inverse is given by $`๐_i=_{a=1}^rm_{ia}๐_a`$. โ
###### Proposition 5.11.
(i) The projection $`\mathrm{Ch}(M_{\mathrm{Mell}})(^{})^r`$ is a ramified covering of degree $`N!\mathrm{Vol}(\mathrm{Conv}(\mathrm{\Sigma }))`$, where $`\mathrm{Conv}(\mathrm{\Sigma })`$ is the convex hull of all the primitive generators $`\stackrel{}{x}_1,\mathrm{},\stackrel{}{x}_{r+N}`$ of 1-dimensional cones in the fan $`\mathrm{\Sigma }`$ defining $`X`$.
(ii) There exist exactly $`dim_{}H^{}(X)`$ branches of the covering $`\mathrm{Ch}(M_{\mathrm{Mell}})(^{})^r`$ corresponding to the subcovering $`\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})|_{q_1\mathrm{}q_r0}(^{})^r`$. These branches are characterized by the condition that $`๐_a0`$ as $`q0`$.
###### Proof.
By Kushnirenkoโs theorem , the dimension of Jacobi ring $`[s_1,\mathrm{},s_N]/_sF_q`$ equals $`N!`$ times the volume of the Newton polytope of $`F_q=_{i=1}^{r+N}(_aq_a^{l_{ia}})s^{\stackrel{}{x}_i}`$. On the other hand, the projection $`\pi :\mathrm{Ch}(M_{\mathrm{Mell}})(^{})^r`$ is a submersion at generic $`๐`$. (In fact, $`d\pi =\mathrm{log}q_a/๐_b=_im_{ia}๐_i^1m_{ib}`$ is positive definite when $`๐_i>0`$.) Hence we obtain (i) by Lemma 5.10. Each branch of $`\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})`$ corresponds to a branch of simultaneous eigenvalues $`(๐_1,\mathrm{},๐_r)`$ of the connection matrices $`(\mathrm{\Omega }_1|_{\mathrm{}=0},\mathrm{},\mathrm{\Omega }_r|_{\mathrm{}=0})`$. Because $`\mathrm{\Omega }_a|_{q=0}`$ is a nilpotent matrix (cup product by $`p_a`$), we see that $`๐_a0`$ as $`q0`$. On the other hand, we have $`\mathrm{Ch}(M_{\mathrm{Mell}})\mathrm{Ch}(FH_0)|_{q_1\mathrm{}q_r0}`$ and the zero-fiber of $`\mathrm{Ch}(FH_0)`$ is the spectrum of $`\sigma (FH_0)/_{a=1}^rq_a\sigma (FH_0)\sigma (FH_0/๐ชFH_0)\sigma (\widehat{FH}_0/๐ช\widehat{FH}_0)H^{}(X)`$. Therefore, there exist only $`dim_{}H^{}(X)`$ branches converging to $`๐=0`$ at $`q=0`$. โ
By Theorem 5.7, we see that $`๐ข๐ช๐^{}(\mathrm{Ch}(QDM_{\mathrm{an}}^{}(X)))\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})`$. Therefore, the generic fiber of the characteristic variety of $`QDM_{\mathrm{an}}^{}(X)`$ consists of $`dim_{}H^{}(X)`$ distinct points. This means that
###### Corollary 5.12.
The quantum cohomology of a smooth, projective toric variety is generically semisimple. (See , Part I, section 3 for the definition of semisimplicity.)
As remarked in the introduction, this corollary together with Kawamataโs result shows that Bayer and Maninโs modified Dubrovinโs conjecture holds for toric varieties.
The proposition below will clarify the role of the nef condition for $`c_1(X)`$.
###### Proposition 5.13.
The following conditions are equivalent.
(i) $`c_1(X)`$ is nef.
(ii) $`N!\mathrm{Vol}(\mathrm{Conv}(\mathrm{\Sigma }))=dim_{}H^{}(X)`$.
(iii) $`\mathrm{Ch}(\stackrel{~}{FH}_{S^1}^{})|_{q_1\mathrm{}q_r0}\mathrm{Ch}(M_{\mathrm{Mell}})`$ on a small neighborhood of $`q=0`$.
(iv) $`\stackrel{~}{FH}_{S^1}^{}FH_0_{[\mathrm{},q]}๐ช_{\mathrm{small}}^{\mathrm{}}(๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}/๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}})`$.
(v) $`FH_{S^1}^{}FH_0_{[\mathrm{},q]}[\mathrm{}][[q]]([\mathrm{}][[q]]\mathrm{}/[\mathrm{}][[q]]I_{\mathrm{poly}})`$.
(vi) The generalized mirror transformation can be done using only convergent power series i.e. $`\mathrm{\Omega }_a(q,\mathrm{})`$ and $`g(q,\mathrm{})`$ in the step (i) and (ii) of Theorem 5.6 are convergent functions of $`q`$ and $`\mathrm{}`$.
###### Proof.
(i) $``$ (ii): The condition $`c_1(X)0`$ is equivalent to that every primitive generator $`\stackrel{}{x}_i`$ of one-dimensional cones lies in the boundary of $`\mathrm{Conv}(\stackrel{}{x}_1,\mathrm{},\stackrel{}{x}_{r+N})`$. The number of top dimensional cones in $`\mathrm{\Sigma }`$ equals $`dim_{}H^{}(X)`$ and each top dimensional cone has volume $`1/N!`$. Hence we obtain the equivalence.
(ii) $``$ (iii): This follows from Proposition 5.11.
(iii) $``$ (iv): Set $`M:=๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}/๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}}`$. Let $`[f(q,\mathrm{},\mathrm{})]`$ be an element of the kernel of the natural surjection $`M\stackrel{~}{FH}_{S^1}^{}`$. We can assume that $`f(q,\mathrm{},\mathrm{})`$ is homogeneous. By the assumption, we can see that there exists an integer $`k>0`$ such that $`(q_1\mathrm{}q_r)^kf(q,๐,0)`$ is contained in $`\sigma (๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}})`$. Because $`\sigma (๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}})=_๐
\{q\}[q^1]\sigma (๐ซ_๐
)\{q\}`$, we can see that $`f(q,๐,0)`$ is also in $`\sigma (๐ช_{\mathrm{poly}}^{\mathrm{}}I_{\mathrm{poly}})`$. Therefore, there exists some $`f_1(q,\mathrm{},\mathrm{})`$ in $`๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}`$ such that $`f(q,\mathrm{},\mathrm{})\mathrm{}f_1(q,\mathrm{},\mathrm{})mod๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}}`$ and $`\mathrm{deg}f_1=\mathrm{deg}f2`$. Then, $`[\mathrm{}f_1]`$ is in the kernel of $`M\stackrel{~}{FH}_{S^1}^{}`$. Because the multiplication by $`\mathrm{}`$ is injective in $`\stackrel{~}{FH}_{S^1}^{}`$, $`[f_1]`$ also lies in the kernel. Repeating this, we have $`f\mathrm{}^nf_nmod๐ช_{\mathrm{small}}^{\mathrm{}}I_{\mathrm{poly}}`$ and $`\mathrm{deg}f_n<0`$ for sufficiently large $`n`$. Because we have already seen that (i) is equivalent to (iii), $`c_1(X)`$ is nef and so $`\mathrm{deg}q_a0`$. Thus $`f_n`$ must be zero and $`[f]=0`$ in $`M`$.
(iv) $``$ (v): Take the tensor product with $`[\mathrm{}][[q]]`$.
(v) $``$ (ii): Because $`FH_{S^1}^{}`$ is a free $`[\mathrm{}][[q]]`$-module of rank $`dim_{}H^{}(X)`$, we see that $`\sigma (FH_0)_{[q]}[[q]]`$ is a free $`[[q]]`$-module of rank $`dim_{}H^{}(X)`$. On the other hand, by Proposition 5.11 $`\sigma (FH_0)[q^1]\sigma (M_{\mathrm{Mell}})`$ becomes a free module of rank $`N!\mathrm{Vol}(\mathrm{Conv}(\mathrm{\Sigma }))`$ when restricted to the complement of the ramification locus. Therefore, we must have $`dim_{}H^{}(X)=N!\mathrm{Vol}(\mathrm{Conv}(\mathrm{\Sigma }))`$.
(i) $``$ (v): We can see from the proof of Proposition 5.5 that the $`I`$-function in (28) is convergent on $`|\mathrm{}|=1`$ if and only if $`c_1(X)`$ is nef. The connection matrix $`\mathrm{\Omega }_a`$ is determined from $`I`$-function by (27). The gauge transformation $`g`$ can be found by applying Birkhoff factorization of the loop group $`LGL(s+1,)`$ to the loop $`\mathrm{}(T_i(\mathrm{})I(q,\mathrm{}))`$ (see ). โ
### 5.5. Example
We study the case of Hirzeburch surface $`๐ฝ_n=(๐ช_^1๐ช_^1(n))`$. This is given by the fan $`\mathrm{\Sigma }`$ whose one-dimensional cones are generated by $`\stackrel{}{x}_1,\mathrm{},\stackrel{}{x}_4`$:
$$\stackrel{}{x}_1=(1,0),\stackrel{}{x}_2=(0,1),\stackrel{}{x}_3=(1,n),\stackrel{}{x}_4=(0,1).$$
We have $`dim_{}H^{}(๐ฝ_n)=4`$ and $`2!\mathrm{Vol}(\mathrm{Conv}(\mathrm{\Sigma }))=\mathrm{max}(4,n+2)`$. Let $`w_1,\mathrm{},w_4`$ be the classes of corresponding toric divisors. Then, $`p_1:=w_1`$ and $`p_2:=w_4`$ form a nef integral basis and we have $`w_2=p_2np_1`$ and $`w_3=p_1`$. The relation of classical cohomology ring is given by $`p_1^2=0`$ and $`p_2^2=np_1p_2`$. The corresponding Mellin system is generated by the following two differential operators:
$$๐ซ_{(1,0)}=q_1W_2(W_2\mathrm{})\mathrm{}(W_2(n1)\mathrm{})W_1W_3,๐ซ_{(0,1)}=q_2W_4W_2,$$
where $`W_1=W_3=\mathrm{}_1`$, $`W_2=\mathrm{}_2n\mathrm{}_1`$, $`W_4=\mathrm{}_2`$.
$$FH_0q_1,q_2,\mathrm{}_1,\mathrm{}_2,\mathrm{}/๐ซ_{(1,0)},๐ซ_{(0,1)},M_{\mathrm{Mell}}FH_0[q_1^1,q_2^2]$$
$$FH_{S^1}^{}[\mathrm{}][[q_1,q_2]]\mathrm{}_1,\mathrm{}_2/\overline{๐ซ_{(1,0)},๐ซ_{(0,1)}}.$$
#### 5.5.1. Fano case $`(n=0,1)`$
In this case, $`FH_0`$ is freely generated by $`[1],[\mathrm{}_1],[\mathrm{}_2],[(\mathrm{}_2)^2]`$ over $`[\mathrm{},q]`$. The small quantum $`D`$-module of $`๐ฝ_n`$ is defined over the polynomial ring $`[\mathrm{},q]`$ and is isomorphic to $`FH_0`$.
#### 5.5.2. Nef but non-Fano case $`(n=2)`$
In this case, $`FH_0`$ is not finitely generated over $`[\mathrm{},q]`$. However, $`FH_0[(14q_1)^1]`$ is freely generated by $`[1],[\mathrm{}_1],[\mathrm{}_2],[(\mathrm{}_2)^2]`$ over $`[\mathrm{},q_1,q_2,(14q_1)^1]`$. For example, we can write $`[(\mathrm{}_1)^2]`$ as
$$[(\mathrm{}_1)^2]=\frac{2q_1q_2}{14q_1}+\frac{2\mathrm{}q_1}{14q_1}[\mathrm{}_1]\frac{\mathrm{}q_1}{14q_1}[\mathrm{}_2]\frac{q_1}{14q_1}[(\mathrm{}_2)^2]$$
The small quantum $`D`$-module is isomorphic to $`FH_0`$ at least on the region $`|4q_1|<1`$ after a suitable coordinate change (see ).
#### 5.5.3. Non-nef case $`(n3)`$
In this case, $`FH_0`$ is not finitely generated over $`[\mathrm{},q]`$. Furthermore, $`FH_0_{[\mathrm{},q]}[\mathrm{}][[q]][\mathrm{}][[q]]\mathrm{}/๐ซ_{(1,0)},๐ซ_{(0,1)}`$ is not finitely generated over $`[\mathrm{}][[q]]`$ either. If it were finitely generated, then $`\sigma (FH_0)_{[q]}[[q]]`$ would be also finitely generated over $`[[q]]`$ and generated by $`1,๐_1,๐_2,๐_2^2`$ by Nakayamaโs lemma. It contradicts that $`\sigma (FH_0)[q_1^1,q_2^1]\sigma (M_{\mathrm{Mell}})`$ is a free module of rank $`n+2>4`$. We must consider the $`q`$-adic closure of the left ideal $`๐ซ_{(1,0)},๐ซ_{(0,1)}`$ which is strictly bigger:
$$๐ซ_{(1,0)},๐ซ_{(0,1)}\overline{๐ซ_{(1,0)},๐ซ_{(0,1)}}\text{ in }[\mathrm{}][[q]]\mathrm{}.$$
Our $`FH_{S^1}^{}`$ is a free $`[\mathrm{}][[q]]`$-module of rank 4 and isomorphic to the restriction of the big quantum $`D`$-module to some non-linear subvariety of $`H^{}(๐ฝ_n)`$. On the other hand, $`M_{\mathrm{Mell}}`$ is of rank $`n+2`$ over $`[\mathrm{},q^\pm ]`$. (In fact, it is generated by $`[1],[\mathrm{}_1],[(\mathrm{}_1)^2],[\mathrm{}_2n\mathrm{}_1],\mathrm{},[(\mathrm{}_2n\mathrm{}_1)^{n1}]`$.) The characteristic variety of $`M_{\mathrm{Mell}}`$ is given by
$$q_1(๐_2n๐_1)^n=๐_1^2,q_2=(๐_2n๐_1)๐_2$$
or equivalently,
$$n^2q_1q_2^n=๐_2^{n2}(๐_2^2q_2)^2,q_2=(๐_2n๐_1)๐_2.$$
It has $`n+2`$ solutions $`(๐_1,๐_2)`$ (counted with multiplicity) for a given $`(q_1,q_2)(^{})^2`$. Out of $`n+2`$ solutions, we have four branches of solutions having the asymptotics:
$$๐_1(q_1(\sqrt{q_2})^n)^{1/2},๐_2\sqrt{q_2}+\frac{n}{2}(q_1(\sqrt{q_2})^n)^{1/2}.$$
These branches satisfy $`(๐_1,๐_2)0`$ as $`(q_1,q_2)0`$ and correspond to the characteristic variety of $`\stackrel{~}{FH}_{S^1}^{}`$. The other $`n2`$ branches have the asymptotics
$$๐_1n^{\frac{n4}{n2}}q_1^{\frac{1}{n2}},๐_2n^{\frac{2}{n2}}q_1^{\frac{1}{n2}}q_2$$
and diverge as $`q_10`$.
By changing the large radius limit $`q0`$, we can construct a $`D`$-module of rank $`n+2`$ which is regular singular along $`\widehat{q}_1\widehat{q}_2=0`$. Set $`\widehat{q}_1=q_1^1`$ and $`\widehat{q}_2=q_2q_1`$. In terms of $`\widehat{q}_1,\widehat{q}_2`$, differential operators of the Mellin system are written as
$$_{(1,0)}=\widehat{q}_1W_1W_3W_2(W_2\mathrm{})\mathrm{}(W_2(n1)\mathrm{}),_{(1,1)}=\widehat{q}_1\widehat{q}_2W_2W_4,$$
$$_{(0,1)}=\widehat{q}_2W_2(W_2\mathrm{})\mathrm{}(W_2(n2)\mathrm{})W_1W_3W_4.$$
Then, the $`D`$-module
$$FH_0^{}=\widehat{q}_1,\widehat{q}_2,\mathrm{}_1,\mathrm{}_2,\mathrm{}/_{(1,0)},_{(1,1)},_{(0,1)}$$
is freely generated by $`[1],[\mathrm{}_1],[(\mathrm{}_1)^2],[\mathrm{}_2n\mathrm{}_1],\mathrm{},[(\mathrm{}_2n\mathrm{}_1)^{n1}]`$ as a $`[\mathrm{},\widehat{q}_1,\widehat{q}_2]`$-module. This becomes an abstract quantum $`D`$-module in the sense of . Therefore, we can find a canonical frame and flat coordinates by reconstructing an $`n+2`$-dimensional base space (see ). It would be interesting to study a geometric meaning of this $`D`$-module.
## 6. $`R`$-conjecture (Virasoro constraints)
The Virasoro constraints are infinite dimensional symmetries of the (all genus, descendant) Gromov-Witten potential conjectured by Eguchi-Hori-Xiong . In , Givental showed that when the target $`X`$ has a torus action with only isolated fixed points and isolated one-dimensional orbits, the Virasoro conjecture is reduced to the $`R`$-conjecture. The $`R`$-conjecture is a conjecture about equivariant quantum cohomology with semisimple non-equivariant counterpart. It states that the asymptotic solution $`R`$ defined in the equivariant theory has non-equivariant limit. For a class of Fano toric varieties, the $`R`$-conjecture was proved by Givental . The $`R`$-conjecture has been proved for complete flag , Grassmannian and partial flag varieties .
In this section, we explain the $`R`$-conjecture and prove it for any smooth projective toric variety. The proof of $`R`$-conjecture will be a first step for the understanding of the mirror oscillatory integral.
### 6.1. Equivariant quantum cohomology
Let $`๐`$ be an $`l`$-dimensional torus ($`(S^1)^l`$) and $`X`$ be a $`๐`$-manifold. In a manner analogous to section 2, we can define the equivariant quantum cohomology algebra $`(QH_๐^{}(X),)`$ which is a deformation of the ring structure of the ordinary equivariant cohomology $`H_๐^{}(X)`$ . We assume that the $`๐`$-action on $`X`$ is Hamiltonian. For projective manifolds, this assumption is equivalent to that the action has at least one fixed point. Then, the equivariant cohomology is of the form $`H_๐^{}(X)H^{}(X)H_๐^{}(\mathrm{pt})`$. This isomorphism is not canonical, but we choose homogeneous equivariant lifts $`p_0,\mathrm{},p_s`$ of the basis in section 2. Let $`t_0,\mathrm{},t_s`$ be linear coordinates dual to $`p_0,\mathrm{},p_s`$ and put $`H_๐^{}(\mathrm{pt})=[\lambda _1,\mathrm{},\lambda _l]`$. Set $`q_a=\mathrm{exp}(t_a)`$ for $`1ar`$. The equivariant quantum cohomology is of the form
$$QH_๐^{}(X)=H_๐^{}(X)\widehat{}[[t_0,q_1,\mathrm{},q_r,t_{r+1},\mathrm{},t_s]]=H^{}(X)[\lambda ][[x]].$$
Here, $`[\lambda ][[x]]`$ is shorthand for $`[\lambda _1,\mathrm{},\lambda _l][[t_0,q_1,\mathrm{},q_r,t_{r+1},\mathrm{},t_s]]`$ and $`\widehat{}`$ means the tensor product completed in the $`x`$-adic topology.
Our $`(QH_๐^{}(X),)`$ defines a quasi-conformal Frobenius structure on the base space $`=\mathrm{Spec}((\lambda )[[x]][q_1^1,\mathrm{},q_r^1])`$. Let $`๐ฏ_{}`$ be the relative tangent sheaf of $`\mathrm{Spec}(\lambda )`$. Then, each element of $`QH_๐^{}(X)`$ gives a section of $`๐ฏ_{}`$ by the correspondence $`p_i_i:=/t_i`$. The quasi-conformal Frobenius structure of $`QH_๐^{}(X)`$ consists of the following data:
1. Flat $`(\lambda )`$-bilinear symmetric pairing $`,_๐`$ on $`๐ฏ_{}`$ defined by $`_i,_j_๐=_X^๐p_ip_j`$.
2. $`(\lambda )`$-bilinear symmetric product $`:๐ฏ_{}๐ฏ_{}๐ฏ_{}`$ satisfying $`_i_j,_k_๐=_i,_j_k_๐`$.
3. Flat unit section $`1=p_0=_0๐ฏ_{}`$.
4. Euler operator $`\stackrel{~}{E}`$ defined by
$$\stackrel{~}{E}=2t_0\frac{}{t_0}+2c_0(\lambda )\frac{}{t_0}+\underset{a=1}{\overset{r}{}}2c_a\frac{}{t_a}+\underset{j=r+1}{\overset{s}{}}(2\mathrm{deg}p_j)t_j\frac{}{t_j}+\underset{j=1}{\overset{l}{}}2\lambda _j\frac{}{\lambda _j},$$
where $`c_0(\lambda )+_{a=1}^rc_ap_a=c_1^๐(TX)`$.
The name quasi-conformal comes from that $`\stackrel{~}{E}`$ is not a section of $`๐ฏ_{}`$ (it contains the derivation $`\lambda _j/\lambda _j`$). As in non-equivariant case, the dual Givental connection $`_j^{\mathrm{}}=\mathrm{}(/t_j)+p_j`$ on $`๐ฏ_{}`$ is flat for any value of $`\mathrm{}`$. This defines the $`๐`$-equivariant quantum $`D`$-module:
$$QDM_๐^{}(X)=(H^{}(X)[\lambda ,\mathrm{}][[x]],^{\mathrm{}}).$$
The Euler operator satisfies the following:
$`[\stackrel{~}{E},V_1V_2]`$ $`=V_1[\stackrel{~}{E},V_2]+[\stackrel{~}{E},V_1]V_2+2V_1V_2,`$
$`\stackrel{~}{E}V_1,V_2_๐`$ $`=[\stackrel{~}{E},V_1],V_2_๐+V_1,[\stackrel{~}{E},V_2]_๐+(42N)V_1,V_2_๐,`$
for $`V_1,V_2๐ฏ_{}`$ and $`N=dim_{}X`$.
We assume some familiarity with the semisimplicity and canonical coordinates for conformal Frobenius manifolds, see e.g. . Here, we review the construction of canonical coordinates in equivariant quantum cohomology. Assume that the $`๐`$ action on $`X`$ has only isolated fixed points. Then, by the localization theorem of equivariant cohomology, we have an isomorphism of rings
$$H_๐^{}(X)_{[\lambda ]}(\lambda )\underset{\sigma X^๐}{}(\lambda )\psi _\sigma ,\psi _\sigma =\frac{i_\sigma ([\sigma ])}{๐^๐(T_\sigma X)}.$$
Here, $`i_\sigma :\{\sigma \}X`$ is the inclusion and $`๐^๐`$ is the $`๐`$-equivariant Euler class. Because $`i_\sigma ^{}(\psi _\tau )=\delta _{\sigma \tau }`$ for $`\sigma ,\tau X^๐`$, we have $`\psi _\sigma \psi _\tau =\delta _{\sigma \tau }\psi _\sigma `$. The localized equivariant quantum cohomology is also semisimple. The idempotent $`\psi _\sigma `$ can be deformed to the idempotent $`\psi _\sigma (x)`$ of $`QH_๐^{}(X)\widehat{}_{[\lambda ]}(\lambda )`$ such that $`lim_{q0}\psi _\sigma (x)=\psi _\sigma `$ and $`\psi _\sigma (x)\psi _\tau (x)=\delta _{\sigma \tau }\psi _\sigma (x)`$. The projection to the $`\psi _\sigma (x)`$-component $`QH_๐^{}(X)\widehat{}_{[\lambda ]}(\lambda )(\lambda )[[x]]`$ corresponds to a closed 1-form on $``$. A primitive $`๐_\sigma (t,\lambda )`$ of this closed 1-form is called a canonical coordinate. Then, we can write $`\psi _\sigma (x)=/๐_\sigma `$. Note that $`d๐_\sigma `$ is characterized by
(33)
$$V\psi _\sigma (x)=d๐_\sigma (V)\psi _\sigma (x),V๐ฏ_{}.$$
The canonical coordinates $`\{๐_\sigma \}_{\sigma X^๐}`$ are determined up to functions in $`\lambda _i`$.
Put $`\alpha (\sigma )=i_\sigma ^{}(\alpha )`$ for $`\alpha H_๐^{}(X)`$. Let $`T_\sigma ^{}X\chi _1(\sigma )\mathrm{}\chi _N(\sigma )`$ be the weight decomposition, where $`\chi _j(\sigma )[\lambda ]`$. Although $``$ does not contain the divisor $`q_a=0`$, the limit $`lim_{q0}d๐_\sigma `$ exists as an element of $`(H^{}(X))^{}(\lambda )`$ because the dual basis $`\psi _\sigma (x)`$ is in $`H^{}(X)(\lambda )[[x]]`$. Since $`lim_{q0}d๐_\sigma `$ is the projection to $`\psi _\sigma `$, we have $`d๐_\sigma =_{i=0}^s(p_i(\sigma )+O(q))dt_i`$, where $`O(q)(\lambda )[[x]]`$. We normalize canonical coordinates by the following classical limit condition:
(34)
$$๐_\sigma (t,\lambda )=\underset{j=1}{\overset{N}{}}(\chi _j(\sigma )+\chi _j(\sigma )\mathrm{log}(\chi _j(\sigma )))+\underset{i=0}{\overset{s}{}}p_i(\sigma )t_i+O(q),$$
where $`O(q)(\lambda )[[x]]`$. This together with (33) determines $`๐_\sigma (t,\lambda )`$ uniquely. Using $`c_1^๐(T_\sigma X)=c_0(\lambda )+_{a=1}^rc_ap_a(\sigma )=_{j=1}^N\chi _j(\sigma )`$, we can check that this $`๐_\sigma `$ automatically satisfies the homogeneity $`\stackrel{~}{E}๐_\sigma =2๐_\sigma `$.
Assume that $`QH_๐^{}(X)`$ has convergent structure constants around $`x=\lambda =0`$. Then $`๐_\sigma (t,\lambda )`$ becomes a multi-valued analytic function. We conjecture the following which resembles the $`R`$-conjecture.
###### Conjecture 6.1.
When the non-equivariant quantum cohomology is generically semisimple, canonical coordinates $`๐_\sigma (t,\lambda )`$ normalized by (34) is regular at $`\lambda =0`$ for a semisimple point $`t`$ of non-equivariant theory.
Let $`๐_\sigma ^{}(t,\lambda )`$ be any canonical coordinate which is homogeneous $`\stackrel{~}{E}๐_\sigma ^{}=2๐_\sigma ^{}`$ and is regular at $`\lambda =0`$ for semisimple $`t`$. If this conjecture is true, such a canonical coordinate is different from the above normalization only by a linear form $`_{i=1}^la_i\lambda _i`$. Note that the non-equivariant limit $`\lambda 0`$ of $`๐_\sigma (t,\lambda )`$ gives the canonical coordinate $`๐(t)`$ of non-equivariant theory satisfying the homogeneity $`E๐(t)=2๐(t)`$. (In non-equivariant theory, homogeneous canonical coordinates are unique up to order.) Later, we will see that this conjecture (together with $`R`$-conjecture stated below) holds for toric variety using the equivariant mirror.
Consider the following differential equation for a section $`s(t,\lambda ,\mathrm{})๐ฏ_{}`$.
(35)
$$\mathrm{}\frac{}{t_i}s(t,\lambda ,\mathrm{})=p_is(t,\lambda ,\mathrm{}).$$
An asymptotic solution $`s_\tau (t,\lambda ,\mathrm{})`$ for this differential equation is the solution of the form
$$s_\tau (t,\lambda ,\mathrm{})=\underset{\sigma }{}R_{\sigma \tau }(t,\lambda ,\mathrm{})e^{๐_\tau (t,\lambda )/\mathrm{}}\mathrm{}_\sigma \frac{}{๐_\sigma },\mathrm{}_\sigma =\frac{}{๐_\sigma },\frac{}{๐_\sigma }_๐^{1/2},$$
where $`R_{\sigma \tau }(t,\lambda ,\mathrm{})=\delta _{\sigma \tau }+R_{\sigma \tau }^{(1)}(t,\lambda )\mathrm{}+R_{\sigma \tau }^{(2)}(t,\lambda )\mathrm{}^2+\mathrm{}`$ is an asymptotic series in $`\mathrm{}`$. The matrix $`R_{\sigma \tau }`$ is uniquely determined by the classical limit condition:
(36)
$$\underset{q0}{lim}R_{\sigma \tau }(t,\lambda ,\mathrm{})=\delta _{\sigma \tau }\mathrm{exp}(b_\sigma (\mathrm{},\lambda )),b_\sigma (\mathrm{},\lambda )=\underset{k=1}{\overset{\mathrm{}}{}}N_{2k1}(\sigma )\frac{B_{2k}}{2k}\frac{\mathrm{}^{2k1}}{2k1},$$
where $`N_{2k1}(\sigma )=_{j=1}^N1/\chi _j(\sigma )^{2k1}`$. Since $`lim_{q0}\mathrm{}_\sigma =๐^๐(T_\sigma X)^{1/2}`$, we can see that $`R_{\sigma \tau }^{(n)}(t,\lambda )`$ belongs to $`(\lambda ,๐^๐(T_\sigma X)^{1/2})[[x]]`$. Under this normalization, $`R_{\sigma \tau }`$ automatically satisfies the following homogeneity and the unitarity:
$$(2\mathrm{}\frac{}{\mathrm{}}+\stackrel{~}{E})R_{\sigma \tau }(t,\lambda ,\mathrm{})=0,\underset{\nu X^๐}{}R_{\nu \sigma }(t,\lambda ,\mathrm{})R_{\nu \tau }(t,\lambda ,\mathrm{})=\delta _{\sigma \tau }.$$
###### Conjecture 6.2 ($`R`$-conjecture ).
When the non-equivariant quantum cohomology is generically semisimple, the asymptotic solution $`R_{\sigma \tau }^{(n)}(t,\lambda )`$ normalized by (36) is regular at $`\lambda =0`$ for a semisimple point $`t`$ of non-equivariant theory.
An asymptotic solution $`R_{\sigma \tau }(t,\lambda ,\mathrm{})`$ satisfying the homogeneity and regularity at $`\lambda =0`$ for a semisimple $`t`$ is unique if it exists. Thus, if $`R`$-conjecture is true, such a solution exists and satisfies the above classical limit condition (36). In this case, the non-equivariant limit $`lim_{\lambda 0}R_{\sigma \tau }`$ gives the homogeneous asymptotic solution of non-equivariant theory.
By the theory of Givental , the $`R`$-conjecture implies the Virasoro constraints for the non-equivariant Gromov-Witten theory of $`X`$.
### 6.2. Equivariant mirror
We review an equivariant version of the mirror of toric variety . We use the same notation as in section 5.1. The torus $`๐:=(S^1)^{r+N}`$ acts on a toric variety $`X=_{}^{r+N}/(^{})^r`$ by $`(z_1,\mathrm{},z_{r+N})(t_1z_1,\mathrm{},t_{r+N}z_{r+N})`$. Here, $`w_1,\mathrm{},w_rH_๐^2(X)`$ denotes the $`๐`$-equivariant Poincarรฉ duals of torus invariant prime divisors $`D_1,\mathrm{},D_{r+N}`$. In other words, $`w_i`$ is the $`๐`$-equivariant first Chern class of the line bundle
$$_{}^{r+N}\times /(z_1,\mathrm{},z_{r+N},v)(t^{w_1}z_1,\mathrm{},t^{w_{r+N}}z_{r+N},t^{w_i}v),(t_1,\mathrm{},t_r)(^{})^r$$
with $`๐`$-action $`(z_1,\mathrm{},z_{r+N},v)(t_1z_1,\mathrm{},t_{r+N}z_{r+N},t_iv)`$. Let $`p_a`$ denote an equivariant lift of the nef integral basis in section 5.1. More precisely, we define $`p_a`$ as the $`๐`$-equivariant first Chern class of the line bundle
$$_{}^{r+N}\times /(z_1,\mathrm{},z_{r+N},v)(t^{w_1}z_1,\mathrm{},t^{w_{r+N}}z_{r+N},t_av),(t_1,\mathrm{},t_r)(^{})^r.$$
with $`๐`$-action $`(z_1,\mathrm{},z_{r+N},v)(t_1z_1,\mathrm{},t_{r+N}z_{r+N},v)`$. Then, $`w_i`$ is written as
$$w_i=\underset{a=1}{\overset{r}{}}m_{ia}p_a\lambda _j.$$
Define an equivariant phase function $`F^๐(๐;\lambda )`$ on the mirror family $`\pi :Y(^{})^r`$ by $`F^๐(๐;\lambda )=_{i=1}^{r+N}(๐_i+\lambda _i\mathrm{log}๐_i)`$. Set $`F_q^๐=F^๐|_{Y_q}`$. Then consider the equivariant oscillatory integral:
$$_\mathrm{\Gamma }(q,\lambda ,\mathrm{})=_{\mathrm{\Gamma }_qY_q}e^{F_q^๐/\mathrm{}}\omega _q.$$
We can easily obtain the equivariant version of Proposition 5.1.
###### Proposition 6.3.
The equivariant oscillatory integral $`_\mathrm{\Gamma }`$ satisfies $`๐ซ_๐^๐_\mathrm{\Gamma }(q,\lambda ,\mathrm{})=0`$ for all $`๐^r`$, where
$$๐ซ_๐
^๐=q^๐
\underset{w_i,๐
<0}{}\underset{k=0}{\overset{w_i,๐
1}{}}\left(\underset{a=1}{\overset{r}{}}m_{ia}\mathrm{}_a\lambda _ik\mathrm{}\right)\underset{w_i,๐
>0}{}\underset{k=0}{\overset{w_i,๐
1}{}}\left(\underset{a=1}{\overset{r}{}}m_{ia}\mathrm{}_a\lambda _ik\mathrm{}\right).$$
We introduce the $`D`$-module $`M_{\mathrm{Mell}}^๐`$ as in section 5.1.
$$M_{\mathrm{Mell}}^๐=q^\pm ,\mathrm{},\mathrm{},\lambda /I_{\mathrm{Mell}}^๐,I_{\mathrm{Mell}}^๐=\underset{๐
^r}{}q^\pm ,\mathrm{},\mathrm{},\lambda ๐ซ_๐
^๐$$
### 6.3. $`๐\times S^1`$-equivariant Floer cohomology
We can construct $`๐\times S^1`$-equivariant Floer cohomology $`FH_{๐\times S^1}^{}`$ in an analogous manner to section 5.2. First, $`H_{๐\times S^1}^{\mathrm{}/2}`$ is defined by the limit of the inductive system ($`H_{๐\times S^1}^{}(L_๐
^{\mathrm{}})`$, $`i_{๐
,๐
^{}}`$), where $`H_{๐\times S^1}^{}(L_๐
^{\mathrm{}})=[P,\mathrm{},\lambda ]`$, $`i_{๐
,๐
^{}}(\alpha )=\alpha _{i=1}^N_{k=w_i,๐
^{}}^{w_i,๐
1}W_{i,k}`$ and $`W_{i,k}=_{a=1}^rm_{ia}P_a\lambda _ik\mathrm{}`$. It has an action of $`P,Q^\pm ,\mathrm{},\lambda `$. Let $`\mathrm{\Delta }^๐`$ be the image of $`1`$ in $`H_{๐\times S^1}^{}(L_0^{\mathrm{}})`$. Let $`FH_0^๐`$ be a $`P,Q,\mathrm{},\lambda `$-submodule of $`H_{๐\times S^1}^{\mathrm{}/2}`$ generated by $`\mathrm{\Delta }^๐`$. Finally, we define $`FH_{๐\times S^1}^{}`$ as the $`Q`$-adic completion of $`FH_0^๐`$. We can check that (by the argument in ) $`FH_{๐\times S^1}^{}`$ is a free module over $`[\mathrm{},\lambda ][[Q]]`$ and has $`\{T_i(P)\mathrm{\Delta }^๐\}_{i=0}^s`$ as a basis, where $`T_i(y_1,\mathrm{},y_r)`$ is the polynomial in Proposition 5.5. By the same argument as Proposition 5.2, 5.4, we obtain
$$H_{๐\times S^1}^{\mathrm{}/2}M_{\mathrm{Mell}}^๐,FH_0^๐q,\mathrm{},\mathrm{},\lambda /I_{\mathrm{poly}}^๐,FH_{๐\times S^1}^{}[\mathrm{},\lambda ][[q]]\mathrm{}/\overline{I_{\mathrm{poly}}^๐},$$
where $`I_{\mathrm{poly}}^๐=I_{\mathrm{Mell}}^๐q,\mathrm{},\mathrm{},\lambda `$ and $`\overline{I_{\mathrm{poly}}^๐}`$ is the closure in the $`q`$-adic topology. The relationship between the $`๐\times S^1`$-equivariant Floer cohomology and $`๐`$-equivariant quantum cohomology is given as follows (c.f. Theorem 5.6).
###### Theorem 6.4.
There exists an embedding $`๐ข๐ช๐:(^r,0)(^{s+1},0)`$ and an isomorphism of $`D`$-modules $`\mathrm{\Phi }_{๐ข๐ช๐}:๐ข๐ช๐^{}(QDM_๐^{}(X))FH_{๐\times S^1}^{}`$. The map $`๐ข๐ช๐`$ is given by the equations
$$\widehat{t}_0=F_0(q;\lambda ),\widehat{q}_1=q_1\mathrm{exp}(F_1(q;\lambda )),\mathrm{},\widehat{q}_r=q_r\mathrm{exp}(F_r(q;\lambda )),\widehat{t}_{r+1}=F_{r+1}(q;\lambda ),\mathrm{},\widehat{t}_s=F_s(q;\lambda )$$
for $`F_i(q;\lambda )[\lambda ][[q]],F_i(0;\lambda )=0`$ and $`\mathrm{\Phi }_{๐ข๐ช๐}|_{q=0}`$ is determined by the canonical isomorphism $`H_๐^{}(X)[\mathrm{}]FH_{๐\times S^1}^{}/_{a=1}^rq_aFH_{๐\times S^1}^{}`$.
This theorem is a generalization of the result of to the $`๐`$-equivariant case. The proof is completely similar and based on a $`๐`$-equivariant version of Coates-Giventalโs quantum Lefschetz theorem . The main points in the proof are (1) any toric variety can be realized as a complete intersection of torus invariant divisors in a Fano toric variety $`\stackrel{~}{X}`$, (2) the $`J`$-function of $`FH_{๐\times S^1}^{}`$ is identical with the Coates-Givental modification of the $`J`$-function of small quantum cohomology of the ambient $`\stackrel{~}{X}`$, and (3) the reconstruction from small to big is unique. We can see that Coates-Giventalโs quantum Lefschetz theorem admits a $`๐`$-equivariant generalization.
### 6.4. Asymptotic solution via equivariant mirror
Using the equivariant mirror, we will construct asymptotic solutions subject to the classical limit condition (36). Then it follows that the $`R`$-conjecture 6.2 holds for toric varieties. Unfortunately, we could not prove the convergence of $`๐`$-equivariant quantum cohomology using the method in this paper. Instead, we use a result in a forthcoming paper which proves the convergence of equivariant quantum cohomology by the localization method.
We start with the construction of canonical coordinates for equivariant mirror. Each fixed point $`\sigma X^๐`$ can be written as an intersection of $`N`$ toric divisors $`_{iI_\sigma }D_i`$, where $`I_\sigma `$ is a subset of $`\{1,2,\mathrm{},r+N\}`$. We can use $`\{๐_i\}_{iI_\sigma }`$ as fiber coordinates of the family $`\pi :Y(^{})^r`$. We write for $`jI_\sigma `$,
$$๐_j=\underset{a=1}{\overset{r}{}}q_a^{l_{aj}^\sigma }\underset{iI_\sigma }{}๐_i^{x_{ji}^\sigma }.$$
Here, the matrix $`l_{aj}^\sigma `$ is the inverse of $`(m_{ja})_{jI^\sigma ,a=1,\mathrm{},r}`$ and $`x_{ji}^\sigma =_{a=1}^rm_{ia}l_{aj}^\sigma `$. Since $`\{p_a\}_{a=1}^r`$ is a nef basis, it follows that the matrix $`l_{aj}^\sigma `$ satisfies $`l_{aj}^\sigma 0`$ and $`_{a=1}^rl_{aj}^\sigma >0`$.
We need the following elementary lemma. See the Appendix for the proof.
###### Lemma 6.5.
Let $`U^l`$ be a neighborhood of $`0`$ and $`DU\{0\}`$ be the complement of an analytic subvariety in $`U`$. Let $`f(q_1,\mathrm{},q_r,\lambda _1,\mathrm{},\lambda _l)`$ be a function holomorphic in the neighborhood of $`\{q=0,\lambda D\}`$. Assume that $`f(,\lambda )`$ can be expanded in the form $`_{๐0}f_๐(\lambda )q^๐`$ for each $`\lambda D`$ and that $`f_๐(\lambda )`$ can be analytically continued to a holomorphic function on $`U`$. Then $`f`$ can be extended to a holomorphic function around $`q=\lambda =0`$.
Put $`\chi _i(\sigma ):=w_i(\sigma )`$. Then we have the following relations.
(37)
$$T_\sigma ^{}X\underset{iI_\sigma }{}\chi _i(\sigma ),\chi _i(\sigma )=\lambda _i+\underset{jI_\sigma }{}\lambda _jx_{ji}^\sigma \text{ for }iI_\sigma ,$$
$$\chi _j(\sigma )=\lambda _j\underset{a=1}{\overset{r}{}}m_{ja}p_a(\sigma )=0\text{ for }jI_\sigma ,$$
###### Lemma 6.6.
(i) For each $`\sigma X^๐`$, there exists a unique branch of critical points $`\{\mathrm{crit}_\sigma (q)\}_q`$ of $`F_q^๐`$ such that in a coordinate system $`\{๐_i\}_{iI_\sigma }`$,
$$๐_i(\mathrm{crit}_\sigma (q))=\chi _i(\sigma )+O(q),iI_\sigma .$$
Here, $`O(q)`$ has an expansion of the form $`_{|๐|>0}f_๐(\lambda )q^๐`$ for $`f_๐(\lambda )[\lambda ,\chi _i(\sigma )^1;iI_\sigma ]`$.
(ii) The critical value $`F_q^๐(\mathrm{crit}_q(\sigma ))`$ equals the pull-back $`๐ข๐ช๐^{}(๐_\sigma )`$ of the canonical coordinate satisfying the classical limit condition $`(\text{34})`$, where $`๐ข๐ช๐`$ is the map in Theorem 6.4.
(iii) The defining equations of the map $`๐ข๐ช๐`$ are convergent power series.
###### Proof.
(i) follows from a direct calculation. For (ii), first note that there exists an isomorphism
$$dF^๐:\underset{q(^{})^r}{}\mathrm{Crit}(F_q^๐)\mathrm{Ch}(M_{\mathrm{Mell}}^๐)$$
as in non-equivariant case (Lemma 5.10). Because we have a surjection
$$M_{\mathrm{Mell}}^๐_{[\mathrm{},\lambda ,q^\pm ]}[\mathrm{},\lambda ][[q]][q^1]FH_{๐\times S^1}^{}[q^1]๐ข๐ช๐^{}(QDM_๐^{}(X))[q^1],$$
$`๐ข๐ช๐^{}(\mathrm{Ch}(QDM_๐^{}(X)))|_{q_a0}`$ is embedded in $`_q\mathrm{Crit}(F_q^๐)`$. The characteristic variety of $`QDM_๐^{}(X)`$ is the union $`_\sigma \mathrm{Graph}(d๐_\sigma )`$ of graphs of the differentials of canonical coordinates. Thus, for a branch $`\{\mathrm{crit}(q)\}`$ contained in $`๐ข๐ช๐^{}(\mathrm{Ch}(QDM_๐^{}(X)))`$, the critical value $`F_q^๐(\mathrm{crit}(q))`$ gives us a pull-back of a canonical coordinate. Furthermore, by the relations (37), it follows that $`F_q^๐(\mathrm{crit}_\sigma (q))`$ satisfies the classical limit condition (34). From this we can see that the branch $`\{\mathrm{crit}_\sigma (q)\}`$ actually corresponds to a branch of $`\mathrm{Ch}(QDM_๐^{}(X))`$ and $`F_q^๐(\mathrm{crit}_\sigma (q))=๐ข๐ช๐^{}(๐_\sigma )(q)`$. In the forthcoming paper , we will prove that $`QH_๐^{}(X)`$ has convergent structure constants. Thus, the natural flat coordinates $`\widehat{x}=\{\widehat{t}_0,\widehat{q}_1,\mathrm{},\widehat{q}_r,\widehat{t}_{r+1},\mathrm{},\widehat{t}_s\}`$ of $`QH_๐^{}(X)`$ can be written as analytic functions in $`๐_\sigma `$ and $`\lambda `$. Because $`F_q^๐(\mathrm{crit}_\sigma (q))`$ is a multi-valued analytic function in $`q`$ and $`\lambda `$, $`๐ข๐ช๐^{}(\widehat{x}_i)`$ is a (possibly multi-valued) analytic function in $`q`$ and $`\lambda `$. It is easy to see that it is holomorphic in the neighborhood of $`\{q=0,\lambda \text{ generic}\}`$ (identically zero at $`q=0`$). Since $`๐ข๐ช๐^{}(\widehat{x}_i)`$ belongs to $`[\lambda ][[q]]`$, (iii) follows from Lemma 6.5. โ
###### Remark 6.7.
(i) The above lemma shows that Conjecture 6.1 holds for toric varieties.
(ii) The branch $`\{\mathrm{crit}_\sigma (q)\}_q`$ in the above lemma corresponds to a branch described in Proposition 5.11 (ii) in the non-equivariant limit.
(iii) Because the map $`๐ข๐ช๐`$ preserves the degree, the homogeneity for $`๐ข๐ช๐^{}(๐_\sigma )=F_q^๐(\mathrm{crit}_\sigma (q))`$ can be written as $`c_0(\lambda )+(_{a=1}^rc_aq_a/q_a+_{j=1}^{r+N}\lambda _j/\lambda _j)F_q^๐(\mathrm{crit}_\sigma (q))=F_q^๐(\mathrm{crit}_\sigma (q))`$ for $`c_0(\lambda )=_{j=1}^{r+N}\lambda _j`$. This can also be shown by a direct calculation.
Let $`\mathrm{\Gamma }(\sigma )`$ be the descending Morse cycle of $`\mathrm{}(F_q^๐/\mathrm{})`$ from the critical point $`\mathrm{crit}_\sigma (q)`$. From now, we write $`๐_\sigma `$ for $`๐ข๐ช๐^{}(๐_\sigma )`$ by abuse of notation. Let $`_\sigma (q,\lambda ,\mathrm{}):=_{\mathrm{\Gamma }(\sigma )}(q,\lambda ,\mathrm{})`$ be the oscillatory integral over $`\mathrm{\Gamma }(\sigma )`$. By the method of stationary phase, $`_\sigma `$ can be expanded in an asymptotic series $`_\sigma ^{\mathrm{asym}}`$. Put $`๐_i=e^{๐ณ_i}๐_i(\mathrm{crit}_\sigma )`$ and $`๐ณ_i=\sqrt{\mathrm{}}๐_i`$. Note that $`๐ณ_j=_{iI_\sigma }x_{ji}^\sigma ๐ณ_i`$ for $`jI_\sigma `$.
$`_\sigma (q,\lambda ,\mathrm{})`$ $`=e^{F_q^๐(\mathrm{crit}_\sigma (q))/\mathrm{}}{\displaystyle _{\mathrm{\Gamma }(\sigma )}}\mathrm{exp}{\displaystyle \frac{F_q^๐(e^๐ณ\mathrm{crit}_\sigma (q))F_q^๐(\mathrm{crit}_\sigma (q))}{\mathrm{}}}{\displaystyle \underset{iI_\sigma }{}}d๐ณ_i`$
(38) $`=e^{๐_\sigma /\mathrm{}}{\displaystyle _{\mathrm{\Gamma }(\sigma )}}\mathrm{exp}{\displaystyle \frac{_{i=1}^{r+N}(e^{๐ณ_i}1)๐_i(\mathrm{crit}_\sigma (q))+\lambda _i๐ณ_i}{\mathrm{}}}{\displaystyle \underset{iI_\sigma }{}}d๐ณ_i`$
$`\mathrm{}^{N/2}`$ $`e^{๐_\sigma /\mathrm{}}{\displaystyle _^N}e^{_{i=1}^{r+N}๐_i(\mathrm{crit}_\sigma )๐_i^2/2}\mathrm{exp}\left({\displaystyle \underset{i=1}{\overset{r+N}{}}}{\displaystyle \underset{n=3}{\overset{\mathrm{}}{}}}\mathrm{}^{n/21}๐_i(\mathrm{crit}_\sigma ){\displaystyle \frac{๐_i^n}{n!}}\right){\displaystyle \underset{iI_\sigma }{}}d๐_i`$
$`=:_\sigma ^{\mathrm{asym}}(q,\lambda ,\mathrm{})=(2\pi \mathrm{})^{N/2}e^{๐_\sigma /\mathrm{}}{\displaystyle \frac{1}{\sqrt{\pm \mathrm{Hess}_\sigma }}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}_{\sigma n}(q,\lambda )\mathrm{}^n.`$
Here, $`_{\sigma 0}=1`$, $`\pm \mathrm{Hess}_\sigma =(1)^Ndet(^2F_q^๐/๐ณ_i๐ณ_j(\mathrm{crit}_\sigma (q)))_{i,jI_\sigma }`$ and we assumed $`๐_i(\mathrm{crit}_\sigma )<0`$ and $`\mathrm{}>0`$. All half-integer powers of $`\mathrm{}`$ disappear due to the antisymmetry. We can see that the analytic function $`_{\sigma n}(q,\lambda )`$ can be expanded in a $`q`$-series of the form $`_๐
f_๐
(\lambda )q^๐
`$, where $`f_๐
(\lambda )[\lambda ,\chi _i(\sigma )^1;iI_\sigma ]`$.
###### Proposition 6.8.
The asymptotic series $`_\sigma ^{\mathrm{asym}}`$ is a formal solution to the $`D`$-module $`FH_{๐\times S^1}^{}`$ for each $`\sigma X^๐`$. More precisely, for any differential operator $`f(q,\lambda ,\mathrm{},\mathrm{})\overline{I_{\mathrm{poly}}^๐}[\mathrm{},\lambda ][[q]]\mathrm{}`$, we have $`f(q,\lambda ,\mathrm{},\mathrm{})_\sigma ^{\mathrm{asym}}(q,\lambda ,\mathrm{})=0`$ as an $`\mathrm{}`$-series.
###### Proof.
Take $`f\overline{I_{\mathrm{poly}}^๐}`$. We put $`f_\sigma ^{\mathrm{asym}}=(2\pi \mathrm{})^{N/2}e^{๐_\sigma /\mathrm{}}_{k=0}^{\mathrm{}}a_k(q,\lambda )\mathrm{}^k`$. Note that $`a_k(q,\lambda )`$ is a formal power series in $`q`$. Let $`f_nI_{\mathrm{poly}}^๐`$ be a sequence converging to $`f\overline{I_{\mathrm{poly}}^๐}`$ such that $`g_n=ff_n=O(q^n)`$. Then we have
(39)
$$g_n_\sigma ^{\mathrm{asym}}=f_\sigma ^{\mathrm{asym}}=(2\pi \mathrm{})^{N/2}e^{๐_\sigma /\mathrm{}}\underset{k=0}{\overset{\mathrm{}}{}}a_k(q,\lambda )\mathrm{}^k$$
by Proposition 6.3. Because we have the following expansions
$$(\pm \mathrm{Hess}_\sigma )^{1/2}=(\underset{iI_\sigma }{}\chi _i(\sigma ))^{1/2}(1+\underset{|๐
|>0}{}b_๐
(\lambda )q^๐
),q_a\frac{๐_\sigma }{q_a}=\underset{๐
0}{}c_๐
(\lambda )q^๐
$$
for some $`b_๐
(\lambda ),c_๐
(\lambda )[\lambda ,\chi _i(\sigma )^1;iI_\sigma ]`$, we can see from (39) that $`a_k(q,\lambda )=O(q^n)`$. Because this holds for all $`n`$, $`a_k`$ must be zero. โ
In non-equivariant case, each coefficient of the asymptotic expansion of $`_\mathrm{\Gamma }`$ in $`\mathrm{}`$ is still analytic function in $`q`$ but not necessarily expanded in $`q`$-series. However, $`f(q,\mathrm{},\mathrm{})_\mathrm{\Gamma }^{\mathrm{asym}}(q,\mathrm{})`$ does make sense for $`f๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}`$ as an $`\mathrm{}`$-series.
###### Corollary 6.9.
Let $`\mathrm{\Gamma }`$ be the descending Morse cycle of $`\mathrm{}(F_q)`$ from a critical point in a branch described in Proposition 5.11 (ii). The asymptotic expansion of the non-equivariant mirror oscillatory integral $`_\mathrm{\Gamma }(q,\mathrm{})`$ in $`\mathrm{}`$ gives a formal solution to $`\stackrel{~}{FH}_{S^1}^{}`$. More precisely, any differential operator in $`\overline{I_{\mathrm{poly}}}๐ช_{\mathrm{small}}^{\mathrm{}}\mathrm{}`$ annihilates the asymptotic expansion of $`_\mathrm{\Gamma }`$ as an $`\mathrm{}`$-series.
For $`vQH_๐^{}(X)`$, there exists a differential operator $`๐[v](q,\lambda ,\mathrm{},\mathrm{})[\mathrm{},\lambda ][[q]]\mathrm{}`$ such that $`\mathrm{\Phi }_{๐ข๐ช๐}(v)=๐[v](Q,\lambda ,P,\mathrm{})\mathrm{\Delta }^๐`$. We can write $`v=๐[v](q,\lambda ,^{\mathrm{}},\mathrm{})\mathrm{\Phi }_{๐ข๐ช๐}^1\mathrm{\Delta }^๐`$, where $`_a^{\mathrm{}}=_{๐ข๐ช๐_{}(_a)}^{\mathrm{}}`$. Define a formal section $`s_\sigma (q,\lambda ,\mathrm{})`$ of $`๐ข๐ช๐^{}๐ฏ_{}`$ by the formula
$$v,(2\pi \mathrm{})^{N/2}s_\sigma _๐=๐[v](q,\lambda ,\mathrm{},\mathrm{})_\sigma ^{\mathrm{asym}}.$$
By Proposition 6.8, the right hand side depends only on $`v`$ and not on a choice of $`๐[v]`$. Using $`/๐_\sigma ,/๐_\tau _๐=\delta _{\sigma \tau }\mathrm{}_\sigma ^2`$, we can also write
$$s_\sigma (q,\lambda ,\mathrm{})=\frac{1}{(2\pi \mathrm{})^{N/2}}\underset{\tau }{}\mathrm{}_\tau (๐[\psi _\tau (q)]_\sigma ^{\mathrm{asym}})\mathrm{}_\tau \frac{}{๐_\tau },$$
where $`\psi _\tau (q)=\psi _\tau (๐ข๐ช๐(q))`$.
###### Theorem 6.10.
(i) The section $`s_\sigma `$ is the pull-back of an asymptotic solution in section 6.1 by the map $`๐ข๐ช๐`$.
(ii) This satisfies the classical limit condition (36).
(iii) The $`R`$-conjecture 6.2 holds for smooth projective toric variety.
###### Proof.
(i): For $`vH_๐^{}(X)`$ and $`\stackrel{~}{s}_\sigma =(2\pi \mathrm{})^{N/2}s_\sigma `$, we have
$`v,\mathrm{}q_a{\displaystyle \frac{}{q_a}}\stackrel{~}{s}_\sigma _๐`$ $`=\mathrm{}_a๐[v]_\sigma ^{\mathrm{asym}}=๐[_{๐ข๐ช๐_{}(_a)}^{\mathrm{}}v]_\sigma ^{\mathrm{asym}}`$
$`=๐ข๐ช๐_{}(_a)v,\stackrel{~}{s}_\sigma _๐=v,๐ข๐ช๐_{}(_a)\stackrel{~}{s}_\sigma _๐.`$
Therefore, $`s_\sigma `$ is a solution to the differential equation (35). We calculate
$$๐[\psi _\tau (q)]_\sigma ^{\mathrm{asym}}=(2\pi \mathrm{})^{N/2}e^{๐_\sigma /\mathrm{}}\left(\frac{๐[\psi _\tau (q)](q,\lambda ,๐_\sigma ,\mathrm{}=0)}{\sqrt{\pm \mathrm{Hess}_\sigma }}+O(\mathrm{})\right).$$
Let $`V(q,\lambda ,\mathrm{},\mathrm{})`$ be a differential operator such that $`V(q,\lambda ,^{\mathrm{}},\mathrm{})1=\mathrm{\Phi }_{๐ข๐ช๐}^1\mathrm{\Delta }^๐`$. Then we have $`\psi _\tau (q)=๐[\psi _\tau (q)](q,\lambda ,^{\mathrm{}},\mathrm{})V(q,\lambda ,^{\mathrm{}},\mathrm{})1=๐[\psi _\tau (q)](q,\lambda ,,0)V(q,\lambda ,,0)1`$, where $`_a`$ is the quantum product by $`๐ข๐ช๐_{}(_a)`$. By using (33), we have
$`\delta _{\sigma \tau }\psi _\sigma (q)=\psi _\tau (q)\psi _\sigma (q)`$ $`=๐[\psi _\tau (q)](q,\lambda ,,0)V(q,\lambda ,,0)\psi _\sigma (q)`$
$`=๐[\psi _\tau (q)](q,\lambda ,๐_\sigma ,0)V(q,\lambda ,๐_\sigma ,0)\psi _\sigma (q).`$
Therefore,
$$\frac{1}{(2\pi \mathrm{})^{N/2}}\mathrm{}_\tau ๐[\psi _\tau (q)]_\sigma ^{\mathrm{asym}}=e^{๐_\sigma /\mathrm{}}\left(\frac{\delta _{\sigma \tau }\mathrm{}_\sigma }{V(q,\lambda ,๐_\sigma ,0)\sqrt{\mathrm{Hess}_\sigma }}+O(\mathrm{})\right).$$
The equality $`\mathrm{}_\sigma =V(q,\lambda ,๐_\sigma ,0)\sqrt{\mathrm{Hess}_\sigma }`$ follows from the differential equation $`\mathrm{}_as_\sigma =_as_\sigma `$, and the classical limit $`lim_{q0}\sqrt{\mathrm{Hess}_\sigma }=\sqrt{_{iI_\sigma }(\chi _i(\sigma ))}=lim_{q0}\mathrm{}_\sigma `$, $`V|_{q=0}=1`$. Therefore, $`s_\sigma `$ is an asymptotic solution in section 6.1.
(ii): By Lemma 6.6 and (38), we have
$`\mathrm{}_\tau ๐[\psi _\tau (q)]_\sigma `$ $`=\mathrm{}_\tau ๐[\psi _\tau (q)]\left(e^{๐_\sigma /\mathrm{}}{\displaystyle _{_0^N}}e^{_{iI_\sigma }\chi _i(\sigma )(๐_i\mathrm{log}๐_i1)/\mathrm{}+O(q)}{\displaystyle \underset{iI_\sigma }{}}d\mathrm{log}๐_i\right)`$
$`=e^{๐_\sigma /\mathrm{}}`$ $`{\displaystyle _{_0^N}}\left(\mathrm{}_\tau ๐[\psi _\tau (q)](0,\lambda ,๐_\sigma ,\mathrm{}){\displaystyle \underset{iI_\sigma }{}}e^{(๐_i1)\chi _i(\sigma )/\mathrm{}}๐_i^{\chi _i(\sigma )/\mathrm{}1}+O(q)\right){\displaystyle \underset{iI_\sigma }{}}d๐_i`$
$`=e^{๐_\sigma /\mathrm{}}`$ $`\left(\delta _{\sigma \tau }\sqrt{{\displaystyle \underset{iI_\sigma }{}}\chi _i(\sigma )}{\displaystyle \underset{iI_\sigma }{}}e^{\chi _i(\sigma )/\mathrm{}}\left({\displaystyle \frac{\chi _i(\sigma )}{\mathrm{}}}\right)^{\chi _i(\sigma )/\mathrm{}}\mathrm{\Gamma }\left({\displaystyle \frac{\chi _i(\sigma )}{\mathrm{}}}\right)+O(q)\right).`$
We put $`๐_i=e^{๐ณ_i}`$ in the first line, used $`_a_b๐_\sigma =O(q)`$ in the second line and $`๐[\psi _\tau (q)](0,\lambda ,๐_\sigma ,\mathrm{})=\delta _{\tau \sigma }`$ in the third line. Note that we are assuming $`\chi _i(\sigma )>0`$. The conclusion follows from the asymptotic expansion of the Gamma function around infinity (Stirlingโs formula).
$$\mathrm{log}\mathrm{\Gamma }(z)\left(z\frac{1}{2}\right)\mathrm{log}zz+\frac{1}{2}\mathrm{log}2\pi +\underset{k=1}{\overset{\mathrm{}}{}}\frac{B_{2k}}{2k(2k1)z^{2k1}},\mathrm{}(z)>0.$$
(iii): First note that if the asymptotic solution is regular at $`\lambda =0`$ for one semisimple point $`t_0`$, then it is regular for any semisimple point. (We can use the value at $`t_0`$ as an initial condition for the differential equation (35).) Thus, it suffices to check that the above solution $`s_\sigma `$ is regular at $`\lambda =0`$ for a semisimple $`q`$. Let $`T_i(y_1,\mathrm{},y_r)`$ be the polynomial in Proposition 5.5. Because $`\{T_i(P)\mathrm{\Delta }^๐\}_{i=0}^s`$ forms a $`[\mathrm{},\lambda ][[q]]`$-basis of $`FH_{๐\times S^1}^{}`$, we can write $`\mathrm{\Phi }_{๐ข๐ช๐}(/t_i)=_{j=0}^sA_{ij}(q,\lambda ,\mathrm{})T_j(P)\mathrm{\Delta }^๐`$ for some $`A_{ij}[\mathrm{},\lambda ][[q]]`$. Let $`\mathrm{}_\tau \psi _\tau (q)=_{i=0}^s\mathrm{\Psi }_{\tau i}(q,\lambda )(/t_i)`$. Because $`QH_๐^{}(X)`$ is convergent, $`\mathrm{\Psi }_{\tau i}(q,\lambda )`$ is an analytic function of $`q,\lambda `$ which is regular at $`\lambda =0`$ for semisimple $`q`$. We can take $`๐[\psi _\tau (q)]`$ as
$$\mathrm{}_\tau ๐[\psi _\tau (q)]=\underset{i,j=0}{\overset{s}{}}\mathrm{\Psi }_{\tau i}(q,\lambda )A_{ij}(q,\lambda ,\mathrm{})T_j(\mathrm{}).$$
Then, the matrix $`R`$ is given by $`(2\pi \mathrm{})^{N/2}e^{๐_\sigma /\mathrm{}}R_{\tau \sigma }=_{i,j=0}^s\mathrm{\Psi }_{\tau i}(q,\lambda )A_{ij}(q,\lambda ,\mathrm{})T_j(\mathrm{})_\sigma ^{\mathrm{asym}}`$. Because we can write $`T_j(\mathrm{})_\sigma ^{\mathrm{asym}}=(2\pi \mathrm{})^{N/2}_{j\sigma }e^{๐_\sigma /\mathrm{}}`$ for some $`_{j\sigma }=_{n0}_{j\sigma }^{(n)}(q,\lambda )\mathrm{}^n`$, we have
$$R_{\tau \sigma }(q,\lambda ,\mathrm{})=\underset{i,j=0}{\overset{s}{}}\mathrm{\Psi }_{\tau i}(q,\lambda )A_{ij}(q,\lambda ,\mathrm{})_{j\sigma }(q,\lambda ,\mathrm{}).$$
On the right hand side, the regularity at $`\lambda =0`$ is clear for $`\mathrm{\Psi }_{\tau i}`$ and $`_{j\sigma }`$. Thus, it suffices to show that $`A_{ij}^{(n)}(q,\lambda )`$ is a holomorphic function around $`q=\lambda =0`$ where $`A_{ij}=_{n0}A_{ij}^{(n)}(q,\lambda )\mathrm{}^n`$. Since $`(A_{ij})=(\mathrm{\Psi }_{\tau i})^1(R_{\tau \sigma })(_{j\sigma })^1`$ and $`R_{\tau \sigma }^{(n)}(q,\lambda )`$ is an analytic function (because $`QH_๐^{}(X)`$ is convergent), we can see that $`A_{ij}^{(n)}(q,\lambda )`$ is convergent around $`q=\lambda =0`$ from Lemma 6.5. โ
## 7. Appendix
### 7.1. Proof of Lemma 4.9
First we study $`\epsilon _1(M)`$. We have
$`\epsilon _1(d,M)`$ $`:={\displaystyle \underset{i=1}{\overset{d1}{}}}{\displaystyle \frac{(d+1)^M}{(di+1)^M(i+1)^M}}`$
$`{\displaystyle _0^d}{\displaystyle \frac{(d+1)^Mdx}{(dx+1)^M(x+1)^M}}=ฯต^{M1}(1ฯต)^M{\displaystyle _ฯต^{1ฯต}}{\displaystyle \frac{dy}{y^M(1y)^M}},`$
where we put $`y=(x+1)/(d+2)`$ and $`ฯต=1/(d+2)`$. Set $`y=\mathrm{sin}^2(\theta /2)`$ and $`ฯต=\mathrm{sin}^2(\alpha /2)`$. Then we have
$$\epsilon _1(d,M)4\mathrm{cos}^2\left(\frac{\alpha }{2}\right)(\mathrm{sin}\alpha )^{2M2}_\alpha ^{\pi /2}\frac{d\theta }{(\mathrm{sin}\theta )^{2M1}}.$$
Because we can assume $`d2`$, we have $`0<\alpha \pi /3`$. Using the formula (see , 2.515)
$$\frac{d\theta }{(\mathrm{sin}\theta )^{2M1}}=\frac{(2M3)!!}{(2M2)!!}\left\{\underset{r=0}{\overset{M2}{}}\frac{(2M42r)!!}{(2M32r)!!}\frac{\mathrm{cos}\theta }{(\mathrm{sin}\theta )^{2M22r}}+\mathrm{log}\left|\mathrm{tan}\frac{\theta }{2}\right|\right\},$$
we have
$`\epsilon _1(d,M)`$ $`4\mathrm{cos}^2\left({\displaystyle \frac{\alpha }{2}}\right){\displaystyle \frac{(2M3)!!}{(2M2)!!}}\{{\displaystyle \underset{r=0}{\overset{M2}{}}}{\displaystyle \frac{(2M42r)!!}{(2M32r)!!}}\mathrm{cos}\alpha (\mathrm{sin}\alpha )^{2r}`$
$`(\mathrm{sin}\alpha )^{2M2}\mathrm{log}\left|\mathrm{tan}{\displaystyle \frac{\alpha }{2}}\right|\}`$
$`4{\displaystyle \frac{(2M3)!!}{(2M2)!!}}\left\{{\displaystyle \frac{1}{1\mathrm{sin}^2\alpha }}+(\mathrm{sin}\alpha )^{2M3}{\displaystyle \frac{2}{e}}\right\}`$
$`4\left(1{\displaystyle \frac{1}{2M2}}\right)\left(1{\displaystyle \frac{1}{2M4}}\right)\mathrm{}\left(1{\displaystyle \frac{1}{2}}\right)\left(4+{\displaystyle \frac{2}{e}}\right).`$
Therefore, we have $`\epsilon _1(M)0`$ as $`M\mathrm{}`$.
Next we study $`\epsilon _2(B_2,M)`$. Set $`C=B_2/B_1`$. We have for $`C>1`$,
$`{\displaystyle \underset{i=1}{\overset{d}{}}}\left({\displaystyle \frac{B_1}{B_2}}\right)^i{\displaystyle \frac{(d+1)^M}{(d+1i)^M}}`$ $`{\displaystyle _0^d}{\displaystyle \frac{(d+1)^Mdx}{C^x(d+1x)^M}}+{\displaystyle \frac{(d+1)^M}{C^d}}`$
$`{\displaystyle _0^{1\frac{1}{d+1}}}{\displaystyle \frac{(d+1)dy}{C^{(d+1)y}(1y)^M}}+\left({\displaystyle \frac{2M}{e\mathrm{log}C}}\right)^M`$
$`{\displaystyle _0^{1/2}}{\displaystyle \frac{(d+1)2^M}{C^{(d+1)y}}}๐y+{\displaystyle _{1/2}^{1\frac{1}{d+1}}}{\displaystyle \frac{(d+1)dy}{C^{(d+1)/2}(1y)^M}}+\left({\displaystyle \frac{2M}{e\mathrm{log}C}}\right)^M`$
$`{\displaystyle \frac{2^M}{\mathrm{log}C}}+{\displaystyle \frac{(d+1)^M}{(M1)C^{(d+1)/2}}}+\left({\displaystyle \frac{2M}{e\mathrm{log}C}}\right)^M`$
$`{\displaystyle \frac{2^M}{\mathrm{log}C}}+{\displaystyle \frac{1}{M1}}\left({\displaystyle \frac{2M}{e\mathrm{log}C}}\right)^M+\left({\displaystyle \frac{2M}{e\mathrm{log}C}}\right)^M.`$
Hence we have $`\epsilon (B_2,N)0`$ as $`B_2\mathrm{}`$.
### 7.2. Proof of Lemma 6.5
We prove the lemma by induction on $`l`$. We write $`\lambda =(\lambda ^{},\lambda _l)^{l1}\times `$. Set $`D^{}=\{\lambda ^{}^{l1}|\{\lambda ^{}\}\times D\text{ is non-empty}\}`$. We will show that $`f`$ is holomorphic in the neighborhood of $`\{q=0,\lambda _l=0,\lambda ^{}D^{}\}`$. We choose $`ฯต(\lambda ^{})>0`$ so that $`\{\lambda ^{}\}\times \{|\lambda _l|=ฯต(\lambda ^{})\}D`$. Then there exists a positive $`C(\lambda ^{})>0`$ such that $`|f_๐
(\lambda ^{},\lambda _l)|C(\lambda ^{})^{|๐
|+1}`$ on $`|\lambda _l|=ฯต(\lambda ^{})`$. For $`|\zeta |<ฯต(\lambda ^{})/2`$, because $`f_๐
(\lambda )`$ is holomorphic on $`U`$,
$$f_๐
(\lambda ^{},\zeta )=\frac{1}{2\pi i}_{|z|=ฯต(\lambda ^{})}\frac{f_๐
(\lambda ^{},z)}{z\zeta }๐z.$$
From this we can see that $`f_๐
(\lambda ^{},\zeta )2C(\lambda ^{})^{|๐
|+1}`$ for $`\lambda ^{}D^{}`$ and $`|\zeta |<ฯต(\lambda ^{})/2`$.
|
warning/0506/math0506425.html
|
ar5iv
|
text
|
# Inรฉgalitรฉs de Milnor-Wood gรฉomรฉtriques
## 1 Introduction
La cรฉlรจbre inรฉgalitรฉ de Milnor-Wood ( et ) affirme que, si
$$E\mathrm{\Sigma }$$
est un fibrรฉ plat en fibres $`S^1`$ sur la surface compacte $`\mathrm{\Sigma }`$ de genre $`\gamma 2`$, alors la caractรฉristique dโEuler de ce fibrรฉ, notรฉe $`\chi (E)`$ vรฉrifie,
$$|\chi (E)||\chi (\mathrm{\Sigma })|=2\gamma 2,$$
lโรฉgalitรฉ ayant lieu si $`E`$ est le fibrรฉ tangent de $`\mathrm{\Sigma }`$.
Un fibrรฉ plat รฉtant dรฉfini par une reprรฉsentation de $`\mathrm{\Pi }_1(\mathrm{\Sigma })`$, lโinรฉgalitรฉ ci-dessus est en fait une restriction imposรฉe ร cette reprรฉsentation ร valeurs dans le groupe des homรฉomorphismes du cercle. Les valeurs possibles de la caractรฉristique dโEuler de $`E`$ sont dรฉcrites dans .
Dans cette article nous envisageons une gรฉnรฉralisation, en dimension supรฉrieure de cette inรฉgalitรฉ. Pour cela nous dรฉfinissons le volume dโune reprรฉsentation. Plus prรฉcisรฉment, soit $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe de dimension $`n`$ et soit $`\rho `$ une reprรฉsentation de son groupe fondamental dans le groupe dโisomรฉtrie dโune variรฉtรฉ symรฉtrique de courbure nรฉgative de dimension $`n`$ et simplement connexe, notรฉe $`\stackrel{~}{X}`$. Considรฉrons une application $`\stackrel{~}{f}`$ du revรชtement universel de $`M`$ dans $`\stackrel{~}{X}`$ รฉquivariante par rapport ร $`\rho `$, alors, si $`\omega `$ dรฉsigne la forme volume de $`\stackrel{~}{X}`$, la forme $`\stackrel{~}{f}^{}\omega `$ passe au quotient sur $`M`$.
###### Definition 1.1
On appelle volume de la reprรฉsentation $`\rho `$ le nombre,
$$vol(\rho )=_M\stackrel{~}{f}^{}\omega .$$
Dans certains cas ce nombre peut-รชtre interprรฉtรฉ comme la classe dโEuler dโun fibrรฉ plat. Des bornes supรฉrieures de $`vol(\rho )`$ existent. Elles reposent souvent sur le choix dโune famille de sections particuliรจres du fibrรฉ plat. Dans , par exemple, K. Corlette utilise des sections harmoniques pour dรฉmontrer un thรฉorรจme de rigiditรฉ sur les reprรฉsentations de volume maximal. Le cas oรน $`\stackrel{~}{X}`$ est hyperbolique rรฉel est abordรฉ par A. Reznikov dans ; lโauteur y prouve une inรฉgalitรฉ optimale et cโest ce type de rรฉsultats que nous รฉtendons dans le prรฉsent travail. Le cas dโรฉgalitรฉ dans lโinรฉgalitรฉ de A. Reznikov est prouvรฉ par N. Dunfield dans et dans , il consiste ร montrer que, si le volume est maximal, la reprรฉsentation est fidรจle et discrรจte. Signalons lโarticle dans lequel lโauteur dรฉcrit une autre notion de volume de reprรฉsentations et construit de nouveaux invariants numรฉriques.
Dans le cas oรน $`\stackrel{~}{X}`$ est lโespace hyperbolique rรฉel nous prouvons, dans cet article, que le volume des reprรฉsentations est constant sur les composantes connexes de lโespace des reprรฉsentations. Cโest un rรฉsultat รฉvident lorsque la dimension est paire car, dans ce cas, le volume est aussi un nombre dโEuler, mais nouveau dans le cas de dimension impaire. Plus prรฉcisรฉment nous prouvons le
###### Thรฉorรจme 1.2
Soit $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe et orientรฉe et $`\rho _t:\mathrm{\Pi }_1(M)Isom(\stackrel{~}{X})`$ une famille de reprรฉsentations qui dรฉpend de maniรจre $`C^1`$ du paramรจtre $`t๐`$, alors le volume $`vol(\rho _t)`$ est constant.
La mรฉthode employรฉe consiste ร utiliser la formule de Schlรคfli (voir aussi ). Il sโagit dโune approche nouvelle dans ce contexte; en fait nous construisons un โpolyรจdreโ gรฉodรฉsique hyperbolique dans $`\stackrel{~}{X}`$ รฉquivariant par rapport ร lโimage de $`\rho `$. Il sโagit dโune rรฉunion de simplexes hyperboliques gรฉodรฉsiques invariants par lโimage de $`\rho `$; les simplexes pouvant se chevaucher ils ne fournissent pas une triangulation de $`\stackrel{~}{X}`$. Nous construisons ensuite une application $`\rho `$-รฉquivariante polyรจdrale. Ceci permet alors de calculer le volume de la reprรฉsentation. La formule de Schlรคfli ainsi quโun peu de thรฉorie du degrรฉ permet alors de montrer la constance de ce volume. Un corollaire frappant est une preuve trรจs simple du rรฉsultat suivant de T. Soma (),
###### Thรฉorรจme 1.3 (T. Soma)
Soit $`Y`$ une variรฉtรฉ diffรฉrentielle fermรฉe de dimension $`3`$. Lโensemble des variรฉtรฉs hyperboliques fermรฉes $`X`$, de dimension $`3`$ telles quโil existe une application continue de degrรฉ non nul de $`Y`$ sur $`X`$, est fini.
La preuve se rรฉsume comme suit. Appelons $`f`$ lโapplication de degrรฉ non nul de $`Y`$ sur $`X`$ et $`\rho `$ la reprรฉsentation induite de $`\mathrm{\Pi }_1(Y)`$ dans $`\mathrm{\Pi }_1(X)`$, alors $`vol(\rho )=\text{deg}(f)vol(X)`$. Le thรฉorรจme 1.4 affirme quโil existe une constante $`C(M)`$ telle que $`vol(\rho )C(M)`$. Par ailleurs le volume des variรฉtรฉs hyperboliques fermรฉes est minorรฉ par une constante universel (qui dรฉpend de la dimension). Ceci montre que le degrรฉ de $`f`$ ne peut prendre quโun nombre fini de valeurs. Le volume de la reprรฉsentation $`\rho `$ รฉtant constant sur les composantes connexes de lโespace des reprรฉsentations de $`\mathrm{\Pi }_1(Y)`$ dans $`\mathrm{\Pi }_1(X)`$ il ne prend รฉgalement quโun nombre fini de valeurs. Finalement le volume de la variรฉtรฉ hyperbolique $`X`$ ne prend quโun nombre fini de valeurs. Un rรฉsultat de W. Thurston affirme alors quโil nโy a quโun nombre fini de variรฉtรฉs $`X`$ possibles. Le lecteur peut consulter les dรฉtails dans le paragraphe 6.
Le thรฉorรจme 1.2 est en fait un corollaire dโun rรฉsultat plus gรฉnรฉral, que nous dรฉcrivons maintenant. Rappelons la dรฉfinition de lโentropie volumique dโune variรฉtรฉ Riemannienne $`(Y,g)`$. Pour $`x\stackrel{~}{Y}`$, on dรฉsigne par $`B(x,R)`$ la boule gรฉodรฉsique de centre $`x`$ et de rayon $`R`$, alors on dรฉfinit
$$Ent(Y,g)=\underset{R\mathrm{}}{lim}\frac{1}{R}\mathrm{log}(vol(B(x,R))).$$
Dans ce qui suit $`\stackrel{~}{X}=_1^p\stackrel{~}{X}_i`$ est le produit des espaces symรฉtriques simplement connexes de courbure strictement nรฉgative, $`\stackrel{~}{X}_i`$. Chacune des variรฉtรฉs $`\stackrel{~}{X}_i`$ est munie dโune mรฉtrique $`\alpha _ig_0^i`$, oรน $`g_0^i`$ est symรฉtrique normalisรฉe (de courbure comprise entre $`4`$ et $`1`$, par exemple) et $`\alpha _i`$ est un rรฉel strictement positif. Parmi tous les choix de nombres $`\alpha _i`$ il en est un qui donne une entropie volumique minimale (voir la proposition 2.4); nous noterons $`g_0`$ la mรฉtrique correspondante sur $`\stackrel{~}{X}`$ qui est de dimension $`n`$. Son entropie volumique est un nombre calculable. Nous prouvons,
###### Thรฉorรจme 1.4
Soit $`Y`$ une variรฉtรฉ riemannienne fermรฉe de dimension $`n`$ et $`\rho `$ une reprรฉsentation de $`\mathrm{\Pi }_1(Y)`$ dans $`Isom(\stackrel{~}{X})`$, alors
* $`vol(\rho )\left(\frac{Ent(Y,g)}{Ent(\stackrel{~}{X},g_0)}\right)^nvol(Y,g).`$
* Lโรฉgalitรฉ dans lโinรฉgalitรฉ ci-dessus a lieu si, et seulement si, la reprรฉsentation $`\rho `$ est injective, $`X=\stackrel{~}{X}/\rho (\mathrm{\Pi }_1(Y))`$ est une variรฉtรฉ compacte et $`(Y,g)`$ est homothรฉtique ร $`(X,g_0)`$
Ce rรฉsultat รฉtait annoncรฉ en 1997 dans et รฉnoncรฉ en 1998 dans . Il gรฉnรฉralise le cas oรน la reprรฉsentation a une image discrรจte et cocompacte, cโest-ร -dire lโanalogue des thรฉorรจmes de pour le cas oรน lโespace localement symรฉtrique compacte est localement un produit dโespaces symรฉtriques de rang 1. Ce dernier rรฉsultat, concernant les produits dโespaces symรฉtriques de rang 1 avec image discrรจte cocompacte, est รฉnoncรฉ par Ch. Connell et B. Farb dans .
La preuve de lโinรฉgalitรฉ se fait en exhibant une famille dโapplications $`\rho `$-รฉquivariantes de $`\stackrel{~}{Y}`$ sur $`\stackrel{~}{X}`$ construites par la mรฉthode introduite dans . Le cas dโรฉgalitรฉ est beaucoup plus difficile car lโimage de $`\rho `$ nโest pas supposรฉe discrรจte; plus prรฉcisรฉment, nous montrons que, dans le cas dโรฉgalitรฉ, la famille dโapplications $`\rho `$-รฉquivariantes que nous construisons converge vers une application harmonique; ceci permet, en particulier, de montrer que la limite est de classe $`C^{\mathrm{}}`$. La combinaison des propriรฉtรฉs liรฉes ร lโharmonicitรฉ et de celles liรฉes ร la construction ci-dessus conduit au rรฉsultat.
Remarquons que les applications $`\rho `$-รฉquivariantes construites sont particuliรจrement adaptรฉes ร lโรฉtude du volume et conduisent ร des rรฉsultats optimaux comparables, dans un cadre plus gรฉnรฉral, ร ceux de N. Dunfield . Signalons รฉgalement un travail rรฉcent de S. Francaviglia et B. Klaff dans lequel les auteurs utilisent une intรฉressante variante de la construction de pour รฉtudier le cas oรน $`Y`$ est une variรฉtรฉ hyperbolique de volume fini.
Enfin, lโinรฉgalitรฉ ci-dessus peut sโinterprรฉter agrรฉablement dans le cadre de la cohomologie bornรฉe (voir ). Le rรฉcent travail de M. Burger, A. Iozzi et A. Wienhard () dรฉveloppe ce point de vue et aboutit ร de trรจs jolis rรฉsultats concernant les reprรฉsentations du groupe fondamental des surfaces.
Nous tenons ร remercier A. Reznikov, M. Boileau, D. Cooper et S. Francaviglia pour leur aide et leurs commentaires lors de la redaction de cet article.
## 2 Gรฉomรฉtrie des espaces produits
ร titre dโexemple, nous dรฉcrirons la gรฉomรฉtrie de lโespace $`(\stackrel{~}{X},g_0^1g_0^2)=(๐^{n_1}\times ๐^{n_2},g_0^1g_0^2)`$ muni de la mรฉtrique produit oรน $`(๐^{n_1},g_0^1)`$ ($`resp.(๐^{n_2},g_0^2)`$) dรฉsigne lโespace hyperbolique simplement connexe de dimension $`n_1`$ ($`resp.n_2`$) (de courbure constante รฉgale ร $`1`$). Pour un exposรฉ gรฉnรฉral sur les espaces symรฉtriques, nous renvoyons ร .
### 2.1 Gรฉodรฉsiques
Soient $`x=(x_1,x_2)\stackrel{~}{X}`$ et $`u=(u_1,u_2)T_{(x_1,x_2)}\stackrel{~}{X}`$ tels que $`u_{g_0^1g_0^2}^2=u_1_{g_0^1}^2+u_2_{g_0^2}^2=1`$, alors la gรฉodรฉsique de $`X`$, notรฉe $`c_u`$, partant de $`x`$ et de vitesse initiale $`u`$ est $`c_u(t)=(c_1(t),c_2(t))`$, oรน $`c_i`$ ($`i=\mathrm{1,2}`$) est la gรฉodรฉsique de $`๐^{n_i}`$ partant de $`x_i`$ et de vitesse initiale $`u_i`$. Une gรฉodรฉsique dรฉfinie par un vecteur $`u=(u_1,u_2)`$ telle que $`u_1=0`$ ou bien $`u_2=0`$ est dite singuliรจre ; ces cas correspondent ร
$$c_u(t)=(x_1,c_2(t))\text{ ou }c_u(t)=(c_1(t),x_2).$$
Une gรฉodรฉsique dรฉfinie par un vecteur $`u=(u_1,u_2)`$ tel que $`u_i0`$, pour $`i=\mathrm{1,2}`$, est dite rรฉguliรจre.
### 2.2 Courbures et plats
La courbure sectionnelle de $`(\stackrel{~}{X},g_0^1g_0^2)`$, qui se calcule aisรฉment, est nรฉgative ou nulle. Soit alors $`x=(x_1,x_2)X`$, $`u=(u_1,u_2)T_x\stackrel{~}{X}`$, un vecteur rรฉgulier, alors lโapplication
$`๐^2`$ $`\stackrel{~}{X}`$
$`(t,s)`$ $`(c_1(t/\alpha _1),c_2(s/\alpha _2))`$
$`\alpha _1=u_1_{g_0^1}`$ et $`\alpha _2=u_2_{g_0^2}`$ rรฉalisent un plongement isomรฉtrique de $`๐^2`$ muni de sa mรฉtrique euclidienne dans $`(\stackrel{~}{X},g_0^1g_0^2)`$. On peut vรฉrifier par le calcul que lโimage de cette application est totalement gรฉodรฉsique (voir , pp. ) ou bien constater que, si $`\sigma _i`$ dรฉsigne la symรฉtrie orthogonale par rapport ร la gรฉodรฉsique $`c_i`$ dans $`(๐^{n_i},g_0^i)`$, lโimage de lโapplication ci-dessus est lโensemble des points fixes de $`\sigma _1\times \sigma _2`$ dans $`\stackrel{~}{X}`$ ; il sโagit donc dโun sous-espace totalement gรฉodรฉsique plat et qui est, de plus, de dimension maximale avec ces propriรฉtรฉs : $`(\stackrel{~}{X},g_0^1g_0^2)`$ est un espace symรฉtrique de rang 2. Nous noterons dรฉsormais $`\overline{g}_0`$ la mรฉtrique $`g_0^1g_0^2`$.
Remarque.Dโune maniรจre gรฉnรฉrale, si $`\stackrel{~}{X}`$ est le produit riemannien de $`p`$ espaces symรฉtriques de courbure strictement nรฉgative, alors $`X`$ est de rang $`p`$.
### 2.3 Mรฉtriques localement symรฉtriques
On peut munir la variรฉtรฉ diffรฉrentielle $`\stackrel{~}{X}`$ dโautres mรฉtriques localement symรฉtriques ; en effet, pour $`\alpha _1`$ et $`\alpha _2`$ deux nombres rรฉels strictement positifs, on dรฉfinit :
$$g_{\alpha _1,\alpha _2}=\alpha _1^2g_0^1\alpha _2^2g_0^2.$$
Contrairement aux espaces symรฉtriques irrรฉductibles, les espaces symรฉtriques produits sont flexibles.
### 2.4 Groupe dโisomรฉtries
On dรฉtermine aisรฉment le groupe dโisomรฉtries de $`(\stackrel{~}{X},g_{\alpha _1,\alpha _2})`$. En effet, si $`n_1n_2`$
$$Isom(\stackrel{~}{X},g_{\alpha _1,\alpha _2})=Isom(๐^{n_1},g_0^1)\times Isom(๐^{n_2},g_0^2).$$
Si $`n_1=n_2`$ et $`\alpha _1=\alpha _2`$, lโรฉchange des deux facteurs est une isomรฉtrie supplรฉmentaire qui est involutive ; le groupe dโisomรฉtrie de $`(\stackrel{~}{X},g_{\alpha _1,\alpha _1})`$ est donc une extension de $`๐/2๐`$ par le groupe $`Isom(๐^{n_1},g_0)\times Isom(๐^{n_2},g_0)`$.
### 2.5 Fonctions de Busemann
On rappelle que, si $`(M,g)`$ est une variรฉtรฉ riemannienne complรจte et si $`c:๐M`$ est une gรฉodรฉsique minimisante sur toute sa longueur et paramรฉtrรฉe par lโabscisse curviligne (cโest-ร -dire, $`c`$ est un plongement isomรฉtrique), alors on dรฉfinit la fonction de Busemann associรฉe ร $`c`$,
$$B_c(x)=\underset{t+\mathrm{}}{lim}d(x,c(t))t=\underset{t+\mathrm{}}{lim}\left(d(x,c(t))d(c(0),c(t))\right).$$
On montre que la limite existe (voir , p. 23). Si $`(M,g)`$ est une variรฉtรฉ simplement connexe de courbure nรฉgative ou nulle son bord ร lโinfini (voir , p. ) sโidentifie ร une sphรจre de dimension $`n1`$, oรน $`n=dimM`$, grรขce au choix dโun point $`OM`$ qui sert dโorigine. Chaque point $`\theta M`$, le bord ร lโinfini de $`M`$, dรฉtermine une gรฉodรฉsique minimisante sur toute sa longueur, ร savoir, lโunique gรฉodรฉsique $`c`$ qui passe par $`O`$ et telle que $`\underset{t+\mathrm{}}{lim}c(t)=\theta `$. La fonction de Busemann correspondante est notรฉe $`B(,\theta )`$. Remarquons quโelle dรฉpend du choix de lโorigine.
Dans notre situation, il est souhaitable de travailler sur une partie du bord qui reflรจte mieux la structure produit. Pour la variรฉtรฉ $`\stackrel{~}{X}`$ ci-dessus le bord ร lโinfini sโidentifie ร $`S^{n_1+n_21}`$ (pour toutes les mรฉtriques $`g_{\alpha _1,\alpha _2}`$) aprรจs le choix dโune origine. Nous utiliserons $`S^{n_11}\times S^{n_21}S^{n_1+n_21}`$ qui sโidentifie dans $`\stackrel{~}{X}`$ ร $`๐^{n_1}\times ๐^{n_2}`$. Plus prรฉcisรฉment, considรฉrons, par exemple, la mรฉtrique $`\overline{g}_0=g_0^1g_0^2`$, appelons $`O=(O_1,O_2)`$ une origine de $`\stackrel{~}{X}=๐^{n_1}\times ๐^{n_2}`$, le bord de $`\stackrel{~}{X}`$ sโidentifie aux rayons gรฉodรฉsiques paramรฉtrรฉs par longueur dโarc et partant de $`O`$ ; nous ne considรฉrerons que les gรฉodรฉsiques $`c=(c_1,c_2)`$$`c_i`$ est une gรฉodรฉsique de $`๐^{n_i}`$, telle que, pour tout $`t๐`$, $`\dot{c}_1(t)_{g_0^1}=\dot{c}_2(t)_{g_0^2}`$ ; nous les appellerons gรฉodรฉsiques diagonales. Elles sont donc paramรฉtrรฉes par un point $`\theta =(\theta _1,\theta _2)`$$`\theta _iS^{n_i1}=๐^{n_i}`$. Il sโagit du bord de Furstenberg (voir ), mais nous nโutiliserons pas sa description probabiliste. Nous le noterons $`_F\stackrel{~}{X}`$. Il est important de noter que nous utiliserons toujours ce bord; en effet, si nous changeons la mรฉtrique en $`g_{\alpha _1,\alpha _2}`$, nous pouvons considรฉrer des $`g_{\alpha _1,\alpha _2}`$-gรฉodรฉsiques $`c=(c_1,c_2)`$ telles que $`\frac{1}{\alpha _1}\dot{c}_1(t)_{g_{\alpha _1}}=\frac{1}{\alpha _2}\dot{c}_2(t)_{g_{\alpha _2}}`$, oรน $`g_{\alpha _i}=\alpha _i^2g_0^i`$; elles dรฉfinissent un bord qui sโidentifie ร $`_F\stackrel{~}{X}`$.
Remarque.Lorsque $`n_1=n_2=2`$ et $`\alpha _1=\alpha _2=1`$, le bord de Furstenberg de $`\stackrel{~}{X}`$, $`S^1\times S^1S^3=\stackrel{~}{X}`$, sโidentifie naturellement ร un tore de Clifford dans $`S^3`$.
Maintenant, pour $`\theta =(\theta _1,\theta _2)S^{n_11}\times S^{n_21}`$, on note $`\overline{B}_0(,\theta )`$ la fonction de Busemann de $`(X,\overline{g}_0)`$ correspondante (lโorigine $`O=(O_1,O_2)`$ รฉtant fixรฉe), et $`B_i(,\theta _i)=B_{O_i}(,\theta _i)`$, $`i=\mathrm{1,2}`$, la fonction de Busemann de $`(๐^{n_i},g_0^i)`$, on a :
###### Lemme 2.1
Avec les notations ci-dessus, si $`x=(x_1,x_2)\stackrel{~}{X}`$
$$\overline{B}_0(x,\theta )=\frac{1}{\sqrt{2}}\left(B_1(x_1,\theta _1)+B_2(x_2,\theta _2)\right).$$
Preuve .Soit $`c`$ la gรฉodรฉsique paramรฉtrรฉe par lโabscisse curviligne dรฉfinie par $`\theta `$ et telle que $`c(0)=O=(O_1,O_2)`$. Alors, si $`c=(c_1,c_2)`$, on a $`\dot{c}_1=\frac{1}{\sqrt{2}}=\dot{c}_2`$, dโoรน
$$d_i(x_i,c_i(t))=\frac{1}{\sqrt{2}}t+B_i(x_i,\theta _i)+\epsilon _i(t),i=\mathrm{1,2}$$
avec $`\epsilon _i(t)\underset{t+\mathrm{}}{}0`$. Ici, $`d_i`$ dรฉsigne la distance dans le facteur $`i=\mathrm{1,2}`$.
Le lemme se dรฉduit alors du dรฉveloppement limitรฉ de
$$d(x,c(t))t=\left(d_1^2(x_1,c_1(t))+d_2^2(x_2,c_2(t))\right)^{1/2}t.$$
De mรชme, si $`B_{\alpha _1,\alpha _2}(,\theta )`$ dรฉsigne la fonction de Busemann de $`(\stackrel{~}{X},g_{\alpha _1,\alpha _2})`$$`\theta `$ est dans le bord dรฉfini ci-dessus, on a :
###### Lemme 2.2
Avec les notations ci-dessus, si $`x=(x_1,x_2)\stackrel{~}{X}`$
$$B_{\alpha _1,\alpha _2}(x,\theta )=\frac{1}{\sqrt{\alpha _1^2+\alpha _2^2}}\left(\alpha _1B_1(x_1,\theta _1)+\alpha _2B_2(x_2,\theta _2)\right).$$
La preuve de ce lemme se fait comme celle du lemme 2.1.
### 2.6 รlรฉment de volume
Si on note $`dv_g`$ lโรฉlรฉment de volume dโune mรฉtrique riemannienne $`g`$, il est immรฉdiat que
$$dv_{g_{\alpha _1,\alpha _2}}=\alpha _1^{n_1}\alpha _2^{n_2}dv_{g_0^1}dv_{g_0^2}$$
$`dv_{g_0^i}`$ dรฉsigne lโรฉlรฉment de volume de $`(๐^{n_i},g_0^i)`$ pour $`i=\mathrm{1,2}`$.
### 2.7 Entropie
On rappelle la dรฉfinition de lโentropie (volumique) dโune variรฉtรฉ riemannienne $`(M,g)`$ que nous supposerons compacte pour simplifier. Soit $`m\stackrel{~}{M}`$ un point du revรชtement universel $`\stackrel{~}{M}`$ de $`M`$ alors la quantitรฉ suivante existe et ne dรฉpend pas de $`x`$,
$$Ent(g)=\underset{R+\mathrm{}}{lim}\frac{1}{R}\mathrm{log}\left(vol(B_{\stackrel{~}{M}}(x,R))\right)$$
$`B_{\stackrel{~}{M}}(x,R)`$ dรฉsigne la boule mรฉtrique de centre $`x`$ et de rayon $`R`$ dans $`\stackrel{~}{M}`$ muni de la mรฉtrique relevรฉ de $`g`$.
Par dรฉfinition $`Ent(g)`$ est lโentropie de la variรฉtรฉ riemannienne $`(M,g)`$, elle ne dรฉpend de $`M`$ quโร travers la relevรฉe de $`g`$ ร $`\stackrel{~}{M}`$. Par abus de langage, nous parlerons de lโentropie de $`g_{\alpha _1,\alpha _2}`$ sur $`\stackrel{~}{X}`$.
###### Proposition 2.3
Pour tous $`\alpha _1,\alpha _2`$ positifs
$$Ent(g_{\alpha _1,\alpha _2})=\sqrt{\frac{(n_11)^2}{\alpha _1^2}+\frac{(n_21)^2}{\alpha _2^2}}.$$
Preuve .Le calcul de lโentropie des espaces symรฉtriques est fait dans . Rappelons que lโentropie dโun produit vรฉrifie
$$Ent(g_{\alpha _1,\alpha _2})^2=\frac{Ent(๐^{n_1},g_0^1)^2}{\alpha _1^2}+\frac{Ent(๐^{n_2},g_0^2)^2}{\alpha _2^2}.$$
Dans cet article on se propose de prouver un thรฉorรจme dโentropie minimale (voir lโintroduction) cโest-ร -dire de minimum de lโentropie ร volume fixรฉ. Dans ce paragraphe nous examinons cette question pour la famille de mรฉtrique $`g_{\alpha _1,\alpha _2}`$. Plus prรฉcisรฉment, soit $`\mathrm{\Gamma }`$ un sous-groupe discret cocompact de $`Isom(๐^{n_1},g_0^1)\times Isom(๐^{n_2},g_0^2)`$, agissant sans points fixes sur $`\stackrel{~}{X}`$. Ce groupe agit par isomรฉtries sur $`\stackrel{~}{X}`$ pour toutes les mรฉtriques $`g_{\alpha _1,\alpha _2}`$, on peut donc munir le quotient $`X=\stackrel{~}{X}/\mathrm{\Gamma }`$ des mรฉtriques induites que nous noterons encore $`g_{\alpha _1,\alpha _2}`$. Par ailleurs,
$$vol(X,g_{\alpha _1,\alpha _2})=\alpha _1^{n_1}\alpha _2^{n_2}vol(X,\overline{g}_0).$$
###### Proposition 2.4
Pour tous $`\alpha _1,\alpha _2`$ strictement positifs tels que $`\alpha _1^{n_1}\alpha _2^{n_2}=1`$, on a
$`Ent(g_{\alpha _1,\alpha _2})`$ $``$ $`\sqrt{n_1+n_2}\left(\left({\displaystyle \frac{n_11}{\sqrt{n_1}}}\right)^{n_1}\left({\displaystyle \frac{n_21}{\sqrt{n_2}}}\right)^{n_2}\right)^{\frac{1}{n_1+n_2}}`$
$`=`$ $`Ent\left(g_{a_1,a_2}\right)`$
$`a_1=\left[\left(\frac{(n_11)\sqrt{n_2}}{\sqrt{n_1}(n_21)}\right)^{n_2}\right]^{\frac{1}{n_1+n_2}}`$ , $`a_2=\left[\left(\frac{(n_21)\sqrt{n_1}}{\sqrt{n_2}(n_11)}\right)^{n_1}\right]^{\frac{1}{n_1+n_2}}`$.
Lโรฉgalitรฉ, dans lโinรฉgalitรฉ ci-dessus, a lieu si et seulement si $`\alpha _i=a_i`$.
Remarque.Lorsque les espaces symรฉtriques sont complexes, quaternioniens ou de Cayley, les calculs sont comparables et sont laissรฉs au lecteur.
Dans la suite nous noterons $`g_0`$ la mรฉtrique $`g_{a_1,a_2}`$.
Preuve .On a
$$Ent(g_{\alpha _1,\alpha _2})^2=(n_1+n_2)\left(\frac{n_1\left(\frac{n_11}{\sqrt{n_1}\alpha _1}\right)^2+n_2\left(\frac{n_21}{\sqrt{n_2}\alpha _2}\right)^2}{n_1+n_2}\right)$$
la fonction $`xx^2`$ รฉtant strictement log-concave
$$Ent(g_{\alpha _1,\alpha _2})^2(n_1+n_2)\left(\frac{n_11}{\sqrt{n_1}}\right)^{\frac{2n_1}{n_1+n_2}}\left(\frac{n_21}{\sqrt{n_2}}\right)^{\frac{2n_2}{n_1+n_2}}\left(\frac{1}{\alpha _1^{n_1}\alpha _2^{n_2}}\right)^{\frac{2}{n_1+n_2}}$$
dโoรน le rรฉsultat
$$Ent(g_{\alpha _1,\alpha _2})\sqrt{n_1+n_2}\left(\left(\frac{n_11}{\sqrt{n_1}}\right)^{n_1}\left(\frac{n_21}{\sqrt{n_2}}\right)^{n_2}\right)^{\frac{1}{n_1+n_2}}.$$
De plus, par stricte log-concavitรฉ, lโรฉgalitรฉ nโa lieu que si et seulement si
$$\frac{n_11}{\sqrt{n_1}\alpha _1}=\frac{n_21}{\sqrt{n_2}\alpha _2}$$
cโest-ร -dire si $`a_i=\alpha _i`$.
Remarques.
* Si $`n_1=n_2`$, alors la mรฉtrique minimisante est homothรฉtique ร $`\overline{g}_0`$ (le facteur dโhomothรฉtie รฉtant calculรฉ de sorte ร avoir un volume 1.
* La courbure de Ricci de la mรฉtrique $`g_{\alpha _1,\alpha _2}`$ est
$$Ricci(g_{\alpha _1,\alpha _2})=(n_11)g_0^1(n_21)g_0^2.$$
La mรฉtrique $`g_{\alpha _1,\alpha _2}`$ nโest donc dโEinstein que si
$$\frac{n_11}{\alpha _1^2}=\frac{n_21}{\alpha _2^2}.$$
Par consรฉquent, en gรฉnรฉral, la mรฉtrique qui minimise la fonctionnelle $`Ent`$, parmi les $`g_{\alpha _1,\alpha _2}`$, nโest pas dโEinstein. Par contre, elle lโest si et seulement si $`n_1=n_2`$.
De mรชme, si $`X`$ est un espace produit gรฉnรฉral, cโest-ร -dire, si $`(X,\overline{g})=(X_1,g_1)\times \mathrm{}\times (X_p,g_p)`$, oรน $`(X_k,g_k)`$ est un espace symรฉtrique de courbure strictement nรฉgative, de dimension $`n_k`$ et dโentropie notรฉe $`E_k`$, on considรจre les mรฉtriques,
$$g_\alpha =\alpha _1^2g_1\mathrm{}\alpha _p^2g_p$$
$`\alpha =(\alpha _1,\mathrm{},\alpha _p)`$ avec $`\alpha _k>0`$. Alors, on a la
###### Proposition 2.5
Pour tous $`\alpha _1,\mathrm{},\alpha _p`$ rรฉels strictement positifs tels que $`\alpha _1^{n_1}\mathrm{}\alpha _p^{n_p}=1`$, on a
$$Ent(g_\alpha )\sqrt{n}\left(\underset{i=1}{\overset{p}{}}\left(\frac{E_i}{\sqrt{n_i}}\right)^{\frac{n_i}{n}}\right)$$
$`n=n_1+\mathrm{}+n_p=dim(X)`$.
Lโรฉgalitรฉ, dans lโinรฉgalitรฉ ci-dessus, a lieu si, et seulement si , pour tout $`i=\mathrm{1,2},\mathrm{},p`$
$$\alpha _i=a_i=\frac{E_i}{\sqrt{n_i}}\left(\underset{k=1}{\overset{p}{}}\left(\frac{\sqrt{n_k}}{E_k}\right)^{\frac{n_k}{n}}\right).$$
### 2.8 Mesure de Patterson-Sullivan
Sur le revรชtement universel dโune variรฉtรฉ de courbure strictement nรฉgative, $`(M,g)`$, on peut dรฉfinir une famille de mesures qui est appelรฉe (par abus de langage) la mesure de Patterson-Sullivan. Elle consiste ร associer ร chaque point $`m\stackrel{~}{M}`$ (le revรชtement universel de $`M`$) une mesure borรฉlienne positive sur $`\stackrel{~}{M}`$, notรฉe $`\mu _m`$. Cette famille est entiรจrement caractรฉrisรฉe par les deux propriรฉtรฉs suivantes :
i) $`{\displaystyle \frac{d\mu _m}{d\mu _m^{}}}(\theta )=\mathrm{exp}\left(Ent(g)(B(m,\theta )B(m^{},\theta ))\right)`$ (on a choisi ici une origine $`O\stackrel{~}{M}`$ afin de dรฉfinir $`B`$). Cette propriรฉtรฉ affirme que pour $`mm^{}`$ les mesures $`\mu _m`$ et $`\mu _m^{}`$ sont absolument continues lโune par rapport ร lโautre et la densitรฉ sโexprime comme ci-dessus.
ii) $`\gamma Isom(\stackrel{~}{M})`$, $`\gamma `$ agit par homรฉomorphisme sur $`\stackrel{~}{M}`$, et
$$\mu _{\gamma (m)}=\gamma _{}(\mu _m)$$
(voir ).
Dans le cas oรน $`\stackrel{~}{M}`$ est un espace symรฉtrique de courbure nรฉgative ou nulle (et pas strictement nรฉgative) une construction est possible (voir , bet ). Dans notre situation, cโest-ร -dire
$$(\stackrel{~}{M},g)=(\stackrel{~}{X},\overline{g}_0)=(๐^{n_1},g_0^1)\times (๐^{n_2},g_0^2)$$
la famille de mesures suivante, portรฉes par $`H^{n_1}xH^{n_2}`$ vรฉrifie des propriรฉtรฉs analogues aux prรฉcรฉdentes : pour $`x=(x_1,x_2)\stackrel{~}{X}`$ et $`\theta =(\theta _1,\theta _2)๐^{n_1}\times ๐^{n_2}`$
$$d\mu _x=e^{(n_11)B_1(x_1,\theta _1)(n_21)B_2(x_2,\theta _2)}d\theta _1d\theta _2.$$
Remarque.Remarquons que la mesure ci-dessus est diffรฉrente de celle utlisรฉe dans les rรฉfรฉrences , et .
En effet,
i) Pour $`O`$ et $`x\stackrel{~}{X}`$, $`d\mu _O`$ et $`d\mu _x`$ sont absolument continues, mais la densitรฉ nโa plus la forme prรฉcรฉdente, elle vaut :
$$\frac{d\mu _x}{d\mu _O}=\mathrm{exp}\left(\left[(n_11)B_1(x_1,\theta _1)+(n_21)B_2(x_2,\theta _2)\right]\right).$$
On remarque que $`\mu _{x_i}^i=e^{(n_i1)B_i(x_i\theta _i)}d\theta _i`$ est la mesure de Patterson-Sullivan de $`(๐^{n_i},g_0^i)`$.
ii) Si $`\gamma =(\gamma _1,\gamma _2)Isom(๐^{n_1},g_0^1)\times Isom(๐^{n_2},g_0^2)`$ alors
$$\mu _{\gamma (x)}=\gamma _{}(\mu _x)$$
car
$`\gamma _{}(\mu _x)`$ $`=(\gamma _1,\gamma _2)_{}(\mu _{x_1}^1\mu _{x_2}^2)=(\gamma _1)_{}(\mu _{x_1}^1)(\gamma _2)_{}(\mu _{x_2}^2)`$
$`=\mu _{\gamma _1(x_1)}^1\mu _{\gamma _2(x_2)}^2=\mu _{\gamma (x)}.`$
De mรชme, si $`n_1=n_2`$, on vรฉrifie aisรฉment que lโisomรฉtrie supplรฉmentaire
$$\zeta (x_1,x_2)=(x_2,x_1)$$
satisfait cette contrainte.
Dans la suite nous travaillerons donc avec cette famille $`\mu _x`$ qui est le produit des mesures de Patterson-Sullivan de chaque facteur. Terminons en remarquant que si $`B_i^{\alpha _i}`$ dรฉsigne la fonction de Busemann de $`(๐^{n_i},\alpha _i^2g_0^i)`$, alors
$$Ent(\alpha _i^2g_0^i)B_i^{\alpha _i}(,)=\frac{1}{\alpha _i}(n_i1)\alpha _iB^i(,);$$
de sorte que la famille $`\mu _x`$ ne dรฉpend ni de $`\alpha _1`$, ni de $`\alpha _2`$.
### 2.9 Barycentre
Nous construisons ici une application inverse de $`x\mu _x`$, cโest-ร -dire une application qui associe ร la plupart des mesures sur $`_F\stackrel{~}{X}`$ un point de $`\stackrel{~}{X}`$ qui est son centre de masse ou barycentre. La construction est analogue ร celle de et ร lโutilisation prรจs de $`_F\stackrel{~}{X}`$ au lieu de $`\stackrel{~}{X}`$.
Soit $`๐ฑ`$ une mesure borรฉlienne positive non nulle sur $`_F\stackrel{~}{X}`$, on considรจre la fonction
$$x\stackrel{~}{X},_{\alpha _1,\alpha _2}(x)=_{_F\stackrel{~}{X}}B_{\alpha _1,\alpha _2}(x,\theta )๐๐ฑ(\theta ).$$
On dรฉfinit les mesures marginales sur $`๐^{n_1}`$ et $`๐^{n_2}`$ par :
i) $`๐ฑ_1(A_1)=๐ฑ(A_1\times ๐^{n_2})=\pi _1(๐ฑ)`$, oรน $`A_1`$ est un borรฉlien de $`๐^{n_1}`$ et $`\pi _1`$ la projection canonique de $`_F\stackrel{~}{X}`$ sur $`๐^{n_1}`$ ; et de mรชme,
ii) $`๐ฑ_2(A_2)=๐ฑ(๐^{n_1}\times A_2)=\pi _2(๐ฑ)`$, oรน $`A_2`$ est un borรฉlien de $`๐^{n_2}`$ et $`\pi _2`$ la projection de $`_F\stackrel{~}{X}`$ sur $`๐^{n_2}`$ .
###### Proposition 2.6
Si $`๐ฑ_1`$ et $`๐ฑ_2`$ sont des mesures non nulles et sans atomes, pour tous $`\alpha _1`$, $`\alpha _2`$ strictement positifs, la fonction $`_{\alpha _1,\alpha _2}`$ est $`๐^{\mathrm{}}`$, strictement convexe sur $`\stackrel{~}{X}`$ et tend vers lโinfini lorsque $`x`$ tend vers lโinfini.
Preuve .Par dรฉfinition de $`๐ฑ_1`$, $`๐ฑ_2`$ et $`B_{\alpha _1,\alpha _2}`$, on a :
$$_{\alpha _1,\alpha _2}(x)=\frac{1}{\sqrt{\alpha _1^2+\alpha _2^2}}\left(\alpha _1_{๐^{n_1}}B_1(x_1,\theta _1)๐๐ฑ_1(\theta _1)+\alpha _2_{๐^{n_2}}B_2(x_2,\theta _2)๐๐ฑ_2(\theta _2)\right).$$
En effet,
$`{\displaystyle _{_F\stackrel{~}{X}}}B_1(x_1,\theta _1)d๐ฑ(\theta _1,\theta _2))`$ $`={\displaystyle _{_F\stackrel{~}{X}}}B_1(x_1,\pi _1(\theta _1,\theta _2))๐๐ฑ(\theta _1,\theta _2)`$
$`={\displaystyle _{๐^{n_1}}}B_1(x_1,\theta )d(\pi _1๐ฑ)(\theta _1)`$
et de mรชme avec lโautre terme. Alors, on applique les rรฉsultats de , et qui montrent que $`x_i_{๐^{n_1}}B_i(x_i,\theta _i)๐๐ฑ(\theta _i)`$ est strictement convexe, pour $`i=\mathrm{1,2}`$, et tend vers lโinfini lorsque $`x_i`$ tend vers lโinfini dans $`๐^{n_i}`$. On rappelle quโune fonction est dite strictement convexe si elle lโest en restriction ร toute gรฉodรฉsique non constante. Il est alors facile de vรฉrifier que $`_{\alpha _1,\alpha _2}`$ est strictement convexe en restriction ร toute gรฉodรฉsique non constante de $`\stackrel{~}{X}=๐^{n_1}\times ๐^{n_2}`$. Les autres conclusions de la proposition sont รฉgalement รฉvidentes.
Remarque.Lโhypothรจse sur la mesure $`๐ฑ`$ est vรฉrifiรฉe, par exemple, dรจs que celle-ci est absolument continue par rapport ร la mesure de Lebesgue sur $`_F\stackrel{~}{X}`$. Par ailleurs, elle peut รชtre affaiblie (voir ).
###### Corollaire 2.7
Sous les mรชmes hypothรจses, la fonction $`_{\alpha _1,\alpha _2}`$ admet un unique minimum sur $`\stackrel{~}{X}`$ que nous appellerons le barycentre de $`๐ฑ`$, notรฉ $`bar(๐ฑ)`$, qui ne dรฉpend pas de $`\alpha _1`$, $`\alpha _2`$ (ร condition quโils soient strictement positifs). De plus $`bar(๐ฑ)=(bar_1(๐ฑ_1),bar_2(๐ฑ_2))`$, oรน $`bar_i(๐ฑ_i)`$ dรฉsigne le barycentre de la mesure $`๐ฑ_i`$ dans $`๐^{n_i}`$.
Preuve .Lโunicitรฉ rรฉsulte de la stricte convexitรฉ de $`_{\alpha _1,\alpha _2}`$ $`(\alpha _1>0,\alpha _2>0)`$ et du fait $`_{\alpha _1,\alpha _2}(x)\underset{x+\mathrm{}}{}+\mathrm{}`$. Le point $`x^{}=(x_1^{},x_2^{})`$ est dรฉfini par lโรฉquation vectorielle,
$$\stackrel{}{_{\alpha _1,\alpha _2}}(x^{})=\stackrel{}{O}$$
cโest-ร -dire $`\alpha _1_{๐^{n_1}}\stackrel{}{_1_1}(x_1^{},\theta _1)๐๐ฑ_1(\theta _1)+\alpha _2_{๐^{n_2}}\stackrel{}{_2_2}(x_2^{},\theta _2)๐๐ฑ_2(\theta _2)=0`$ (ici $`_i`$ dรฉsigne le gradient dโune fonction dรฉfinie sur $`๐^{n_i}`$).
Si $`\overline{x}_i=bar_i(๐ฑ_i)`$ (voir ), alors
$$_{๐^{n_i}}_iB_i(\overline{x}_i,\theta _i)๐๐ฑ(\theta _i)=\stackrel{}{O}\text{ pour }i=\mathrm{1,2}.$$
Par unicitรฉ on a donc $`\overline{x}_i=x_i^{}`$ ($`i=\mathrm{1,2}`$), cโest-ร -dire
$$bar(๐ฑ_{\alpha _1,\alpha _2})=(bar_1(๐ฑ_1),bar_2(๐ฑ_2)).$$
## 3 Le thรฉorรจme principal
Dans ce chapitre, nous nous proposons de prouver le thรฉorรจme principal sur lโentropie, analogue, dans cette situation, des rรฉsultats prouvรฉs dans et . Nous donnons lโรฉnoncรฉ et la preuve dans un cas particulier afin dโรฉviter des lourdeurs dans les notations ; le cas gรฉnรฉral est rigoureusement identique.
Nous considรฉrons, comme prรฉcรฉdemment, $`\stackrel{~}{X}=๐^{n_1}\times ๐^{n_2}`$, oรน $`n_i3`$ et nous munissons $`X`$ de la mรฉtrique $`g_0=g_{a_1,a_2}`$, oรน les nombres $`a_i`$ sont ceux calculรฉs dans la proposition 1.4. La mรฉtrique $`g_0`$ minimise lโentropie normalisรฉe, sur $`X`$, parmi les mรฉtriques $`g_{\alpha _1,\alpha _2}`$ (voir la proposition 2.4). Soit $`\mathrm{\Gamma }`$ un sous-groupe du groupe dโisomรฉtries de $`(\stackrel{~}{X},g_0)`$ tel que $`X=\stackrel{~}{X}/\mathrm{\Gamma }`$ est une variรฉtรฉ compacte ($`\mathrm{\Gamma }`$ est un rรฉseau co-compact et sans torsion).
###### Thรฉorรจme 3.1
Soit $`(Y,g)`$ une variรฉtรฉ riemannienne compacte de dimension $`n=n_1+n_2`$ et $`f:YX`$ une application continue, alors
* $`(Ent(Y,g))^nvol(Y,g)|\text{deg}f|Ent(X,g_0)^nvol(X,g_0)`$ ;
* lโรฉgalitรฉ, dans lโinรฉgalitรฉ ci-dessus, a lieu si, et seulement si, $`f`$ est homotope ร un revรชtement riemannien.
Remarque.Ce rรฉsultat est vrai dans la version gรฉnรฉrale donnรฉe en introduction. Sa preuve est analogue ร celle de .
Preuve de lโinรฉgalitรฉ i).
Nous donnons une preuve inspirรฉe de la technique dรฉveloppรฉe dans . On note $`\mathrm{\Gamma }=\pi _1(Y)`$, le groupe fondamental de $`Y`$, $`\stackrel{~}{Y}`$ le revรชtement universel de $`Y`$. Lโapplication continue $`f:YX`$ induit un morphisme $`\rho :\mathrm{\Gamma }\mathrm{\Gamma }_0=\pi _1(X)`$. On appelle $`\mu _O`$ la mesure (canonique) $`d\theta _1d\theta _2`$ sur $`_F\stackrel{~}{X}`$ et $`(_F\stackrel{~}{X})`$ lโespace des mesures de Radon positives sur $`_F\stackrel{~}{X}`$. Soit $`O\stackrel{~}{Y}`$ un point fixรฉ (une origine) ; considรฉrons lโapplication
$`\stackrel{~}{Y}`$ $`(_F\stackrel{~}{X})`$
$`y`$ $`\mu _{y,\epsilon }={\displaystyle \underset{\gamma \mathrm{\Gamma }}{}}e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}\rho (\gamma )_{}(\mu _O)`$
$`\epsilon >0`$. Cette application est รฉquivariante; en effet, pour tout $`\alpha \mathrm{\Gamma }`$,
$$\mu _{\alpha (y),\epsilon }=\rho (\alpha )_{}(\mu _{y,\epsilon }).$$
On dรฉfinit alors
$`\stackrel{~}{F}_\epsilon :\stackrel{~}{Y}`$ $`\stackrel{~}{X}`$
$`y`$ $`bar(\mu _{y,\epsilon }).`$
Notons que chaque mesure $`\mu _{y,\epsilon }`$ est sans atome. La notion de barycentre รฉtant indรฉpendante des coefficients $`\alpha _i`$ servant ร dรฉfinir la mรฉtrique, nous utiliserons, pour simplifier, la mรฉtrique $`\overline{g}_0=g_0^1g_0^2`$ (voir le corollaire 2.7).
Alors, par รฉquivariance, $`\stackrel{~}{F}_\epsilon `$ donne une famille dโapplications
$$F_\epsilon :YX.$$
Par ailleurs, le barycentre sur $`\stackrel{~}{X}`$ se dรฉcompose (cf. corollaire 2.7) et donc รฉgalement la fonction $`\stackrel{~}{F}_\epsilon `$
$`\stackrel{~}{F}_\epsilon :\stackrel{~}{Y}`$ $`\stackrel{~}{X}=๐^{n_1}\times ๐^{n_2}`$
$`y`$ $`(\stackrel{~}{F}_{1,\epsilon }(y),\stackrel{~}{F}_{2,\epsilon }(y))`$
$`\stackrel{~}{F}_{i,\epsilon }(y)=bar_i(\pi _i(\mu _{y,\epsilon }))`$.
Nous notons $`F_\epsilon `$, $`F_{1,\epsilon }`$ et $`F_{2,\epsilon }`$ les applications correspondantes de $`Y`$ dans $`X`$.
###### Lemme 3.2
Pour tout $`\epsilon >0`$, les fonctions $`F_\epsilon `$, $`F_{1,\epsilon }`$ et $`F_{2,\epsilon }`$ sont lipschitziennes.
Preuve .
i) La sรฉrie $`\underset{\gamma \mathrm{\Gamma }}{}e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}`$ converge dรจs que $`\epsilon >0`$ ; en effet, puisque $`Y`$ est compacte, elle est comparable ร lโintรฉgrale
$$I_\epsilon =_{\stackrel{~}{Y}}e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,z)}๐v_g(z)$$
qui converge. Pour vรฉrifier ce dernier point il suffit dโรฉcrire $`I_\epsilon `$ en coordonnรฉes polaires et dโappliquer la dรฉfinition de $`Ent(Y,g)`$.
ii) La fonction $`\stackrel{~}{F}_\epsilon `$ est dรฉfinie par lโรฉquation implicite
$$_{_F\stackrel{~}{X}}\stackrel{}{\overline{B}_0}_{|(F_\epsilon (y),\theta )}(\underset{\gamma \mathrm{\Gamma }}{}e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}\rho (\gamma )_{}(d\mu _O))=\stackrel{}{0}$$
$`\overline{B}_0`$ dรฉsigne la fonction de Busemann de la mรฉtrique $`\overline{g}_0=g_{\mathrm{1,1}}`$ sur $`\stackrel{~}{X}`$. Ici $`\theta =(\theta _1,\theta _2)`$. Lโรฉquation ci-dessus peut se rรฉcrire en
$$0=L(x,y)=_{_F\stackrel{~}{X}}\underset{\gamma \mathrm{\Gamma }}{}\stackrel{}{\overline{B}_0}_{(x,\rho (\gamma )\theta )}e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}d\mu _O$$
$`x=(x_1,x_2)\stackrel{~}{X}`$. La fonction $`L`$ est $`๐^{\mathrm{}}`$ en $`x`$, lipschitzienne en $`y`$ et chaque diffรฉrentielle partielle en $`x`$ est lipschitzienne en $`y`$. Dans cette situation, on peut utiliser le thรฉorรจme des fonctions implicites pour conclure que $`\stackrel{~}{F}_\epsilon `$ est lipschitzienne en $`y`$ (voir ). Remarquons que la condition sur la diffรฉrentielle partielle en $`x`$ qui est nรฉcessaire pour appliquer le thรฉorรจme des fonctions implicites est exactement celle qui prouve lโexistence du barycentre (voir le chapitre prรฉcรฉdent), cโest-ร -dire la stricte convexitรฉ de $`_{\mathrm{1,1}}`$.
###### Lemme 3.3
Pour $`\epsilon >0`$ et pour tout $`y\stackrel{~}{Y}`$, on a
$$|Jac\stackrel{~}{F}_\epsilon (y)|(1+\epsilon )^n\left(\frac{Ent(Y,g)}{Ent(X,g_0)}\right)^n$$
$`n=dimY=dimX=n_1+n_2`$ et le Jacobien est calculรฉ ร lโaide des mรฉtriques $`g`$ sur $`Y`$ et $`g_0`$ sur $`X`$.
Preuve .Comme nous lโavons remarquรฉ dans le chapitre prรฉcรฉdent la notion de barycentre, et donc la dรฉfinition de lโapplication $`\stackrel{~}{F}`$, ne dรฉpend pas de $`\alpha _1`$, $`\alpha _2`$. Nous pouvons donc utiliser sur $`\stackrel{~}{X}`$ la mรฉtrique $`\overline{g}_0=g_0^1g_0^2`$ (on rappelle que $`g_0^i`$ dรฉsigne ici la mรฉtrique de courbure constante รฉgale ร $`1`$ sur $`๐^{n_i}`$). Rappelons รฉgalement la notation $`g_0=a_1g_0^1a_2g_0^2`$$`a_i`$ sont les valeurs calculรฉes dans la section prรฉcรฉdente, telles que $`g_0`$ minimise lโentropie normalisรฉe parmi les mรฉtriques $`g_{\alpha _1,\alpha _2}`$. Nous noterons $`\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))`$ le dรฉterminant de la diffรฉrentielle de $`\stackrel{~}{F}_\epsilon `$ en $`y`$ calculรฉ ร lโaide des mรฉtriques $`g`$ sur $`\stackrel{~}{Y}`$ et $`\overline{g_0}`$ sur $`\stackrel{~}{X}`$ ; par ailleurs $`Jac\stackrel{~}{F}_\epsilon (y)=a_1^{n_1}a_2^{n_2}\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))`$ est le dรฉterminant de $`D\stackrel{~}{F}_\epsilon (y)`$ calculรฉ ร lโaide des mรฉtriques $`g`$ sur $`\stackrel{~}{Y}`$ et $`g_0`$ sur $`\stackrel{~}{X}`$. Notons que $`g_0`$ est normalisรฉe par $`a_1^{n_1}a_2^{n_2}=1`$, de sorte que $`Jac\stackrel{~}{F}_\epsilon (y)=a_1^{n_1}a_2^{n_2}\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))=\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))`$. Nous distinguerons toutefois les deux expressions afin dโรฉviter les confusions entre les mรฉtriques $`g_0`$ et $`\overline{g}_0`$.
Estimation de $`\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))`$; ici, tous les calculs se font ร lโaide de la mรฉtrique $`\overline{g}_0`$ sur $`\stackrel{~}{X}`$. Rappelons que nous dรฉsignons par $`B_i`$ les fonctions de Busemann sur $`๐^{n_i}`$ muni de la mรฉtrique $`g_0^i`$. Comme dans , page 155, nous posons
$`k_{y,\epsilon }(v,v)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}Dd\overline{B}_0{}_{|(\stackrel{~}{F}_\epsilon (y),\theta )}{}^{}(v,v)d\mu _{y,\epsilon }(\theta )=\overline{g}_0(K_{y,\epsilon }(v),v)`$
$`h_{y,\epsilon }(v,v)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_\epsilon (y),\theta )}{}^{}(v)\right)^2d\mu _{y,\epsilon }(\theta )`$
$`=\overline{g}_0(H_{y,\epsilon }(v),v)`$
$`vT_{\stackrel{~}{F}_\epsilon (y)}\stackrel{~}{X}`$. Ici, comme dans la section prรฉcรฉdente,
$$\overline{B}_0(x,\theta )=\frac{1}{\sqrt{2}}\left(B_1(x_1,\theta _1)+B_2(x_2,\theta _2)\right).$$
Enfin,
$`h_{y,\epsilon }^{}(u,u)`$ $`={\displaystyle \underset{\gamma \mathrm{\Gamma }}{}}{\displaystyle _{_F\stackrel{~}{X}}}d_{\stackrel{~}{Y}|(y,\gamma (O))},u^2e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}d(\rho (\gamma )_{}\mu _O)(\theta )`$
$`=\left(\mu _O(_F\stackrel{~}{X})\right){\displaystyle \underset{\gamma \mathrm{\Gamma }}{}}d_{\stackrel{~}{Y}|(y,\gamma (O))},u^2e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}`$
$`=g(H_{y,\epsilon }^{}u,u)`$
pour $`uT_y\stackrel{~}{Y}`$. Nous utiliserons les mรชmes notations pour les formes bilinรฉaires associรฉes. En diffรฉrenciant lโรฉquation implicite qui dรฉfinit $`\stackrel{~}{F}_\epsilon `$, nous obtenons, pour $`uT_y\stackrel{~}{Y}`$ et $`vT_{\stackrel{~}{F}_\epsilon (y)}\stackrel{~}{X}`$,
$$\begin{array}{c}k_{y,\epsilon }(v,D\stackrel{~}{F}_\epsilon (y)(u))=(1+\epsilon )Ent(Y,g)\underset{\gamma \mathrm{\Gamma }}{}_{_F\stackrel{~}{X}}g(d_{\stackrel{~}{Y}|(y,\gamma (O))},u)\overline{g}_0(\overline{B}_{0|(\stackrel{~}{F}_\epsilon (y),\theta )},v)\hfill \\ \hfill e^{Ent(Y,g)(1+\epsilon )d_{\stackrel{~}{Y}}(y,\gamma (O))}d(\rho (\gamma )_{}\mu _O)(\theta )\end{array}$$
et, en utilisant lโinรฉgalitรฉ de Cauchy-Schwarz,
$$k_{y,\epsilon }(D\stackrel{~}{F}_\epsilon (y)(u),v)(1+\epsilon )Ent(Y,g)\left(h_{y,\epsilon }(v,v)\right)^{1/2}\left(h_{y,\epsilon }^{}(u,u)\right)^{1/2}.$$
$`(2.5)`$
Un lemme รฉlรฉmentaire dโalgรจbre linรฉaire (cf. , lemme 5.4) donne, ร partir de (2.5),
$$\text{dรฉt}(K_{y,\epsilon })\overline{\text{dรฉt}}(D\stackrel{~}{F}_\epsilon (y))\left((1+\epsilon )Ent(Y,g)\right)^{n_1+n_2}\left(\text{dรฉt}H_{y,\epsilon }\right)^{1/2}\left(\text{dรฉt}H_{y,\epsilon }^{}\right)^{1/2}.$$
Nous devons maintenant remarquer que la notion de barycentre ne change pas lorsque lโon multiplie une mesure par un nombre strictement positif, de sorte que
$$\stackrel{~}{F}_\epsilon (y)=bar(\mu _{y,\epsilon })=bar\left(\frac{\mu _{y,\epsilon }}{\mu _{y,\epsilon }(_F\stackrel{~}{X})}\right).$$
On peut donc supposer que la famille de mesures que lโon considรจre est normalisรฉe (de masse totale รฉgale ร 1 pour tout $`y\stackrel{~}{Y}`$ et $`\epsilon >0`$). La trace dโune forme quadratique $`\phi `$ (calculรฉe dans une base orthonormรฉe par rapport ร une structure euclidienne $`g`$) รฉtant notรฉe $`trace_g\phi `$, en injectant dans la dรฉfinition de $`h_{y,\epsilon }^{}`$ le fait que $`d_{\stackrel{~}{Y}}_g=1`$, nous obtenons
$$trace(H_{y,\epsilon }^{})=trace_g(h_{y,\epsilon }^{})=1,$$
dโoรน
$$\left(\text{dรฉt}H_{y,\epsilon }^{}\right)^{1/2}\left(\frac{1}{\sqrt{n_1+n_2}}\right)^{n_1+n_2}.$$
Maintenant la dรฉfinition de $`h_{y,\epsilon }`$ (et $`H_{y,\epsilon }`$) montre que
$$H_{y,\epsilon }=\left(\begin{array}{cc}H_1& \\ & H_2\end{array}\right)$$
$`H_i`$ dรฉsigne la restriction de $`H_{y,\epsilon }`$ ร $`๐^{n_i}`$ ; plus prรฉcisรฉment, pour $`i=\mathrm{1,2}`$ et $`v_iT_{F_{i,\epsilon }(y)}๐^{n_i}`$
$`g_0^i(H_iv_i,v_i)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_{0|(\stackrel{~}{F}_\epsilon (y),\theta )}(v_i)\right)^2๐\mu _{y,\epsilon }(\theta )`$
$`={\displaystyle _{_F\stackrel{~}{X}}}{\displaystyle \frac{1}{2}}\left(dB_{i|(\stackrel{~}{F}_{i,\epsilon }(y),\theta _i)}(v_i)\right)^2๐\mu _{y,\epsilon }(\theta )`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _{๐^{n_i}}}\left(dB_{i|(\stackrel{~}{F}_{i,\epsilon }(y),\theta _i)}(v_i)\right)^2d\left(\pi _i(\mu _{y,\epsilon })\right)(\theta _i).`$
Remarquons que, puisque $`dB_i_{g_0^i}=1`$ et $`(\pi _i)_{}\mu _{y,\epsilon }`$ est une probabilitรฉ, nous avons
$$trace(2H_i)=1.$$
De mรชme,
$$K_{y,\epsilon }=\left(\begin{array}{cc}K_1& 0\\ 0& K_2\end{array}\right)$$
avec, pour $`i=\mathrm{1,2}`$ et $`v_i=T_{\stackrel{~}{F}_{i,\epsilon }(y)}๐^{n_i}`$,
$$g_0^i(K_iv_i,v_i)=\frac{1}{\sqrt{2}}_{๐^{n_i}}D๐B_{i|(\stackrel{~}{F}_{i,\epsilon }(y),\theta _i)}(v_i,v_i)d\left(\pi _i(\mu _{y,\epsilon })\right)(\theta _i).$$
###### Lemme 3.4
Avec les notations prรฉcรฉdentes, nous avons
* $`\text{dรฉt}(K_{y,\epsilon })=\text{dรฉt}(K_1)\text{dรฉt}(K_2);`$
* $`\text{dรฉt}(H_{y,\epsilon })\text{dรฉt}(H_1)\text{dรฉt}(H_2)`$.
Lโรฉgalitรฉ i) est รฉvidente et lโinรฉgalitรฉ ii) est classique (voir , p. 63) pour les matrices symรฉtriques.
Par ailleurs, sur les espaces hyperboliques $`๐^{n_i}`$, la relation suivante est vรฉrifiรฉe (voir , p. 751), pour $`i=\mathrm{1,2}`$,
$$DdB_i=g_0^idB_idB_i$$
qui se traduit en
$$K_i=\frac{1}{\sqrt{2}}(I_i2H_i)$$
$`I_i`$ dรฉsigne lโidentitรฉ de $`T_{F_{i,\epsilon }(y)}๐^{n_i}`$. En regroupant ces remarques, nous obtenons, ร partir de 2.5,
$$\overline{\text{dรฉt}}(D\stackrel{~}{\stackrel{~}{F}_\epsilon }(y))\left(\frac{(1+\epsilon )Ent(Y,g)}{\sqrt{n_1+n_2}}\right)^{n_1+n_2}\frac{(\text{dรฉt}2H_1)^{1/2}}{\text{dรฉt}(I2H_1)}\frac{(\text{dรฉt}2H_2)^{1/2}}{\text{dรฉt}(I2H_2)}.$$
Alors, un lemme algรฉbrique donne (cf. , appendice B),
$$\frac{(\text{dรฉt}2H_i)^{1/2}}{\text{dรฉt}(I2H_i)}\left(\frac{\sqrt{n_i}}{n_i1}\right)^{n_i}$$
lโรฉgalitรฉ nโayant lieu que si, et seulement si, $`2H_i=\frac{1}{n_i}I_i`$ (on rappelle que $`trace(2H_i)=1`$).
En regroupant ces inรฉgalitรฉs, il vient
$`Jac\stackrel{~}{\stackrel{~}{F}_\epsilon }(y)`$ $`=a_1^{n_1}a_2^{n_2}(\overline{\text{dรฉt}}(D\stackrel{~}{\stackrel{~}{F}_\epsilon }(y))`$
$`\left((1+\epsilon )Ent(Y,g)\right)^{n_1+n_2}{\displaystyle \frac{a_1^{n_1}a_2^{n_2}}{(\sqrt{n_1+n_2})^{n_1+n_2}}}\left({\displaystyle \frac{\sqrt{n_1}}{n_11}}\right)^{n_1}\left({\displaystyle \frac{\sqrt{n_2}}{n_21}}\right)^{n_2}`$
$`=\left((1+\epsilon ){\displaystyle \frac{Ent(Y,g)}{Ent(X,g_0)}}\right)^n`$
dโaprรจs la proposition 2.4. Ce qui prouve le lemme 3.3.
Lโinรฉgalitรฉ i) du thรฉorรจme 3.1 sโen dรฉduit par intรฉgration et passage ร la limite en $`\epsilon =0`$.
Le cas dโรฉgalitรฉ sera traitรฉ, dans un cadre plus gรฉnรฉral, dans le paragraphe suivant.
Remarques sur le cas gรฉnรฉral. Si $`\stackrel{~}{X}=\stackrel{~}{X}_1\times \mathrm{}\times \stackrel{~}{X}_p`$ et $`\overline{g}_0=g_1\mathrm{}g_p`$$`(\stackrel{~}{X}_k,g_k)`$ est un espace symรฉtrique de courbure strictement nรฉgatif et de dimension $`n_k`$, on munit $`X`$ de la mรฉtrique $`g_0=a_1g_1\mathrm{}a_pg_p`$, oรน les nombres $`a_i`$ sont ceux calculรฉs dans la proposition 2.5. La mรฉtrique $`g_0`$ minimise lโentropie normalisรฉe parmi les mรฉtriques $`g_\alpha `$ (voir la proposition 2.5).
Alors, comme ci-dessus, on pose $`\overline{g}_0=g_1\mathrm{}g_p`$. On suppose de plus que la courbure sectionnelle de $`(\stackrel{~}{X}_k,g_k)`$ est normalisรฉe de sorte quโelle soit รฉgale ร $`1`$ si $`(\stackrel{~}{X}_k,g_k)`$ est hyperbolique rรฉelle et comprise entre $`4`$ et $`1`$ dans les autres cas. Le calcul de lโentropie dโune telle mรฉtrique est donnรฉ dans , p. 740.
Pour $`x=(x_1,\mathrm{},x_p)\stackrel{~}{X}`$ et $`\theta =(\theta _1,\mathrm{},\theta _p)_F\stackrel{~}{X}`$ ($`_F\stackrel{~}{X}=\stackrel{~}{X}_1\times \mathrm{}\times \stackrel{~}{X}_p`$), la fonction de Busemann de $`(\stackrel{~}{X},\overline{g}_0)`$ est
$$\overline{B}_0(x,\theta )=\frac{1}{\sqrt{p}}\left(B_1(x_1,\theta _1)+\mathrm{}+B_p(x_p,\theta _p)\right)$$
et on a les dรฉcompositions
$$H_{y,\epsilon }=\left(\begin{array}{cccc}H_1& & & \\ & H_2& & \\ & & \mathrm{}& \\ & & & H_p\end{array}\right),K_{y,\epsilon }=\left(\begin{array}{cccc}K_1& 0& 0& 0\\ 0& K_2& 0& 0\\ 0& 0& \mathrm{}& 0\\ 0& 0& 0& K_p\end{array}\right)$$
avec $`trace(pH_k)=1`$ pour $`k=\mathrm{1,2},\mathrm{},p`$.
La relation qui lie $`K_i`$ et $`H_i`$ dรฉpend du type dโespace considรฉrรฉ (hyperbolique rรฉel, complexe, quaternionien ou de Cayley) et est dรฉcrite dans , p. 751. On peut vรฉrifier aisรฉment que
$$trace(\sqrt{p}K_k)=E_k=\text{ entropie de }(\stackrel{~}{X}_k,g_k)$$
pour $`k=1,\mathrm{},p`$.
Dans lโappendice $`B`$ de nous montrons que
$$\frac{\text{dรฉt}(pH_k)^{1/2}}{\text{dรฉt}(\sqrt{p}K_k)}\left(\frac{\sqrt{n_k}}{E_k}\right)^{n_k}.$$
On conclut, alors, grรขce ร la proposition 2.5, comme ci-dessus.
## 4 Le volume des reprรฉsentations
Nous donnons dans ce paragraphe une application de la technique introduite dans aux reprรฉsentations du groupe fondamental dโune variรฉtรฉ compacte.
Dans ce qui suit $`\stackrel{~}{X}`$ est un produit fini dโespaces symรฉtriques simplement connexe de courbure strictement nรฉgative. Chaque facteur est supposรฉ de dimension supรฉrieure ou รฉgale ร 3. On munit $`\stackrel{~}{X}`$ de la mรฉtrique $`g_0`$ dรฉcrite dans la proposition 2.4, cโest-ร -dire celle qui rรฉalise lโentropie minimale pour tous les quotients compacts de $`\stackrel{~}{X}`$. Par ailleurs, $`(Y,g)`$ est une variรฉtรฉ riemannienne compacte dont le groupe fondamental est notรฉ $`\mathrm{\Gamma }`$. On considรจre
$$\rho :\mathrm{\Gamma }Isom(\stackrel{~}{X},g_0)$$
une reprรฉsentation. Il existe toujours des applications รฉquivariantes $`f:\stackrel{~}{Y}\stackrel{~}{X}`$ car $`\stackrel{~}{X}`$ est contractile (dans la suite nous donnerons un exemple explicite dโune telle application). Elle vรฉrifie donc
$$\phi \mathrm{\Gamma },y\stackrel{~}{Y},f(\gamma (y))=\rho (\gamma )f(y).$$
On peut toujours la supposer $`C^1`$, quitte ร la rรฉgulariser. Si on note $`\omega _0`$ la forme volume de $`(\stackrel{~}{X},g_0)`$ alors,
###### Definition 4.1
On appelle volume de la reprรฉsentation $`\rho `$, le nombre
$$vol(\rho )=_Yf^{}(\omega _0).$$
Remarques.
* La dรฉfinition ci-dessus a un sens car, $`f`$ รฉtant $`C^1`$, $`f^{}(\omega _0)`$ est une forme continue sur $`\stackrel{~}{Y}`$ qui de plus est invariante par $`\mathrm{\Gamma }`$. Par ailleurs, il est immรฉdiat de vรฉrifier que $`vol(\rho )`$ ne dรฉpend pas du choix de lโapplication รฉquivariante $`f`$.
* Il faut interprรฉter $`vol(\rho )`$ comme lโanalogue de la quantitรฉ $`|\text{deg}f|vol(X)`$ du thรฉorรจme 3.1. En effet, lorsque $`\rho (\mathrm{\Gamma })`$ est discret et cocompact, agissant sans point fixe, nous nous trouvons dans la situation du paragraphe 3 oรน $`X=\stackrel{~}{X}/\rho (\mathrm{\Gamma })`$ et $`vol(X)|\text{deg}f|=vol(\rho )`$ par dรฉfinition du degrรฉ de lโapplication $`f`$.
Nous prouvons donc un thรฉorรจme analogue :
###### Thรฉorรจme 4.2
Avec les notations ci-dessus :
* $`vol(\rho )\left({\displaystyle \frac{Ent(Y,g)}{Ent(\stackrel{~}{X},g_0)}}\right)^nvol(Y,g)`$.
* Lโรฉgalitรฉ, dans lโinรฉgalitรฉ ci-dessus a lieu si, et seulement si, la reprรฉsentation $`\rho `$ est injective, $`X=\stackrel{~}{X}/\rho (\mathrm{\Gamma })`$ est une variรฉtรฉ compacte et $`(Y,g)`$ est homothรฉtique ร $`(X,g_0)`$.
Remarques.
* Ce rรฉsultat est un premier pas dans la comprรฉhension des reprรฉsentations des groupes fondamentaux de variรฉtรฉs compactes dans des groupes dโisomรฉtries dโespaces symรฉtriques de type non compact.
* Les exemples de telles reprรฉsentations sont rares et nous discuterons ce point plus loin dans le texte. Plus rares encore sont les exemples dont le volume est non nul.
* Seul le cas de dimension 2, oรน notre mรฉthode ne sโapplique pas, est complรจtement compris (cf. ). En particulier, le thรฉorรจme 4.2 est une gรฉnรฉralisation de la cรฉlรจbre inรฉgalitรฉ de Milnor-Wood (cf. , et ).
Preuve .Lโinรฉgalitรฉ est รฉlรฉmentaire et sa preuve est celle du thรฉorรจme 3.1, i). Le cas dโรฉgalitรฉ par contre est beaucoup plus difficile car nous ne possรฉdons pas de quotient compact de $`\stackrel{~}{X}`$ ($`\stackrel{~}{X}/\rho (\mathrm{\Gamma })`$ nโest mรชme pas un espace sรฉparรฉ, en gรฉnรฉral) sur lequel sโappuyer afin dโutiliser la thรฉorie du degrรฉ (voir la preuve du cas dโรฉgalitรฉ de ).
Afin de traiter ce cas dโรฉgalitรฉ difficile nous devons considรฉrer une autre application รฉquivariante que celle introduite dans le paragraphe 3. Soit $`f`$ une premiรจre application continue et $`\rho `$-รฉquivariante,
$$f:\stackrel{~}{Y}\stackrel{~}{X},$$
par exemple, nous pouvons prendre comme prรฉcรฉdemment
$$\stackrel{~}{f}(y)=bar\left(\underset{\gamma \mathrm{\Gamma }}{}e^{Ent(Y,g)(1+\epsilon )d(y,\gamma (O))}\rho (\gamma )_{}d\mu \right)$$
les notations รฉtant, ici, celles du paragraphe 2.
On rappelle que si $`\theta _F\stackrel{~}{X}`$ et $`z\stackrel{~}{X}`$, $`P_0(z,\theta )`$ dรฉsigne le noyau de Poisson de $`\stackrel{~}{X}`$, normalisรฉ en une origine $`O_0\stackrel{~}{X}`$ de sorte que
$$P_0(O_0,)1.$$
Nous construisons une autre application, comme dans , dรฉfinie, pour tout$`c>Ent(Y,g)`$, par
$$\stackrel{~}{F}_c(y)=bar(\left(_{\stackrel{~}{Y}}e^{cd(y,z)}P_0(\stackrel{~}{f}(z),\theta )dv_g(z)\right)d\theta ).$$
La preuve de lโinรฉgalitรฉ i) du thรฉorรจme 4.2 est rigoureusement identique ร celle donnรฉe dans le paragraphe 3. Nous ne la reproduirons donc pas. Notons quโelle peut รชtre faite ร lโaide de la fonction $`\stackrel{~}{f}`$ dรฉfinie e quโil nโest pas nรฉcessaire dโutiliser la fonction $`\stackrel{~}{F}_c`$; cette derniรจre est toutefois beaucoup plus aisรฉe ร manipuler dans la preuve du cas dโรฉgalitรฉ; elle est, par exemple plus rรฉguliรจre que $`f`$.
Posons comme dans , pour $`c>Ent(Y,g)`$
$$\psi (c,y,\theta )=_{\stackrel{~}{Y}}e^{cd(y,z)}P_0(f(z),\theta )๐v_g(z)$$
et
$$\mathrm{\Phi }(c,y,\theta )=\frac{\psi (c,y,\theta )}{_{_F\stackrel{~}{X}}\psi (c,y,\theta )๐\theta }$$
qui est de norme $`L^1(_F\stackrel{~}{X},d\theta )`$ รฉgale ร 1.
###### Lemme 4.3
Lโapplication $`(c,y)\mathrm{\Phi }(c,y,)`$ est de classe $`C^1`$ de lโintervalle $`]Ent(Y,g),+\mathrm{}[\times \stackrel{~}{Y}`$ dans $`L^1(_F\stackrel{~}{X})`$.
Preuve .โIl nโest pas possible de montrer le lemme ci-dessus par simple application du thรฉorรจme de dรฉrivation sous le signe somme. Toutefois, dans , nous prouvons, comme corollaire du thรฉorรจme de convergence dominรฉe, que $`y\mathrm{\Phi }(c,y,)`$ est de classe $`C^1`$$`c`$ fixรฉ) et, si $`uT_y\stackrel{~}{Y}`$, sa diffรฉrentielle est donnรฉe par
$$(u\psi )(c,y,\theta )=c_{\stackrel{~}{Y}}e^{cd(y,z)}(ud)(y,z)P_0(\stackrel{~}{f}(z),\theta )๐v_g(z)$$
la continuitรฉ en $`c`$ de cette quantitรฉ est รฉvidente en remarquant que $`|ud|u_g`$, que $`P_0`$ est strictement positif et que, pour $`y`$ et $`z`$ fixรฉs, $`ce^{cd(y,z)}`$ est dรฉcroissante en $`c`$ ; ceci permet dโappliquer une nouvelle fois le thรฉorรจme de convergence dominรฉe.
De mรชme, pour $`y`$ et $`\theta `$ fixรฉs, on peut appliquer le thรฉorรจme de dรฉrivation sous le signe somme afin de montrer la diffรฉrentiabilitรฉ en $`c`$$`y`$ et $`\theta `$ fixรฉ). En effet,
$$0d(y,z)e^{cd(y,z)}e^{c^{}d(y,z)}$$
pour tout $`c^{}<c`$. Ceci montre que
$$\frac{\psi }{c}(c,y,\theta )=_{\stackrel{~}{Y}}d(y,z)e^{cd(y,z)}P_0(\stackrel{~}{f}(z),\theta )๐v_g(z)$$
existe et, encore grรขce au thรฉorรจme de convergence dominรฉe, est continue en $`(c,y)`$. Ceci prouve le lemme ci-dessus. On remarque que le mรชme type dโargument que ceux utilisรฉs dans montrent que $`\frac{\psi }{c}`$ est de classe $`C^1`$ comme fonction de $`y`$ ร valeurs dans $`L^1(_F\stackrel{~}{X})`$.
De mรชme $`\psi `$ est de classe $`๐^{\mathrm{}}`$ en $`c`$ et chaque dรฉrivรฉe en $`c`$ est de classe $`C^1`$ en $`y`$ comme fonction de $`\stackrel{~}{Y}`$ ร valeurs dans $`L^1(_F\stackrel{~}{X})`$. Lโassertion du lemme concernant $`\mathrm{\Phi }`$ sโen dรฉduit.
###### Lemme 4.4
Lโapplication
$`\stackrel{~}{F}:]Ent(Y,g),+\mathrm{}[\times \stackrel{~}{Y}`$ $`\stackrel{~}{X}`$
$`(c,y)`$ $`\stackrel{~}{F}_c(y)`$
est de classe $`C^1`$.
Preuve .Il sโagit dโune simple application du thรฉorรจme des fonctions implicites (voir ). Rappelons la preuve de ce fait. Soit $`\{e_i(x)\}_{i=1,\mathrm{},n}`$ une base orthonormรฉe de $`T_x\stackrel{~}{X}`$ dรฉpendant de maniรจre $`๐^{\mathrm{}}`$ de $`x\stackrel{~}{X}`$. Dรฉfinissons les fonctions
$$G_i(x,c,y)=_{_F\stackrel{~}{X}}๐\overline{B}_0(x,\theta )(e_i(x))\mathrm{\Phi }(c,y,\theta )๐\theta $$
(on rappelle que $`\overline{B}_0(x,\theta )`$ dรฉsigne la fonction de Busemann de $`(\stackrel{~}{X},g_0)`$ normalisรฉe en $`O_0`$ et $`d\theta `$ la mesure canonique de $`_F\stackrel{~}{X}`$), et
$`G:\stackrel{~}{X}\times ]Ent(Y,g),+\mathrm{}[\times \stackrel{~}{Y}`$ $`๐^n`$
$`(x,c,y)`$ $`(G_1(x,c,y),\mathrm{},G_n(x,c,y)).`$
Alors, la fonction $`\stackrel{~}{F}`$ est dรฉfinie par lโรฉquation implicite
$$G(\stackrel{~}{F}_c(y),c,y)=0.$$
Le thรฉorรจme des fonctions implicites est alors facile ร vรฉrifier car la condition quโil requiert est exactement celle qui assure lโexistence du barycentre.
La fonction $`G`$ รฉtant $`C^1`$ en $`(x,c,y)`$ le lemme est prouvรฉ. En fait $`F`$ est, pour les mรชmes raisons que prรฉcรฉdemment, $`๐^{\mathrm{}}`$ en $`c`$.
Preuve du cas dโรฉgalitรฉ ii). du thรฉorรจme 4.2
La preuve commence comme dans le paragraphe 7 de . Pour fixer le facteur dโhomothรฉtie supposons que $`g`$ est normalisรฉe de sorte que
$$Ent(Y,g)=Ent(\stackrel{~}{X},g_0)=E_0.$$
On suppose donc que $`vol(\rho )=vol(Y,g)`$. Le travail porte sur lโรฉtude des formes quadratiques, dรฉjร introduites au paragraphe prรฉcรฉdent,
$`h_{y,c}(,)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_{0|(\stackrel{~}{F}_c(y),\theta )}()\right)^2\mathrm{\Phi }(y,c,\theta )๐\theta `$
$`k_{y,c}(,)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}D๐\overline{B}_{0|(\stackrel{~}{F}_c(y),\theta )}(,)\mathrm{\Phi }(y,c,\theta )๐\theta `$
et des endomorphismes symรฉtriques et dรฉfinis positifs correspondants, $`H_{y,c}`$ et $`K_{y,c}`$ (ici, $`c`$ joue le rรดle de $`Ent(Y,g)(1+\epsilon )=E_0(1+\epsilon )`$). La plus grande valeur propre de $`H_{y,c}`$ est notรฉe $`\mu _n^c(y)`$ et vรฉrifie,
$$0<\mu _n^c(y)<1,$$
en effet, lโendomorphisme symรฉtrique $`H_{y,c}`$ est de $`\overline{g}_0`$-trace รฉgale ร $`1`$ et est dรฉfini positif. On rappelle รฉgalement que $`trace(K_{y,c})=Ent(\stackrel{~}{X},g_0)=E_0`$ (ceci car $`\mathrm{\Phi }`$ est normalisรฉe).
1รจre รฉtape : convergence presque sรปre de $`H_{y,c}`$.
La preuve de lโinรฉgalitรฉ i) du thรฉorรจme 4.2 consiste (comme dans le paragraphe 3) ร montrer que,
$$y\stackrel{~}{Y},c>E_0,|Jac\stackrel{~}{F}_c(y)|\left(\frac{c}{E_0}\right)^n.$$
###### Lemme 4.5
Il existe une suite $`c_k`$ tendant vers $`E_0`$, telle que $`Jac\stackrel{~}{F}_{c_k}(y)\underset{k+\mathrm{}}{}1`$ presque sรปrement sur $`\stackrel{~}{Y}`$.
Preuve .Comme dans , lemme 7.3, posons $`f_c(y)=Jac\stackrel{~}{F}_c(y)1`$ et $`f_c^\pm =sup(0,\pm f_c)`$ ; la fonction $`f_c^+`$ tend uniformรฉment vers $`0`$ lorsque $`c`$ tend vers $`E_0`$ car,
$$y\stackrel{~}{Y}\mathrm{,\; 0}f_c^+(y)\left(\frac{c}{E_0}\right)^n1.$$
Par ailleurs, pour tout $`c>E_0`$,
$`vol(\rho )={\displaystyle _Y}\stackrel{~}{F}_{c}^{}{}_{}{}^{}(\omega _0)`$ $`={\displaystyle _Y}Jac\stackrel{~}{F}_c(y)dv_g`$
$`\left({\displaystyle \frac{c}{E_0}}\right)^nvol(Y,g){\displaystyle _Y}f_c^{}๐v_g`$
lโhypothรจses $`vol(\rho )=vol(Y,g)`$ implique que $`f_c^{}`$ tend vers $`0`$ dans $`L^1(Y,g)`$ lorsque $`c`$ tend vers $`E_0`$, dโoรน lโexistence dโune sous-suite $`c_k`$ telle que $`f_{c_k}^{}`$ tende vers zรฉro presque sรปrement.
Lorsque $`(\stackrel{~}{Y},\stackrel{~}{g})=(\stackrel{~}{X},\overline{g}_0)`$ et la mesure $`\mu _0`$ est la mesure canonique du bord de Furstenberg $`_F\stackrel{~}{X}`$, alors lโendomorphisme $`H_{y,\epsilon }`$ prend une forme particuliรจre ; en effet, pour tout $`x\stackrel{~}{X}`$ et pour $`\epsilon =0`$
$$H_{x\mathrm{,0}}=\left(\begin{array}{cccc}\frac{1}{pn_1}I_1& 0& 0& 0\\ 0& \frac{1}{pn_2}I_2& 0& 0\\ 0& 0& \mathrm{}& 0\\ 0& 0& 0& \frac{1}{pn_p}I_p\end{array}\right),$$
$`x=(x_1,\mathrm{},x_p)`$ et $`I_k`$ dรฉsigne lโidentitรฉ de $`T_{x_k}\stackrel{~}{X}_k`$. Dรฉsormais nous noterons $`H_0`$ lโendomorphisme $`H_{x\mathrm{,0}}`$. De mรชme, les termes $`K_i`$ (voir le paragraphe prรฉcรฉdent) qui se calculent en fonctions de $`H_i=\frac{1}{pn_i}I_i`$ et valent $`K_i=\frac{E_i}{\sqrt{p}n_i}I_i`$. Nous noterons $`K_0`$ lโendomorphisme $`K_{x\mathrm{,0}}`$ correspondant.
ร partir de maintenant nous considรจrerons une suite $`c_k\underset{k+\mathrm{}}{}E_0`$ telle que $`Jac\stackrel{~}{F}_{c_k}(y)\underset{k+\mathrm{}}{}1`$ presque sรปrement en $`y\stackrel{~}{Y}`$.
###### Lemme 4.6
Pour presque tout $`y\stackrel{~}{Y}`$, $`lim_{k+\mathrm{}}H_{y,c_k}=H_0`$.
Preuve .Pour tout $`y\stackrel{~}{Y}`$ et pour tout $`x>E_0`$
$$|Jac\stackrel{~}{F}_c(y)|\left(\frac{c}{\sqrt{n}}\right)^n\frac{(\text{dรฉt}H_{y,c})^{1/2}}{\text{dรฉt}(K_{y,c})}\left(\frac{c}{E_0}\right)^n.$$
Soit $`yY`$ tel que $`|Jac\stackrel{~}{F}_{c_k}(y)|\underset{k+\mathrm{}}{}1`$, la quantitรฉ $`\frac{(\text{dรฉt}H_{y,c_k})^{1/2}}{\text{dรฉt}(K_{y,c_k})}`$ tend vers sa valeur maximale, ร savoir $`\left(\frac{\sqrt{n}}{E_0}\right)^n`$. On rappelle que $`\underset{i=1}{\overset{p}{}}a_i^{n_i}=1`$ (voir le paragraphe 1).
Par une preuve en tout point analogue ร celle donnรฉe dans lโappendice B, proposition B5 de , nous montrons lโexistence dโune constante $`A>0`$ telle que
$$\frac{(\text{dรฉt}H_{y,c})^{1/2}}{\text{dรฉt}(K_{y,c})}\left(\frac{\sqrt{n}}{E_0}\right)^n\left(1AH_{y,c}H_0_{\overline{g}_0}^2\right)$$
de sorte que
$$H_{y,c}H_0_{\overline{g}_0}^2\frac{1}{A}\left(1\left(\frac{E_0}{c}\right)^n|Jac\stackrel{~}{F}_c(y)|\right)$$
et, si $`|Jac\stackrel{~}{F}_{c_k}(y)|\underset{k+\mathrm{}}{}1`$, alors
$$H_{y,c_k}\underset{k+\mathrm{}}{}H_0.$$
2รจme รฉtape : convergence uniforme de $`H_{y,c_k}`$ vers $`H_0`$.
Nous reprenons les รฉtapes de la preuve du cas dโรฉgalitรฉ de , paragraphe 7.
Soit $`c_k\underset{k+\mathrm{}}{}E_0`$ une sous-suite telle que $`Jac\stackrel{~}{F}_{c_k}1`$ presque sรปrement et $`H_{y,c_k}`$ tende presque sรปrement vers $`H_0`$. Pour simplifier les notations nous utiliserons lโindice $`k`$ en lieu et place de lโindice $`c_k`$.
###### Lemme 4.7
Soient $`y`$ et $`y^{}`$ deux points de $`\stackrel{~}{Y}`$ tels que $`\mu _n^k1\frac{1}{n}`$ en tout point dโune $`g`$-gรฉodรฉsique minimisante $`\alpha `$ qui joint $`y`$ ร $`y^{}`$, alors
$$d_{\overline{g}_0}(\stackrel{~}{F}_k(y),\stackrel{~}{F}_k(y^{}))K_1d_g(y,y^{}).$$
On rappelle que $`\mu _n^k(y)=\mu _n^{c_k}(y)`$ est la plus grande valeur propre de $`H_{y,c_k}`$.
Preuve .On tire, comme dans le paragraphe 3, de lโรฉquation implicite qui dรฉfinit $`\stackrel{~}{F}_k`$, pour tous $`uT_y\stackrel{~}{Y}`$ et $`vT_{\stackrel{~}{F}_k(y)}\stackrel{~}{X}`$,
$`\overline{g}_0(K_{y,k}D_y\stackrel{~}{F}_k(u),v)`$ $`={\displaystyle _{_F\stackrel{~}{X}}}d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(v)d\mathrm{\Phi }_k{}_{|(y,\theta )}{}^{}(u)d\theta `$
$`=2{\displaystyle _{_F\stackrel{~}{X}}}d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(v)\sqrt{\mathrm{\Phi }_k}(y,\theta )d\sqrt{\mathrm{\Phi }_k}_{|(y,\theta )}(u)d\theta `$
$`2\overline{g}_0(H_{y,k}(v),v)^{1/2}\left({\displaystyle _{_F\stackrel{~}{X}}}\left(d\sqrt{\mathrm{\Phi }_k}{}_{|(y,\theta )}{}^{}(u)\right)^2d\theta \right)^{1/2}.`$
Un calcul immรฉdiat montre que
$$\left(_{_F\stackrel{~}{X}}\left(d\sqrt{\mathrm{\Phi }_k}_{|(y,\theta )}(u)\right)^2d\theta \right)^{1/2}\frac{c_k}{2}g(u,u)^{1/2}.()$$
Si $`u`$ et $`v`$ sont de norme 1, dans leur espace respectif, alors
$`\overline{g}_0(K_{y,k}D_y\stackrel{~}{F}_k(u),v)`$ $`c_k\overline{g}_0(H_{y,k}(v),v)^{1/2}`$
$`c_k\sqrt{\mu _n^k(y)}.`$
Maintenant, si $`\stackrel{~}{X}`$ est un produit dโespaces symรฉtriques de rang 1, de courbure comprise entre $`1`$ et $`4`$, il est facile de vรฉrifier (voir , appendice B) que, au sens des formes quadratiques, pour tout $`i=\mathrm{1,2},\mathrm{},p`$,
$$K_iI_iH_i(1\mu _n^k(y))I_i.$$
On rappelle que $`H_i`$ (resp. $`K_i`$) dรฉsigne la restriction de $`H_{y,k}`$ (resp. $`K_{y,k})`$ ร $`\stackrel{~}{X}_i`$. En prenant $`v=\frac{D_y\stackrel{~}{F}_k(u)}{D_y\stackrel{~}{F}_k(u)_{\overline{g}_0}}`$ si $`D_y\stackrel{~}{F}_k(u)0`$, il vient
$$D_y\stackrel{~}{F}_k(u)_{\overline{g}_0}c_k\frac{\sqrt{\mu }_n^k(y)}{1\mu _n^k(y)}()$$
(si $`D_y\stackrel{~}{F}_k(u)=0`$, lโinรฉgalitรฉ est trivialement vraie). Soit $`\alpha `$ la $`g`$-gรฉodรฉsique de $`y`$ ร $`y^{}`$ le long de laquelle $`\mu _n^k(\alpha (t))1\frac{1}{n}`$, on a, pour tout $`uT_{\alpha (t)}\stackrel{~}{Y}`$, de norme 1
$$D_{\alpha (t)}\stackrel{~}{F}_k(u)_{\overline{g}_0}2nE_0=K_1$$
(si $`k`$ est assez grand pour que $`c_k2E_0`$). Par le thรฉorรจme des accroissements finis
$$d_{\overline{g}_0}(\stackrel{~}{F}_k(y),\stackrel{~}{F}_k(y^{}))K_1d_g(y,y^{}).$$
###### Lemme 4.8
Avec les mรชmes notations que prรฉcรฉdemment, si $`P`$ dรฉsigne le transport parallรจle de $`\stackrel{~}{F}_k(y)`$ ร $`\stackrel{~}{F}_k(y^{})`$ le long de la $`\overline{g}_0`$-gรฉodรฉsique minimisante qui les joint, on a
$$h_{y,k}Ph_{y^{},k}_{\overline{g}_0}K_2\left[d_g(y,y^{})+d_{\overline{g}_0}(\stackrel{~}{F}_k(y),\stackrel{~}{F}_k(y^{}))\right].$$
Preuve .Nous dรฉsignons par $`\beta (t)`$ lโunique $`\overline{g}_0`$-gรฉodรฉsique, qui est minimisante, allant de $`\stackrel{~}{F}_k(y)`$ ร $`\stackrel{~}{F}_k(y^{})`$ et par $`Z`$ un champ de vecteurs parallรจle, le long de $`\beta `$, de norme 1. Pour simplifier, posons $`Z_1=Z(\stackrel{~}{F}_k(y))`$ et $`Z_2=Z(\stackrel{~}{F}_k(y^{}))`$. Alors
$`h_{y^{},k}(Z_2,Z_2)h_{y,k}`$ $`(Z_1,Z_1)`$
$`={\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y^{}),\theta )}{}^{}(Z_2)\right)^2\mathrm{\Phi }_k(y^{},\theta )d\theta `$
$`{\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(Z_1)\right)^2\mathrm{\Phi }_k(y,\theta )d\theta `$
$`={\displaystyle _{_F\stackrel{~}{X}}}[\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y^{}),\theta )}{}^{}(Z_2)\right)^2\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(Z_1)\right)^2]\mathrm{\Phi }_k(y^{},\theta )d\theta `$
$`+{\displaystyle _{_F\stackrel{~}{X}}}\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(Z_1)\right)^2(\mathrm{\Phi }_k(y^{},\theta )\mathrm{\Phi }_k(y,\theta ))d\theta .`$
Des formules explicites de $`Dd\overline{B}_0`$ et du fait que $`d\overline{B}_0{}_{|(x,\theta )}{}^{}()_{\overline{g}_0}1`$, nous tirons lโinรฉgalitรฉ
$$|\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y^{}),\theta )}{}^{}(Z_2)\right)^2\left(d\overline{B}_0{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(Z_1)\right)^2|K_2^{}d_{\overline{g}_0}(\stackrel{~}{F}_k(y^{}),\stackrel{~}{F}_k(y)).$$
De mรชme, comme $`\mathrm{\Phi }_k(y,)`$ est de norme 1 dans $`L^1(_F\stackrel{~}{X},d\theta )`$ et en utilisant lโinรฉquation $`()`$
$`{\displaystyle _{_F\stackrel{~}{X}}}(d\overline{B}_0`$ $`{}_{|(\stackrel{~}{F}_k(y),\theta )}{}^{}(Z_1)\left)^2\right[\left(\sqrt{\mathrm{\Phi }_k(y^{},\theta )}\right)^2\left(\sqrt{\mathrm{\Phi }_k(y,\theta )}\right)^2]d\theta `$
$`\left(\sqrt{\mathrm{\Phi }_k}(y,)\sqrt{\mathrm{\Phi }_k}(y^{},)_{L^2(_F\stackrel{~}{X})}\right)\left(\sqrt{\mathrm{\Phi }_k}(y,)+\sqrt{\mathrm{\Phi }_k}(y^{},)_{L^2(_F\stackrel{~}{X})}\right)`$
$`c_kd_g(y,y^{}).`$
Le lemme dรฉcoule de lโaddition de ces inรฉgalitรฉs.
###### Lemme 4.9
La suite $`H_{k,y}`$ converge uniformรฉment par rapport ร $`y\stackrel{~}{Y}`$ vers $`H_0`$ lorsque $`k`$ tend vers $`+\mathrm{}`$.
Preuve .Le comportement de $`H_k`$ vis-ร -vis de lโaction de $`\mathrm{\Gamma }`$ sur $`\stackrel{~}{Y}`$ montre quโil suffit de prouver la convergence uniforme sur un domaine fondamental $`D\stackrel{~}{Y}`$. Le thรฉorรจme dโEgoroff (, p. 77) et le lemme 4.6 attestent que, pour tout $`\eta >0`$, il existe un ensemble mesurable $`K`$ tel que
* $`vol_g(DK)<\eta `$ ;
* sur $`K`$, la suite $`yH_{k,y}`$ converge uniformรฉment vers $`H_0`$.
Fixons $`\epsilon >0`$ petit, on peut choisir $`\eta `$ tel que $`DK`$ ne contienne aucune $`g`$-boule de rayon $`\epsilon `$, car, en effet, le volume dโune telle boule sur $`\stackrel{~}{Y}`$ est minorรฉ (la mรฉtrique de $`\stackrel{~}{Y}`$ est pรฉriodique). On choisit aussi $`N\text{N}`$ de sorte que
* pour tout $`kN`$, $`E_0<c_k<E_0+\epsilon `$ ;
* pour tout $`kN`$ et pour tout $`yK`$, $`H_{y,k}H_0_{\overline{g}_0}<\epsilon `$.
Par ailleurs, si $`yK`$, $`d_g(y,K)<\epsilon `$. Rappelons que les valeurs propres de $`H_0`$ sont les nombres $`\frac{1}{pn_i}`$, $`i=\mathrm{1,2},\mathrm{},p`$. Posons $`K_3=K_2(K_1+1)+1`$ et supposons $`\epsilon `$ assez petit pour que $`K_3\epsilon 1\underset{i}{sup}\left({\displaystyle \frac{1}{pn_i}}\right){\displaystyle \frac{1}{n}}`$. Nous allons montrer que si $`kN`$, alors
$$yD,H_{y,k}H_0_{\overline{g}_0}<K_3\epsilon .$$
Si ce nโest pas vrai, il existe $`y^{}D`$ tel que
$$H_{y^{},k}H_0_{\overline{g}_0}K_3\epsilon ,$$
soit alors $`yK`$ tel que $`d_g(y^{},y)<\epsilon `$. Par continuitรฉ de lโapplication $`yH_{y,k}`$, il existe un premier point $`y^{\prime \prime }`$ sur le segment gรฉodรฉsique $`[y,y^{}]`$ tel que $`H_{y^{\prime \prime },k}H_0=K_3\epsilon `$. Le choix de $`K_3`$ montre que, sur le segment gรฉodรฉsique $`[y,y^{\prime \prime }]`$,
$$\mu _n^k\underset{i}{sup}\left(\frac{1}{pn_i}\right)+K_3\epsilon 1\frac{1}{n}.$$
Dโaprรจs les lemmes 4.7 et 4.8 ceci conduit ร
$$h_{y,k}Ph_{y^{\prime \prime },k}_{\overline{g}_0}K_2(K_1+1)\epsilon $$
et comme $`H_{y,k}H_0_{\overline{g}_0}<\epsilon `$ ceci conduit ร
$$H_{y^{\prime \prime },k}H_0_{\overline{g}_0}<\left(K_2(K_1+1)+1\right)\epsilon =K_3\epsilon $$
qui est une contradiction.
Remarquons que la convergence uniforme de $`H_{y,k}`$ vers $`H_0`$ implique la convergence uniforme de $`K_{y,k}`$ vers $`K_0`$.
3รจme รฉtape : convergence uniforme dโune sous-suite de $`\stackrel{~}{F}_k`$.
###### Lemme 4.10
Il existe une sous-suite de la suite $`\stackrel{~}{F}_k`$ qui converge uniformรฉment vers une application $`\stackrel{~}{F}:\stackrel{~}{Y}\stackrel{~}{X}`$ continue et รฉquivariante.
Preuve .Pour $`\epsilon >0`$ donnรฉ, il existe $`M\text{N}`$ tel que si $`kM`$
$$y\stackrel{~}{Y},H_{y,k}H_0_{\overline{g}_0}<\epsilon .$$
Dโoรน
$$H_{y,k}H_0+\epsilon I$$
et par une remarque prรฉcรฉdente
$$K_{y,k}K_0\epsilon I.$$
Ces deux inรฉgalitรฉs รฉtant ร comprendre au sens des formes quadratiques. On dรฉduit alors, avec $`()`$, quโil existe un nombre rรฉel $`C>0`$ tel que, pour tout $`y\stackrel{~}{Y}`$ et $`uT_y\stackrel{~}{Y}`$,
$$D_y\stackrel{~}{F}_k(u)_{\overline{g}_0}C$$
(si $`\epsilon `$ est assez petit).
La suite dโapplication $`\stackrel{~}{F}_k:\stackrel{~}{Y}\stackrel{~}{X}`$ est donc รฉquicontinue.
Supposons quโil existe $`y_0`$ tel que $`\stackrel{~}{F}_k(y_0)`$ ne reste dans aucun compact. Quitte ร extraire une sous-suite, on peut supposer que $`\stackrel{~}{F}_k(y_0)\underset{k+\mathrm{}}{}\theta \stackrel{~}{X}`$ (le bord gรฉomรฉtrique de $`\stackrel{~}{X}`$). Pour tout $`y\stackrel{~}{Y}`$, alors
$$d_{\overline{g}_0}(\stackrel{~}{F}_k(y),\stackrel{~}{F}_k(y_0))Cd_g(y,y_0)$$
de sorte que $`\stackrel{~}{F}_k(y)\underset{k+\mathrm{}}{}\theta `$ par dรฉfinition du bord gรฉomรฉtrique de $`\stackrel{~}{X}`$. Lโรฉquivariance de $`\stackrel{~}{F}_k`$ donne
$$\stackrel{~}{F}_k(\gamma y_0)=\rho (\gamma )\stackrel{~}{F}_k(y_0)$$
et donc en passant ร la limite en $`k`$
$$\theta =\rho (\gamma )\theta $$
cโest-ร -dire, la reprรฉsentation $`\rho `$ fixe un point de $`\stackrel{~}{X}`$.
###### Lemme 4.11
Si $`\rho `$ fixe un point $`\theta _0`$ de $`\stackrel{~}{X}`$, alors $`vol(\rho )=0`$.
Preuve .Soit $`\overline{B}_0(,\theta _0)`$ la fonction de Busemann dรฉfinie par le point $`\theta _0\stackrel{~}{X}`$. Supposons dโabord que $`\theta _0`$ est dans le bord de Fรผrstenberg. Posons
$$Z(x)=\overline{B}_0(x,\theta _0)$$
alors le champ de vecteurs $`Z`$ est invariant par $`\rho `$. En effet, lโรฉgalitรฉ
$$\overline{B}_0(\alpha (x),\theta _0)=\overline{B}_0(x,\alpha ^1(\theta _0))+\overline{B}_0(\alpha (O_0),\theta _0)$$
pour $`\alpha Isom(\stackrel{~}{X})`$, conduit ร
$$\overline{B}_0(\rho (\gamma )(x),\theta _0)=\overline{B}_0(x,\theta _0)+\overline{B}_0(\rho (\gamma )(O_0),\theta _0)$$
pour tout $`\gamma \mathrm{\Gamma }`$ ; ce qui donne en diffรฉrenciant
$$Z\left(\rho (\gamma )(x)\right)=\rho (\gamma )(Z(x)).$$
Par ailleurs, pour tout $`x\stackrel{~}{X}`$
$$div(Z)(x)=\mathrm{\Delta }\left(\overline{B}_0(,\theta )\right)=E_0.$$
Donc la forme diffรฉrentielle $`\omega =div(Z)\omega _0=E_0\omega _0`$ est invariante par $`\rho (\gamma )`$, pour tout $`\gamma \mathrm{\Gamma }`$. En consรฉquence, pour $`c>E_0`$, $`\stackrel{~}{F}_c^{}(\omega )`$ est invariante par $`\gamma `$, pour tout $`\gamma \mathrm{\Gamma }`$. La dรฉfinition de la divergence conduit ร lโรฉgalitรฉ
$$div(Z)\omega _0=d\left(i(Z)\omega _0\right)$$
$`i(Z)\omega _0`$ dรฉsigne le produit intรฉrieur de $`\omega _0`$ par le champ de vecteurs $`Z`$. Dโoรน
$`\stackrel{~}{F}_c^{}(\omega )`$ $`=\stackrel{~}{F}_c^{}\left(d(i(Z)\omega _0)\right)`$
$`=d\left(\stackrel{~}{F}_c^{}(i(Z)\omega _0)\right)`$
et
$$vol(\rho )=_Y\stackrel{~}{F}_c^{}(\omega _0)=\frac{1}{E_0}_Y\stackrel{~}{F}_c^{}(\omega )=0.$$
Si $`\theta _0`$ nโest pas dans le bord de Fรผrstenberg la mรชme preuve peut รชtre faite car
$$div(Z)(x)=\mathrm{\Delta }\left(\overline{B}_0(,\theta )\right)0.$$
Puisque nous sommes dans le cas dโรฉgalitรฉ, $`vol(\rho )0`$, et la suite $`\stackrel{~}{F}_k(y_0)`$ reste donc dans un compact de $`\stackrel{~}{X}`$. On peut alors appliquer le thรฉorรจme dโAscoli pour dรฉduire quโil existe une sous-suite, notรฉe encore $`\stackrel{~}{F}_k`$, qui converge uniformรฉment sur $`D\stackrel{~}{Y}`$ vers une application continue $`F:D\stackrel{~}{X}`$. Lโรฉquivariance de $`\stackrel{~}{F}_k`$, pour tout $`k`$, montre que $`\stackrel{~}{F}_k`$ converge uniformรฉment sur $`\stackrel{~}{Y}`$ et que la limite $`F`$ est รฉgalement รฉquivariante.
4รจme รฉtape : $`\stackrel{~}{F}`$ est une isomรฉtrie.
###### Lemme 4.12
Lโapplication $`\stackrel{~}{F}:(\stackrel{~}{Y},\stackrel{~}{g})(\stackrel{~}{X},g_0)`$ contracte les distances, cโest-ร -dire, pour tout $`y`$, $`y^{}`$ dans $`Y^{}`$
$$d_{g_0}(\stackrel{~}{F}(y),\stackrel{~}{F}(y^{}))d_g(y,y^{})$$
et $`D\stackrel{~}{F}(y)`$ est une isomรฉtrie entre $`(T_y\stackrel{~}{Y},g)`$ et $`(T_{\stackrel{~}{F}(y)}\stackrel{~}{X},g_0)`$ pour presque tout $`y\stackrel{~}{Y}`$.
Preuve .Pour $`\epsilon >0`$ donnรฉ, on peut choisir $`k`$ assez grand pour que, pour tout $`y\stackrel{~}{Y}`$,
$`H_{y,k}`$ $`H_0+\epsilon I`$
$`K_{y,k}`$ $`K_0\epsilon I.`$
Alors, lโinรฉgalitรฉ 2.5 nous conduit ร lโestimation suivante, pour $`uT_y\stackrel{~}{Y}`$ et $`vT_{\stackrel{~}{F}_k(y)}\stackrel{~}{X}`$ :
$$\overline{g}_0(K_{y,c}(D\stackrel{~}{F}_k(y)(u)),v)(1+\epsilon )E_0\left(\overline{g}_0(H_{y,c}(v),v)\right)^{1/2}\left(h_{y,k}^{}(u,u)\right)^{1/2}.$$
On rappelle que la dรฉfinition des fonctions $`\stackrel{~}{F}_k`$ est indรฉpendante des coefficients choisis pour dรฉfinir la mรฉtrique de rรฉfรฉrence, cโest-ร -dire quโelle donne la mรชme fonction quโon utilise $`\overline{g}_0=\underset{i=1}{\overset{p}{}}g_0^i`$ ou bien $`g_0=\underset{i=1}{\overset{p}{}}a_i^2g_0^i`$. Nous avons choisi dโutiliser $`\overline{g}_0`$ pour dรฉfinir le barycentre sur $`\stackrel{~}{X}`$, en consรฉquence les matrices $`H_{y,k}`$, $`H_0`$, $`K_{y,k}`$ et $`K_0`$ sont dรฉfinies รฉgalement grรขce ร la mรฉtrique $`\overline{g}_0`$.
On rappelle รฉgalement que $`G_{i,k}`$, $`i=\mathrm{1,2},\mathrm{},p`$, dรฉsigne la $`i`$-iรจme composante de $`\stackrel{~}{F}_k`$ dans la dรฉcomposition $`\stackrel{~}{X}=\stackrel{~}{X}_1\times \mathrm{}\times \stackrel{~}{X}_p`$ et que $`H_i`$ (resp. $`K_i`$) dรฉsigne la restriction de $`H_{y,k}`$ (resp. $`K_{y,k}`$) ร $`T_{\stackrel{~}{F}_{i,k}(y)}\stackrel{~}{X}_i`$ (ici on omet volontairement les indices $`y`$ et $`k`$ dans $`H_i`$ et $`K_i`$ afin dโallรฉger les notations). Si $`v=(v_1,\mathrm{},v_n)`$ est tangent ร $`\stackrel{~}{X}_i`$, cโest-ร -dire si $`v_j=0`$ pour tout $`ji`$, alors, grรขce ร la forme diagonale par blocs de $`K_{y,c}`$ nous obtenons
$$\overline{g}_0(K_i(D\stackrel{~}{F}_{i,k}(y)(u)),v_i)(1+\epsilon )E_0\left(\overline{g}_0(H_iv_i,v_i)\right)^{1/2}\left(h_{y,k}^{}(u,u)\right)^{1/2}$$
(on identifie, par abus de langage $`v`$ ร sa composante $`v_i`$).
En utilisant lโinรฉgalitรฉ prรฉcรฉdente sur $`H_{y,k}`$,
$$\overline{g}_0^i(K_i(D\stackrel{~}{F}_{i,k}(y)(u)),v_i)(1+\epsilon )E_0\left(\frac{1}{pn_i}+\epsilon \right)^{1/2}v_i_{g_0^i}\left(h_{y,k}^{}(u,u)\right)^{1/2}.$$
En prenant le supremum en $`v_i`$ de norme 1, nous obtenons,
$$K_{y,c}\left(D\stackrel{~}{F}_{i,k}(y)(u)\right)_{\overline{g}_0}=K_i\left(D\stackrel{~}{F}_{i,k}(y)(u)\right)_{g_0^i}\frac{E_0}{\sqrt{p}\sqrt{n_i}}\left(h_{y,k}^{}(u,u)\right)^{1/2}(1+o(\epsilon )).$$
Les inรฉgalitรฉs prรฉcรฉdentes donnent encore,
$$\frac{E_i}{\sqrt{pn_i}}(1+o(\epsilon ))D\stackrel{~}{F}_{i,k}(y)(u)_{\overline{g}_0}\frac{E_0}{\sqrt{p}\sqrt{n_i}}\left(h_{y,k}^{}(u,u)\right)^{1/2}(1+o(\epsilon )),$$
or les coefficients $`a_i`$ apparaissant dans la dรฉfinition de la mรฉtrique $`g_0`$ valent :
$$a_i=\frac{E_i\sqrt{n}}{\sqrt{n_i}E_0}$$
dโoรน, pour tout $`uT_y\stackrel{~}{Y}`$
$$a_iD\stackrel{~}{F}_{i,k}(y)(u)_{g_0^i}\sqrt{n}\left(h_{y,k}^{}(u,u)\right)^{1/2}(1+o(\epsilon ))$$
et, pour tout $`uT_y\stackrel{~}{Y}`$
$$\stackrel{~}{F}_k^{}g_0(u,u)=D\stackrel{~}{F}_k(y)(u)_{g_0}^2=\underset{i=1}{\overset{p}{}}a_i^2D\stackrel{~}{F}_{i,k}(y)(u)_{g_0^i}^2nh_{y,k}^{}(u,u)(1+o(\epsilon )).$$
On peut alors calculer la trace du tenseur symรฉtrique $`\stackrel{~}{F}_k^{}g_0`$ par rapport ร la mรฉtrique $`g`$ sur $`\stackrel{~}{Y}`$ en $`y\stackrel{~}{Y}`$.
$$trace_g(\stackrel{~}{F}_k^{}g_0)n(1+o(\epsilon )).$$
En effet, on rappelle que $`trace_g(h_{y,k}^{})=1`$ (voir le paragraphe 2).
Par ailleurs le dรฉterminant de $`\stackrel{~}{F}_k^{}g_0`$ relativement ร $`g`$, cโest-ร -dire $`|Jac\stackrel{~}{F}_k|^2`$, tend presque sรปrement vers $`1`$ sur $`\stackrel{~}{Y}`$. Alors si $`A_{k,y}`$ dรฉsigne la matrice de $`\stackrel{~}{F}_k^{}g_0`$ dans une base $`g`$-orthonormรฉe, nous avons, pour $`k`$ assez grand,
$$1\epsilon (\text{dรฉt}A_{k,y})^{1/n}\frac{1}{n}trace(A_{k,y})1+\epsilon $$
ce qui implique que
$$A_{k,y}(\text{dรฉt}A_{k,y})^{1/n}Id=o(\epsilon ).$$
En conclusion, $`D\stackrel{~}{F}_k`$ converge presque sรปrement sur $`\stackrel{~}{Y}`$ vers une isomรฉtrie.
Alors, lโapplication $`\stackrel{~}{F}`$ est limite uniforme dโune suite dโapplications lipschitzienne $`\stackrel{~}{F}_k`$ dont les diffรฉrentielles $`D\stackrel{~}{F}_k`$ sont uniformรฉment bornรฉes et convergent presque sรปrement vers une isomรฉtrie ; le lemme 7.8 de montre que, dans ce cas, lโapplication $`F`$ est $`1`$-lipschitzienne. Nous ne reproduisons pas la preuve de ce fait.
Lโapplication $`\stackrel{~}{F}`$ est presque partout diffรฉrentiable par le thรฉorรจme de Rademacher et, comme elle est $`1`$-lipschitzienne, on a, pour presque tout $`y\stackrel{~}{Y}`$
$$|Jac\stackrel{~}{F}(y)|1.$$
Par ailleurs,
$$vol(\rho )=_Y\stackrel{~}{F}^{}(\omega _0)=_Y\left(Jac\stackrel{~}{F}(y)\right)๐v_g=vol(Y,g).$$
Dโoรน, pour presque tout $`y\stackrel{~}{Y}`$, $`Jac\stackrel{~}{F}(y)=1`$.
Enfin, pour presque tout $`y\stackrel{~}{Y}`$, pour tout $`uT_y\stackrel{~}{Y}`$, le caractรจre $`1`$-lipschitzien de $`\stackrel{~}{F}`$ implique que
$$D_y\stackrel{~}{F}(u)_{g_0}u_g.$$
Ceci, combinรฉ au fait que pour presque tout $`y\stackrel{~}{Y}`$, $`Jac\stackrel{~}{F}(y)=1`$, montre que la diffรฉrentielle de $`\stackrel{~}{F}`$, $`D_y\stackrel{~}{F}`$, est presque partout sur $`\stackrel{~}{Y}`$ une isomรฉtrie (entre $`T_y\stackrel{~}{Y}`$ et $`T_{\stackrel{~}{F}(y)}\stackrel{~}{X}`$).
Le lemme est prouvรฉ.
###### Lemme 4.13
Lโapplication $`\stackrel{~}{F}`$ minimise la fonctionnelle $`E_p(h)=\frac{1}{vol(Y)}_YDh_{g,g_0}^p`$ parmi toutes les applications $`h`$ de $`\stackrel{~}{Y}`$ dans $`\stackrel{~}{X}`$, $`\rho `$-รฉquivariantes et lipschitziennes, pour tout $`pn`$. Ici $`Dh_{g,g_0}^p`$ est calculรฉe ร lโaide de la mรฉtrique $`g`$ sur $`\stackrel{~}{Y}`$ et $`g_0`$ sur $`\stackrel{~}{X}`$.
Preuve .Notons que, par lโรฉquivariance de $`h`$, lโintรฉgrand dans lโexpression de $`E_p(h)`$ est invariant par $`\mathrm{\Gamma }`$ et est donc une quantitรฉ dรฉfinie sur $`Y`$. Si $`\{e_i\}`$ est une base $`g`$-orthonormรฉe en $`y\stackrel{~}{Y}`$.
$$Dh(y)_{g,g_0}=\left(\frac{1}{n}\underset{i=1}{\overset{n}{}}Dh(y)(e_i)_{g_0}^2\right)^{1/2}.$$
Cette quantitรฉ est dรฉfinie pour presque tout $`y\stackrel{~}{Y}`$. On a donc, pour presque tout $`y\stackrel{~}{Y}`$,
$$|Jach(y)|^{p/n}Dh(y)_{g,g_0}^p$$
pour tout $`p0`$. Maintenant si $`pn`$
$`1=\left({\displaystyle \frac{vol(\rho )}{vol(Y)}}\right)^{p/n}=\left({\displaystyle \frac{1}{vol(Y)}}{\displaystyle _Y}Jach(y)dv_g(y)\right)^{p/n}`$ $`{\displaystyle \frac{1}{vol(Y)}}{\displaystyle |Jach(y)|^{p/n}๐v_g(y)}`$
$`E_p(h).`$
Si $`h`$ est remplacรฉe par $`\stackrel{~}{F}`$, en utilisant le fait que $`D\stackrel{~}{F}(y)`$ est une isomรฉtrie pour presque tout $`y\stackrel{~}{Y}`$, il vient
$$1=E_p(\stackrel{~}{F})E_p(h).$$
###### Corollaire 4.14
Lโapplication $`\stackrel{~}{F}`$ est de classe $`๐^{\mathrm{}}`$.
Preuve .En fait, nous prouvons que $`\stackrel{~}{F}`$ est harmonique, la rรฉgularitรฉ sโen dรฉduit.
De maniรจre heuristique nous pouvons dire que lโรฉquation dโEuler associรฉe ร la fonctionnelle $`E_p`$, $`pn`$ sโรฉcrit
$$div\left(D\stackrel{~}{F}_{g,g_0}^{p2}D\stackrel{~}{F}\right)=0$$
oรน la divergence est ร comprendre comme celle dโune 1-forme sur $`\stackrel{~}{Y}`$ ร valeurs dans $`T\stackrel{~}{X}`$ (voir , page 6). Mais $`\stackrel{~}{F}`$ a une diffรฉrentielle qui est presque partout une isomรฉtrie, de sorte que $`D\stackrel{~}{F}_{g,g_0}=1`$ presque partout sur $`\stackrel{~}{Y}`$, et lโรฉquation devient
$$div(D\stackrel{~}{F})=0$$
cโest-ร -dire $`\stackrel{~}{F}`$ est harmonique.
Plus prรฉcisรฉment, $`D\stackrel{~}{F}`$ est interprรฉtรฉe comme une 1-forme sur $`\stackrel{~}{Y}`$ ร valeurs dans le fibrรฉ $`\stackrel{~}{F}^1(T\stackrel{~}{X})`$, cโest-ร -dire un รฉlรฉment de $`C_\rho ^0\left(T^{}(\stackrel{~}{Y})\stackrel{~}{F}^1(T\stackrel{~}{X})\right)`$, qui est de plus $`\rho `$-รฉquivariante (voir , page 8) ; soit alors $`Z`$ un champ de vecteurs $`๐^{\mathrm{}}`$ le long de $`\stackrel{~}{F}`$, qui satisfait รฉgalement la relation de $`\rho `$-รฉquivariance adรฉquate, cโest-ร -dire qui est un รฉlรฉment de $`๐_\rho ^{\mathrm{}}(\stackrel{~}{Y},\stackrel{~}{F}^1(T\stackrel{~}{X}))`$; alors il existe une variation ร un paramรจtre de $`\stackrel{~}{F}`$, notรฉe $`\stackrel{~}{F}_t`$, $`\rho `$-รฉquivariante, telle que
$$y\stackrel{~}{Y},\frac{d}{dt}\stackrel{~}{F}_t(y)_{|t=0}=Z(y)$$
(voir , page 397).
Comme $`\stackrel{~}{F}`$ minimise $`E_p`$, pour $`pn`$, on a
$$\frac{d}{dt}_{|t=0}E_p(\stackrel{~}{F}_t)=0$$
cโest-ร -dire,
$$\frac{d}{dt}_{|t=0}\frac{1}{vol(Y)}_YD\stackrel{~}{F}_t(y)_{g,g_0}^p๐v_g(y)=0$$
mais
$`{\displaystyle \frac{d}{dt}}({\displaystyle \frac{1}{vol(Y)}}{\displaystyle _Y}D\stackrel{~}{F}_t(y)_{g,g_0}^p`$ $`dv_g(y))`$
$`={\displaystyle \frac{1}{vol(Y)}}{\displaystyle _Y}{\displaystyle \frac{d}{dt}}\left(D\stackrel{~}{F}_t(y)_{g,g_0}^p\right)๐v_g(y)`$
$`={\displaystyle \frac{1}{vol(Y)}}{\displaystyle _Y}{\displaystyle \frac{p}{2}}D\stackrel{~}{F}_t(y)_{g,g_0}^{p2}{\displaystyle \frac{d}{dt}}\left(D\stackrel{~}{F}_t(y)_{g,g_0}^2\right)๐v_g(y).`$
En $`t=0`$, comme $`D\stackrel{~}{F}(y)_{g,g_0}=1`$ pour presque tout $`y\stackrel{~}{Y}`$, on a
$$0=\frac{p/2}{vol(Y)}_Y\frac{d}{dt}_{|t=0}\left(D\stackrel{~}{F}_t(y)_{g,g_0}^2\right)๐v_g(y)$$
cโest-ร -dire, $`\stackrel{~}{F}`$ est un point critique de la fonctionne $`E_2`$. Lโapplication $`\stackrel{~}{F}`$ est donc faiblement harmonique (au sens des distributions, voir , page 397). Dโaprรจs les thรฉorรจmes de rรฉgularitรฉ classiques (voir , 3.10, page 397), $`\stackrel{~}{F}`$ รฉtant continue, elle est de classe $`๐^{\mathrm{}}`$.
Remarque.Nous avons montrรฉ que $`\stackrel{~}{F}`$ est un point critique de $`E_2`$, mais en fait elle minimise cette fonctionnelle car lโespace รฉtant de courbure nรฉgative ou nulle la fonctionnelle $`E_2`$ est convexe.
Nous pouvons alors terminer la preuve du thรฉorรจme 4.2 ii). Lโapplication $`\stackrel{~}{F}`$ a une diffรฉrentielle $`D\stackrel{~}{F}(y)`$ qui est continue en $`y`$ et est donc une isomรฉtrie pour tout $`y\stackrel{~}{Y}`$ ; la variรฉtรฉ $`\stackrel{~}{Y}`$ รฉtant connexe et complรจte, $`\stackrel{~}{X}`$ รฉtant connexe et simplement connexe nous dรฉduisons de cela que $`\stackrel{~}{F}`$ est une isomรฉtrie surjective de $`\stackrel{~}{Y}`$ sur $`\stackrel{~}{X}`$ (cโest en effet un exercice classique, voir , 2.108, exercice a), page 97). En particulier $`\rho (\mathrm{\Gamma })`$ est un sous-groupe discret cocompact de $`Isom(\stackrel{~}{X})`$ agissant sans points fixes et la reprรฉsentation $`\rho `$ est injective.
Remarques.
* Le lemme 4.11 peut sโรฉtendre et donne lieu ร la proposition suivante :
###### Proposition 4.15
Sโil existe une mesure de Radon finie et non nulle $`\mu `$, dรฉfinie sur $`\stackrel{~}{X}`$, invariante par $`\rho (\mathrm{\Gamma })`$, alors $`vol(\rho )=0`$.
Preuve .La preuve est identique ร celle du lemme 4.11, en posant
$$Z(x)=_{\stackrel{~}{X}}B(x,\theta )๐\mu (\theta ).$$
* Par ailleurs, si $`vol(\rho )0`$ le groupe $`\rho (\mathrm{\Gamma })`$ ne peut pas fixer (globalement) un sous-espace strict et totalement gรฉodรฉsique de $`\stackrel{~}{X}`$, car, sinon, nous pourrions choisir une application รฉquivariante $`f`$ ร valeurs dans ce sous-espace, et la chute de dimension entraรฎnerait que $`vol(\rho )=0`$, une contradiction. En utilisant le critรจre gรฉomรฉtrique รฉnoncรฉ dans , nous montrons donc (ร lโaide de la remarque ii) et de la proposition 4.15) que
###### Proposition 4.16
Si $`vol(\rho )0`$ alors $`\rho (\mathrm{\Gamma })`$ est rรฉductif.
On rappelle que $`\rho (\mathrm{\Gamma })`$ est dit rรฉductif si son adhรฉrence de Zariski lโest, cโest-ร -dire si cette derniรจre a un radical unipotent trivial.
Notons que dans , la rรฉductivitรฉ de $`\rho (\mathrm{\Gamma })`$ est prouvรฉe รชtre une condition nรฉcessaire et suffisante ร lโexistence dโune application harmonique $`\rho `$-รฉquivariante.
Enfin, le thรฉorรจme 4.2 conduit au
###### Corollaire 4.17
Si $`\rho `$ est une reprรฉsentation de $`\mathrm{\Gamma }=\pi _1(Y)`$ dans $`Isom(\stackrel{~}{X},g_0)`$, oรน $`Y`$ est une variรฉtรฉ compacte, alors
$$minvol(Y)\left(\frac{Ent(\stackrel{~}{X},g_0)}{n1}\right)^nvol(\rho ).$$
Preuve .On rappelle que
$$minvol(Y)=inf\{vol(Y,g)g\text{ mรฉtrique sur }Y\text{ telle que }|K_g|1\}$$
et que si la courbure sectionnelle $`K_g`$ de la mรฉtrique $`g`$ vรฉrifie $`K_g1`$ alors on a $`Ent(Y,g)n1`$ (voir ).
Remarque.
i) En particulier, sโil existe une reprรฉsentation $`\rho `$ telle que $`vol(\rho )0`$ alors $`minvol(Y)>0`$.
ii) On pourrait remplacer le volume minimal $`minvol(Y)`$ par
$$minvol_{Ricci}(Y)=inf\left\{vol(Y,g)Ricci_g(n1)g\right\}.$$
## 5 Applications
Dans ce paragraphe nous nous intรฉressons au cas oรน $`(\stackrel{~}{Y},\stackrel{~}{g})`$ est elle-mรชme un produit fini dโespaces symรฉtriques simplement connexe de courbure strictement nรฉgative. Comme prรฉcรฉdemment un tel espace sera notรฉ $`(\stackrel{~}{X},\stackrel{~}{g}_0)`$, oรน $`g_0`$ est la mรฉtrique dรฉfinie au paragraphe 2 et qui minimise lโentropie. De mรชme, $`\mathrm{\Gamma }`$ dรฉsigne un rรฉseau cocompact et sans torsion de $`Isom(\stackrel{~}{X},\stackrel{~}{g}_0)`$, et $`\rho `$ est un morphisme
$$\rho :\mathrm{\Gamma }Isom(\stackrel{~}{X},\stackrel{~}{g}_0).$$
Des exemples de telles reprรฉsentations sont rares et le but de ce paragraphe est, en particulier, de rappeler quelques unes des constructions classiques.
Dans cette situation, le thรฉorรจme 4.2 sโรฉcrit
$$vol(\rho )vol(X,g_0)$$
$`X=\stackrel{~}{X}/\mathrm{\Gamma }`$. Lโรฉgalitรฉ, dans cette inรฉgalitรฉ, nโa lieu que si et seulement si $`(\stackrel{~}{X}/\rho (\mathrm{\Gamma }),g_0)`$ est une variรฉtรฉ isomรฉtrique ร $`(X,g_0)`$, cโest-ร -dire si $`\rho (\mathrm{\Gamma })`$ est un rรฉseau cocompact de $`Isom(\stackrel{~}{X},\stackrel{~}{g}_0)`$. Nous rรฉpondons, dans ce paragraphe ร la question :
###### Question 5.1
Existe-t-il des reprรฉsentations, comme ci-dessus, telles que $`0<vol(\rho )<vol(X,g_0)`$ ?
Rappelons quโun rรฉseau $`\mathrm{\Gamma }`$ dans un groupe de Lie $`G`$, semi-simple connexe sans facteur compact est dit rรฉductible si $`G`$ possรจde des sous-groupes normaux $`H`$ et $`H^{}`$ tels que $`G=H.H^{}`$, $`HH^{}`$ est discret et $`\mathrm{\Gamma }/(\mathrm{\Gamma }H).(\mathrm{\Gamma }H^{})`$ est fini (voir page 86). $`\mathrm{\Gamma }`$ est dit irrรฉductible sโil nโest pas rรฉductible
Alors, lorsque $`\mathrm{\Gamma }`$ est irrรฉductible, le thรฉorรจme de super-rigiditรฉ de Margulis (, chapitre VII) fournit une rรฉponse nรฉgative complรจte ร la question ci-dessus.
###### Proposition 5.2
Avec les notations ci-dessus, si $`\mathrm{\Gamma }`$ est irrรฉductible et $`vol(\rho )0`$ alors $`\rho (\mathrm{\Gamma })`$ est un rรฉseau cocompact de $`Isom(\stackrel{~}{X},\stackrel{~}{g}_0)`$ et donc $`vol(\rho )=vol(X,g_0)`$.
Preuve .On se propose dโappliquer le thรฉorรจme 6.16 de , p. 332. On note $`G=Isom(\stackrel{~}{X},\stackrel{~}{g}_0)`$, cโest un groupe algรฉbrique dรฉfini sur $`๐`$ et semi-simple. Pour utiliser le rรฉsultat 6.16 de il faut travailler avec des groupes de Lie connexe, or $`\mathrm{\Gamma }`$ est un sous-groupe de $`G_+`$, le sous-groupe de $`G`$ constituรฉ des isomรฉtries prรฉservant lโorientation et $`G_+`$ nโest pas nรฉcessairement connexe. En effet, si $`\gamma =(\gamma _1,\mathrm{},\gamma _n)`$, oรน $`\gamma _iIsom(\stackrel{~}{X}_i,\stackrel{~}{g}_0^i)`$ , et si un nombre pair de $`\gamma _i`$ renverse lโorientation alors $`\gamma G_+`$, nรฉanmoins $`\gamma `$ ne peut pas รชtre connectรฉ ร lโidentitรฉ.
On rappelle que $`G`$ ร un nombre fini de composantes connexes car cโest un groupe algรฉbrique. Soit $`G^0`$ la composante de lโรฉlรฉment neutre et $`\mathrm{\Gamma }^0=G^0\mathrm{\Gamma }`$.
Il est aisรฉ de vรฉrifier que $`G^0=Isom_+(\stackrel{~}{X}_1,\stackrel{~}{g}_0^1)\times \mathrm{}\times Isom_+(\stackrel{~}{X}_p,\stackrel{~}{g}_0^p)`$$`Isom_+`$ dรฉsigne le groupe (connexe) dโisomรฉtries directes.
Les quatre lemmes qui suivent nโutilisent pas lโirrรฉductibilitรฉ de $`\mathrm{\Gamma }`$. Cette hypothรจse ne sera utilisรฉe que pour appliquer le thรฉorรจme de super-rigiditรฉ.
###### Lemme 5.3
Le groupe $`\mathrm{\Gamma }^0`$ est un rรฉseau cocompact de $`G^0`$ ainsi que de $`G`$.
Preuve .Lโapplication naturelle $`\mathrm{\Gamma }/\mathrm{\Gamma }^0G/G^0`$ est injective, $`\mathrm{\Gamma }^0`$ est donc dโindice fini dans $`\mathrm{\Gamma }`$ et est un rรฉseau cocompact de $`G`$. Par ailleurs, $`G^0/\mathrm{\Gamma }^0`$ est une composante connexe de $`G/\mathrm{\Gamma }^0`$, donc est compacte. Un thรฉorรจme gรฉnรฉral est prouvรฉ dans , p. 23 (thรฉorรจme 1.13).
Pour allรฉger les notations nous dรฉsignerons maintenant par $`\rho `$ la reprรฉsentation restreinte ร $`\mathrm{\Gamma }^0`$. Soit $`\mathrm{\Gamma }^1=\rho ^1\left(\rho (\mathrm{\Gamma }^0)G^0\right)`$.
###### Lemme 5.4
Le groupe $`\mathrm{\Gamma }^1`$ est dโindice fini dans $`\mathrm{\Gamma }^0`$.
Preuve .Lโapplication $`\mathrm{\Gamma }^0/\mathrm{\Gamma }^1G/G^0`$ induite par $`\rho `$ est injective, dโoรน le rรฉsultat.
Le groupe $`\mathrm{\Gamma }^1`$ est donc un rรฉseau cocompact de $`G`$ (et de $`G^0`$) qui de plus, comme $`\mathrm{\Gamma }`$, est irrรฉductible. La restriction de $`\rho `$ ร $`\mathrm{\Gamma }^1`$ est un homomorphisme
$$\rho :\mathrm{\Gamma }^1G^0$$
ร valeurs dans le groupe semi-simple, connexe $`G^0`$.
###### Lemme 5.5
Les groupes $`G`$ et $`G^0`$ nโont pas de centre.
Preuve .Si $`aG^0`$ est dans le centre de $`G^0`$, $`a`$ doit commuter avec tous les รฉlรฉments de $`G^0`$ ; or, pour $`x\stackrel{~}{X}`$ fixรฉ $`x=(x_1,\mathrm{},x_n)`$ les isomรฉtries du type $`(\gamma _1,\mathrm{},\gamma _n)`$, oรน $`\gamma _i`$ est une isomรฉtrie directe fixant $`x_i`$, sont dans $`G^0`$. Lโรฉlรฉment $`a`$ doit donc fixer $`x`$, pour tout $`x`$, cโest donc lโidentitรฉ.
Dans la terminologie de , le groupe $`G^0`$ est adjoint (il nโa pas de centre et est dรฉfini sur $`๐`$, voir , p. 13).
###### Lemme 5.6
Le groupe $`\rho (\mathrm{\Gamma }^1)`$ est Zariski-dense dans $`G^0`$.
Preuve .Rappelons que $`\mathrm{\Gamma }^1`$ est dโindice fini dans $`\mathrm{\Gamma }`$, on voit alors, de maniรจre รฉlรฉmentaire, que
$$vol(\rho )=[\mathrm{\Gamma }:\mathrm{\Gamma }^1]vol(\rho _{|\mathrm{\Gamma }^1})$$
de sorte que lโhypothรจse de la proposition LABEL:4.1 implique que $`vol(\rho _{|\mathrm{\Gamma }^1})0`$. La proposition 4.16, qui est un corollaire de la remarque 1.4 i) de , montre que $`\rho (\mathrm{\Gamma }^1)`$ est rรฉductif. Soit $`H`$ son adhรฉrence de Zariski, alors $`H`$ est รฉgalement rรฉductif. Comme $`H`$ est algรฉbrique, quitte ร restreindre ร un rรฉseau dโindice fini dans $`\mathrm{\Gamma }^1`$, on peut supposer que $`H`$ est connexe.
Lโalgรจbre de Lie de $`H`$, cโest-ร -dire $`๐ฅ`$ est une sous-algรจbre rรฉductive algรฉbrique de $`๐ค^0`$, alors dโaprรจs le thรฉorรจme 4, p. 261 de , il existe une involution de Cartan de $`๐ค^0`$ qui stabilise $`๐ฅ`$. Plus prรฉcisรฉment, lโespace symรฉtrique $`\stackrel{~}{X}`$ est identifiรฉ ร $`G^0/K^0`$ par le choix dโune dรฉcomposition de Cartan de $`G^0`$ (ici, $`K^0`$ dรฉsigne un sous-groupe compact maximal de $`G^0`$) ; alors, si $`\sigma ^0`$ dรฉsigne lโinvolution de Cartan correspondante, il existe $`gG^0`$ tel que lโinvolution $`g\sigma ^0g^1`$ prรฉserve $`h`$. Soit $`x\stackrel{~}{X}`$ le point correspondant ร la classe de $`g`$ dans $`G^0/K^0`$, alors dโaprรจs la proposition 2.6.2 de , la sous-variรฉtรฉ $`Hx=Y`$ est totalement gรฉodรฉsique dans $`(\stackrel{~}{X},g_0)`$ et $`\rho (\mathrm{\Gamma }^1)`$ invariante.
On peut donc, pour calculer le volume de la reprรฉsentation $`\rho _{|\mathrm{\Gamma }^1}`$ ($`\rho `$ restreinte ร $`\mathrm{\Gamma }^1`$) utiliser une application $`C^1`$, $`\rho _{|\mathrm{\Gamma }^1}`$-รฉquivariante de $`\stackrel{~}{X}`$ dans $`Y`$. Si $`HG^0`$ alors $`dimY<dim\stackrel{~}{X}`$ ce qui implique $`vol(\rho _{|\mathrm{\Gamma }^1})=0`$ et $`vol(\rho )=0`$. Ceci est en contradiction avec lโhypothรจse de la proposition.
Nous sommes maintenant en situation pour appliquer le thรฉorรจme de super-rigiditรฉ 6.16 b) de , p. 332 (le groupe $`G^0`$, qui est le groupe de dรฉpart et dโarrivรฉe nโa aucune composante simple compacte, cโest-ร -dire nโa pas de facteur $`๐`$-anisotrope). La reprรฉsentation $`\rho `$ se prolonge en un (unique) homomorphisme continu
$$\stackrel{~}{\rho }:G^0G^0$$
qui est donc analytique (, p. 117, thรฉorรจme 2.6). Le noyau $`Ker\stackrel{~}{\rho }`$ est un sous-groupe de Lie de $`G^0`$ (car fermรฉ). On rappelle que $`G^0=\underset{i=1}{\overset{p}{}}G_i`$, oรน $`G_i=Isom_+(\stackrel{~}{X}_i,\stackrel{~}{g}_0^i)`$ est un groupe simple. Comme $`Ker\stackrel{~}{\rho }`$ est normal, il est produit de certains $`G_i`$ de la liste prรฉcรฉdente : $`Ker\stackrel{~}{\rho }=\underset{k=1}{\overset{q}{}}G_k`$ pour $`qp`$.
De plus, lโimage $`\stackrel{~}{\rho }(G^0)`$ est un groupe de Lie isomorphe ร $`G^0/Ker\stackrel{~}{\rho }`$, cโest-ร -dire isomorphe ร $`\underset{k=q+1}{\overset{p}{}}G_k`$ si $`q<p`$, ร $`\{e\}`$ sinon. En particulier $`\stackrel{~}{\rho }(G^0)`$ est un sous-groupe semi-simple de $`G^0`$, invariant par $`\rho (\mathrm{\Gamma }^1)`$. Lโargument du lemme 5.6 montre que lโhypothรจse $`vol(\rho )0`$ implique que $`\stackrel{~}{\rho }(G^0)`$ doit รชtre รฉgal ร $`G^0`$, cโest-ร -dire que $`\stackrel{~}{\rho }`$ doit รชtre un automorphisme (analytique). En particulier $`\stackrel{~}{\rho }`$ est un diffรฉomorphisme et $`\stackrel{~}{\rho }(\mathrm{\Gamma }^1)`$ est un groupe discret et cocompact. Le thรฉorรจme de Mostow permet de conclure que les variรฉtรฉs localement symรฉtriques $`\stackrel{~}{X}/\mathrm{\Gamma }^1`$ et $`\stackrel{~}{X}/\stackrel{~}{\rho }(\mathrm{\Gamma }^1)`$ sont isomรฉtriques et donc que
$$vol(\rho _{|\mathrm{\Gamma }^1})=vol(\stackrel{~}{X}/\mathrm{\Gamma }^1).$$
Comme $`\mathrm{\Gamma }^1`$ est dโindice fini dans $`\mathrm{\Gamma }`$, on en dรฉduit que
$$vol(\rho )=vol(X,g_0).$$
Nous allons maintenant รฉtudier les cas oรน le rรฉseau $`\mathrm{\Gamma }`$ est rรฉductible. On rappelle quโun rรฉseau $`\mathrm{\Gamma }^1`$ de $`G^0`$ qui est rรฉductible vรฉrifie les propriรฉtรฉs suivantes (voir , p. 86, 5.22) : il existe une famille finie de sous-groupes normaux et connexes de $`G^0`$, $`H_1,\mathrm{},H_k`$ telle que :
* $`H_i\underset{ji}{}H_j`$ est discret pour tout $`i\{1,\mathrm{},k\}`$.
* $`G^0=\underset{i=1}{\overset{k}{}}H_i`$.
* $`\mathrm{\Gamma }_i^1=H_i\mathrm{\Gamma }^1`$ est un rรฉseau irrรฉductible de $`H_i`$.
* $`\underset{i=1}{\overset{k}{}}\mathrm{\Gamma }_i^1`$ est un sous-groupe normal dโindice fini de $`\mathrm{\Gamma }^1`$.
Comme prรฉcรฉdemment nous pouvons travailler ร un sous-groupe dโindice fini prรจs et donc supposer que $`\underset{i=1}{\overset{k}{}}\mathrm{\Gamma }_i^1=\mathrm{\Gamma }^1`$. De mรชme, chaque $`H_i`$ doit รชtre un produit de facteurs simples composant $`G^0`$, cโest-ร -dire
$$H_i=\underset{s=p_i}{\overset{p_i+r_i}{}}G_s.$$
En particulier, si $`ij`$, $`H_i`$ et $`H_j`$ commutent.
###### Proposition 5.7
Avec les notations ci-dessus, si $`\mathrm{\Gamma }`$ est rรฉductible et $`vol(\rho )0`$ et si, pour tout $`i`$, $`H_i`$ est super-rigide alors $`\rho (\mathrm{\Gamma })`$ est un rรฉseau cocompact et donc $`vol(\rho )=vol(X,g_0)`$.
Preuve .Par groupe super-rigide nous entendons un groupe auquel nous pouvons appliquer le thรฉorรจme de super-rigiditรฉ, cโest-ร -dire , dans notre situation, soit $`H_i`$ est de rang supรฉrieur oรน รฉgal ร 2 ($`r_i1`$) ou bien $`H_i`$ est le groupe dโisomรฉtries directes dโun espace hyperbolique quaternionien ou du plan hyperbolique de Cayley.
Comme prรฉcรฉdemment nous travaillons avec le sous-groupe $`\mathrm{\Gamma }^1`$. Dรฉfinissons $`\rho _i=\rho _{|\mathrm{\Gamma }_i^1}`$, pour $`i=1,\mathrm{},k`$ ; ici nous commettons un abus de langage et identifions $`\mathrm{\Gamma }_i^1`$ et $`\{e\}\times \mathrm{}\times \{e\}\times \mathrm{\Gamma }_i^1\times \{e\}\times \mathrm{}\times \{e\}`$. Dรฉfinissons les groupes $`K_i=\overline{\rho _i(\mathrm{\Gamma }_i^1)}`$, lโadhรฉrence de Zariski de $`\rho _i(\mathrm{\Gamma }_i^1)`$ ; ce sont des sous-groupes algรฉbriques de $`G^0`$ et le thรฉorรจme 6.15 i), a) de , p. 332 affirme que, si rang $`H_i2`$, $`K_i`$ est un groupe semi-simple. Insistons sur le fait que $`H_i`$ est considรฉrรฉ comme un sous-groupe du groupe de dรฉpart de la reprรฉsentation $`\rho `$ et $`K_i`$ comme un sous-groupe du groupe dโarrivรฉe. Le mรชme rรฉsultat pour le cas oรน $`H_i`$ est le groupe dโisomรฉtries (directes) de lโespace hyperbolique quaternionien ou du plan hyperbolique de Cayley est prouvรฉ dans . Dans tous les cas, donc, $`K_i`$ est un groupe semi-simple.
###### Lemme 5.8
Les groupes $`K_i`$ sont sans facteurs compacts et $`K_i`$ est semi-simple.
Preuve .Quitte ร passer ร un sous-groupe dโindice fini de $`\mathrm{\Gamma }_i^1`$ nous pouvons supposer que $`K_i`$ est un produit de groupes simples. Supposons quโil contienne un facteur compact, soit
$$K_i=L_i^1\times \mathrm{}\times L_i^{k_i}\times U$$
$`U`$ est un groupe simple compact ; alors $`U`$ est normal dans $`K_i`$. Par ailleurs $`\rho _i(\mathrm{\Gamma }_i^1)`$ et $`\rho _j(\mathrm{\Gamma }_j^1)`$ commutent si $`ij`$, car $`\mathrm{\Gamma }_i^1`$ et $`\mathrm{\Gamma }_j^1`$ commutent dans $`G^0`$ (nous faisons ici lโabus de langage signalรฉ prรฉcรฉdemment) ; les groupes $`K_i`$ et $`K_j`$ commutent donc รฉgalement si $`ji`$. Le groupe $`U`$ est donc normal dans le produit $`\underset{i=1}{\overset{k}{}}K_i`$; remarquons que ce produit est dรฉfini comme le groupe engendrรฉ par les produits dโรฉlรฉments de $`K_i`$. Il est dรฉfini sans ambiguรฏtรฉ car les groupes commutent deux ร deux. Par ailleurs $`\underset{i=1}{\overset{k}{}}K_i`$ est Zariski-dense donc รฉgal ร $`G_0`$; en effet, il contient $`\rho (\mathrm{\Gamma }^1)`$ qui Zariski dense car $`vol(\rho )0`$ (voir le lemme 5.6). Le groupe $`U`$ est donc normal dans $`G^0`$ ; il est alors รฉgal ร une des composantes de $`G^0`$ ou bien rรฉduit ร lโรฉlรฉment neutre ; aucune des composantes de $`G^0`$ nโรฉtant compacte $`U`$ est trivial.
Enfin les arguments prรฉcรฉdents montrent que les composantes simples de $`K_i`$ et $`K_j`$ sont distincts et donc que $`\underset{i=1}{\overset{k}{}}K_i`$ est semi-simple (cโest le produit de toutes les composantes des groupes $`K_i`$).
Nous pouvons donc appliquer le thรฉorรจme de super-rigiditรฉ de , (6.16 c), p. 332) pour les composantes $`H_i`$ de rang $`2`$ et celui de pour les autres et affirmer que les reprรฉsentations $`\rho _i`$ se prolongent en des morphismes continus
$$\phi _i:H_iK_i.$$
On construit alors un prolongement de $`\rho `$ en
$`\phi :G^0`$ $`{\displaystyle \underset{i=1}{\overset{k}{}}}K_iG^0`$
$`(\gamma _1,\mathrm{},\gamma _k)`$ $`(\phi _1(\gamma _1),\mathrm{},\phi _k(\gamma _k))`$
Les morphismes $`\phi _i`$ commutent et $`\phi `$ est bien dรฉfini et est un morphisme continu. On termine donc la preuve de la proposition refreductible par les mรชmes arguments que ceux de la preuve de la proposition 5.2.
Nous nous intรฉressons maintenant au cas oรน $`G_0`$ possรจde des composantes simples non super-rigides. Supposons donc que $`\mathrm{\Gamma }^1=\mathrm{\Gamma }_1^1\mathrm{\Gamma }_2^1`$ (produit libre) oรน $`\mathrm{\Gamma }_1^1`$ est un rรฉseau cocompact dโun groupe $`H_1`$ extension finie dโun produit de groupes super-rigides et $`\mathrm{\Gamma }_2^1`$ est un rรฉseau cocompact de $`H_2`$ produit de copies de $`PO(k\mathrm{,1})`$ et $`PU(k^{}\mathrm{,1})`$. Les arguments qui prรฉcรจdent sโappliquent pour montrer que
i) $`\rho (\mathrm{\Gamma })`$ est Zariski dense dans $`G^0`$ si $`vol(\rho )0`$.
ii) Soit $`K_i`$, $`i=\mathrm{1,2}`$, lโadhรฉrence de Zariski de $`\rho _i(\mathrm{\Gamma }_i^1)`$. La densitรฉ de $`\rho (\mathrm{\Gamma })`$ implique la densitรฉ (pour la topologie de Zariski) de $`K_1K_2`$ ; les deux groupes commutent. En dรฉcomposant $`\mathrm{\Gamma }_1^1`$ en produits de rรฉseaux cocompacts irrรฉductibles on voit, en utilisant les arguments de la preuve du lemme 5.8, que $`K_1`$ est semi-simple sans facteurs compacts. Si $`G^0=\underset{i=1}{\overset{p}{}}G_k`$, oรน les $`G_k`$ sont des groupes dโisomรฉtries directes dโespaces symรฉtriques de rang 1 et de type non compact, alors $`K_1=\underset{k=1}{\overset{q}{}}G_k`$ (par exemple) ; en effet, $`K_1`$ est un sous-groupe normal de $`G^0`$. Le groupe $`K_2`$ est donc inclus dans $`\underset{q+1}{\overset{p}{}}G_k`$ et comme $`K_1K_2`$ est Zariski dense, $`K_2=\underset{q+1}{\overset{p}{}}G_k`$ (en particulier il est semi-simple). On peut choisir une application รฉquivariante $`f_i`$ de $`H_iK_i`$ et un calcul immรฉdiat montre que
$$vol(\rho )=vol(\rho _1)vol(\rho _2),$$
de sorte que $`vol(\rho _1)0`$. Le thรฉorรจme de super-rigiditรฉ (appliquรฉ comme prรฉcรฉdemment aux composantes irrรฉductibles de $`\mathrm{\Gamma }_1^1`$) permet dโรฉtendre $`\rho _1`$ en un morphisme continu
$$\phi _1:H_1K_1$$
et la non nullitรฉ de $`vol(\rho _1)`$ montre que $`\phi _1`$ est un isomorphisme et donc montre que $`\rho (\mathrm{\Gamma }_1^1)`$ est isomorphe ร $`\mathrm{\Gamma }_1^1`$ ce qui conduit ร
$$vol(\rho _1)=vol(H_1/\mathrm{\Gamma }_1^1).$$
Le groupe $`K_2`$ contient les composantes non super-rigides de $`G^0`$, de mรชme que $`H_2`$, ils sont donc isomorphes. Cโest la seule composante non triviale de $`\rho `$.
Nous donnons maintenant un exemple de reprรฉsentation du groupe fondamental dโune variรฉtรฉ hyperbolique rรฉelle, de volume non nul et dโimage non discrรจte.
Exemple : produit amalgamรฉ.
Soit $`X`$ une variรฉtรฉ hyperbolique compacte de dimension $`n3`$. Supposons quโil existe dans $`X`$ une hypersurface compacte plongรฉe totalement gรฉodรฉsique notรฉe $`\mathrm{\Sigma }`$ et incompressible, cโest-ร -dire telle que lโapplication induite : $`\pi _1(\mathrm{\Sigma })\mathrm{\Pi }_1(X)`$ soit une injection. Nous supposons de plus que cette hypersurface sรฉpare $`X`$ en deux composantes connexes $`X_A`$ et $`X_B`$ de groupe fondamental respectif $`A`$ et $`B`$. En posons $`C=\pi _1(\mathrm{\Sigma })`$, le thรฉorรจme de Van Kampen montre que
$$\pi _1(X)=A_CB$$
produit amalgamรฉ de $`A`$ et $`B`$ sur $`C`$. Les groupes $`\pi _1(X)`$, $`A`$, $`B`$ et $`C`$ sont des sous-groupes de $`PO(n\mathrm{,1})`$ et agissent donc sur lโespace hyperbolique $`๐^n`$. Choisissons un relevรฉ $`\stackrel{~}{\mathrm{\Sigma }}`$ de $`\mathrm{\Sigma }`$ dans $`๐^n`$ ; $`\stackrel{~}{\mathrm{\Sigma }}`$ est une hypersurface totalement gรฉodรฉsique. On identifie $`C`$ au sous-groupe de $`\pi _1(X)`$ qui fixe $`\stackrel{~}{\mathrm{\Sigma }}`$. Soit $`s`$ la symรฉtrie par rapport ร $`\stackrel{~}{\mathrm{\Sigma }}`$, on dรฉfinit
$`\rho :\pi _1(X)`$ $`PO(n\mathrm{,1})`$
$`aA`$ $`a`$
$`bB`$ $`sbs^1.`$
###### Lemme 5.9
Lโapplication $`\rho `$ dรฉfinit une reprรฉsentation de $`\pi _1(X)`$ dans $`PO(n\mathrm{,1})`$.
Preuve .Le groupe $`\pi _1(X)`$ est le quotient du produit libre $`AB`$ par les relations qui consistent ร identifier un รฉlรฉment de $`C`$ dans $`A`$ avec le mรชme รฉlรฉment dans $`B`$. Comme $`\rho `$ est un morphisme en restriction ร $`A`$ et ร $`B`$ respectivement, il suffit de vรฉrifier la compatibilitรฉ avec les relations. Or, si $`cC`$
$$scs^1=c$$
dโoรน le rรฉsultat.
Afin de calculer le volume de cette reprรฉsentation il faut trouver une application lipschitzienne $`f:๐^n๐^m`$, $`\rho `$-รฉquivariante.
###### Proposition 5.10
Avec les notations ci-dessus on a,
$$vol(\rho )=vol(X_A)vol(X_B).$$
Preuve .Nous allons dรฉcrire $`f`$ de maniรจre prรฉcise et le calcul du volume sโensuivra. Le fait que $`\pi _1(X)`$ soit un produit amalgamรฉ est รฉquivalent (, p. 48) ร lโexistence dโun arbre $`T`$, sur lequel $`\pi _1(X)`$ opรจre (sans inversion) en sorte que le quotient soit un segment (deux sommets joints par une arรชte). Les sous-groupes $`A`$, $`B`$ et $`C`$ sont alors les stabilisateurs respectifs des deux sommets et de lโarรชte de lโarbre quotient. Nous allons donner une description gรฉomรฉtrique de cet arbre $`T`$. Nous avons choisi un relevรฉ $`\stackrel{~}{\mathrm{\Sigma }}`$ de lโhypersurface compacte $`\mathrm{\Sigma }`$ plongรฉe dans $`X`$ ; $`\mathrm{\Sigma }`$ รฉtant une sous-variรฉtรฉ plongรฉe, sans auto-intersection, les translatรฉs $`\gamma \stackrel{~}{\mathrm{\Sigma }}`$ de $`\stackrel{~}{\mathrm{\Sigma }}`$ par les รฉlรฉments $`\gamma \pi _1(X)`$ sont deux ร deux disjoints ; ils sรฉparent donc $`๐^n`$ en une infinitรฉ de composantes connexes. Les deux composantes connexes dont lโadhรฉrence contient $`\stackrel{~}{\mathrm{\Sigma }}`$ sont des revรชtements universels de $`X_A`$ et $`X_B`$ respectivement, que nous noterons $`\stackrel{~}{X}_A`$ et $`\stackrel{~}{X}_B`$. Les autres composantes connexes sont les translatรฉs par les รฉlรฉments de $`\pi _1(X)`$ de $`\stackrel{~}{X}_A`$ et $`\stackrel{~}{X}_B`$. Les sous-groupes $`A`$ et $`B`$ prรฉservent $`\stackrel{~}{X}_A`$ et $`\stackrel{~}{X}_B`$ respectivement (aprรจs un choix convenable dโun point base et dโun de ses relevรฉs).
Maintenant, choisissons un point $`x_a\stackrel{~}{X}_A`$ et un point $`x_b\stackrel{~}{X}_B`$, les sommets de lโarbre $`T`$ sont les $`\gamma (x_a)`$ et $`\gamma (x_b)`$, oรน $`\gamma `$ parcourt $`\pi _1(X)`$ ; on joint deux sommets $`\gamma (x_a)`$ et $`\gamma ^{}(x_b)`$ (de type diffรฉrent) si, et seulement si, les composantes connexes correspondantes $`\gamma \stackrel{~}{X}_A`$ et $`\gamma ^{}\stackrel{~}{X}_B`$ sont telles que $`\overline{\gamma \stackrel{~}{X}_A}\overline{\gamma ^{}\stackrel{~}{X}_B}\mathrm{}`$.
$`\stackrel{~}{\mathrm{\Sigma }}`$$`\stackrel{~}{X}_A`$$`\stackrel{~}{X}_B`$$``$$``$$``$$``$
Soit alors $`x\stackrel{~}{X}`$, il appartient ร une composante connexe du complรฉmentaire de $`\underset{\gamma \mathrm{\Gamma }}{}\gamma \stackrel{~}{\mathrm{\Sigma }}`$ qui correspond ร un sommet de lโarbre prรฉcรฉdent. Dans cet arbre il existe un unique chemin joignant la composante $`\stackrel{~}{X}_A`$ ร celle de $`x`$ ; ce chemin est une succession dโarรชtes $`e_1,e_2,\mathrm{},e_k`$ prises dans lโordre, de la composante $`\stackrel{~}{X}_A`$ ร celle de $`x`$. Chacune de ces arรชtes correspond ร une image de $`\stackrel{~}{\mathrm{\Sigma }}`$ et nous noterons $`s_{e_i}`$ la symรฉtrie orthogonale hyperbolique par rapport ร cette hypersurface totalement gรฉodรฉsique.
###### Definition 5.11
On pose $`f(x)=s_{e_1}s_{e_2}\mathrm{}s_{e_k}(x)`$
Lโapplication $`f`$ est bien dรฉfinie. Elle est $`๐^{\mathrm{}}`$ par morceaux et continue ; en effet, la seule ambiguรฏtรฉ dans la formule ci-dessus est lorsque $`x`$ est sur lโhypersurface dรฉfinie par $`e_k`$, mais dans ce cas $`s_{e_k}(x)=x`$.
###### Lemme 5.12
Lโapplication $`f`$ est $`\rho `$-รฉquivariante.
Preuve .Il suffit de vรฉrifier lโรฉquivariance pour les รฉlรฉments de $`A`$ et ceux de $`B`$ qui engendrent le groupe fondamental de $`X`$.
a) Si $`aA`$, le chemin dans lโarbre joignant la composante $`\stackrel{~}{X}_A`$ ร celle de $`ax`$ est constituรฉ des arรชtes $`ae_1,\mathrm{},ae_k`$ ; en effet, puisque $`aA`$, $`a\stackrel{~}{X}_A=\stackrel{~}{X}_A`$ et lโarรชte $`ae_1`$ a son origine dans $`\stackrel{~}{X}_A`$. Dโoรน
$`f(ax)`$ $`=s_{ae_1}\mathrm{}s_{ae_k}(ax)`$
$`=as_{e_1}a^1\mathrm{}as_{e_k}a^1(ax)`$
$`=af(x)=\rho (a)f(x)`$
b) si $`bB`$, le chemin joignant la composante $`\stackrel{~}{X}_A`$ ร celle de $`bx`$ est constituรฉ du chemin dans lโarbre joignant $`\stackrel{~}{X}_A`$ ร $`b\stackrel{~}{X}_A`$ suivi de lโimage par $`b`$ du chemin prรฉcรฉdent. Rappelons que les sommets de lโarbre sont les รฉlรฉments de $`\mathrm{\Gamma }/A`$ et $`\mathrm{\Gamma }/B`$ et les arรชtes sont les รฉlรฉments de $`\mathrm{\Gamma }/C`$ (voir ). Par exemple, la composante connexe $`\stackrel{~}{X}_A`$ correspond ร $`eA`$ (classe de lโรฉlรฉment neutre $`e`$), celle de $`\stackrel{~}{X}_B`$ ร $`eB`$ ; elles sont
$`eA`$$`eB`$$`bA`$$`eC`$$`bC`$$``$$``$$``$
reliรฉes par lโarรชte $`eC`$. Par ailleurs $`b\stackrel{~}{X}_A`$ correspond ร la classe $`bA`$ reliรฉe ร $`eB`$ par lโarรชte $`bC`$. En conclusion, nous avons
$$f(bx)=ss_{be}(bs_{e_1}\mathrm{}s_{e_k}b^1)(bx)$$
$`e`$ dรฉsigne par abus de langage lโarรชte $`eC`$ et $`s_e=s`$. Dโoรน
$`f(bx)`$ $`=(sbsb^1)(bs_{e_1}\mathrm{}s_{e_k}b^1)(bx)`$
$`=\rho (b)f(x).`$
Fin de la preuve de la proposition . La fin de la preuve est รฉvidente ; en effet $`f`$ renverse lโorientation sur $`\stackrel{~}{X}_B`$ et est lโidentitรฉ sur $`\stackrel{~}{X}_A`$, il suffit donc de choisir un domaine fondamental dans la rรฉunion $`\stackrel{~}{X}_A\stackrel{~}{X}_B`$ pour lequel $`\mathrm{\Sigma }`$ se relรจve sur $`\stackrel{~}{\mathrm{\Sigma }}`$.
Pour รชtre complet, il faut construire des variรฉtรฉs $`X`$ hyperboliques admettant une hypersurface connexe sรฉparante qui sรฉpare la variรฉtรฉ en deux parties de volume distinct. Cette construction nous a รฉtรฉ suggรฉrรฉe par N. Bergeron. Soit $`M_1`$ une variรฉtรฉ compacte de dimension 3, hyperbolique ร bord totalement gรฉodรฉsique qui est une surface compacte connexe notรฉe $`\mathrm{\Sigma }`$. De tels exemples existent (voir , , th. 4.3 et ). Considรฉrons le double $`M`$ obtenu par recollement de deux copies de $`M_1`$ le long de $`\mathrm{\Sigma }`$. La variรฉtรฉ compacte $`M`$ sans bord est hyperbolique car $`\mathrm{\Sigma }`$ est totalement gรฉodรฉsique. Le thรฉorรจme 2 de montre que lโon peut construire un revรชtement fini $`\widehat{M}`$ de $`M`$ tel que $`\mathrm{\Sigma }`$ se relรจve isomรฉtriquement ร $`\widehat{M}`$ en une sous-variรฉtรฉ totalement gรฉodรฉsique $`\widehat{\mathrm{\Sigma }}`$ non sรฉparante. On dรฉcoupe alors $`\widehat{M}`$ le long de $`\widehat{\mathrm{\Sigma }}`$ pour obtenir une variรฉtรฉ ร bord dont les deux composantes du bord, notรฉe $`\widehat{\mathrm{\Sigma }}_1`$, $`\widehat{\mathrm{\Sigma }}_2`$, sont isomรฉtriques ร $`\widehat{\mathrm{\Sigma }}\mathrm{\Sigma }`$ et on recolle ร chacune de ces composantes une copie de $`M_1`$. Alors, $`\widehat{\mathrm{\Sigma }}_1`$ (et $`\widehat{\mathrm{\Sigma }}_2`$) dรฉcoupe la nouvelle variรฉtรฉ hyperbolique en deux composantes lโune de volume รฉgal ร $`vol(M_1)`$ et lโautre de volume รฉgal ร $`vol(\widehat{M})+vol(M_1)>vol(M_1)`$.
Il serait intรฉressant de disposer de tels exemples en dimension $`n4`$. Remarquons, par ailleurs, que lโensemble des valeurs de $`vol(\rho )`$ ainsi obtenu est discret (pour une variรฉtรฉ donnรฉe) ; une explication prรฉcise ร ce phรฉnomรจne est fournie par le chapitre suivant.
## 6 Volume et dรฉformations
Nous avons dรฉjร remarquรฉ que, lorsque la dimension de $`X`$ est paire, le volume dโune reprรฉsentation ($`\stackrel{~}{X}`$ est supposรฉe symรฉtrique) est le nombre dโEuler du fibrรฉ plat correspondant. En particulier, ce nombre est constant le long des dรฉformations continues de reprรฉsentations. Nous allons prouver un rรฉsultat analogue dans le cas oรน la dimension de $`X`$ est impaire. De telles dรฉformations existent en dimension $`3`$ () et nous en donnons des exemples. La constance du volume est prouvรฉe en dimension $`3`$ par S. Reznikov ; nous donnons ici une preuve, valable en toute dimension, qui repose sur le formule de Schlรคfli. Dans ce qui suit $`M`$ dรฉsigne une variรฉtรฉ riemannienne fermรฉe et orientรฉe de dimension $`n`$ et $`\stackrel{~}{X}`$ lโespace hyperbolique rรฉel simplement connexe de dimension $`n`$.
###### Thรฉorรจme 6.1
Soit $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe et orientรฉe et $`\rho _t:\mathrm{\Pi }_1(M)Isom(\stackrel{~}{X})`$ une famille de reprรฉsentations qui dรฉpend de maniรจre $`C^1`$ du paramรจtre $`t๐`$, alors le volume $`vol(\rho _t)`$ est constant.
La preuve repose sur un lemme technique dont le but est de construire une application รฉquivariante affine par morceaux particuliรจre. Par application affine nous entendons une application affine le long de toute gรฉodรฉsique.
###### Lemme 6.2
Sous les hypothรจses du thรฉorรจme 6.1, il existe une triangulation $`๐ฏ`$ de $`\stackrel{~}{M}`$ et une application continue et affine par morceaux $`\stackrel{~}{f_0}:\stackrel{~}{M}\stackrel{~}{X}`$ qui est $`\rho _0`$-รฉquivariante et non dรฉgรฉnรฉrรฉe au sens oรน lโimage par $`\stackrel{~}{f_0}`$ dโun simplexe de la triangulation $`๐ฏ`$ est un simplexe gรฉodรฉsique de $`\stackrel{~}{X}`$ non dรฉgรฉnรฉrรฉ.
Preuve .Un thรฉorรจme classique affirme que toute variรฉtรฉ lisse compacte $`M`$ est homรฉomorphe ร un complexe simplicial $`K`$; plus prรฉcisรฉment $`K`$ est un espace triangulรฉ muni dโune mรฉtrique euclidienne par morceaux (que lโon peut rรฉaliser dans $`๐^๐ง`$). Cet homรฉomorphisme peut, de plus, รชtre choisi Lipschitzien. Le volume de toute reprรฉsentation de $`\mathrm{\Pi }_1(M)=\mathrm{\Pi }_1(K)`$ peut donc se calculer en intรฉgrant sur $`M`$ ou bien sur $`K`$. Dans la suite nous noterons รฉgalement $`M`$ ce complexe euclidien par morceaux et toute triangulation sera une subdivision de la dรฉcomposition de $`K`$ en simplexes.
Choisissons alors une triangulation suffisamment fine de $`M`$ et appelons $`๐ฏ_{\stackrel{~}{M}}`$ la triangulation invariante par $`\mathrm{\Pi }_1(M)`$ sur $`\stackrel{~}{M}`$ qui sโen dรฉduit par image rรฉciproque. Soit $`D`$ un domaine fondamental (de Dirichlet) dans $`\stackrel{~}{M}`$ pour lโaction de $`\mathrm{\Pi }_1(M)`$. Quitte ร modifier un peu $`๐ฏ_{\stackrel{~}{M}}`$ ou bien $`D`$ on peut supposer quโaucun sommet de la triangulation nโest sur $`D`$.
Notons $`(m_1,\mathrm{},m_N)`$ la liste des sommets de $`๐ฏ_{\stackrel{~}{M}}`$ qui sont dans lโintรฉrieur de $`D`$, N est alors le cardinal des sommets de la triangulation de dรฉpart sur $`M`$. Choisissons maintenant $`N`$ points dans $`\stackrel{~}{X}`$, notรฉs $`(y_1,\mathrm{},y_N)`$ de sorte que si $`(m_{i_1},\mathrm{},m_{i_{k+1}})`$ est un $`k`$-simplexe de $`๐ฏ`$ alors le simplexe gรฉodรฉsique de $`\stackrel{~}{X}`$ de sommets $`(y_{i_1},\mathrm{},y_{i_{k+1}})`$ est non dรฉgรฉnรฉrรฉ pour tout $`k\text{dim}M+1`$. Ceci est toujours possible car, pour chaque sommet $`y_j`$, la rรฉunion des conditions de dรฉgรฉnรฉrescence des simplexes contenant $`y_j`$ est un ensemble fermรฉ dโintรฉrieur vide (une rรฉunion finie de $`k1`$-plans). Ces choix รฉtant fait, il existe autour de chaque point $`y_j`$ un petit voisinage $`V_j`$ en sorte que, pour nโimporte quel choix de points $`y_1^{},\mathrm{},y_N^{}`$ avec $`y_j^{}V_j`$, la propriรฉtรฉ de non dรฉgรฉnรฉrescence ci-dessus soit encore vรฉrifiรฉe. Par la suite nous aurons รฉgalement besoin de choisir les point $`y_j`$ de sorte que
$$\gamma \mathrm{\Pi }_1(M),ji,y_i\rho _0(\gamma )y_j,$$
ceci est toujours possible car la rรฉunion des points de lโorbite des $`y_j`$, pour $`ji`$, qui sont dans $`V_i`$ est un ensemble dรฉnombrable. On procรจde dons par rรฉcurrence, $`y_1`$ รฉtant fixรฉ on choisit $`y_2V_2`$ dans lโensemble partout dense qui est le complรฉmentaire de lโorbite de $`y_1`$ , puis $`y_3`$ dans le complรฉmentaire des orbites de $`y_1`$ et $`y_2`$ et ainsi de suite.
On dรฉfinit alors $`\stackrel{~}{f_0}`$ par :
$$\gamma \mathrm{\Pi }_(M),i=1,\mathrm{},N\stackrel{~}{f_0}(\gamma m_i)=\rho _0(\gamma )y_i,$$
et on รฉtend $`\stackrel{~}{f_0}`$ ร lโintรฉrieur dโun simplexe $`(m_{i_1},\mathrm{},m_{i_N})`$ en une application affine sur le simplexe gรฉodรฉsique engendrรฉ par les points $`y_{i_1},\mathrm{},y_{i_N}`$; on utilise pour cela la mรฉtrique euclidienne sur les simplexes de $`\stackrel{~}{M}`$ et la mรฉtrique hyperbolique sur ceux de $`\stackrel{~}{X}`$. Par le choix des points $`y_i\stackrel{~}{X}`$, tous les simplexes dont les sommets sont dans lโintรฉrieur de $`D`$ sont transformรฉs par $`\stackrel{~}{f_0}`$ en des simplexes non dรฉgรฉnรฉrรฉs. Considรฉrons maintenant le cas oรน certains sommets sont dans lโintรฉrieur de $`D`$ et dโautres ร lโextรฉrieur. Soit $`(m_{i_1},\mathrm{},m_{i_p},\gamma _{j_1}m_{j_1},\mathrm{},\gamma _{j_q}m_{j_q})`$ un tel simplexe et supposons que son image par $`\stackrel{~}{f_0}`$, cโest-ร -dire le simplexe notรฉ $`(y_{i_1},\mathrm{},y_{i_p},\rho (\gamma _{j_1})y_{j_1},\mathrm{},\rho (\gamma _{j_q})y_{j_q})`$, soit dรฉgรฉnรฉrรฉ; cela signifie quโil existe $`1kq`$ tel que $`\rho (\gamma _{j_k})y_{j_k}`$ appartienne au sous-espace totalement gรฉodรฉsique $`E`$ engendrรฉ par les points $`(y_{i_1},\mathrm{},y_{i_p},\rho (\gamma _{j_1})y_{j_1},\mathrm{},\rho (\gamma _{j_{k1}})y_{j_{k1}})`$ (rappelons que, par construction, $`(y_{i_1},\mathrm{},y_{i_p})`$ est un $`(p1)`$-simplexe non dรฉgรฉnรฉrรฉ); on dรฉplace alors $`y_{j_k}`$ ร lโintรฉrieur de $`V_{j_k}`$ pour le sรฉparer de $`\rho (\gamma _{j_k})^1E`$; ceci est possible si $`E`$ reste fixe lorsque lโon dรฉplace $`y_{j_k}`$, cโest-ร -dire si aucun des points $`y_{i_1},\mathrm{},y_{i_p},\rho (\gamma _{j_1})y_{j_1},\mathrm{},\rho (\gamma _{j_{k1}})y_{j_{k1}}`$ nโest dans lโorbite de $`y_{j_k}`$. Par le choix des $`y_i`$ ceci ne peut se produire que si $`y_{j_k}=y_{j_l}\text{ avec }l=1,\mathrm{},q\text{ et }lk`$ ou bien $`y_{i_k}=y_{i_l},l=1,\mathrm{},p`$. Les points $`y_i`$ รฉtant en bijection avec les points $`m_i`$ cela impliquerait que dans le simplexe $`(m_{i_1},\mathrm{},m_{i_p},\gamma _{j_1}m_{j_1},\mathrm{},\gamma _{j_q}m_{j_q})`$ deux des points $`m_l`$ coรฏncident et donc quโau quotient sur $`M`$ il se projette sur un simplexe dรฉgรฉnรฉrรฉ ce qui est impossible. On peut donc sรฉparer $`y_{j_k}`$ du sous-espace totalement gรฉodรฉsique $`\rho (\gamma _{j_k})^1E`$. On utilise ensuite lโargument de densitรฉ pour choisir le nouveau point $`y_{j_k}`$ disjoint de la rรฉunion des orbites par $`\rho (\mathrm{\Pi }_1(M))`$ des autres points $`y_l`$. On procรจde alors par rรฉcurrence sur les simplexes considรฉrรฉs qui sont en nombre fini.
Les autres simplexes sont des images par un รฉlรฉment $`\rho (\gamma )`$, pour $`\gamma \mathrm{\Pi }_1(M)`$ des simplexes dโun des deux types prรฉcรฉdents. Ceci prouve le lemme 6.2.
preuve du thรฉorรจme
Nous noterons $`๐ฏ_{\stackrel{~}{X}}`$ la collection des simplexes de $`\stackrel{~}{X}`$ ainsi obtenue. Soit $`F`$ une face de codimension $`2`$ de $`๐ฏ_{\stackrel{~}{M}}`$ et $`F^{}`$ son image dans $`๐ฏ_{\stackrel{~}{X}}`$. Lโรฉtoile de $`F`$ dans $`๐ฏ_{\stackrel{~}{M}}`$ contient un nombre fini de $`n`$-simplexes $`s_1,\mathrm{},s_k`$ dont les images sont notรฉes $`s_1^{},\mathrm{},s_k^{}`$. Le link autour de $`F`$ est un cercle. Prรฉcisรฉment, considรฉrons un voisinage tubulaire de rayon assez petit, notรฉ Tub$`(F)`$, de cette face $`F`$ de codimension $`2`$. Alors le bord de Tub$`(F)`$ est diffรฉomorphe ร $`F\times S^1`$. La variรฉtรฉ $`M`$ est supposรฉe orientรฉe, et donc aussi $`\stackrel{~}{M}`$. Sur le bord de Tub$`(F)`$ nous choisissons une courbe $`๐`$ gรฉnรฉrateur de $`H_1(\text{Tub}(F),\text{Z})\text{Z}`$; nous pouvons, par exemple, prendre lโintersection de $`\text{Tub}(F)`$ avec un hyperplan orthogonal ร $`F`$ en un point (on peut dรฉfinir un tel hyperplan bien que la mรฉtrique sur $`\stackrel{~}{M}`$, qui est euclidienne sur chaque simplexe, soit singuliรจre en $`F`$). Si nous choisissons arbitrairement une orientation sur chaque face de codimension $`2`$, donc en particulier sur $`F`$, cela fournit une orientation du cercle $`๐`$ compatible avec celle de $`\stackrel{~}{M}`$.
Lโapplication $`\stackrel{~}{f_0}`$, linรฉaire par morceaux, envoie $`F`$ sur $`F^{}`$ (par construction) et donc $`\text{Tub}(F)`$ sur un cylindre topologique que lโon peut projeter, ร partir de $`F^{}`$, sur le bord $`\text{Tub}(F^{})`$ dโun petit voisinage tubulaire de $`F^{}`$ (pour la mรฉtrique hyperbolique). Cela induit une application,
$$\stackrel{~}{f_0}:H_1(\text{Tub}(F),\text{Z})H_1(\text{Tub}(F^{}),\text{Z})$$
et on appelle degrรฉ transverse de $`\stackrel{~}{f_0}`$ en $`F`$, lโimage par $`\stackrel{~}{f_0}_{}`$ du gรฉnรฉrateur de $`H_1(\text{Tub}(F),\text{R})`$; cette classe est un multiple entier de la classe fondamentale de $`H_1(\text{Tub}(F^{}),\text{Z})`$ et nous pouvons donc, par abus de langage, identifiรฉ le degrรฉ transverse ร un nombre entier relatif. On peut รฉgalement dรฉfinir ce degrรฉ en utilisant le cercle $`๐`$ tracรฉ sur $`\text{Tub}(F)`$ et un cercle $`๐^{}`$ analogue sur $`\text{Tub}(F^{})`$ sur lequel on projette $`\stackrel{~}{f_0}(๐)`$.
Soit $`\theta (F,s)`$ (resp. $`\theta ^{}(F^{},s^{})`$) lโangle diรฉdral (euclidien) du simplexe $`s๐ฏ_{\stackrel{~}{M}}`$ en la face $`F`$ (resp. du simplexe $`s^{}๐ฏ_{\stackrel{~}{X}}`$ en la face $`F^{}`$). Les nombres $`\theta `$ et $`\theta ^{}`$ sont choisis positifs. Lโapplication $`\stackrel{~}{f_0}`$ dโun simplexe $`s`$ sur un simplexe $`s^{}`$ peut prรฉserver ou renverser lโorientation (on rappelle que cette application est affine en restriction ร $`s`$) et nous poserons $`ฯต(s)=ฯต(s^{})=\pm 1`$ suivant le cas considรฉrรฉ.
###### Lemme 6.3
Soit $`F^{}`$ une face de codimension $`2`$ image de $`F`$, le degrรฉ transverse de $`\stackrel{~}{f_0}`$ en $`F`$, notรฉ $`\text{deg}_F\stackrel{~}{f_0}`$, vรฉrifie,
$$2\pi \text{deg}_F\stackrel{~}{f_0}=\pm \underset{s^{}/F^{}s^{}}{}ฯต(s^{})\theta ^{}(F^{},s^{}).$$
Preuve .
Pour $`F`$ telle que $`\stackrel{~}{f_0}(F)=F^{}`$ et $`s๐ฏ_{\stackrel{~}{M}}`$ tels que $`Fs`$, $`\stackrel{~}{f_0}(๐s)`$ se projette sur $`๐^{}s^{}`$ (oรน $`s^{}=\stackrel{~}{f_0}(s)`$) qui est un arc dโangle de valeur absolue $`\theta ^{}(F^{},s^{})`$. On peut choisir les orientations de $`\stackrel{~}{X}`$ et $`F^{}`$ sont telles que lโangle orientรฉ de la projection de $`\stackrel{~}{f_0}(๐s)`$ est $`+\theta ^{}(F^{},s^{})`$ si $`ฯต(S^{})=+1`$, et $`\theta ^{}(F^{},s^{})`$ si $`\stackrel{~}{f_0}`$ renverse lโorientation de $`s`$. La quantitรฉ $`\underset{s^{}/F^{}s^{}}{}ฯต(s^{})\theta ^{}(F^{},s^{})`$ reprรฉsente donc lโangle orientรฉ total de la projection de $`\stackrel{~}{f_0}(๐)`$ sur $`๐^{}`$, cโest-ร -dire $`2\pi \text{deg}_F^{}\stackrel{~}{f_0}`$. Si lโorientation de $`\stackrel{~}{X}`$ est renversรฉe la relation devient $`2\pi \text{deg}_F^{}\stackrel{~}{f_0}=\underset{s^{}/F^{}s^{}}{}ฯต(s^{})\theta ^{}(F^{},s^{})`$.
Considรฉrons alors une dรฉformation de $`\rho _0`$, soit $`\rho _t`$, que nous supposerons $`C^1`$ en $`t`$. Nous construisons lโapplication $`\stackrel{~}{f_t}`$ de la maniรจre suivante :
$`i=1,\mathrm{},N,`$ $`\stackrel{~}{f_t}(m_i)`$ $`=y_i`$
$`\gamma \mathrm{\Pi }_1(M),`$ $`\stackrel{~}{f_t}(\gamma m_i)`$ $`=\rho _t(\gamma )y_i`$
et ensuite on รฉtend $`\stackrel{~}{f_t}`$ de maniรจre affine dans chaque simplexe. La collection des simplexes images et leurs sommets varient de maniรจre $`C^1`$ en $`t`$. Nous noterons cette collection $`๐ฏ_{\stackrel{~}{X}}(t)`$. Tous les simplexes de $`๐ฏ_{\stackrel{~}{X}}(t)`$ sont non dรฉgรฉnรฉrรฉs, pour $`t`$ assez petit; en effet, il suffit de nโen considรฉrer quโun nombre fini, les autres sโen dรฉduisant par รฉquivariance. Notons รฉgalement que, par construction, $`\stackrel{~}{f_t}`$ dรฉpend de maniรจre $`C^1`$ en $`t`$, en particulier, le volume hyperbolique dโun simplexe de $`๐ฏ_{\stackrel{~}{X}}(t)`$ est une fonction $`C^1`$ de $`t`$. Soit $`F`$ une face de codimension $`2`$ de $`๐ฏ_{\stackrel{~}{M}}`$ et $`F^{}(t)`$ son image par $`\stackrel{~}{f_t}`$. Pour $`s๐ฏ_{\stackrel{~}{X}}(t)`$, nous noterons $`\theta ^{}(t;F^{},s^{})`$ lโangle (positif) diรฉdral de $`s^{}`$ en $`F^{}`$. Nous ne mentionnerons pas la dรฉpendance en $`t`$ des simplexes de $`๐ฏ_{\stackrel{~}{X}}(t)`$ et de leurs faces de codimension $`2`$ sโil nโy a pas dโambiguรฏtรฉ. Par ailleurs si $`s^{}(t)=\stackrel{~}{f_t}(s)`$ et $`t`$ est assez petit $`ฯต(s^{}(t))`$ ne dรฉpend pas de $`t`$.
###### Lemme 6.4
$$\frac{d}{dt}\left(\underset{s^{}/F^{}s^{}}{}ฯต(s^{})\theta ^{}(t;F^{},s^{})\right)=0$$
Preuve .Pour $`t`$ assez petit, la face $`F^{}(t)`$ est homรฉomorphe ร $`F^{}(0)`$; de mรชme $`s^{}(t)`$ est homรฉomorphe ร $`s^{}(0)`$ si $`s^{}(t)๐ฏ_{\stackrel{~}{X}}(t)`$ et ils sont tous non-dรฉgรฉnรฉrรฉs. Les voisinages tubulaires de $`F^{}(t)`$ et $`F^{}(0)`$ sont aussi homรฉomorphes et on peut dรฉfinir le degrรฉ transverse de $`\stackrel{~}{f_t}`$ grรขce ร $`F^{}(0)`$. Alors, par constance du degrรฉ par dรฉformation, pour $`t`$ assez petit, on a $`\text{deg}_{F^{}(t)}(\stackrel{~}{f_t})=\text{deg}_{F^{}(0)}(\stackrel{~}{f_0})`$.
Rappelons la formule de Schlรคfli (cf. ). Soit $`s^{}`$ un simplexe hyperbolique gรฉodรฉsique et $`F^{}`$ une de ses faces de codimension $`2`$; si $`s^{}(t)`$ est une dรฉformation de classe $`C^1`$ de $`s^{}=s^{}(0)`$, alors
$$\frac{d}{dt}_{|t=0}vol(s^{}(t))=\underset{F^{}s^{}}{}\frac{d}{dt}_{|t=0}(\theta ^{}(t;F^{}(t),s^{}(t))vol_{n2}(F^{}(t))$$
$`vol_{n2}`$ dรฉsigne le volume $`(n2)`$-dimensionnel de la face considรฉrรฉ.
Pour $`\overline{s}๐ฏ_M`$ choisissons un relevรฉ $`s๐ฏ_{\stackrel{~}{M}}`$; alors $`\stackrel{~}{f_t}`$ identifie de maniรจre $`๐^{\mathrm{}}`$ jusquโau bord $`s`$ avec un simplexe hyperbolique de $`๐ฏ_{\stackrel{~}{X}}`$. Lโรฉquivariance de $`\stackrel{~}{f_t}`$ permet de dรฉfinir de maniรจre unique une mรฉtrique hyperbolique sur $`\overline{s}`$ dont la collection produit une mรฉtrique $`\overline{g}(t)`$ sur $`M`$ qui est continue et hyperbolique par morceaux. En particulier le volume des faces de codimension $`2`$ et les angles diรฉdraux en celles-ci sont ceux du simplexe hyperbolique $`\stackrel{~}{f_t}(s)`$. Soit $`\omega `$ la forme volume hyperbolique de $`\stackrel{~}{X}`$, alors
$`vol(\rho _t)`$ $`={\displaystyle _M}\stackrel{~}{f_t}^{}(\omega )={\displaystyle \underset{\overline{s}๐ฏ_M}{}}{\displaystyle _{\overline{s}}}\stackrel{~}{f_t}^{}(\omega )`$
$`={\displaystyle \underset{\overline{s}๐ฏ_M}{}}ฯต(s)vol(\stackrel{~}{f_t}(s))={\displaystyle \underset{\overline{s}๐ฏ_M}{}}ฯต(\overline{s})vol(\overline{s},\overline{g}(t))`$
en dรฉfinissant $`ฯต(\overline{s})=ฯต(s)=ฯต(\stackrel{~}{f_t}(s))`$. Ici on a identifiรฉ, par abus de langage, $`\stackrel{~}{f_t}^{}(\omega )`$ avec une forme diffรฉrentielle sur $`M`$ grรขce ร lโรฉquivariance de $`\stackrel{~}{f_t}`$. La formule de Schlรคfli donne,
$`{\displaystyle \frac{d}{dt}}vol(\rho _t)`$ $`={\displaystyle \underset{\overline{s}๐ฏ_M}{}}{\displaystyle \frac{d}{dt}}(ฯต(\overline{s})vol(\overline{s},\overline{g}(t)))`$
$`={\displaystyle \underset{\overline{s}๐ฏ_M}{}}{\displaystyle \underset{\overline{F}\overline{s}}{}}ฯต(\overline{s}){\displaystyle \frac{d}{dt}}(\overline{\theta }(t;\overline{s},\overline{F}))vol_{n2}(\overline{F},\overline{g}(t))`$
$`\overline{\theta }(t;\overline{s},\overline{F})`$ dรฉsigne lโangle diรฉdral en $`\overline{F}`$ du simplexe $`\overline{s}`$ mesurรฉ ร lโaide de la mรฉtrique $`\overline{g}(t)`$. Il est รฉgal ร $`\theta ^{}(t;\stackrel{~}{f_t}(s),\stackrel{~}{f_t}(F))`$$`s`$ et $`F`$ sont des relevรฉs respectifs de $`\overline{s}`$ et $`\overline{F}`$.
$$\frac{d}{dt}vol(\rho _t)=\underset{\overline{F}}{}\left(\underset{\overline{s}/\overline{F}\overline{s}}{}ฯต(\overline{s})\frac{d}{dt}(\overline{\theta }(t;\overline{F},\overline{s}))\right)vol_{n2}(\overline{F},\overline{g}(t))$$
La quantitรฉ entre parenthรจse peut se calculer sur $`M`$ ou bien sur $`\stackrel{~}{M}`$ car elle ne concerne que lโรฉtoile dโune face $`\overline{F}`$; elle peut รฉgalement se calculer sur $`\stackrel{~}{X}`$ par dรฉfinition de $`\overline{g}(t)`$. Le lemme prรฉcรฉdent montre que, pour toute face $`\overline{F}`$,
$$\underset{\overline{s}\overline{F}}{}ฯต(\overline{s})\frac{d}{dt}(\overline{\theta }(t;\overline{F},\overline{s}))=0$$
Ce qui prouve que $`\frac{d}{dt}vol(\rho _t)=0`$.
Une consรฉquence immรฉdiate du thรฉorรจme 6.1 est le corollaire suivant. Notons $`(\mathrm{\Pi }_1(M),Isom(\stackrel{~}{X}))`$ lโespace des reprรฉsentations du groupe fondamental dโune variรฉtรฉ $`M`$ dans le groupe dโisomรฉtries de lโespace hyperbolique.
###### Corollaire 6.5
Soit $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe et orientรฉe, alors la fonctionnelle,
$$vol:(\mathrm{\Pi }_1(M),Isom(\stackrel{~}{X}))\text{R}^+$$
prend un nombre fini de valeurs.
Preuve .Le groupe dโisomรฉtries $`Isom(\stackrel{~}{X})=PO(n\mathrm{,1})`$ est un groupe algรฉbrique; par ailleurs, $`\mathrm{\Pi }_1(M)`$ est de prรฉsentation finie donc $`(\mathrm{\Pi }_1(M),Isom(\stackrel{~}{X}))`$ est une variรฉtรฉ algรฉbrique (avec singularitรฉs) et possรจde un nombre fini de composantes connexes. Le thรฉorรจme 6.1 affirme que la fonctionnelle $`vol`$ est constante sur chaque composante connexe.
Remarque.Ce rรฉsultat est รฉnoncรฉ dans , toutefois la preuve est incomplรจte sauf, peut-รชtre, en dimension $`3`$. Celle prรฉsentรฉe ci-dessus nous a รฉtรฉ suggรฉrรฉe par J.-P. Otal (voir ).
Considรฉrons alors les variรฉtรฉs hyperboliques fermรฉes de dimension $`n`$. Un thรฉorรจme de Wang affirme que, pour $`n4`$ et $`V>0`$ le nombre de variรฉtรฉs hyperboliques fermรฉes de volume infรฉrieur ร $`V`$ est fini. Ce rรฉsultat est notoirement faux en dimension $`3`$ et en dimension $`2`$. Si $`X`$ dรฉsigne une variรฉtรฉ hyperbolique fermรฉe et $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe, nous dirons (voir ) que $`M`$ domine $`X`$ sโil existe une application continue de degrรฉ non nul de $`M`$ sur $`X`$. Le thรฉorรจme 6.1 permet de donner une preuve trรจs simple du rรฉsultat suivant :
###### Thรฉorรจme 6.6 (T. Soma )
Soit $`M`$ une variรฉtรฉ diffรฉrentielle fermรฉe de dimension $`3`$, alors il nโexiste quโun nombre fini de variรฉtรฉs hyperboliques de dimension $`3`$ fermรฉes dominรฉes par $`M`$.
Preuve .Dรฉsignons par $`f:MX`$ lโapplication continue de degrรฉ non nul de $`M`$ sur $`X`$, oรน $`X`$ est une variรฉtรฉ hyperbolique fermรฉe. Lโapplication $`f`$ induit un morphisme $`f_{}:\mathrm{\Pi }_1(M)\mathrm{\Pi }_1(X)`$, cโest-ร -dire une reprรฉsentation $`\rho `$ de $`\mathrm{\Pi }_1(M)`$ dans $`PO(n\mathrm{,1})`$. Par dรฉfinition du degrรฉ de $`f`$ nous avons,
$$vol(\rho )=\text{deg}(f)vol(X).$$
Par ailleurs, si on munit $`M`$ dโune mรฉtrique riemannienne quelconque, le thรฉorรจme 4.2 montre que ce volume est bornรฉ par un nombre ne dรฉpendant que de $`M`$ (et de cette mรฉtrique) que nous noterons $`C(M)`$. Nous avons donc,
$$\text{deg}(f)vol(X)C(M)$$
cโest-ร -dire, $`\text{deg}(f)C(M)/vol(X)`$. Le volume dโune variรฉtรฉ hyperbolique compacte est bornรฉe infรฉrieurement par une constante universelle $`v_n`$ ne dรฉpendant que de la dimension $`n`$ grรขce au lemme de Margulis (voir ). En consรฉquence,
$$\text{deg}(f)C(M)/v_3.$$
Il nโy a donc quโun nombre fini de valeurs possibles pour le degrรฉ de lโapplication $`f`$. De mรชme $`\text{deg}(f)vol(X)=vol(\rho )`$ ne prend quโun nombre fini de valeurs dโaprรจs le corollaire 6.5 . Le volume des variรฉtรฉs $`X`$ fermรฉes dominรฉes par une variรฉtรฉ fermรฉe fixe $`M`$ ne peut donc prendre quโun nombre fini de valeurs ce qui, dโaprรจs une rรฉsultat de W. Thurston (), montre quโil ne peut y avoir quโun nombre fini de telles variรฉtรฉs.
Nous terminons en donnant un exemple de telles dรฉformations, montrant la pertinence du thรฉorรจme 6.1. Il nous a รฉtรฉ communiquรฉ par Daryl Cooper par lโintermรฉdiaire de Michel Boileau.
Exemple(D. Cooper) Soit $`N`$ une variรฉtรฉ hyperbolique fermรฉe de dimension $`3`$. Considรฉrons la somme connexe de $`N`$ avec $`S^1\times S^2`$, notรฉe $`N\mathrm{}(S^1\times S^2)`$, le groupe fondamental de cette variรฉtรฉ est le produit libre $`\mathrm{\Pi }_1(M)\text{Z}`$. Soit $`k`$ un noeud homotopiquement nul dans $`N\mathrm{}(S^1\times S^2)`$ qui rencontre $`S^1\times S^2`$ en au moins deux points. Dโaprรจs R. Myers () on peut trouver de tels noeuds en sorte quโune chirurgie de Dehn autour de $`k`$ transforme $`N\mathrm{}(S^1\times S^2)`$ en une variรฉtรฉ hyperbolique fermรฉe $`M`$ (voir aussi page 797). La proposition 3.2 de permet de construire une application continue $`f:MN\mathrm{}(S^1\times S^2)`$ de degrรฉ $`1`$. Par ailleurs il existe รฉgalement une application continue, $`h:N\mathrm{}(S^1\times S^2)N`$ de degrรฉ $`1`$ qui consiste ร รฉcraser $`S^1\times S^2`$ en un point. Nous obtenons donc une application continue de degrรฉ $`1`$,
$$hf:MN$$
et une reprรฉsentation $`\rho =h_{}f_{}:\mathrm{\Pi }_1(M)Pi_1(N)PO(\mathrm{3,1})`$. Le volume de cette reprรฉsentation est,
$$vol(\rho )=vol(N)>0$$
car $`hf`$ est degrรฉ $`1`$ et lโimage de $`\rho `$ est le groupe fondamental de $`N`$. Par ailleurs $`\rho `$ se dรฉcompose en,
$$\rho :\mathrm{\Pi }_1(M)\stackrel{f_{}}{}\mathrm{\Pi }_1(M)\text{Z}\stackrel{h_{}}{}PO(\mathrm{3,1}).$$
Le facteur libre Z permet alors de dรฉformer $`h_{}`$ sans contrainte et donc de produire des dรฉformations non triviales (ce fait est รฉlรฉmentaire et sa vรฉrification est laissรฉe au lecteur). En augmentant le nombre de facteurs $`S^1\times S^2`$ nous pouvons aisรฉment augmenter le nombre de paramรจtres disponibles pour dรฉformer $`\rho `$.
Remarque.Il serait intรฉressant de construire de telles dรฉformations en dimension supรฉrieure ou รฉgale ร $`4`$. Il est facile dโen construire de volume nul, mais des exemples de volume non nul restent ร dรฉcrire.
Gรฉrard BESSON
INSTITUT FOURIER
Laboratoire de Mathรฉmatiques
UMR5582 (UJF-CNRS)
BP 74
38402 St MARTIN DโHรRES Cedex (France)
G.Besson@fourier.ujf-grenoble.fr
Gilles COURTOIS
รCOLE POLYTECHNIQUE
Centre de mathรฉmatiques
UMR7640 (CNRS)
91128 PALAISEAU Cedex (France)
Courtois@math.polytechnique.fr
Sylvestre GALLOT
INSTITUT FOURIER
Laboratoire de Mathรฉmatiques
UMR5582 (UJF-CNRS)
BP 74
38402 St MARTIN DโHรRES Cedex (France)
Sylvestre.Gallot@fourier.ujf-grenoble.fr
|
warning/0506/astro-ph0506508.html
|
ar5iv
|
text
|
# Why is the Fast Solar Wind Fast and the Slow Solar Wind Slow? A Survey of Geometrical Models
## 1 Introduction
The intertwined nature of solar wind acceleration and the โcoronal heating problemโ has been known since Parker (1958) postulated a transonic flow solution made possible only by the high gas pressure of the corona. Mariner 2 confirmed the existence of a continuous supersonic solar wind in interplanetary space just a few years after Parkerโs initially controversial work (for a first-hand account of the discovery, see Neugebauer 1997). Mariner also showed that the wind exists in two relatively distinct states: slow (300โ500 km/s) and fast (600โ800 km/s). The slow component was initially believed to be the โambientโ background state (e.g., Hundhausen 1972), but it was eventually realized that the fast component was in general more quiet and steady (Feldman et al. 1976; Axford 1977). The polar passes of Ulysses in the 1990s confirmed this revised paradigm (Gosling 1996; Marsden 2001).
In the 1970s and 1980s it became increasingly evident that even the most sophisticated solar wind models could not produce a fast wind without the deposition of heat or momentum in some form into the corona (e.g., Hartle and Sturrock 1968; Holzer and Leer 1980). It also was realized that the geometry of the flowโi.e., whether the magnetic flux tubes were radially expanding cones or superradially flaring trumpetsโcould have a significant impact on the mass flux and wind speed. This paper briefly surveys geometry-related explanations for the observed distribution of solar wind speeds, and presents a prediction for coronal heating rates in fast vs. slow flux tubes as a consistency check on these ideas.
## 2 Coronal Source Regions
Even after several decades of ever-improving in situ and remote-sensing observations, there is still no universal agreement concerning the full range of coronal sources of the solar wind. It is clear that strong connections exist between large coronal holes and the highest-speed wind streams (Wilcox 1968; Krieger et al. 1973; Noci 1973; Zirker 1977; see, however, Habbal & Woo 2001). The more chaotic slow wind, though, may come from a multiplicity of source regions. Two regions that are frequently cited as sources of slow wind are: (1) boundaries between coronal holes and large streamers that undergo strong superradial expansion in the corona, and (2) narrow plasma sheets that extend out from the tops of streamer cusps (Wang et al. 2000; Strachan et al. 2002). However, during active phases of the solar cycle, there is evidence that slow wind also emanates from small coronal holes (e.g., Nolte et al. 1976; Neugebauer et al. 1998) and active regions (Hick et al. 1995; Liewer et al. 2004). During the rising phase of solar activity, there seems to be an abrupt ($`<6`$ month) change in the magnetic connectivity between field lines in the in situ ecliptic plane and the Sun (see Figure 5 of Luhmann et al. 2002). At minimum, a large fraction of these field lines map into the high-latitude northern and southern polar hole/streamer boundaries, but at maximum nearly all the field lines map into low-latitude active regions and small coronal holes. The majority of the most recent transition time was not observed by SOHO because of its 4-month mission interruption in 1998.
The remainder of this paper is concerned mainly with the dichotomy between high-speed wind that emerges from the central regions of large coronal holes and low-speed wind that emerges from the hole/streamer boundary regions. More work is needed to apply the ideas presented below to other potential slow-wind source regions (e.g., strong-field flux tubes rooted in or near active regions; see Wang 1994).
## 3 โGeometry is Destiny?โ
There is a strong empirical relationship between the solar wind speed $`u`$ measured in situ and the inferred lateral expansion of magnetic flux tubes near the Sun. Levine et al. (1977) and Wang & Sheeley (1990) found that the asymptotic wind speed is inversely correlated with the amount of transverse flux-tube expansion between the solar surface and a reference point in the mid-corona (see also Arge & Pizzo 2000; Poduval & Zhao 2004; Fujiki, these proceedings). As illustrated in Figures 1a and 1b, the field lines in the central regions of coronal holes undergo a relatively slow and gradual rate of superradial expansion, but the more distorted field lines at the hole/streamer boundaries undergo more rapid expansion. It should be noted, though, that the eventual flux tube expansion (i.e., between the Sun and $`r\mathrm{}`$) for polar coronal holes is likely to exceed that of the streamer edges, despite the opposite trend seen when the expansion factor $`f`$ is measured between $`R_{}`$ and a coronal source surface.
Several potential explanations for the observed anticorrelation between wind speed and flux-tube expansion have been proposed (see ยง 4). However, it is worthwhile to begin examining such a relationship from the standpoint of the the equation of momentum conservation along a solar wind flux tube:
$$\left(u\frac{a_{}^2}{u}\right)\frac{du}{dr}=\frac{dF}{dr}$$
(1)
where, for a plasma dominated by protons and electrons, the effective one-fluid most-probable speeds are defined as $`a_/^2=k_\mathrm{B}(T_{p/}+T_e)/m_p`$ and collisions and external sources of momentum are neglected. The function $`F(r)`$ appearing on the right-hand side is defined as
$$F(r)\frac{GM_{}}{r}a_{}^2+_R_{}^r๐r^{}a_{}^2\left(\frac{2}{r^{}}+\frac{1}{f}\frac{df}{dr^{}}\right)$$
(2)
and $`f(r)`$ is the dimensionless flux-tube expansion factor (which is proportional to $`B^1r^2`$ measured along a flux tube; see also Kopp & Holzer 1976).
Local extrema in $`F(r)`$ satisfy the Parker (1958) critical point condition. Vรกsquez et al. (2003) found that only the global minimum in $`F(r)`$ gives a sonic/critical point location that allows a consistent and continuous solution for $`u(r)`$ over the full range of distances from the Sun to 1 AU. For monotonically increasing expansion factors like those over the poles, $`F(r)`$ tends to exhibit a single minimum in the low corona ($`r2R_{}`$). For streamer-like expansion factors that peak near the cusp, another minimum in $`F(r)`$ appears at a height well above the cusp; this new point tends to be the global minimum. The latter kind of flux tubeโi.e., one that allows a more distant critical point radiusโseems to correspond directly to the slow-speed wind measured in situ (see also Wang 1994; Bravo & Stewart 1997; Chen & Hu 2002).
Figure 1c shows the radial locations of minima in $`F(r)`$ along individually mapped flux tubes that range from the pole to the edge of the streamer belt (see corresponding labels A $``$ D in the other panels). Eq. (2) was solved using the magnetic field model of Banaszkiewicz et al. (1998) and an isothermal corona ($`T_p=T_p=T_e=`$ 1.75 MK) for simplicity. The outer critical point appears only for field lines having latitudes at $`r\mathrm{}`$ less than about 23 above and below the equator. In more physically realistic models that include radial and latitudinal temperature variations (e.g., Vรกsquez et al. 2003), the outermost minimum in $`F(r)`$ is the global minimum, and thus as one moves from the centers of coronal holes to their edges, the critical point moves outwards abruptly from $`<2R_{}`$ to 3โ6 $`R_{}`$ at a latitude still rather far removed from the streamer cusp.
## 4 Heating Above & Below the Critical Point
Why does the height of the critical point matter? Physically, the critical or singular point (equivalent to the sonic point for a hydrodynamic pressure-driven wind) is the location where the subsonic (i.e., nearly hydrostatic) coronal atmosphere gives way to the kinetic-energy-dominated supersonic flow.<sup>1</sup><sup>1</sup>1The idealized Parker critical point loses some of its mathematical importance when solving the time-dependent momentum equation (e.g., Suess 1982) or when including the effects of viscosity (Axford & Newman 1967). However, the critical transonic โbranchโ remains the robust stable time-steady solution in nearly all models with varying levels of sophistication (e.g., Holzer & Leer 1997; Velli 2001). Whether the critical point lies above or below the regions where most of the energy deposition occurs is a key factor in determining the nature of the wind:
1. If substantial heating occurs in the subsonic corona, its primary impact is to โpuff upโ the scale height, drawing more particles into the accelerating wind and thus increasing the mass flux. Roughly, the increase in energy flux due to the heating can be balanced by the increase in mass flux, so that the eventual kinetic energy per particle is relatively unaffected and the wind speed may not change (relative to an unheated model). In some scenarios the mass flux increase can be stronger than the energy flux increase, and the asymptotic wind speed decreases.
2. If substantial heating occurs in the supersonic corona, the subsonic temperature is unaffected and the mass flux is unchanged. The local increase in energy flux has nowhere else to go but into the kinetic energy of the wind, and the flow speed increases.
(Leer & Holzer 1980; Pneuman 1980; Leer et al. 1982). The above dichotomy is often modeled by changing the height at which the bulk of the energy is deposited, but it can also occur if the heating remains the same and the height of the critical point changes (as discussed in ยง 3).
A natural link can be made between geometry-related changes in the flow topology and the heating-related changes in the wind. Wang & Sheeley (1991) proposed that the observed anticorrelation between $`u`$ and $`f`$ is a by-product of equal amounts of Alfvรฉn wave flux emitted at the bases of all flux tubes (see also earlier work by Kovalenko 1978, 1981). Near the Sun, the Alfvรฉn wave flux $`F_A`$ is proportional to $`\rho V_A\delta V_{}^2`$. The density dependence in the product of Alfvรฉn speed $`V_A`$ and the squared Alfvรฉn wave amplitude $`\delta V_{}^2`$ cancels almost exactly with the linear factor of $`\rho `$ in the wave flux, thus leaving $`F_A`$ proportional mainly to the radial magnetic field strength $`B`$. The ratio of $`F_A`$ at the critical point to its value at the photosphere thus scales as the ratio of $`B`$ at the critical point to its value at the photosphere. The latter ratio of field strengths is proportional to $`1/f`$, where $`f`$ is the coronal expansion factor as defined by Wang and Sheeley. For equal wave fluxes at the photosphere for all regions, coronal holes (with low $`f`$) will thus have a larger flux of Alfvรฉn waves at and above the critical point compared to streamers (that have high $`f`$).
To summarize, for streamers \[coronal holes\], more of the Alfvรฉnic energy flux should be deposited below \[above\] the critical point. This effect is complementary to the change in height of the critical point discussed above; i.e., for streamers \[holes\] the critical point is high \[low\].
## 5 Turbulent Heating: fast vs. slow
One aspect of the Wang/Sheeley/Kovalenko hypothesis that needs further clarification is the link between an increased Alfvรฉn wave flux and increased coronal heating. Alfvรฉn waves can exert a dissipationless wave-pressure force that can accelerate the wind (e.g., Isenberg & Hollweg 1982), but their ability to heat the plasma is less well understood. One idea that has received much recent attention is that low-frequency Alfvรฉn waves can be damped in the corona by undergoing a turbulent cascade from large to small scales. Here we present an empirically constrained model of Alfvรฉnic turbulence and predict the contrast in extended heating that occurs between a polar coronal hole flux tube and a near-equatorial streamer edge flux tube.
Cranmer & van Ballegooijen (2005) built a comprehensive model of MHD turbulence in a polar coronal hole flux tube. This model follows the radial evolution of the power spectrum of non-WKB Alfvรฉn waves (i.e., waves propagating both outwards and inwards along the flux tube) from the photosphere to 4 AU, and allows the turbulent energy injection rate (and thus the heating rate) to be derived as a function of height. The Alfvรฉn waves have their origin in the transverse shaking of strong-field ($``$1500 G) thin flux tubes in the photosphere. The bottom boundary condition on the wave power spectrum was derived from measurements of G-band bright point motions in the photosphere (e.g., Nisenson et al. 2003). Below the mid-chromosphere, where the bright-point flux tubes are isolated and thin, the linear wave properties are computed using a generalized form of the kink-mode wave equations derived by Spruit (1981). Above the mid-chromosphere, where the flux tubes have merged into a more homogeneous network โfunnelโ (e.g., Tu et al. 2005), the wind-modified non-WKB transport equations of Heinemann & Olbert (1980) are solved.
Figure 2 shows a summary of results from the published coronal-hole model (corresponding to $`\theta =0`$ in Figure 1) and the new streamer-edge model (corresponding to field line D). Below the transition region, the coronal-hole and streamer-edge models were assumed to be identical except for the mass flux. Figure 2a plots the adopted wind speed for both flux tubes, as constrained by mass flux conservation, empirical electron densities (e.g., Sittler & Guhathakurta 1999) and the Banaszkiewicz et al. (1998) magnetic field model, modified by including the thin tubes and funnels at low heights. Also shown are the frequency-integrated Alfvรฉn wave amplitudes $`\delta V_{}`$ as computed from the non-WKB wave transport equations, with turbulent damping as described below.
Figure 2b gives the derived heating rateโexpressed as the energy flux density $`Q`$ per unit mass density $`\rho `$โfor the two flux tube models. The slowly-varying ratio $`Q/\rho `$ is plotted for convenience, because $`Q`$ itself drops by more than 15 orders of magnitude between the transition region ($`r1.003R_{}`$) and 1 AU. In some ways, though, the plot is deceiving because it seems as if the streamer flux tube is heated more than the polar-hole model only below $`r1.02R_{}`$. In fact, $`Q_{\mathrm{streamer}}>Q_{\mathrm{hole}}`$ everywhere below $`r1.4R_{}`$, but this is masked in the plot because $`\rho _{\mathrm{streamer}}>\rho _{\mathrm{hole}}`$. The heating rate is assumed to be equal to the energy cascade rate that has been derived for anisotropic MHD turbulence:
$$Q=\rho \frac{Z_{}^2Z_++Z_+^2Z_{}}{4\mathrm{}_{}}$$
(3)
(Hossain et al. 1995; Matthaeus et al. 1999; Dmitruk et al. 2001, 2002), where $`\mathrm{}_{}`$ is a transverse outer-scale correlation length and Elsasser (1950) variables are used to distinguish between outwardly propagating waves ($`Z_{}`$) and inwardly propagating waves ($`Z_+`$), with $`Z_\pm \delta V\pm \delta B/\sqrt{4\pi \rho }`$. The correlation length is assumed to expand with the transverse width of the flux tube (i.e., $`\mathrm{}_{}^2B=\mathrm{const};`$ see Hollweg 1986), and its normalization is specified by Cranmer & van Ballegooijen (2005).
Figure 2c shows the result of integrating a one-fluid internal energy equation (e.g., eq. 3 of Leer & Holzer 1980) to compute the mean temperatures $`T(r)`$ that result from the heating rates discussed above. These results should be interpreted as preliminary because: (1) they are not the result of a self-consistent calculation of all fluid variables, and (2) a simple choice was made for the electron heat flux (i.e., $`q_e=q_{\mathrm{SH}}/10`$, where $`q_{\mathrm{SH}}`$ is the classical Spitzer-Hรคrm value) in order to split the difference between the known strong conduction at low heights and the collisionless inhibition of $`q_e`$ at large heights. However, the overall trend in Figure 2c (i.e., the streamer being heated more at low heights, and less above the critical point, than the coronal hole) is likely to remain valid when the above approximations are corrected.
Several observations are also shown in Figure 2c, and the general agreement in the hole/streamer contrast lends credence to the overall validity of the approach outlined in this work. UVCS/SOHO measured H I Ly$`\alpha `$ resonance line profiles in the extended corona which provide a probe of proton velocity distributions. For off-limb observations, the line of sight samples directions that are mainly perpendicular to the $``$radial field lines, and the $`1/e`$ line width $`V_{1/e}`$ arises from two primary types of motion:
$$V_{1/e}^2=\frac{2k_\mathrm{B}T_p}{m_p}+\delta V_{}^2.$$
(4)
The two terms on the right represent random thermal motions and unresolved transverse wave motions. Using measured values of $`V_{1/e}`$ and the modeled values of $`\delta V_{}`$ for the waves, the above equation was solved for the plotted $`T_p`$. The UVCS curve for a solar-minimum coronal hole (dark gray) comes from an amalgam of data from Kohl et al. (1997), Cranmer et al. (1999), Esser et al. (1999), Zangrilli et al. (1999), and Antonucci et al. (2000). The UVCS streamer data (light gray) came from Kohl et al. (1997) and Strachan et al. (2004). Also shown in Figure 2c are Helios data points which were computed by averaging together proton (Marsch et al. 1982) and electron (Pilipp et al. 1990) measurements, and isotropizing from bi-Maxwellian fits; i.e., taking $`T(T_{}+2T_{})/3`$.
When examining predictions for the plasma temperatures in fast vs. slow solar wind, it is worthwhile to compare with different, but potentially complementary ideas. Recently, Fisk (2003) and Schwadron & McComas (2003) discussed the origins of correlations between the eventual wind speed and observed properties of emerging loops in the low corona (see also the related footpoint diffusion model of Fisk & Schwadron 2001). Their prediction for there to be more basal coronal heating (and a higher mass flux) in the slow wind seems to be in accord with the results shown in Figure 2 (see also Matthaeus et al., these proceedings). There still seems to be a disconnect, though, between theories of coronal heating via flux emergence and theories that invoke magnetic footpoint shaking (which in turn generates waves). The relative contributions of these processes in various coronal regions needs to be quantified further.
## 6 Proton vs. electron heating
The above analysis did not account explicitly for differences between heating the various particle species in the plasma. Even in a perfectly โcollisionally coupledโ plasma, there can be macroscopic dynamical consequences depending on how the energy is deposited into protons, electrons, and possibly heavy ions as well.
Hansteen & Leer (1995) demonstrated these effects for a 1D solar wind model: when all of the heat goes into electrons, there is substantially more downward conduction compared to a proton-heated model. An electron-heated wind thus has a lower mid-corona temperature and a lower wind speed than a proton-heated wind. A 2D simulation of streamers by Endeve et al. (2004) showed that the stability of closed-field regions is closely related to this kinetic partitioning of heat. When protons are heated strongly, the modeled streamers become unstable to the ejection of massive plasmoids; when the electrons are heated, the streamers are stable. Any sufficiently predictive model of fast and slow solar wind must take these effects into account (see also Cranmer & van Ballegooijen 2003).
## 7 Conclusions and Future Missions
Our understanding of the dominant physics of solar wind acceleration has progressed rapidly in the SOHO era, and many of the insights embedded in the above analysis would not have been possible without SOHO. In particular, the strong preferential heating and acceleration of heavy ions seen in coronal holes by UVCS has sharpened theoretical efforts to understand kinetic energy deposition in the collisionless extended corona (see reviews by Axford et al. 1999; Hollweg & Isenberg 2002; Cranmer 2002a; Marsch 2004).
Despite these advances, the diagnostic capabilities of the SOHO instruments were limited and the most fundamental questions have not yet been answered. If the kinetic properties of additional ions were to be measured in the extended corona (i.e., a wider sampling of charge/mass combinations) we could much better constrain the specific kinds of waves that are present as well as the specific collisionless damping modes (Cranmer 2001, 2002b). Measuring the coronal electron temperature above $``$1.5 $`R_{}`$ (never done directly before) would allow us to determine the bulk-plasma heating rate in different solar wind structures, thus putting the firmest ever constraints on models of why the fast/slow wind is fast/slow. Measuring non-Maxwellian velocity distributions of electrons and positive ions would allow us to test specific models of MHD turbulence, cyclotron resonance, and velocity filtration. New capabilities such as these would be enabled by greater photon sensitivity, an expanded wavelength range, and the use of measurements that heretofore have only been utilized in a testing capacity (e.g., Thomson-scattered H I Ly$`\alpha `$ to obtain $`T_e`$). Spectroscopy is key for the above measurementsโespecially in combination with coronagraph occultationโin order to measure detailed plasma properties out into the windโs acceleration region (see also Kohl et al., these proceedings).
This work is supported by NASA under grants NAG5-11913, NAG5-10996, NNG04GE77G, and NNG04GE84G to the Smithsonian Astrophysical Observatory, by Agenzia Spaziale Italiana, and by the Swiss contribution to ESAโs PRODEX program.
|
warning/0506/hep-th0506076.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
A tachyon of closed string theory is said to be a bulk tachyon if it lives throughout spacetime. In the presence of a tachyon one has an instability and two important and related questions arise:
1. Is there a ground state of the theory without the instability ?
2. What is the end-result of the physical decay process associated with the instability ?
The answers are presently known for the open string theory tachyons that live on the world-volume of unstable D-branes . The ground state, or tachyon vacuum, is a state without the D-brane and without open strings โ it is in fact the vacuum state of closed strings. In the associated physical decay process the D-brane dissappears but the result is not quite the closed string vacuum but rather an excited state of closed strings that carries the original energy of the D-brane. The decay process is not simply a transition from the unstable to the stable vacuum.
The purpose of the present paper is to study the physical decay induced by the bulk closed string tachyon of bosonic closed string theory โ the second question above applied to bulk tachyons (for localized closed string tachyons, see .). Our work was prompted by new information about the first question: recently-found evidence that the closed string field theory โtachyon potentialโ has a critical point โ a candidate for a closed string tachyon vacuum . A set of considerations suggests that in such vacuum closed string states would not propagate and spacetime would cease to be dynamical. Our analysis of the physical decay aims to illuminate the nature of the tachyon vacuum. This may be possible because the physical decay turns out to be rather insensitive to the specific details of the tachyon potential, about which little is known.
We study here the low-energy field equations that couple the metric, the dilaton, and the tachyon. These equations are motivated by the conditions of conformal invariance of sigma models and are expected to provide solutions that capture relevant features of exact string theory solutions. The low-energy field equations have been used in many papers to study all kinds of dynamical and cosmological issues (for a review and references see ). Few of these works, however, deal with key features of our present problem: a standard minimal coupling of the dilaton to other fields, an unstable closed string vacuum with zero cosmological constant, a tachyon potential that is not positive, and a rolling process induced by the tachyon. The related problem of light (bulk) tachyons that arise from circle compactification has been studied by Dine et.al. and Suyama . These authors compute quadratic and quartic terms in the tachyon potential and consider a cosmological evolution that involves the metric, the dilaton, the radion, and the tachyon. The authors of state that numerical studies show that rather general initial conditions lead to a radius that evolves to make the tachyon more tachyonic and a dilaton that evolves to make the system strongly coupled (the simplicity of a light tachyon seems to be illusory). The author of freezes the radion and discusses explicitly the simplified system, showing in a numerical solution how it appears to be driven to strong coupling.
Our analysis assumes arbitrary tachyonic potentials $`V(T)=\frac{1}{2}m^2T^2+๐ช(T^3)`$ and reveals a few surprises. We have found that if the rolling process is triggered by the tachyon the string metric does not evolve. Moreover, the dilaton expectation value $`\mathrm{\Phi }`$ will always increase as time goes by. If the tachyon history $`T(t)`$ is such that the potential is negative, $`V(T(t))0`$, the evolution reaches a singular point in finite time: both $`\dot{T}`$ and $`\dot{\mathrm{\Phi }}`$ become infinite, and so do $`T`$ and $`\mathrm{\Phi }`$. In the string frame this is a system with infinite string coupling, while in the Einstein frame the universe undergoes a big crunch. While negative potentials help accelerate its occurrance, a crunch occurs at finite time (both in the string and Einstein frames) for a wide class of potentials that grow arbitrarily large and positive for large $`T`$, $`V(T)=T^2+T^4`$, for example. The growing dilaton acts on the tachyon like anti-friction, a force proportional to the tachyon velocity in the direction of the velocity. This generally enables the tachyon to reach infinite value in finite time, even if it has to climb an infinite potential.
This paper is organized as follows. In Section 2 we write the relevant coupled equations and examine them in the cosmological setting. We use the string metric and emphasize how the dilaton time derivative plays a role similar to that of minus the Hubble parameter $`H(t)`$. In Section 3 we define tachyon-induced rolling and show that it results in a constant string metric. The rolling problem simplifies considerably and becomes the coupled dynamics of a dilaton and a tachyon. Analytic solutions are possible if one can solve a certain first-order nonlinear differential equation. Up to numerical constants, the dilaton-tachyon equations can be mapped to those that describe a single scalar field rolling in Einsteinโs theory. In Section 4 we establish that the Einstein metric crunches in finite time if the tachyon potential is negative throughout the rolling solution. In Section 5 we consider potentials that can be positive and develop tools to decide if there is a big crunch and if it occurs in finite time. Conclusions are offered in Section 6.
## 2 The coupled system of rolling fields
Consider the action that describes the low-energy dynamics of the metric, the dilaton, and the tachyon:
$$S=\frac{1}{2\kappa ^2}d^{d+1}x\sqrt{g}e^{2\mathrm{\Phi }}\left(R+4(_\mu \mathrm{\Phi })^2(_\mu T)^22V(T)\right).$$
(2.1)
Here $`g_{\mu \nu }`$ is the string metric, $`\mathrm{\Phi }`$ is the dilaton, and $`T`$ is the tachyon, with potential $`V(T)`$. The number of spatial dimensions is $`d`$. We are following the conventions of , with their dilaton $`\varphi `$ replaced by $`(2\mathrm{\Phi })`$. The metricโdilaton part of the action is that in . The equations of motion are:
$$\begin{array}{cc}\hfill R_{\mu \nu }+2_\mu _\nu \mathrm{\Phi }(_\mu T)(_\nu T)& =0,\hfill \\ \hfill ^2T2(_\mu \mathrm{\Phi })(^\mu T)V^{}(T)& =0,\hfill \\ \hfill ^2\mathrm{\Phi }2(_\mu \mathrm{\Phi })^2V(T)& =0.\hfill \end{array}$$
(2.2)
To evaluate the action on-shell we multiply the first equation by $`g^{\mu \nu }`$, use the third equation to eliminate $`^2\mathrm{\Phi }`$, and find that $`R+4(_\mu \mathrm{\Phi })^2(_\mu T)^2=2V(T)`$. Using this,
$$S_{onshell}=\frac{1}{2\kappa ^2}d^{d+1}x\sqrt{g}e^{2\mathrm{\Phi }}\left(4V(T)\right).$$
(2.3)
We look for solutions of (2.2) that represent a rolling tachyon field $`T(t)`$ accompanied by a time dependent dilaton $`\mathrm{\Phi }(t)`$ and a time dependent string metric of the form
$$ds^2=(dt)^2+a^2(t)\left(dx_1^2+dx_2^2+\mathrm{}dx_d^2\right),H(t)\frac{\dot{a}(t)}{a(t)}.$$
(2.4)
With this metric, the gravitational equations of motion (first line in (2.2)) give two equations
$$d\frac{\ddot{a}}{a}+\dot{T}^22\ddot{\mathrm{\Phi }}=0,\frac{\ddot{a}}{a}+(d1)\left(\frac{\dot{a}}{a}\right)^22\frac{\dot{a}}{a}\dot{\mathrm{\Phi }}=0.$$
(2.5)
The equations of motion for the dilaton and the tachyon are:
$`\ddot{\mathrm{\Phi }}+\left(dH2\dot{\mathrm{\Phi }}\right)\dot{\mathrm{\Phi }}+V(T)`$ $`=`$ $`0,`$ (2.6)
$`\ddot{T}+\left(dH2\dot{\mathrm{\Phi }}\right)\dot{T}+V^{}(T)`$ $`=`$ $`0.`$ (2.7)
We recognize the familiar Hubble โfrictionโ term that couples $`H`$ to the field velocity. Indeed, for $`H>0`$ the force is opposite to the velocity and slows down the field. Similarly, the dilaton velocity $`\dot{\mathrm{\Phi }}`$ is anti-friction: if $`\dot{\mathrm{\Phi }}>0`$ the force is in the direction of the velocity and accelerates the field. The dilaton is driven by $`V(T)`$; it will tend to go to strong coupling while $`V(T)<0`$.
The gravity equations (2.5) can be rearranged into two equivalent equations:
$`\frac{1}{2}(d1)\dot{H}`$ $`=`$ $`\frac{1}{2}\dot{T}^2+\ddot{\mathrm{\Phi }}H\dot{\mathrm{\Phi }},`$ (2.8)
$`\frac{1}{2}d(d1)H^2`$ $`=`$ $`\frac{1}{2}\dot{T}^2\ddot{\mathrm{\Phi }}+dH\dot{\mathrm{\Phi }}.`$ (2.9)
It can be shown that if equation (2.9) holds at some time, equations (2.7), (2.6), and (2.8) guarantee that it holds for all times.
It is instructive to compare of the previous equations with those that govern the dynamics of a scalar field $`\varphi `$ with potential $`V(\varphi )`$ coupled to gravity without a dilaton:
$`\frac{1}{2}(d1)\dot{H}`$ $`=`$ $`\frac{1}{2}\dot{\varphi }^2,`$ (2.10)
$`\frac{1}{2}d(d1)H^2`$ $`=`$ $`\frac{1}{2}\dot{\varphi }^2+V(\varphi ),`$ (2.11)
$`\ddot{\varphi }+dH\dot{\varphi }+V^{}(\varphi )`$ $`=`$ $`\mathrm{\hspace{0.17em}\hspace{0.17em}0}.`$ (2.12)
Note that $`\dot{H}\dot{\varphi }^2<0`$, which means decelerating expansion or accelerating contraction. On the other hand, the analogous equation in the presence of a dilaton, (2.8), allows the possibility that $`\dot{H}`$ vanishes. Equation (2.11) is analogous to (2.9). Comparison of (2.12) with (2.7) confirms that the rolling scalar is only affected by the addition of the dilaton-induced anti-friction ($`\dot{\mathrm{\Phi }}>0`$).
The Einstein metric $`g_{\mu \nu }^E`$ is determined by the string metric and the dilaton: $`g_{\mu \nu }^E=\mathrm{exp}(\frac{4}{d1}\mathrm{\Phi })g_{\mu \nu }`$. For a fixed string metric, the Einstein metric goes to zero if the dilaton expectation value goes to infinity. This corresponds to infinite string coupling.
## 3 Tachyon-driven rolling and the string metric
We now consider a general class of potentials $`V(T)`$ for a tachyon $`T`$ that satisfy the condition $`V(0)=0`$ and can be written as
$$V(T)=\frac{1}{2}m^2T^2+๐ช(T^3).$$
(3.13)
We build a solution where $`T0`$ for $`t\mathrm{}`$, and the field rolls to positive values:
$$T(t)=e^{mt}+\underset{n2}{}t_ne^{nmt}.$$
(3.14)
The first term in this ansatz is the solution to the linearized tachyon equation of motion. The arbitrary constant multiplying this term can be absorbed, as we did, by a redefinition of time. The exponentials in the sum are subleading to $`e^{mt}`$ for large negative $`t`$. We say that the tachyon drives the rolling if the other fields, in this case $`H(t)`$ and $`\mathrm{\Phi }(t)`$, have solutions with exponentials subleading to $`e^{mt}`$:
$$\mathrm{\Phi }(t)=\underset{n2}{}\varphi _ne^{nmt},H(t)=\underset{n2}{}h_ne^{nmt}.$$
(3.15)
Given (3.14), the dilaton equation (2.6) gives
$$\mathrm{\Phi }(t)=\frac{1}{8}e^{2mt}+๐ช(e^{3mt}).$$
(3.16)
This leading behavior is valid for all potentials of the form (3.13). The dilaton begins to run towards stronger coupling. Evaluating the right-hand side of equation (2.8) we see that
$$\dot{T}^2+2\ddot{\mathrm{\Phi }}=m^2e^{2mt}+2\frac{1}{8}(4m^2)e^{2mt}+๐ช(e^{3mt})=0e^{2mt}+๐ช(e^{3mt}).$$
(3.17)
Since the other term on the right-hand side, $`H\dot{\mathrm{\Phi }}e^{4mt}`$, we deduce that $`\dot{H}e^{3mt}`$ and therefore the contribution of order $`e^{2mt}`$ to $`H`$ vanishes: $`h_2=0`$. The string metric is not affected to this order. This is actually the beginning of a pattern: we now prove that $`H(t)`$ vanishes identically for tachyon-induced rolling. Adding equations (2.8) and (2.9) we find
$$\dot{H}=(dH2\dot{\mathrm{\Phi }})H.$$
(3.18)
Now assume that $`h_2=h_3=\mathrm{}=h_N=0`$ for some $`N2`$. Since $`\dot{\mathrm{\Phi }}e^{2mt}`$, the above equation gives $`\dot{H}e^{(N+3)mt}`$, which implies that $`h_{N+1}=0`$. By induction, $`H(t)`$ vanishes identically.
We now reconsider the equations of motion with $`H=0`$. The gravitational equations (2.8) and (2.9) give a single equation, $`\ddot{\mathrm{\Phi }}=\frac{1}{2}\dot{T}^2`$. Additionally, we have the equations of motion (2.7) and (2.6). With small rearrangements, the equations are:
$`\ddot{\mathrm{\Phi }}`$ $`=`$ $`\frac{1}{2}\dot{T}^2,`$ (3.19)
$`2\dot{\mathrm{\Phi }}^2`$ $`=`$ $`\frac{1}{2}\dot{T}^2+V(T),`$ (3.20)
$`\ddot{T}2\dot{\mathrm{\Phi }}\dot{T}+V^{}(T)`$ $`=`$ $`0.`$ (3.21)
Since $`\ddot{\mathrm{\Phi }}0`$, the dilaton velocity $`\dot{\mathrm{\Phi }}(t)`$ never decreases. Given that $`\dot{\mathrm{\Phi }}(t)>0`$ for sufficiently early times (see (3.16)), the dilaton $`\mathrm{\Phi }(t)`$ increases without bound. If the evolution is regular, $`\mathrm{\Phi }\mathrm{}`$ as $`t\mathrm{}`$ (the universe takes infinite time to crunch). More generally, the evolution produces a singular point at some finite time for which, as we shall see, both $`\dot{\mathrm{\Phi }}`$ and $`\mathrm{\Phi }`$ become infinite. Note also the complete correspondance between the above equations and equations (2.10), (2.11), and (2.12) for an ordinary scalar coupled to gravity. The sets of equations match, up to constants, when we set $`H\dot{\mathrm{\Phi }}`$. Out of the three equations above, the last two suffice. Taking the time derivative of (3.20) and using (3.21), we find that (3.19) holds as long as $`\dot{\mathrm{\Phi }}0`$. The rolling of ordinary scalars with negative potentials was studied by Felder et.al., who noted that the final state is roughly independent of the shape of the potential. Given the correspondance with dilaton/tachyon rolling, this is also true in our problem.
We derived the final equations (3.21) and (3.20) using a class of initial conditions that implied $`H=0`$. These equations, viewed as the original equations with the ansatz $`H=0`$, allow more general initial conditions. For an initial time $`t_i`$ we can take arbitrary $`T(t_i)`$ and $`\dot{T}(t_i)`$ as long as
$$(\dot{T}^2+2V(T))|_{t_i}0.$$
(3.22)
The evolution is fixed by choosing a square-root branch for $`\dot{\mathrm{\Phi }}`$ in (3.20). Since $`\dot{\mathrm{\Phi }}`$ is positive for tachyon-driven rolling, we take
$$2\dot{\mathrm{\Phi }}=\sqrt{\dot{T}^2+2V(T)}.$$
(3.23)
This enables us to rewrite (3.21) as a second-order nonlinear differential equation for the tachyon alone, an equation that is quite convenient for numerical integration:
$$\ddot{T}\sqrt{\dot{T}^2+2V(T)}\dot{T}+V^{}(T)=0.$$
(3.24)
The general rolling problem can be reduced to the problem of solving a first-order nonlinear differential equation. For this we consider the โenergyโ $`E`$ defined as
$$E^2\frac{1}{2}\dot{T}^2+V(T).$$
(3.25)
One readily checks that
$$\frac{dE}{dt}=\dot{T}\dot{T}(2\dot{\mathrm{\Phi }}).$$
(3.26)
Since $`\dot{\mathrm{\Phi }}>0`$, $`E`$ can only increase. The desired equation arises by rewriting (3.26) as
$$\frac{dE}{dt}=\pm \sqrt{EV}\frac{dT}{dt}\mathrm{\hspace{0.17em}2}\sqrt{E}\frac{dE}{dT}=\pm 2\sqrt{E(EV)}.$$
(3.27)
This is an equation for $`E(T)`$. The sign choice arises from solving for $`\dot{T}`$ in terms of $`E`$ and $`V`$. During evolution the sign must be changed each time $`\dot{T}`$ goes through zero. We will use the above mostly when $`\dot{T}>0`$, so we will take the plus sign. The equation becomes a little simpler in terms of $`=\sqrt{E}`$:
$$\frac{d}{dT}=\sqrt{^2V}.$$
(3.28)
Equipped with $`(T)`$, one finds $`T(t)`$ by solving the first-order linear equation that follows from (3.25).
A reverse engineering problem can also be solved. Suppose we are given a tachyon rolling solution specified by a function $`T(t)`$ that has an inverse $`t(T)`$. It is then possible to find the associated dilaton $`\mathrm{\Phi }(t)`$ and the potential $`V(T)`$. We use (3.19) to find $`\dot{\mathrm{\Phi }}(t)`$ by integration, and (3.20) to find $`V(t)`$, which gives the potential $`V(t(T))`$. As a simple illustration we take the leading solution in (3.14) to be exact: $`T(t)=e^{mt}`$. Setting integration constants to zero we find
$$T(t)=e^{mt},\mathrm{\Phi }(t)=\frac{1}{8}e^{2mt},V(T)=m^2\left(\frac{1}{2}T^2+\frac{1}{8}T^4\right).$$
(3.29)
In this solution the crunch happens at infinite string time (but finite Einstein time). Related rolling solutions have been considered using two-dimensional Liouville field theory to provide conformal invariant sigma model with spacetime background fields that typically include a linear dilaton and a constant string metric . In some of these solutions $`T(t)=e^{mt}`$ and the linear dilaton vanishes. This is unexpected given our analysis, which shows that the dilaton is sourced. It would be interesting to use this discrepancy to find constraints on the form of the effective action for the coupled system of fields.
## 4 Finite-time crunch with negative scalar potentials
We now show that for non-positive potentials $`V(T)0`$, if $`\dot{\mathrm{\Phi }}(t_0)>0`$ for some time $`t_0`$ then $`\dot{\mathrm{\Phi }}(t_{})=\mathrm{}`$ for some finite time $`t_{}>t_0`$. To do this we combine (3.19) and (3.20) to write
$$\frac{\ddot{\mathrm{\Phi }}}{\dot{\mathrm{\Phi }}^2}=2+\frac{V(T)}{\dot{\mathrm{\Phi }}^2}.$$
(4.30)
Integrating both sides of the equation from an initial time $`t_0`$ up to a time $`t`$ we find
$$\frac{1}{\dot{\mathrm{\Phi }}(t)}=\frac{1}{\dot{\mathrm{\Phi }}(t_0)}2(tt_0)+_{t_0}^t\frac{V(T(t^{}))}{\dot{\mathrm{\Phi }}^2(t^{})}๐t^{}.$$
(4.31)
To have a divergent $`\dot{\mathrm{\Phi }}`$ we need the terms on the right-hand side to add up to zero. If we ignore the integral on the right-hand side, the first two terms cancel for $`t=t_1`$, with $`t_1t_0=1/(2\dot{\mathrm{\Phi }}(t_0))`$. Since the integral vanishes at $`t=t_0`$ and can only decrease afterwards, the cancellation will actually occur for a time earlier than $`t_1`$:
$$t_{}t_0+\frac{1}{2\dot{\mathrm{\Phi }}(t_0)}.$$
(4.32)
This is what we wanted to prove. It follows from (3.23) and $`V0`$ that as $`\dot{\mathrm{\Phi }}\mathrm{}`$ we also have $`\dot{T}\pm \mathrm{}`$. The time evolution reaches a singular point at finite time.
To understand how the dilaton $`\mathrm{\Phi }`$ itself diverges we can do an estimate that proves to be self-consistent. The integral term in (4.31) is assumed to be negligible. This is certainly the case if $`V(T)`$ is also bounded below since then, the integral is negligible for sufficiently large $`\dot{\mathrm{\Phi }}`$. In fact, we will see that the integral is negligible under far more general circumstances. It then follows that
$$\dot{\mathrm{\Phi }}(t)\frac{1}{2(t_{}t)}\text{and}\dot{T}(t)\frac{\pm 1}{(t_{}t)}.$$
(4.33)
For such solutions the tachyon and dilaton diverge logarithmically:
$$\mathrm{\Phi }(t)=\frac{1}{2}\mathrm{ln}(t_{}t)+\mathrm{\Phi }_0T(t)=\mathrm{ln}(t_{}t)+T_0.$$
(4.34)
For any polynomial potential $`V(T)`$ the integrand in (4.31) is of the form $`(t_{}t)^2V(\mathrm{ln}(t_{}t))`$ and goes to zero as we approach collapse, thus justifying our approximations. Note that $`\dot{T}^2`$ is much larger than both $`V(T)`$ and $`V^{}(T)`$. This fact alone implies finite time collapse: the tachyon equation (3.24) becomes $`\ddot{T}\dot{T}^20`$, whose general solution describes a $`\dot{T}`$ that diverges at an adjustable finite time. Note that the dilaton prefactor in the spacetime action, $`e^{2\mathrm{\Phi }}e^{2\mathrm{\Phi }_0}(t_{}t)`$, vanishes linearly with time as we approach the collapse. The string coupling becomes infinity, the Einstein metric crunches, and the value of the on-shell action (2.3) goes to zero. Since the Einstein metric becomes much smaller than the string metric as the dilaton diverges, the collapse also occurs in finite time in Einstein frame.
## 5 Crunching with arbitrary potentials
In this section we consider rolling solutions for rather general potentials $`V(T)`$, not necessarily tachyonic. As before, we take $`H=0`$ and $`\dot{\mathrm{\Phi }}>0`$; these conditions ensure that we are dealing with a problem qualitatively related to tachyon-induced rolling. As a warmup we consider the case where the potential is positive and bounded and show that for a sufficiently large initial tachyon velocity the Einstein metric crunches in finite time. We then discuss a related question for more general potentials: is there an initial tachyon velocity $`\dot{T}>0`$ for which $`T=\mathrm{}`$ (and crunching) is reached in finite time even if $`V(T\mathrm{})\mathrm{}`$? We discuss a set of tools that enable one to approach this question systematically, at least in a case by case basis. We find that potentials of the form $`V(T)\mathrm{exp}(nT)`$ with $`n2`$ are too steep, and no positive tachyon velocity allows the tachyon to reach $`T=\mathrm{}`$. We also analyze in detail the simple potential $`V=\frac{1}{2}T^2+\frac{1}{8}T^4`$.
Crunching with bounded potentials. We claim that for a bounded potential $`0V(T)<\beta ^2`$ with bounded derivative $`V^{}(T)<\gamma ^2`$, there is an initial tachyon velocity $`\dot{T}(t=0)`$ for which crunching occurs in finite time. Here is a short proof. Take $`\dot{T}(0)=\sqrt{\alpha ^2+2\beta ^2}`$ with $`\alpha ^2>\gamma ^2`$. Since the energy $`E`$ (see (3.26)) cannot decrease in time, for $`t>0`$:
$$2E(t)=\dot{T}^2(t)+2V(T(t))>2E(0)=\alpha ^2+2\beta ^2+2V(T(0))\alpha ^2+2\beta ^2.$$
(5.1)
Therefore, $`\dot{T}^2(t)>\alpha ^2+2\beta ^22V(T(t))>\alpha ^2`$ and, as a result, $`\dot{T}^2(t)>\alpha ^2`$ for all times. The tachyon equation of motion (3.24) then gives
$$\ddot{T}(t)>\sqrt{\alpha ^2+2V(T)}\alpha \gamma ^2\alpha ^2\gamma ^2>0.$$
(5.2)
Since $`\ddot{\mathrm{\Phi }}=\frac{1}{2}\dot{T}^2`$, one finds $`\stackrel{\dot{}\dot{}\dot{}}{\mathrm{\Phi }}(t)=\dot{T}(t)\ddot{T}(t)>0`$, so $`\dot{\mathrm{\Phi }}(t)`$ is convex and grows without bound. On the other hand, with our bounds, equation (4.31) gives
$$\frac{1}{\dot{\mathrm{\Phi }}(t)}<\frac{1}{\dot{\mathrm{\Phi }}(t_0)}\left(2\frac{\beta ^2}{\dot{\mathrm{\Phi }}^2(t_0)}\right)(tt_0).$$
(5.3)
Since $`\dot{\mathrm{\Phi }}`$ grows without bound, there is a time $`t_0`$ for which $`2\frac{\beta ^2}{\dot{\mathrm{\Phi }}^2(t_0)}>0`$. Then at some time $`t_1`$, the right hand side of equation (5.3) vanishes. Therefore, at some finite time $`t_{}<t_1`$, $`\dot{\mathrm{\Phi }}(t_{})=\mathrm{}`$.
General techniques. We now consider a general class of potentials $`V(T)`$, well-defined for all $`T`$, and unbounded above as $`T`$ grows positive and large. We examine an initial configuration with some fixed value $`T(t_0)`$ and variable initial velocity $`\dot{T}(t_0)>0`$. We wish to find out if the tachyon reaches $`T=\mathrm{}`$ and if it does so in finite time, causing the dilaton to diverge and the Einstein metric to crunch. We find that, typically, there is a critical tachyon velocity for which it takes infinite time to reach $`T=\mathrm{}`$. For velocities larger than critical, $`T=\mathrm{}`$ is reached in finite time. For velocities smaller than critical the tachyon evolution gives a turning point.
Three curves can be defined in the $`(T,)`$ plane and help us understand the integral curves $`(T)`$ that solve our first-order differentail equation $`\frac{d}{dT}=\sqrt{^2V(T)}h(T,)`$:
* $`h(T,)=0`$ is the *turning point* curve.
* $`\frac{d}{dT}h(T,)=0`$ is the *inflection* curve. It separates a region where the integral curves are convex from a region where they are concave.
* $`^2V(T)=f(T)`$, with $`f`$ specified below, is the *separating* curve. Any integral curve starting above the separating curve will remain above it.
Since $`\frac{d}{dT}=\sqrt{^2V}=\frac{1}{\sqrt{2}}|\dot{T}(t)|`$ the turning point curve is the locus of points where we get turning points for the tachyon time evolution. If an integral curve hits the turning point curve, the time evolution of the tachyon has a turning point. Moreover, since $`\frac{d}{dT}\dot{T}=\frac{\ddot{T}}{\dot{T}}`$, the inflection curve also controls the convexity or concavity of $`T(t)`$.
Consider the curve $`^2V(T)=f(T)`$. On this curve the slope of the integral curve is $`\sqrt{f(T)}`$. Moreover, the slope of this curve itself is $`\frac{f^{}+V^{}}{2\sqrt{f+V}}`$. In order to be a separating curve we require the former to be larger than the latter:
$$\sqrt{f(T)}\frac{f^{}+V^{}}{2\sqrt{f+V}}.$$
(5.4)
The equality gives an integral curve โ integral curves separate because they cannot cross, but are hard to find. If $`V(T)0`$, a suitable $`f(T)`$ is obtained by setting:
$$\sqrt{f(T)}=\frac{f^{}+V^{}}{2\sqrt{f}}2ff^{}=V^{}.$$
(5.5)
For a polynomial $`V(T)`$, a convenient choice is $`f(T)=_{n=1}\frac{1}{2^n}\frac{d^nV}{dT^n}`$.
A worked out example. We illustrate the above discussion with the potential $`V=\frac{1}{2}T^2+\frac{1}{8}T^4`$, which is tachyonic near $`T=0`$, vanishes for $`T=\pm 2`$, and grows arbitrarily large for large $`|T|`$. We consider arbitrary velocities for a tachyon for which $`T=2`$ for $`t=0`$. The potential was chosen so that it has an easily obtained solution with critical velocity: $`T(t)=2\mathrm{exp}(t)`$. This is, in fact, the tachyon-induced rolling solution that starts at $`T=0`$ for $`t=\mathrm{}`$. Here $`T(0)=2`$, $`\dot{T}(0)=2`$, $`(T=2)=\sqrt{2}`$, and the solution reaches $`T=\mathrm{}`$ at $`t=\mathrm{}`$. We have verified numerically that solutions with larger initial velocity reach $`T=\mathrm{}`$ in finite time, while solutions with lesser initial velocity encounter a turning point. In Fig. 1, we show the critical trajectory $`T(t)`$ and two additional solutions corresponding to initial velocities slightly higher and slightly lower than critical.
For the potential in question, the turning point curve $`(T)=\sqrt{V(T)}`$ lies below the inflection curve $`(T)=(\frac{V+\sqrt{V^2+V^2}}{2})^{1/2}`$, which in turn lies below the separating curve ($`T2`$) defined with $`f(T)=\frac{1}{16}(1+2T6T^24T^3)`$. In Fig. 2 we plot these three curves along with three solutions. The solution with lowest initial energy has velocity smaller than critical: it crosses the inflection curve and hits the turning point curve. We also show the critical solution and a solution with velocity larger than critical that lies above the separating curve. By construction any solution above the separating curve cannot have a turning point.
Potentials of the form $`V(T)=\mathrm{exp}(nT)`$. We now show that for $`V(T)=\mathrm{exp}(nT)`$, with $`n2`$, there is no initial (positive) tachyon velocity for which the tachyon can reach $`T=\mathrm{}`$. When $`V(T)=\mathrm{exp}(nT)`$ the differential equation (3.28) is solvable. Setting $`=\sqrt{V}g(T)`$, we find
$$g^{}+\frac{V^{}}{2V}g=\sqrt{g^21}g^{}=\sqrt{g^21}\frac{n}{2}g.$$
(5.6)
We have a turning point at $`T_{}`$ if $`g(T_{})=1`$. Dividing both sides of the equation by $`g`$ and integrating,
$$\mathrm{ln}g(T)=\mathrm{ln}g(T_0)\frac{n}{2}(TT_0)+_{T_0}^T๐T^{}\left(1\frac{1}{g^2(T^{})}\right)^{1/2}<\mathrm{ln}g(T_0)(\frac{n}{2}1)(TT_0).$$
(5.7)
For $`n>2`$, the right hand side vanishes at some $`T_1>T_0`$. There is therefore some $`T_{}<T_1`$ with $`g(T_{})=1`$, and thus a turning point, as we wanted to show. The above argument in fact applies for any potential $`V`$ such that $`\frac{V^{}}{V}>2`$ for sufficiently large $`T`$. For $`n=2`$, the solution of (5.6) is
$$T=T_0+\frac{1}{2}(F(g)F(g_0)),F(g)\mathrm{ln}(g+\sqrt{g^21})g(g+\sqrt{g^21}).$$
(5.8)
It is readily checked that $`F(1)=1`$ and $`F(g)`$ decreases monotonically for $`g>1`$. Therefore, $`g(T^{})=1`$ for $`T^{}=T_0\frac{1}{2}(1+F(g_0))>T_0`$. This proves that all solutions have a turning point.
## 6 Conclusions
Our analysis of tachyon-induced rolling has revealed two general facts: 1) the string metric is constant and, 2) the dilaton rolls toward stronger coupling. These facts match precisely the properties of the candidate tachyon vacuum identified in . Consider fact one. In the tachyon vacuum both the tachyon and the dilaton take expectation values, and so do an infinite number of massive fields. The string metric, however, is not sourced and need not acquire an expectation value โ this is guaranteed by rather general string field theory universality arguments . The cosmological constancy of the string metric appears to be the sigma model version of the universality result. Consider now fact two. It was shown in that the dilaton expectation value in the candidate solution corresponds to stronger string coupling. The qualitative agreement makes it plausible that the rolling solutions discussed here represent rolling towards the tachyon vacuum conjectured in . In the case of open strings, the end-product of the rolling solution is different but somewhat related to the tachyon vacuum. Our rolling solutions represent an Einstein metric big crunch or closed strings at infinite coupling. The tachyon vacuum may represent the dissappearance of dynamical spacetime. We feel these two states could be related. The crunch certainly lies beyond the applicability of the action (2.1), which should be supplemented by terms of higher order in $`\alpha ^{}`$. The generality of the evolution and the almost complete independence on the details of the tachyon potential<sup>1</sup><sup>1</sup>1For comments on the specific form of the tachyon potential in closed string theory see Figure 1 and the conclusion section of . suggest to us that the cosmological solutions presented here are relevant, modulo some stringy resolution of the big crunch singularity. It would be rather interesting if the stringy resolution would push the crunch to infinite time. A big crunch, followed by a big bang, is the key element in cyclic universe models . The crunch is induced by a scalar field rolling down a negative potential with a steep region โ the rest of the potential is largely undetermined. Negative, initially steep potentials, are the hallmark of bulk closed string tachyons. We found that generally a big crunch ensues, although in our case the gravitational part of the solution is carried by the dilaton, and thus the crunch has the alternative interpretation of a closed string theory at infinite coupling. It is tempting to speculate that closed string tachyons may play a role in cyclic universe scenarios โ the central difficulty remaining the mysterious transition from a big crunch to a big bang. In such studies it would be useful to focus on tachyonic heterotic models and Type-0 strings.
If the vacuum of the bulk closed string tachyon truly represents the demise of fluctuating spacetime, understanding properly this state and how it fits into a consistent cosmology would give invualuable insight into the mechanisms by which a universe could come into existence. The tachyon vacuum would be roughly imagined to be the state of a universe before the Big Bang.
Acknowledgements We are indebted to Justin Khoury for many illuminating discussions and significant guidance on the subject of cosmological solutions. We would also like to acknowledge useful conversations with K. Hashimoto, M. Headrick, H. Liu, A. Sen, A. Tseytlin, and Y. Okawa.
|
warning/0506/math0506392.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The notion of Lie algebroid, which may be regarded as a natural generalization of the tangent bundle to a manifold, allows one to treat in a unified manner several geometric structures, such as Poisson manifolds, connections on principal bundles, foliations, etc. Since in many situations one can associate a Lie algebroid to a singular foliation, Lie algebroids seem to provide a way for generalizing several constructions (e.g., connections) to singular settings. Recently several field-theoretic models based on Lie algebroids have been proposed, see e.g. .
Every Lie algebroid intrinsically defines a cohomology theory. On the other hand, the theory of $`G`$-differential complexes developed in encompasses the case of the *equivariant Lie algebroid cohomology* that one can define when a Lie algebroid carries a (possibly infinitesimal) action of a Lie group $`G`$ compatible with an action of $`G`$ on the base manifold $`M`$. A natural question is whether one can generalize to this setting the usual localization formula for equivariant (de Rham) cohomology. This seems to have also a certain interest for the applications; for instance, recently the localization formula has been used to compute the partition function of $`N=2`$ super Yang-Mills theory . In this case the relevant cohomology is the equivariant de Rham cohomology of the instanton moduli space โ the moduli space of framed self-dual connections on $`^4`$ (i.e., the relevant Lie algebroid is the tangent bundle to the instanton moduli space). It seems plausible that super Yang-Mills theories with a different number of supersymmetry charges may be treated in a similar way with a different choice of a Lie algebroid on the instanton moduli space.
In this paper we present a localization formula for equivariant Lie algebroid cohomology. When a Lie algebroid $`A`$ on a (compact oriented) manifold $`M`$ carries a (infinitesimal) action of a Lie group, one can introduce an equivariant Lie algebroid cohomology. If one twists such cohomology by means of the orientation bundle naturally associated with $`A`$, equivariant cocycles can be integrated on $`M`$. If the group action has only isolated fixed points, the value of the integral can be calculated as a finite sum of suitably defined residues at the fixed points.
This localization formula of course reduces to the usual one for equivariant de Rham cohomology when the Lie algebroid $`A`$ is the tangent bundle $`TM`$. In a similar way, it encompasses a number of classical localization formulas, providing new proofs for them. For instance, it implies a generalization to the Lie algebroid setting of Bottโs theorem about zeroes of vector fields (and also related formulas due to Cenkl and Kubarski ). This generalizes Bottโs formula exactly in the same sense as the notion of Lie algebroid generalizes that of tangent bundle. Moreover, when $`M`$ is complex and $`A`$ is the Atiyah algebroid of a holomorphic vector bundle $`E`$ on $`M`$, our localization formula can be specialized so as to produce localization formulas due to Baum-Bott , Chern , Carrell and Lieberman and K. Liu . These aspects will be developed in . Another localization formula which is generalized is the one given in . But many more examples can be given.
The paper is structured as follows. In Section 2 we review the basic definitions and some constructions concerning Lie algebroid cohomology. Section 3 introduces the equivariant Lie algebroid cohomology, basically within the framework of the theory of *$`G`$-differential complexes* developed in . Moreover, we prove our localization formula. In Section 4, following , we review one of the several approaches to connections on Lie algebroids and use this to construct equivariant characteristic classes for Lie algebroids. This will be used in Section 5 to prove a Bott-type formula.
Acknowledgements. The authors thank the referees and the editor A. A. Rosly for several useful remarks which helped to substantially improve the presentation and the wording of many basic concepts. We are grateful to P. Bressler and Y. Kosmann-Schwarzbach for their attention to this work.
## 2 Lie algebroid cohomology
Let $`M`$ be a smooth manifold. We shall denote by $`๐(M)`$ the space of vector fields on $`M`$ equipped with the usual Lie bracket $`[,]`$.
###### Definition 2.1.
An algebroid $`A`$ over $`M`$ is a vector bundle on $`M`$ together with a vector bundle morphism $`a:ATM`$ (called the anchor) and a structure of Lie algebra on the space of global sections $`\mathrm{\Gamma }(A)`$, such that
1. $`a:\mathrm{\Gamma }(A)๐(M)`$ is a Lie algebra homomorphism;
2. the following Leibniz rule holds true for every $`\alpha `$, $`\beta \mathrm{\Gamma }(A)`$ and every function $`f`$:
$$\{\alpha ,f\beta \}=f\{\alpha ,\beta \}+a(\alpha )(f)\beta $$
(we denote by $`\{,\}`$ the bracket in $`\mathrm{\Gamma }(A)`$).
Morphisms between two Lie algebroids $`(A,a)`$ and $`(A^{},a^{})`$ on the same base manifold $`M`$ are defined in a natural way, i.e., they are vector bundle morphisms $`\varphi :AA^{}`$ such that the map $`\varphi :\mathrm{\Gamma }(A)\mathrm{\Gamma }(A^{})`$ is a Lie algebra homomorphism, and the obvious diagram involving the two anchors commutes.
To any Lie algebroid $`A`$ one can associate the cohomology complex $`(C_A^{},\delta )`$, with $`C_A^{}=\mathrm{\Gamma }(\mathrm{\Lambda }^{}A^{})`$ and differential $`\delta `$ defined by
$$\begin{array}{c}(\delta \xi )(\alpha _1,\mathrm{},\alpha _{p+1})=\underset{i=1}{\overset{p+1}{}}(1)^{i1}a(\alpha _i)(\xi (\alpha _1,\mathrm{},\widehat{\alpha }_i,\mathrm{},\alpha _{p+1}))\hfill \\ \hfill +\underset{i<j}{}(1)^{i+j}\xi (\{\alpha _i,\alpha _j\},\mathrm{},\widehat{\alpha }_i,\mathrm{},\widehat{\alpha }_j,\mathrm{},\alpha _{p+1})\end{array}$$
if $`\xi C_A^p`$ and $`\alpha _i\mathrm{\Gamma }(A),1ip+1`$. The resulting cohomology is denoted by $`H^{}(A)`$ and is called the cohomology of the Lie algebroid $`A`$.
###### Remark 2.2.
One should notice that if $`A`$ is a vector bundle and $`\delta `$ is a derivation of degree $`+1`$ of the graded algebra $`\mathrm{\Gamma }(\mathrm{\Lambda }^{}A^{})`$ which satisfies $`\delta ^2=0`$, out of $`\delta `$ one can construct an anchor $`a:ATM`$ and a Lie bracket on $`\mathrm{\Gamma }(A)`$ making $`A`$ into a Lie algebroid. One simply defines
$$a(\alpha )(f)=\delta f(\alpha )\text{for}\alpha \mathrm{\Gamma }(A),fC^{\mathrm{}}(M),$$
$$\xi (\{\alpha ,\beta \})=a(\alpha )(\xi (\beta ))a(\beta )(\xi (\alpha ))\delta \xi (\alpha ,\beta )\text{for}\alpha ,\beta \mathrm{\Gamma }(A),\xi \mathrm{\Gamma }(A^{}).$$
$`\mathrm{}`$
We recall some examples of Lie algebroids.
###### Example 2.3.
An involutive distribution inside the tangent bundle (i.e., a foliation) is a Lie algebroid, whose anchor is injective. $`\mathrm{}`$
###### Example 2.4.
Let $`๐=(M,)`$ be a supermanifold; in particular, $``$ is a sheaf of $`_2`$-graded commutative $``$-algebras on the differentiable manifold $`M`$ that can be realized as the sheaf of sections of the exterior algebra bundle $`\mathrm{\Lambda }^{}(E)`$ for a vector bundle $`E`$. Let $`D`$ be an odd supervector field on $`๐`$ squaring to zero. Then $`E^{}`$, with an anchor and a Lie algebra structure on $`\mathrm{\Gamma }(\mathrm{\Lambda }^{}E^{})`$ given according to Remark 2.2 by $`D`$ regarded as a differential for the complex $`\mathrm{\Gamma }(\mathrm{\Lambda }^{}E)`$, is a Lie algebroid. Of course, starting from a Lie algebroid we can construct a supermanifold with an odd supervector field on it squaring to zero, so that the two sets of data are equivalent. $`\mathrm{}`$
###### Example 2.5.
Let $`(M,\mathrm{\Pi })`$ be a Poisson manifold, where $`\mathrm{\Pi }`$ a Poisson tensor. In this case $`A=T^{}M`$ with the Lie bracket of differential forms
$$\{\alpha ,\beta \}=_{\mathrm{\Pi }(\alpha )}\beta _{\mathrm{\Pi }(\beta )}(\alpha )d\mathrm{\Pi }(\alpha ,\beta )$$
(1)
(where $``$ is the Lie derivative) and the anchor is the Poisson tensor. The cohomology of $`A`$ is the Lichnerowicz-Poisson cohomology of $`(M,\mathrm{\Pi })`$. $`\mathrm{}`$
###### Example 2.6.
Let $`P\stackrel{๐}{}M`$ be a principal bundle with structure group $`G`$. One has the Atiyah exact sequence of vector bundles on $`M`$
$$0\text{ad}(P)TP/GTM0$$
(2)
(a connection on $`P`$ is a splitting of this sequence). Sections of the vector bundle $`TP/G`$ are in a one-to-one correspondence with $`G`$-invariant vector fields on $`P`$. On the global sections of $`TP/G`$ there is a natural Lie algebra structure, and taking the surjection $`TP/GTM`$ as anchor map, we obtain a Lie algebroid โ the *Atiyah algebroid* associated with the principal bundle $`P`$.
If $`E`$ is a vector bundle on $`M`$, we can also associate an Atiyah algebroid with it: in this case indeed one has an exact sequence
$$0\text{End}(E)\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}(E)TM0$$
(3)
where $`\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}(E)`$ is the bundle of 1-st order differential operators on $`E`$ having scalar symbol , and again $`\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}E`$, with the natural Lie algebra structure on its global sections and the natural map $`\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}(E)TM`$ as anchor, is a Lie algebroid. The two notions of Atiyah algebroid coincide when $`P`$ is the bundle of linear frames of a vector bundle $`E`$ (indeed, an element in $`T_uP`$ is given by an endomorphism of the fibre $`E_{p(u)}`$ and a vector in $`T_{p(u)}M`$).
Lie algebroids whose anchor map is surjective, as in the case of the Atiyah algebroids, are called *transitive*. $`\mathrm{}`$
Following we now describe a twisted form of the Lie algebroid cohomology together with a natural pairing between the two cohomologies. This will be another ingredient for the localization formula. Let $`Q_A`$ be the line bundle $`^rA\mathrm{\Omega }_M^m`$, where $`r=\text{rk}A`$ and $`m=dimM`$, and $`\mathrm{\Omega }_M^m`$ is the bundle of differential $`m`$-forms on $`M`$ (later on we shall denote by $`\mathrm{\Omega }^m(M)`$ the global sections of this bundle). For every $`s\mathrm{\Gamma }(A)`$ define a map $`L_s=\{s,\}=:\mathrm{\Gamma }(^{}A)\mathrm{\Gamma }(^{}A)`$ by letting
$$L_s(s_1\mathrm{}s_k)=\underset{i=1}{\overset{k}{}}s_1\mathrm{}\{s,s_i\}\mathrm{}s_k.$$
Moreover define a map
$`D:\mathrm{\Gamma }(Q_A)`$ $``$ $`\mathrm{\Gamma }(A^{}Q_A)=\mathrm{\Gamma }(A^{})_{C^{\mathrm{}}(M)}\mathrm{\Gamma }(Q_A)`$
$`D\tau (s)`$ $`=`$ $`L_s(X)\mu +X_{a(s)}\mu `$
if $`\tau =X\mu \mathrm{\Gamma }(Q_A)`$ and $`s\mathrm{\Gamma }(A)`$. We consider the twisted complex $`\stackrel{~}{C}_A^{}=\mathrm{\Gamma }(^{}A^{}Q_A)=C_A^{}_{C^{\mathrm{}}(M)}\mathrm{\Gamma }(Q_A)`$ with the differential $`\stackrel{~}{\delta }`$ defined by
$$\stackrel{~}{\delta }(\xi \tau )=\delta \xi \tau +(1)^{\mathrm{deg}(\xi )}\xi D\tau $$
and $`\xi C_A^{}=_{k=0}^r\mathrm{\Gamma }(^k(A^{}).`$ We denote the resulting cohomology $`H^{}(A,Q_A)`$.
There is a naturally defined map<sup>1</sup><sup>1</sup>1We use interchangeably the notations $`\psi \mathrm{}`$ and $`i_\psi `$ to denote the inner product with an element $`\psi `$, according to typographical convenience. $`p:\stackrel{~}{C}_A^{}\mathrm{\Omega }^{r+m}(M)`$
$$p(\psi X\mu )=(a(\psi \mathrm{}X))\mathrm{}\mu .$$
###### Proposition 2.7.
The morphism $`p`$ is a chain map,<sup>2</sup><sup>2</sup>2We are thankful to A. Rosly for pointing out this fact, and for calling to our attention the following proof. in the sense that the diagram
(4)
commutes up to a sign, i.e., on $`C_A^k`$ one has
$$p\stackrel{~}{\delta }=(1)^kdp.$$
(5)
###### Proof.
If we define a Lie derivative on $`\stackrel{~}{C}_A^{}`$ by letting $`L_s=i_s\stackrel{~}{\delta }+\stackrel{~}{\delta }i_s`$ for elements $`s\mathrm{\Gamma }(A)`$, this satisfies the commutation relation
$$pL_s=_{a(s)}p$$
(6)
on $`\stackrel{~}{C}_A^k`$. By using this identity we may prove (5) by descending induction on $`k`$. For $`k=r`$ the identity reduces to $`0=0`$. For smaller values of $`k`$ it is enough to prove the identity for those $`c^{}\stackrel{~}{C}_A^k`$ that can be represented as $`c^{}=i_sc`$ for some $`s\mathrm{\Gamma }(A)`$ and $`c\stackrel{~}{C}_A^{k+1}`$. In this case the result follows from a simple computation which uses (6). โ
Let $`M`$ be compact and oriented, and note that since $`\stackrel{~}{C}_A^r\mathrm{\Omega }^m(M)`$ (canonically), one can integrate elements of $`\stackrel{~}{C}_A^r`$ over $`M`$. There is a nondegenerate pairing
$`C_A^k_{C^{\mathrm{}}(M)}\stackrel{~}{C}_A^{rk}`$ $``$ $``$
$`\xi (\psi X\mu )`$ $``$ $`{\displaystyle _M}(\xi \psi ,X)\mu .`$
A version of Stokesโ theorem holds for the complex $`\stackrel{~}{C}_A^{}`$ : if $`c\stackrel{~}{C}_A^{r1}`$, then
$$_M\stackrel{~}{\delta }c=0.$$
This formula follows from the identity (5) for $`k=r1`$. It implies that the pairing descends to cohomology, yielding a bilinear map
$$H^{}(A)H^r(A,Q_A).$$
(7)
This pairing in general may be degenerate.
One also has a natural morphism $`C_A^{}_{C^{\mathrm{}}(M)}\stackrel{~}{C}_A^{}\stackrel{~}{C}_A^{}`$ which is compatible with the degrees. Again this descends to cohomology and provides a cup product
$$H^i(A)H^j(A,Q_A)H^{i+j}(A,Q_A).$$
(8)
## 3 Equivariant cohomology and localization
In this section we introduce an equivariant cohomology for Lie algebroids, basically following the pattern exploited in to define equivariant cohomology for Poisson manifolds (and falling within the general theory of equivariant cohomology for $`G`$-differential complexes developed there). Moreover, we write a localization formula for the equivariant Lie algebroid cohomology (Theorem 3.2).
We assume that there is an action of a Lie algebra $`๐ค`$ on the Lie algebroid $`A`$, i.e., that there is a Lie algebra map
$$b:๐ค\mathrm{\Gamma }(A).$$
(9)
By composing this with the anchor map we obtain a Lie algebra homorphism $`\stackrel{~}{\rho }=ab:๐ค๐(M)`$, i.e., an action of $`๐ค`$ on $`M`$. Lie algebra maps like our $`b`$ have been introduced in in the case of the Atiyah algebroids under the name of โderivation representations.โ
###### Example 3.1.
(Cf. ) Let $`\mathrm{\Pi }`$ be a regular Poisson tensor on $`M`$ (i.e., $`\mathrm{\Pi }`$ has constant rank), and let $`๐ฎ=\text{Im}(\mathrm{\Pi })`$ be the associated symplectic distribution. The family of symplectic forms defined on the leaves of the distribution $`๐ฎ`$ yields an isomorphism $`๐ฎ๐ฎ^{}`$. Then $`๐ฎ^{}`$ is a subbundle of $`TM`$ and $`\mathrm{\Gamma }(\mathrm{ker}\mathrm{\Pi })`$ is an ideal in $`\mathrm{\Omega }^1(M)`$ with respect to the Lie algebroid structure in $`T^{}M`$ given by the bracket in Eq. (1). Moreover, $`๐ฎ^{}`$ is a Lie subalgebroid of $`TM`$; its cohomology is called the *tangential Lichnerowicz-Poisson cohomology*. Assume now that $`M`$ carries the action $`\rho `$ of a Lie group $`G`$; if $`๐ค`$ is the Lie algebra of $`G`$, for every $`\xi ๐ค`$ let us denote by
$$\xi ^{}=\frac{d}{dt}\rho _{\mathrm{exp}(t\xi )|t=0}$$
the corresponding fundamental vector field (thus, we have a Lie algebra homomorphism $`\stackrel{~}{\rho }:๐ค๐(M)`$, $`\stackrel{~}{\rho }(\xi )=\xi ^{}`$). If the $`G`$-action is tangent to $`๐ฎ^{}`$, we obtain an infinitesimal action $`b:๐ค\mathrm{\Gamma }(๐ฎ^{})`$ (this is what in is called a *cotangential lift*). $`\mathrm{}`$
If for a $`\xi ๐ค`$ the point $`xM`$ is a zero of $`\xi ^{}`$ then we have the usual endomorphism
$$L_{\xi ,x}:T_pMT_pM,L_{\xi ,x}(v)=[\xi ^{},v].$$
We consider the graded vector space
$$๐^{}=\text{Sym}^{}(๐ค^{})\mathrm{\Gamma }(^{}A^{})$$
with the grading
$$\mathrm{deg}(๐ซ\beta )=2\mathrm{deg}(๐ซ)+\mathrm{deg}\beta $$
where $`๐ซ\text{Sym}^{}(๐ค^{})`$ and $`\beta \mathrm{\Gamma }(^{}A^{}).`$
We will consider $`๐ซ`$ as a polynomial function on $`๐ค`$ and define the equivariant differential $`\delta _๐ค:๐^{}๐^{+1}`$
$$(\delta _๐ค(๐ซ\beta ))(\xi )=๐ซ(\xi )\left(\delta (\beta )i_{b(\xi )}\beta \right),$$
(10)
where both sides have been evaluated on an element $`\xi ๐ค`$. If we denote $`๐_G^{}=\mathrm{ker}\delta _๐ค^2`$, then $`(๐_G^{},\delta _๐ค)`$ is a complex, whose cohomology we denote $`H_G^{}(A)`$ and call the *equivariant cohomology* of the Lie algebroid $`A`$ (to be more precise, of the pair $`(A,b)`$).
By considering the graded vector space
$$๐^{}=๐^{}\mathrm{\Gamma }(Q_A)=\text{Sym}^{}(๐ค^{})\mathrm{\Gamma }(^{}A^{}Q_A)$$
with a differential $`\stackrel{~}{\delta }_๐ค`$ obtained by coupling $`\delta _๐ค`$ with the differential $`D`$, and letting $`๐_G^{}=\mathrm{ker}\stackrel{~}{\delta }_๐ค^2`$, one also has a twisted equivariant cohomology $`H_G^{}(Q_A)`$, and there is a cup product
$$H_G^i(A)H_G^k(Q_A)H_G^{i+k}(Q_A).$$
We write now a localization formula. In view of Proposition 2.7, its right-hand side can be computed by means of the usual localization formula in equivariant de Rham cohomology; the integral of an equivariantly closed $`๐_G^{}`$-cocycle $`\gamma (\xi )`$ is actually the integral of the differential form $`p(\gamma (\xi ))`$, and $`p(\gamma )`$ is a cocycle in the equivariant de Rham complex due to Proposition 2.7. It is indeed quite easy to prove the identity
$$p(\stackrel{~}{\delta }_๐ค(\gamma ))=(1)^kd_๐ค(p(\gamma ))$$
(11)
when $`\gamma ๐_G^k`$. Here $`d_๐ค`$ is the differential in the usual equivariant de Rham cohomology. This follows from Proposition 2.7 and the equalities
$`i_\xi ^{}p(\gamma )`$ $`=`$ $`i_\xi ^{}\left[a(\psi \mathrm{}X)\mathrm{}\mu \right]=(\xi ^{}a(\psi \mathrm{}X))\mathrm{}\mu `$
$`=`$ $`a(b(\xi )(\psi \mathrm{}X))\mathrm{}\mu =(1)^{k1}a\left((i_{b(\xi )}\psi )\mathrm{}X\right)\mathrm{}\mu `$
$`=`$ $`(1)^{k1}p(i_{b(\xi )}\gamma )`$
having set $`\gamma =\psi X\mu `$.
Let $`M`$ be a closed manifold which carries the action $`\rho `$ of a compact Lie group $`G`$. We also assume that $`M`$ is oriented, and that a $`\xi ๐ค`$ has been chosen such that $`\xi ^{}=\stackrel{~}{\rho }(\xi )`$ (this has been defined in Example 3.1) only has isolated zeroes. We denote by $`M_\xi `$ the set of such zeroes. Note that due to the compactness of $`G`$, we have $`det(L_{\xi ,x})0`$ at every isolated zero $`x`$, and the dimension $`m`$ of $`M`$ is necessarily even (as we shall assume henceforth).
If the rank $`r`$ of $`A`$ is smaller than the dimension $`m`$ of $`M`$, then for every equivariantly closed $`\gamma ๐^{}`$ we have $`_M\gamma (\xi )=0`$ for dimensional reasons, since $`p(\gamma )_0=0`$ in that case. (Here the subscript $`0`$ denotes the piece of degree 0 in the usual de Rham grading). We may therefore henceforth assume that $`rm`$.
###### Theorem 3.2.
Let $`M`$ be a closed oriented $`m`$-dimensional manifold on which a compact Lie group $`G`$ acts. Let $`A`$ be a rank $`r`$ Lie algebroid on $`M`$, with $`rm`$, and assume that a Lie algebra homomorphism $`b:๐ค\mathrm{\Gamma }(A)`$ exists making the diagram
(12)
commutative; in other terms, $`\stackrel{~}{\rho }`$ is the Lie algebra homomorphism $`\stackrel{~}{\rho }(\xi )=\xi ^{}`$ (here $`๐ค`$ is the Lie algebra of $`G`$). Moreover, assume that $`\xi ๐ค`$ is such that the associated fundamental vector field $`\xi ^{}`$ has only isolated zeroes. Finally, let $`\gamma ๐^{}`$ be equivariantly closed, $`\stackrel{~}{\delta }_๐ค\gamma =0`$.
Then the following localization formula holds:
$$_M\gamma (\xi )=(2\pi )^{m/2}\underset{xM_\xi }{}\frac{p(\gamma (\xi ))_0(x)}{\mathrm{det}^{1/2}L_{\xi ,x}}.$$
(13)
###### Proof.
Since in the left-hand side we are actually integrating the differential form $`p(\gamma (\xi ))`$, the formula follows from the identity (11) and the usual localization formula. โ
One easily checks that in the case of the โtrivialโ algebroid given by the tangent bundle with the identity map as anchor, this reduces to the ordinary localization formula for the equivariant de Rham cohomology (see e.g. ).
###### Remark 3.3.
If $`rm`$ and the rank of the linear morphism $`a`$ at the point $`p`$ is not maximal (i.e., less than $`m`$), then $`p(\gamma (\xi ))_0(x)=0`$. $`\mathrm{}`$
###### Remark 3.4.
As a particular case of Theorem 3.2, one can state a localization formula related to the action of a vector field on $`M`$. Let $`M`$ be a compact oriented $`m`$-dimensional manifold, and $`X\mathrm{\Gamma }(\mathrm{T}M)`$ a vector field on $`M`$ with isolated zeroes which generates a circle action. Let $`A`$ be a rank $`r`$ Lie algebroid on $`M`$ such that there exists $`\stackrel{~}{X}\mathrm{\Gamma }(A)`$ with $`a(\stackrel{~}{X})=X`$ (where $`a`$ is the anchor map). Then, for the integration of a form $`\gamma \mathrm{\Gamma }(^{}A^{}^rA^m\mathrm{T}^{}M)`$ such that $`\stackrel{~}{\delta }_X\gamma :=(\stackrel{~}{\delta }i_{\stackrel{~}{X}})\gamma =0`$, the localization formula (3.2) holds.
One can replace the assumption that $`X`$ generates a circle action by assuming that $`X`$ is an isometry of a Riemannian metric on $`M`$. $`\mathrm{}`$
## 4 Connections and characteristic classes for Lie algebroids
Several applications of the localization formula may be given by using a notion of a characteristic class for a Lie algebroid. We start by introducing the concept of *$`A`$-connection,* cf. .
Let $`A`$ be a Lie algebroid with anchor $`a`$, and let $`P\stackrel{๐}{}M`$ be a principal bundle with structure group $`K`$. Note that the pullback $`p^{}A=A\times _MP`$ carries a natural $`K`$-action, and $`Ap^{}A/K`$. Also the tangent bundle $`TP`$ carries a natural $`K`$-action, and one has $`p_{}(vk)=p_{}(v)`$ for $`vTP`$ and $`kK`$, so that one has an induced map $`p_{}:TP/KTM`$ which is the anchor of the Atiyah algebroid associated with $`P`$, see Eq. (2).
###### Definition 4.1.
An $`A`$-connection on $`P`$ is a bundle map $`\eta :p^{}ATP`$ such that:
1. the following diagram commutes:
2. $`\eta `$ is $`K`$-equivariant, i.e., $`\eta (uk,\alpha )=R_k\eta (u,\alpha )`$ for all $`kK`$, $`uP`$, $`\alpha A`$. (Here $`R_k`$ denotes the structural right action of an element $`kK`$ on $`P`$.)
If $`P`$ is the bundle of linear frames of $`A`$, $`\eta `$ is called an $`A`$-linear connection.
By its equivariance, an $`A`$-connection $`\eta `$ for a principal $`K`$-bundle $`P`$ defines a bundle map $`\omega _\eta :ATP/K`$, called the *connection 1-section* of $`\eta `$. One has $`p_{}\omega _\eta =a`$.
###### Remark 4.2.
1. The usual notion of connection is recovered by taking $`A=TM`$, and $`\eta `$ is then the corresponding horizontal lift $`\eta :p^{}TMTP`$.
2. An ordinary connection on $`P`$ (regarded as the associated horizontal lift $`\zeta :p^{}TMTP`$) defines an $`A`$-connection $`\eta `$ for $`P`$ by letting $`\eta =\zeta p^{}a`$. $`\mathrm{}`$
If $`E\stackrel{p_E}{}M`$ is a vector bundle associated with $`P`$ via a representation of $`K`$ on a linear space, an $`A`$-connection on $`P`$ defines a similar structure on $`E`$, that is, a bundle map $`\eta _E:p_E^{}ATE`$ which makes the diagram
commutative. The $`A`$-connection $`\eta _E`$ defines in the usual way a covariant derivative $`:\mathrm{\Gamma }(A)_{}\mathrm{\Gamma }(E)\mathrm{\Gamma }(E)`$: if $`\varphi :TP/K\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}(E)`$ is the natural map,<sup>3</sup><sup>3</sup>3A section $`X`$ of $`TP/K`$ is a $`K`$-invariant vector field on $`P`$. Since there is an obvious map $`TP\text{Diff}_{\mathrm{\hspace{0.17em}0}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}1}}(P\times V)`$, where $`V`$ is the standard fibre of $`E`$, by equivariance $`X`$ yields a differential operator on $`E`$ whose symbol is scalar. one sets $`_\alpha =(\varphi \omega _\eta )(\alpha )`$. This covariant derivative satisfies the Leibniz rule
$$_\alpha (fs)=f_\alpha (s)+a(\alpha )(f)s$$
(14)
for all functions $`f`$ on $`M`$.
Let us introduce the notion of a *$`G`$-equivariant* $`A`$-connection. Assume that a Lie group $`G`$ acts on $`M`$, and that this action $`\rho `$ lifts to an action $`\widehat{\rho }`$ on $`A`$; this means that for every $`gG`$ we have a commutative diagram
Moreover, we assume that $`\rho `$ also lifts to an action $`\stackrel{~}{\rho }`$ on the principal $`K`$-bundle $`P`$ which commutes with the structural $`K`$-action.
###### Definition 4.3.
A $`G`$-equivariant $`A`$-connection $`\eta `$ on $`P`$ is an $`A`$-connection $`\eta `$ such that the diagram
commutes for every $`gG`$.
Since the action of $`G`$ on $`P`$ commutes with the action of $`K`$, we have an induced action $`\stackrel{~}{\rho }_{}`$ on $`TP/K`$, and the condition for $`\eta `$ to be $`G`$-equivariant may be stated in terms of the connection 1-section $`\omega _\eta `$ as the commutativity of the diagram
The *curvature* $`_\eta `$ of an $`A`$-connection $`\eta `$ on a principal $`K`$-bundle $`P`$ may be defined in terms of the map $`\omega _\eta `$ as the element in $`\mathrm{\Gamma }(\mathrm{\Lambda }^2A^{}TP/K)`$ given by
$$_\eta (\alpha ,\beta )=[\omega _\eta (\alpha ),\omega _\eta (\beta )]\omega _\eta (\{\alpha ,\beta \}).$$
As a matter of fact $`p_{}_\eta =0`$, so that $`_\eta `$ is an element in $`\mathrm{\Gamma }(\mathrm{\Lambda }^2A^{}\text{ad}(P))`$. The curvature $`_\eta `$ satisfies analogues of the *structure equations* and *Bianchi* identities. These identities are conveniently stated in terms of the so-called *exterior $`A`$-derivative*
$`D_A`$ $`:`$ $`\mathrm{\Gamma }(\mathrm{\Lambda }^{}A^{}TP/K)\mathrm{\Gamma }(\mathrm{\Lambda }^{+1}A^{}TP/K)`$
$`(D_A\chi )(\alpha _1,\mathrm{},\alpha _{p+1})`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{p+1}{}}}(1)^{i1}[\omega _\eta (\alpha _i),\chi (\alpha _1\mathrm{},\widehat{\alpha }_i,\mathrm{},\alpha _{p+1})]`$
$`+`$ $`{\displaystyle \underset{i<j}{}}(1)^{i+j}\chi (\{\alpha _i,\alpha _j\},\mathrm{},\widehat{\alpha }_i,\mathrm{},\widehat{\alpha }_j,\mathrm{},\alpha _{p+1})`$
in the form
$$_\eta =D_A\omega _\eta \frac{1}{2}[\omega _\eta ,\omega _\eta ],D_A_\eta =0.$$
(15)
Note that the equivariance of the connection may be expressed by the condition
$$[D_A,_{\stackrel{~}{\xi }^{}}]=0$$
for all $`\xi ๐ค`$. Here $`_{\stackrel{~}{\xi }^{}}`$ is the Lie derivative of sections of $`TP/K`$ with respect to the section $`\stackrel{~}{\xi }^{}`$ of $`TP/K`$ induced by the vector field on $`P`$ which generates the action of $`G`$ (i.e., $`_{\stackrel{~}{\xi }^{}}(v)=[\stackrel{~}{\xi }^{},v]`$).
If the connection $`\eta `$ is equivariant, we may equivariantize these relations, in particular by defining the equivariant exterior $`A`$-derivative $`D_A^๐ค`$, which acts on $`\mathrm{Sym}^{}(๐ค^{})\mathrm{\Gamma }(\mathrm{\Lambda }^{}A^{}TP/K)`$ as
$$(D_A^๐ค\chi )(\xi )=D_A(\chi (\xi ))(i_{\stackrel{~}{\xi }^{}}\mathrm{id})(\chi (\xi )).$$
Moreover we define the equivariant curvature $`R_\eta ^๐ค`$ of $`\eta `$ as
$$(R_\eta ^๐ค\chi )(\xi )=R_\eta (\chi (\xi ))+_{\stackrel{~}{\xi }^{}}(\chi (\xi ))[D_A,i_{\stackrel{~}{\xi }^{}}\mathrm{id}](\chi (\xi ))=R_\eta (\chi (\xi ))+\mu (\chi (\xi ))$$
where the last equality defines the โmoment mapโ $`\mu `$. Moreover, the square brackets in this equation denote an anticommutator. An easy calculation shows that the equivariant curvature satisfies the equivariant Bianchi identity
$$D_A^๐คR_\eta ^๐ค=0.$$
(16)
We can also write the identities (15) and (16) in a local form, involving the *local connection 1-sections* defined in the following way. Let $`\{U_i\}`$ be an open cover of $`M`$ over which the bundle $`P`$ trivializes. Then one has local isomorphisms
$$\psi _j:(TP/K)_{|U_j}TU_j\times ๐จ,$$
where $`๐จ`$ is the Lie algebra of $`K`$. The local connection 1-sections are defined by the condition
$$\omega _j(\alpha )=\text{pr}_2\psi _j\omega _\eta (\alpha )$$
(where $`\text{pr}_2`$ is the projection onto the second factor of $`TU_j\times ๐จ`$), and one analogously defines the local curvature 2-sections $`_j`$. The identities (15) are now written in the form
$$_j=\delta \omega _j+\frac{1}{2}[\omega _j,\omega _j],\delta _j+[\omega _j,_j]=0.$$
(17)
In the same way, the equivariant curvature may be represented by local 2-sections $`_j^๐ค`$ which, in view of Eq. (17), satisfy the identities
$$\delta _๐ค_j^๐ค+[\omega _j,_j^๐ค]=0.$$
(18)
The Chern-Weil homomorphism is defined as follows. Let $`I^{}(๐จ)=(\text{Sym}^{}๐จ^{})^K`$ , choose an $`A`$-connection $`\eta `$ for $`P`$, and for any polynomial $`QI^{\mathrm{}}(๐จ)`$ of degree $`\mathrm{}`$ define the element $`\lambda _QC_A^2\mathrm{}`$
$$\lambda _Q(\alpha _1,\mathrm{},\alpha _2\mathrm{})=\underset{\sigma }{}(1)^\sigma \stackrel{~}{Q}(_\eta (\alpha _{\sigma _1},\alpha _{\sigma _2}),\mathrm{},_\eta (\alpha _{\sigma _{2\mathrm{}1}},\alpha _{\sigma _2\mathrm{}}))$$
where the summation runs over the permutations of $`2\mathrm{}`$ objects and $`\stackrel{~}{Q}`$ is the polarization of $`Q`$, i.e., the unique Ad-invariant symmetric function of $`\mathrm{}`$ variables in $`๐จ`$ such that $`\stackrel{~}{Q}(\chi ,\mathrm{},\chi )=Q(\chi )`$ for all $`\chi ๐จ`$. One proves that this cochain is $`\delta `$-closed, and that the resulting cohomology class $`[\lambda _Q]`$ does not depend on the connection, thus defining a graded ring homomorphism $`\lambda :I^{}(๐จ)H^2(A)`$. If $`\stackrel{~}{\lambda }:I^{}(๐จ)H_{dR}^2(M)`$ is the usual Chern-Weil homomorphism to the de Rham cohomology of $`M`$, then there is a commutative diagram
(19)
Using the Chern-Weil homomorphism one can introduce various sorts of characteristic classes for the Lie algebroid $`A`$. However due to diagram (19), and somehow unpleasantly, these characteristic classes are nothing more that the image under $`a^{}`$ of the corresponding characteristic classes for the bundle $`E`$. To show this, choose any (ordinary) connection on $`P`$ to compute a characteristic classes in $`H_{dR}^{}(M)`$, and the $`A`$-connection associated with it (see Remark 4.2 (2)) to compute a characteristic class in $`H^{}(A)`$. The two characteristic classes are then related by the morphism $`a^{}`$ as in diagram (19).
In the following we shall use โPontryagin-likeโ characteristic classes: namely, we take $`๐จ=\text{gl}(r,)`$, so that $`P`$ is the bundle of linear frames of a complex vector bundle $`E`$. We assume that $`E`$ is the complexification of a real vector bundle. Let $`Q_i`$ be the $`i`$-th elementary Ad-invariant polynomial, and denote by $`\lambda _i`$ the corresponding characteristic class. These characteristic classes vanish when $`i`$ is odd (to check this, take for instance a connection on $`P`$ which is compatible with a fibre metric on $`E`$, and compute the characteristic classes by means of the induced $`A`$-connection).
If $`QI^{}(\text{gl}(r,))`$ is an Ad-invariant homogeneous symmetric polynomial of degree $`2i`$ on the Lie algebra $`\text{gl}(r,)`$, one has $`Q(_\eta ^๐ค)๐_G^{4i}`$. The following statement is easily proved.
###### Proposition 4.4.
The element $`Q(_\eta ^๐ค)`$ is $`\delta _๐ค`$-closed. The corresponding cohomology class $`\lambda _Q^๐ค(A)H_G^{4i}(A)`$ does not depend on the equivariant connection $`\eta `$.
###### Proof.
First one shows that the quantity $`Q(_\eta ^๐ค)`$ is $`\delta _๐ค`$-closed by using the equations (18). To prove the second claim, if $`\eta `$, $`\eta ^{}`$ are two equivariant $`A`$-connections for the principal bundle $`P`$, define the 1-parameter family of connections
$$\eta _t=t\eta ^{}+(1t)\eta $$
with $`0t1`$, and let
$$\begin{array}{c}q(\eta ,\eta ^{})(\alpha _1,\mathrm{},\alpha _{2i})=\hfill \\ \hfill i\underset{\sigma }{}(1)^\sigma _0^1\left[\stackrel{~}{Q}(\frac{d}{dt}\omega _{\eta _t}(\alpha _{\sigma _1},\alpha _{\sigma _2}),_{\eta _t}^๐ค(\alpha _{\sigma _3},\alpha _{\sigma _4}),\mathrm{}_{\eta _t}^๐ค(\alpha _{\sigma _{2i1}},\alpha _{\sigma _{2i}}))\right]๐t\end{array}$$
where $`\stackrel{~}{Q}`$ is the polarization of the polynomial $`Q`$. A straightforward computation which uses the identity (18) and the identity
$$\frac{d}{dt}_{\eta _t}^๐ค=D_A^๐ค\frac{d}{dt}\omega _{\eta _t}$$
now shows that
$$Q(_\eta ^{}^๐ค)Q(_\eta ^๐ค)=\delta _๐คq(\eta ,\eta ^{})$$
whence the claim follows. โ
When $`Q`$ is the standard $`i`$-th elementary Ad-invariant polynomial $`\varsigma _i`$ on $`\text{gl}(r,)`$ we shall denote by $`\lambda _i^๐ค(A)`$ $`(i=1,\mathrm{},r)`$ the corresponding equivariant characteristic class (vanishing for odd $`i`$). As we discussed previously, these are just the images under the morphism $`a^{}`$ (the adjoint of the anchor map) of the equivariant Chern classes of the (complexified) vector bundle $`E`$.
## 5 A Bott-type formula
As an application of our localization theorem we prove a result which generalizes the classical Bott formula as well as similar results by Cenkl and Kubarski . Bottโs formula comes in different flavours, according to the assumptions that one makes on the vector field entering the formula. The case we consider here extends the usual Bott formula for a vector field which generates a circle action and has isolated critical points.
Let $`\mathrm{\Phi }`$ be a monomial in $`n=[r/4]`$ variables. Let us denote by $`W_\mathrm{\Phi }`$ its *total weight*, defined by assigning weight $`4i`$ to its $`i`$-th variable. We can use the monomial $`\mathrm{\Phi }`$ to attach a real number to the algebroid $`A`$. Assume that $`W_\mathrm{\Phi }r`$, and let $`\mathrm{\Xi }^๐ค`$ be an equivariant twisted coycle of degree $`rW_\mathrm{\Phi }`$, i.e., $`\mathrm{\Xi }^๐ค๐_G^{rW_\mathrm{\Phi }}`$. Let $`\mathrm{\Xi }=\mathrm{\Xi }^๐ค(0)\stackrel{~}{C}_A^{rW_\mathrm{\Phi }}`$ be the zero degree term of $`\mathrm{\Xi }^๐ค`$ as a function of $`\xi ๐ค`$; it satisfies $`\stackrel{~}{\delta }\mathrm{\Xi }=0`$. We may define
$$\mathrm{\Phi }_\mathrm{\Xi }(A)=(2\pi )^{m/2}_M\mathrm{\Phi }(\lambda _2(A),\mathrm{},\lambda _{2n}(A))\mathrm{\Xi }$$
where the $`\lambda _i`$ are the characteristic classes of the vector bundle $`A`$ as defined in the previous section. This number only depends on the Lie algebroid $`A`$ and on the cohomology class $`\mathrm{\Xi }`$. We also define an element in $`\text{Sym}^{}(๐ค^{})`$
$`\mathrm{\Phi }_\mathrm{\Xi }^๐ค(A)`$ $`=`$ $`(2\pi )^{m/2}{\displaystyle _M}\mathrm{\Phi }(\lambda _2^๐ค(A),\mathrm{},\lambda _{2n}^๐ค(A))\mathrm{\Xi }^๐ค`$ (20)
$`=`$ $`(2\pi )^{m/2}{\displaystyle _M}\mathrm{\Phi }(\varsigma _2(_\eta +\mu ),\mathrm{},\varsigma _{2n}(_\eta +\mu ))\mathrm{\Xi }^๐ค.`$
Of course, $`\mathrm{\Phi }_\mathrm{\Xi }(A)=\mathrm{\Phi }_\mathrm{\Xi }^๐ค(A)(0)`$. One has the following Bott-type result, which follows straightforwardly from Theorem 3.2.
###### Theorem 5.1.
Let $`A`$ be a rank $`r`$ Lie algebroid on an $`m`$-dimensional compact oriented manifold $`M`$, and let $`\alpha \mathrm{\Gamma }(A)`$ be any section which generates a circle action on $`A`$, and is such that $`a(\alpha )`$ has isolated zeroes. Let $`\mathrm{\Phi }`$ be a polynomial in $`n=[r/4]`$ variables whose monomials have total weight $`W_\mathrm{\Phi }`$. Then if $`rmW_\varphi `$ one has
$$\mathrm{\Phi }_\mathrm{\Xi }(A)=\underset{xM_{a(\alpha )}}{}\frac{\mathrm{\Phi }(c_2(L_{a(\alpha ),x}),\mathrm{},c_{2n}(L_{a(\alpha ),x}))p(\mathrm{\Xi }^๐ค)_0}{\mathrm{det}^{1/2}L_{a(\alpha ),x}}$$
(21)
where the classes $`c_i(L_{a(\alpha ),x})`$ are the equivariant Chern classes of the endomorphism $`L_{a(\alpha ),x}`$ acting on the tangent space $`T_xM`$ at a zero $`x`$ of $`a(\alpha )`$ (cf. ).
###### Proof.
The r.h.s. of Eq. (21) is computed from the r.h.s. of the formula (13), taking into account two facts. First, we can evaluate the equivariant characteristic classes involved in $`\mathrm{\Phi }_\mathrm{\Xi }^๐ค(A)`$ by choosing in the principal bundle $`GL(A)`$ an equivariant $`A`$-connection induced by an ordinary equivariant connection $`\zeta `$ on the vector bundle $`A`$. In this way, we have
$$\mathrm{\Phi }(\lambda _2^๐ค(A),\mathrm{},\lambda _{2n}^๐ค(A))=a^{}(\mathrm{\Phi }(\nu _2^๐ค(A),\mathrm{},\nu _{2n}^๐ค(A))$$
where the classes $`\nu _i^๐ค`$ are equivariant Chern classes for the complexification of the vector bundle $`A`$. Secondly, if $`R_\zeta ^๐ค`$ is the equivariant curvature of $`\zeta `$, for every symmetric elementary function $`\varsigma _i`$ and every zero $`x`$ of $`a(\alpha )`$, we have
$$\left(\varsigma _i(R_\zeta ^๐ค)\right)_0(x)=c_i(L_{a(\alpha ),x}).$$
When $`A=TM`$, formula (21) reduces to the ordinary Bott formula.
## 6 Final remarks
As we shall show in a forthcoming paper , Theorem 5.1 generalizes several localization formulas, for instance associated with the action of a holomorphic vector field on a complex manifold which is equivariantly lifted to an action on a holomorphic vector bundle, see , and reproduces in particular Grothendieckโs residue theorem.
On the other hand, our formula can be generalized in several directions. One of these would be a localization formula for an equivariant cohomology associated with the action of a Lie group on a Courant algebroid. This should encompass several formulas recently appeared in the literature, mostly concerned with generalized Calabi-Yau structures and should reproduce our formula when the Courant algebroid reduces to a Lie algebroid.
|
warning/0506/math0506353.html
|
ar5iv
|
text
|
# Quasistatic crack growth for a cohesive zone model with prescribed crack path
## 1. INTRODUCTION
In this paper we present a variational model for quasistatic crack growth in the presence of a cohesive force exerted between the lips of the crack.
The evolution of the crack is governed by an energy which is the sum of three terms: the bulk energy of the uncracked part, the energy dissipated in the fracture process, and the work of the external loads. The main mathematical difficulty is given by the fact that the fracture energy depends on the opening of the crack. For this reason we cannot apply directly the tools developed so far in the applications to fracture mechanics of the theory of free discontinuity problems (see , , , , , , ).
To simplify the mathematical difficulties, we assume that the crack path is prescribed, and we focus only on the time evolution. This allows us to consider very general bulk and crack energies, which may include constraints on the crack opening, related to the infinitesimal noninterpenetration of matter. The evolution of the crack is defined (see Definition 3.4 below) in the framework of Mielkeโs approach to a variational theory of rate-independent processes (see , ).
We prove an existence result for the quasistatic evolution, by approximating the continuous-time problem by discrete-time problems, for which the evolution is defined by solving incremental minimum problems. The irreversibility of the crack process leads to introduce an auxiliary time-dependent function $`t\gamma (t)`$ (see Section 2 below), defined on the prescribed crack path, which takes into account the local history of the crack up to time $`t`$. The main mathematical difficulty in the proof is the compactness of the approximating functions $`t\gamma _k(t)`$. This is solved by introducing a new notion of convergence of functions related to the problem, with good compactness and semicontinuity properties.
## 2. SETTING
The reference configuration is a bounded open set $`\mathrm{\Omega }`$ of $`^n`$ with Lipschitz boundary $`\mathrm{\Omega }`$, which can be written as the union of two disjoint Borel sets $`_0\mathrm{\Omega }`$ and $`_1\mathrm{\Omega }`$, with $`^{n1}(_0\mathrm{\Omega })>0`$ and $`_1\mathrm{\Omega }`$ relatively open. Here and henceforth $`^{n1}`$ denotes the $`(n1)`$-dimensional Hausdorff measure. On $`_0\mathrm{\Omega }`$, the Dirichlet part of the boundary, we will assign the boundary deformation, while on $`_1\mathrm{\Omega }`$, the Neumann part of the boundary, we will prescribe surface forces.
We assume that the cracks are contained in a compact $`C^1`$-orientable $`(n1)`$-dimensional manifold $`M\mathrm{\Omega }`$ with boundary $`M`$, such that $`\mathrm{\Omega }M`$ is connected. Therefore it is reasonable to take the deformation $`u`$ as a function in the space $`W^{1,p}(\mathrm{\Omega }M;^m)`$, so that the essential discontinuity points of $`u`$ are contained in $`M`$. Although the natural choice is $`m=n`$, there are no mathematical difficulties in considering an arbitrary $`m1`$. The case $`m=1`$ is used in the study of antiplane shears. The number $`p>1`$ depends on the bounds on the energy density considered below.
We take into account prescribed time-dependent boundary deformations $`t\psi (t)`$, with $`\psi (t)W^{1,p}(\mathrm{\Omega };^m)`$, in the sense that for each time $`t[0,T]`$ we consider only deformations $`uW^{1,p}(\mathrm{\Omega }M;^m)`$ such that
$$u=\psi (t)\text{on }_0\mathrm{\Omega },$$
where the previous equality has to be considered in the sense of traces. We assume also that, as a function of time, $`t\psi (t)`$ is absolutely continuous from $`[0,T]`$ into $`W^{1,p}(\mathrm{\Omega };^m)`$.
Thus the time derivative $`t\dot{\psi }(t)`$ belongs to the space $`L^1([0,T];W^{1,p}(\mathrm{\Omega };^m))`$ and its spatial gradient $`t\dot{\psi }(t)`$ belongs to the space $`L^1([0,T];L^p(\mathrm{\Omega };๐^{m\times n}))`$.
We assume that the uncracked part of the body is hyperelastic and that its bulk energy relative to the deformation $`uW^{1,p}(\mathrm{\Omega }M;^m)`$ is of the form
$$_{\mathrm{\Omega }M}W(x,u)๐x,$$
where $`W(x,\xi )`$ is a given Carathรฉodory function $`W:(\mathrm{\Omega }M)\times ๐^{m\times n}`$ such that
* $`\xi W(x,\xi )`$ is quasiconvex and $`C^1`$ for every $`x\mathrm{\Omega }M;`$
* there are two positive constants $`a_0,a_1`$ and two nonnegative functions $`b_0,b_1L^1(\mathrm{\Omega }M)`$ such that
$$a_0\left|\xi \right|^pb_0(x)W(x,\xi )a_1\left|\xi \right|^p+b_1(x),$$
(2.1)
for every $`(x,\xi )(\mathrm{\Omega }M)\times ๐^{m\times n}`$.
Since $`\xi W(x,\xi )`$ is rank-one convex on $`๐^{m\times n}`$ for every $`x\mathrm{\Omega }M`$, we can deduce from $`(\text{2.1})`$ an estimate for the partial gradient of $`W`$ with respect to $`\xi `$, $`_\xi W:(\mathrm{\Omega }M)\times ๐^{m\times n}๐^{m\times n}`$. More precisely, there are a positive constant $`a_2`$ and a nonnegative function $`b_2L^1(\mathrm{\Omega }M)`$ such that
$$\left|_\xi W(x,\xi )\right|a_2\left|\xi \right|^{p1}+b_2(x),$$
(2.2)
for every $`(x,\xi )(\mathrm{\Omega }M)\times ๐^{m\times n}`$.
To shorten the notation we introduce the function $`๐ฒ:L^p(\mathrm{\Omega }M;๐^{m\times n})`$ defined by
$$๐ฒ(\mathrm{\Psi }):=_{\mathrm{\Omega }M}W(x,\mathrm{\Psi })๐x,$$
for every $`\mathrm{\Psi }L^p(\mathrm{\Omega }M;๐^{m\times n})`$. By $`(\text{2.1})`$ and $`(\text{2.2})`$ the functional $`๐ฒ`$ is of class $`C^1`$ on $`L^p(\mathrm{\Omega }M;๐^{m\times n})`$ and its differential $`๐ฒ:L^p(\mathrm{\Omega }M;๐^{m\times n})L^q(\mathrm{\Omega }M;๐^{m\times n})`$, $`p^1+q^1=1`$, is given by
$$๐ฒ(\mathrm{\Psi }),\mathrm{\Phi }=_{\mathrm{\Omega }M}_\xi W(x,\mathrm{\Psi }):\mathrm{\Phi }dx,$$
for every $`\mathrm{\Phi }`$, $`\mathrm{\Psi }L^p(\mathrm{\Omega }M;๐^{m\times n})`$, where $`,`$ denotes the duality pairing between the spaces $`L^q(\mathrm{\Omega }M;๐^{m\times n})`$ and $`L^p(\mathrm{\Omega }M;๐^{m\times n})`$, and $`_\xi W(x,\mathrm{\Psi }):\mathrm{\Phi }`$ denotes the scalar product between the two matrices $`_\xi W(x,\mathrm{\Psi })`$ and $`\mathrm{\Phi }`$.
By the assumptions on $`W`$, the functions $`๐ฒ`$ and $`๐ฒ`$ satisfy the following properties: there are two positive constants $`\alpha _0,\alpha _1`$ and two nonnegative constants $`\beta _0,\beta _1`$ such that
$$\alpha _0\mathrm{\Psi }_p^p\beta _0๐ฒ(\mathrm{\Psi })\alpha _1\mathrm{\Psi }_p^p+\beta _1,$$
(2.3)
for every $`\mathrm{\Psi }L^p(\mathrm{\Omega }M;๐^{m\times n})`$, and there is a positive constant $`\alpha _2`$ such that
$$๐ฒ(\mathrm{\Psi }),\mathrm{\Phi }\alpha _2(1+\mathrm{\Psi }_p^{p1})\mathrm{\Phi }_p,$$
(2.4)
for every $`\mathrm{\Psi }`$, $`\mathrm{\Phi }L^p(\mathrm{\Omega }M;๐^{m\times n})`$.
For a fixed time $`t[0,T]`$, we assume that the external time-dependent loads $`(t)`$ belong to $`(W^{1,p}(\mathrm{\Omega }M;^m))^{}`$, the dual space of $`W^{1,p}(\mathrm{\Omega }M;^m)`$. The duality product $`(t),u`$ is interpreted as the work done by the loads on the deformation $`u`$.
Let us fix an orientation of $`M`$ and let $`u^{}`$ be the trace of $`u`$ on the positive side of $`M`$, and $`u^{}`$ be the trace of $`u`$ on the negative side of $`M`$. The most general form of the work done by the external loads is given by
$$\begin{array}{cc}\hfill (t),u=& _{\mathrm{\Omega }M}f(t)u๐x+_{\mathrm{\Omega }M}H(t):udx+\hfill \\ & +_{_1\mathrm{\Omega }}g(t)u๐^{n1}+_M(g^{}(t)u^{}+g^{}(t)u^{})๐^{n1},\hfill \end{array}$$
(2.5)
where $`f(t)L^q(\mathrm{\Omega }M;^m)`$, $`H(t)L^q(\mathrm{\Omega }M;๐^{m\times n})`$, $`g(t)L^q(_1\mathrm{\Omega };^m)`$, $`g^{}(t)`$ and $`g^{}(t)L^q(M;^m)`$, with $`p^1+q^1=1`$. Actually the representation theorem for $`(W^{1,p}(\mathrm{\Omega }M;^m))^{}`$ shows that it is enough to use just the terms of the first line of (2.5). The terms in the second line have been added in order to write in an explicit way the contribution of the surface forces acting on the Neumann part of the boundary and on one or both sides of $`M`$.
With these assumptions we do not exclude the possibility that $`H(t)`$ could be discontinuous on $`M`$. Moreover, observe that if $`f(t),H(t),g(t),g^{}(t)`$ and $`g^{}(t)`$ are sufficiently regular, then
$$f(t)\text{ div }H(t)$$
plays the role of the volume forces on $`\mathrm{\Omega }M`$,
$$g(t)+H(t)\nu $$
plays the role of the surface forces on $`_1\mathrm{\Omega }`$, and
$$g^{}(t)H^{}(t)\nu \text{and}g^{}(t)+H^{}(t)\nu $$
play the role of the surface forces acting on the positive (respectively negative) side of $`M`$, where $`\nu `$ is the outer unit normal to $`(\mathrm{\Omega }M)`$. We observe that, by our positions, $`\nu `$ turns out to be the inner normal on the positive side of $`M`$; this is why in the last formula we take the minus sign in front of $`H^{}(t)\nu `$.
We assume that, as a function of time, $`t(t)`$ is absolutely continuous from $`[0,T]`$ into $`(W^{1,p}(\mathrm{\Omega }M;^m))^{}`$. Thus the time derivative $`t\dot{}(t)`$ belongs to the space $`L^1([0,T];(W^{1,p}(\mathrm{\Omega }M;^m))^{})`$. If $`(t)`$ is represented by (2.5), then the absolute continuity of $`t(t)`$ follows from the absolute continuity of the functions $`tf(t)`$, $`tH(t)`$, $`tg(t)`$, $`tg^{}(t)`$, and $`tg^{}(t)`$.
If the deformation $`u`$ has a nonzero jump $`[u]=u^{}u^{}`$ on $`M`$, then the body has a crack on (part of) $`M`$. More precisely the crack is given by the set
$$\{xM:[u](x)0\}.$$
Let us consider now the work done to produce a crack. If we neglect for a moment the problem of irreversibility, we may assume that this work can be written in the form
$$_M\phi (x,[u])๐^{n1},$$
where $`\phi :M\times ^m[0,+\mathrm{}]`$ satisfies the following properties
* $`\phi `$ is a Borel function;
* $`\phi (x,0)=0`$ for $`^{n1}`$-a.e. $`xM`$;
* the function $`y\phi (x,y)`$ is lower semicontinuous on $`^m`$ for $`^{n1}`$-a.e. $`xM`$.
A simple example is given by the function
$`\phi (x,y)`$ $`:=`$ $`\{\begin{array}{cc}a+b|y|\hfill & \text{if }y^m\{0\},\hfill \\ 0\hfill & \text{if }y=0,\hfill \end{array}`$ (2.6)
where $`a0`$ and $`b0`$ are real constants. The constant $`a`$ plays the role of an activation energy; if $`b>0`$, there is also an energy term proportional to the amplitude of the crack opening. The classical Griffithโs model corresponds to the case $`a>0`$ and $`b=0`$.
Let $`L^0(M)`$ be the set of extended real valued measurable functions on $`M`$ and let $`L^0(M)^+`$ be the set of functions $`wL^0(M)`$ such that $`w0`$ $`^{n1}`$-a.e. on $`M`$.
We introduce the function $`\varphi :L^p(M;^m)L^0(M)^+`$ defined by
$$\varphi (w)(x):=\phi (x,w(x)),$$
for every $`wL^p(M;^m)`$ and for $`^{n1}`$-a.e. $`xM`$.
Given an arbitrary family $`(w_i)_{iI}`$ in $`L^0(M)^+`$ the essential supremum
$$w=\underset{iI}{\mathrm{ess}\mathrm{sup}}w_i$$
of the family is defined as the unique (up to $`^{n1}`$-equivalence) function in $`L^0(M)^+`$ such that
* $`ww_i`$ $`^{n1}`$-a.e. on $`M`$ for all $`iI`$;
* if $`zL^0(M)^+`$ and $`zw_i`$ $`^{n1}`$-a.e. on $`M`$, then $`zw`$ $`^{n1}`$-a.e. on $`M`$.
For the existence of such a function see, for instance, \[14, Proposition VI-1-1\].
Suppose now that the deformation $`u`$ depends on time, i.e., we have a map $`tu(t)`$ from $`[0,T]`$ into $`W^{1,p}(\mathrm{\Omega }M;^m)`$. If no crack is present until time $`0`$ and
$$\varphi ([u(s)])\varphi ([u(t)])^{n1}\text{-a.e. on }M$$
for every $`s[0,t]`$, then the energy dissipated in the crack process in the time interval $`[0,t]`$ is given, in our model, by
$$_M\varphi ([u(t)])๐^{n1}.$$
This happens for instance when $`s\varphi ([u(s)])`$ is monotonically increasing $`^{n1}`$-a.e. on $`M`$.
In the general case, the irreversibility of the fracture process leads to introduce an auxiliary function $`t\beta (t)`$ from $`[0,T]`$ to $`L^1(M)`$, which takes into account the history of the system up to time $`t`$. We assume that for every $`0t_1t_2T`$ we have
$$\beta (t_2)=\beta (t_1)\underset{t_1st_2}{\mathrm{ess}\mathrm{sup}}\varphi ([u(s)])^{n1}\text{-a.e. on }M\text{,}$$
(2.7)
so that
$$\beta (t_2)\beta (t_1)=\underset{t_1st_2}{\mathrm{ess}\mathrm{sup}}(\varphi ([u(s)])\beta (t_1))^+^{n1}\text{-a.e. on }M,$$
where for every $`a`$, $`a^+:=a0`$ denotes the positive part of $`a`$.
In particular
* $`t\beta (t)`$ is increasing, i.e., $`\beta (t_1)\beta (t_2)`$ $`^{n1}`$-a.e. on $`M`$ for $`0t_1t_2T`$;
* $`\beta (t)\varphi ([u(t)])`$ $`^{n1}`$-a.e. on $`M`$ for every $`t[0,T]`$.
In our model the energy dissipated in the time interval $`[t_1,t_2]`$ is given by
$$\beta (t_2)\beta (t_1)_{1,M}:=_M(\beta (t_2)\beta (t_1))๐^{n1}.$$
According to this assumption there is no dissipation in the intervals $`[t_1,t_2]`$ where $`\varphi ([u(s)])\beta (t_1)`$ $`^{n1}`$-a.e. on $`M`$ for every $`s[t_1,t_2]`$, while the dissipation is given by
$$_M\left(\varphi ([u(t_2)])\varphi ([u(t_1)])\right)๐^{n1}$$
whenever $`\beta (t_1)\varphi ([u(s)])\varphi ([u(t_2)])`$ for every $`s[t_1,t_2]`$.
It follows from (2.7) that $`\beta (t)`$ is uniquely determined by $`\beta (0)`$ and by the history of the deformation $`su(s)`$ in the interval $`[0,t]`$. Since it is difficult to deal with (2.7) directly, we prefer to define the notion of quasistatic evolution by considering a more general internal variable $`t\gamma (t)`$ which is assumed to satisfy the following weaker conditions:
* $`t\gamma (t)`$ is increasing, i.e., $`\gamma (t_1)\gamma (t_2)`$ $`^{n1}`$-a.e. on $`M`$ for $`0t_1t_2T`$;
* $`\gamma (t)\varphi ([u(t)])`$ $`^{n1}`$-a.e. on $`M`$ for every $`t[0,T]`$.
We do not assume from the beginning that $`t\gamma (t)`$ satisfies (2.7). This property will be a nontrivial consequence of the other conditions considered in the definition of quasistatic evolution (see Theorem 3.7).
Given functions $`\psi W^{1,p}(\mathrm{\Omega };^m)`$ and $`\gamma L^0(M)^+`$, it is convenient to introduce the set $`AD(\psi ,\gamma )`$ of admissible deformations with boundary value $`\psi `$ on $`_0\mathrm{\Omega }`$ and internal variable $`\gamma `$. It is defined by
$$AD(\psi ,\gamma ):=\{uW^{1,p}(\mathrm{\Omega }M;^m):\varphi ([u])\gamma \text{ on }M\text{, and }u=\psi \text{ on }_0\mathrm{\Omega }\},$$
where equalities and inequalities are considered $`^{n1}`$-a.e., and the last equality refers to the traces of $`u`$ and $`\psi `$ on $`_0\mathrm{\Omega }`$.
An *admissible configuration* with boundary value $`\psi `$ on $`_0\mathrm{\Omega }`$ is a pair $`(u,\gamma )`$, with $`\gamma L^1(M)^+:=L^1(M)L^0(M)^+`$ and $`uAD(\psi ,\gamma )`$.
## 3. Definition and properties of quasistatic evolutions
For every $`t[0,T]`$, the total energy of an admissible configuration $`(u,\gamma )`$ at time $`t`$ is defined as
$$(t)(u,\gamma ):=๐ฒ(u)(t),u+\gamma _{1,M},$$
where $`_{1,M}`$ denotes the $`L^1`$-norm on $`M`$.
We now introduce the following definition in the spirit of Griffithโs original theory on the crack propagation.
###### Definition 3.1.
A pair $`(u,\gamma )W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ is globally stable at time $`t[0,T]`$ if $`uAD(\psi (t),\gamma )`$ and
$$(t)(u,\gamma )(t)(v,\delta )$$
(3.1)
for every $`\delta \gamma `$ and for every $`vAD(\psi (t),\delta )`$.
In other words, the total energy of $`(u,\gamma )`$ at time $`t`$ cannot be reduced by increasing the internal variable $`\gamma `$ or by choosing a new admissible deformation with the same boundary condition.
###### Remark 3.2.
For every $`t[0,T]`$ let $`(u(t),\gamma (t))W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ be globally stable at time $`t`$. By Definition 3.1 we can deduce an a priori estimate on $`u(t)`$. Indeed, by comparing $`(t)(u(t),\gamma (t))`$ with $`(t)(\psi (t),\gamma (t))`$, which is bounded uniformly with respect to $`t`$, we get that $`๐ฒ(u(t))(t),u(t)`$ is bounded uniformly in time. Next, by the assumption (2.3) on $`๐ฒ`$ and the boundedness of $`(t)`$ in $`(W^{1,p}(\mathrm{\Omega }M;^m))^{}`$, we obtain that the $`W^{1,p}`$-norm of $`u(t)`$, $`u(t)_{1,p}`$, is bounded uniformly with respect to $`t`$. Furthermore from this fact and by Definition 3.1 we get that the crack term $`\gamma (t)_{1,M}`$ is bounded uniformly in time, too.
###### Remark 3.3.
Condition $`(\text{3.1})`$ is equivalent to
$$(t)(u,\gamma )(t)(v,\gamma \varphi ([v])),$$
for every $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`v=\psi (t)`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$. This is equivalent to
$$๐ฒ(u)(t),u๐ฒ(v)(t),v+(\varphi ([v])\gamma )^+_{1,M}$$
(3.2)
for every $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`v=\psi (t)`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$. This implies that if $`(u,\gamma )`$ is globally stable at time $`t`$ and $`\stackrel{~}{\gamma }L^1(M)^+`$ satisfies $`\varphi ([u])\stackrel{~}{\gamma }\gamma `$ $`^{n1}`$-a.e. on $`M`$, then $`(u,\stackrel{~}{\gamma })`$ is globally stable at time $`t`$.
###### Definition 3.4.
An irreversible quasistatic evolution of minimum energy configurations is a function $`t(u(t),\gamma (t))`$ from $`[0,T]`$ into $`W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ which satisfies the following conditions:
* global stability: for every $`t[0,T]`$ the pair $`(u(t),\gamma (t))`$ is globally stable at time $`t`$;
* irreversibility: $`\gamma (s)\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$ for every $`0stT`$;
* energy balance: the function $`t(t)(u(t),\gamma (t))`$ is absolutely continuous on $`[0,T]`$ and
$$\frac{d}{dt}((t)(u(t),\gamma (t)))=๐ฒ(u(t)),\dot{\psi }(t)(t),\dot{\psi }(t)\dot{}(t),u(t),$$
for a.e. $`t[0,T]`$.
###### Remark 3.5.
Condition (c) is equivalent to the following one:
* energy balance in integral form: the function $`t๐ฒ(u(t)),\dot{\psi }(t)\dot{}(t),u(t)`$ belongs to $`L^1([0,T])`$ and
$`(t)(u(t),\gamma (t))(0)(u(0),\gamma (0))=`$
$`={\displaystyle _0^t}\left(๐ฒ(u(s)),\dot{\psi }(s)(s),\dot{\psi }(s)\dot{}(s),u(s)\right)๐s`$
for every $`t[0,T]`$.
This can be written in the form
$`๐ฒ(u(t))๐ฒ(u(0))+\gamma (t)\gamma (0)_{1,M}=`$
$`={\displaystyle _0^t}\left(๐ฒ(u(s)),\dot{\psi }(s)(s),\dot{\psi }(s)\right)๐s+`$ (3.3)
$`+(t),u(t)(0),u(0){\displaystyle _0^t}\dot{}(s),u(s)๐s,`$
for every $`t[0,T]`$. The first line is the increment in stored energy plus a term which will be interpreted as the energy dissipated by the crack process in the time interval $`[0,t]`$, as we shall see in Remark 3.8. Using the divergence theorem we can show that the second line represents the work done in the same time interval by the forces which act on $`_0\mathrm{\Omega }`$ to produce the imposed deformation. The third line represents the work done by the imposed forces in the interval $`[0,t]`$; this follows from an integration by parts when $`tu(t)`$ is regular enough, and can be obtained by approximation in the other cases.
If $`t(u(t),\gamma (t))`$ satisfies condition (a), then $`(u(t),\gamma (t))`$ is bounded in $`W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ by Remark 3.2. Therefore in condition (cโ) it is enough to assume that $`t๐ฒ(u(t)),\dot{\psi }(t)\dot{}(t),u(t)`$ is measurable.
In the following theorem we prove one inequality of the energy balance.
###### Theorem 3.6.
Let $`t(u(t),\gamma (t))`$ be a function from $`[0,T]`$ into $`W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ which satisfies the global stability condition $`(a)`$ and the irreversibility condition $`(b)`$ of Definition 3.4. Assume that $`t๐ฒ(u(t)),\dot{\psi }(t)\dot{}(t),u(t)`$ is measurable. Then
$`(t)(u(t),\gamma (t))(0)(u(0),\gamma (0))`$
$`{\displaystyle _0^t}\left(๐ฒ(u(s)),\dot{\psi }(s)(s),\dot{\psi }(s)\dot{}(s),u(s)\right)๐s`$
for every $`t[0,T]`$.
###### Proof.
We note that $`t๐ฒ(u(t)),\dot{\psi }(t)\dot{}(t),u(t)`$ belongs to $`L^1([0,T])`$ by the estimates of Remark 3.2. The result can now be obtained arguing as in (see the proof of Lemma 7.1 and the final part of the proof of Theorem 3.15). โ
Now we prove that for a quasistatic evolution $`t(u(t),\gamma (t))`$, the internal variable $`t\gamma (t)`$ satisfies a condition analogous to (2.7).
###### Theorem 3.7.
Let $`t(u(t),\gamma (t))`$ be a quasistatic evolution. Then
$$\gamma (t_2)=\gamma (t_1)\underset{t_1st_2}{\mathrm{ess}\mathrm{sup}}\varphi ([u(s)])^{n1}\text{-a.e. on }M,$$
(3.4)
for every $`0t_1t_2T`$.
###### Proof.
It is enough to prove that
$$\gamma (t)=\gamma (0)\underset{0st}{\mathrm{ess}\mathrm{sup}}\varphi ([u(s)])^{n1}\text{-a.e. on }M,$$
(3.5)
for every $`t[0,T]`$. Let $`\stackrel{~}{\gamma }(t)`$ be the right-hand side of (3.5). Since $`t\gamma (t)`$ is increasing and $`\varphi ([u(t)])\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$ for every $`t[0,T]`$, it follows that $`\stackrel{~}{\gamma }(t)\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$ for every $`t[0,T]`$. As $`\varphi ([u(t)])\stackrel{~}{\gamma }(t)`$ $`^{n1}`$-a.e. on $`M`$, by Remark 3.3 the pair $`(u(t),\stackrel{~}{\gamma }(t))`$ is globally stable at time $`t`$ for every $`t[0,T]`$. Since $`t\stackrel{~}{\gamma }(t)`$ is increasing, we can apply Theorem 3.6 and we obtain
$`(t)(u(t),\stackrel{~}{\gamma }(t))(0)(u(0),\gamma (0))`$
$`{\displaystyle _0^t}\left(๐ฒ(u(s)),\dot{\psi }(s)(s),\dot{\psi }(s)\dot{}(s),u(s)\right)๐s`$
for every $`t[0,T]`$. By the energy balance (c) it follows that $`(t)(u(t),\stackrel{~}{\gamma }(t))(t)(u(t),\gamma (t))`$, i.e.,
$`๐ฒ(u(t))(t),u(t)+\stackrel{~}{\gamma }(t)_{1,M}๐ฒ(u(t))(t),u(t)+\gamma (t)_{1,M},`$
which implies $`\stackrel{~}{\gamma }(t)_{1,M}\gamma (t)_{1,M}`$. As $`\stackrel{~}{\gamma }(t)\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$, we deduce that $`\stackrel{~}{\gamma }(t)=\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$ for every $`t[0,T]`$, which concludes the proof. โ
Theorem 3.7 can be used to explain the mechanical meaning of the internal variable $`\gamma `$ in the model case $`\phi (x,y):=|y|`$. Indeed, if $`t(u(t),\gamma (t))`$ is a quasistatic evolution with $`\gamma (0)=0`$ and $`\phi (x,y):=|y|`$, then (3.4) shows that $`\gamma (t)(x)`$ coincides with the maximum modulus of the amplitude of the opening reached by the crack at $`x`$ up to time $`t`$.
###### Remark 3.8.
As $`t\gamma (t)`$ satisfies (2.7) by Theorem 3.7, the mechanical interpretation given in Section 2 shows that the term $`\gamma (t)\gamma (0)_{1,M}`$ in (3.5) represents the energy dissipated in the crack process in the time interval $`[0,t]`$.
###### Remark 3.9.
In our model, the dissipation term in the energy functional comes from the expression $`\gamma \varphi ([v])\gamma _{1,M}`$ and is nonlinear in $`\gamma `$. This turns out to be the main mathematical difference between our model and the model considered by Mielke and Mainik and Mielke in \[12, Section 4.2\] and \[11, Section 6.2\], where the dissipation term is linear.
We are now in a position to state our main result.
###### Theorem 3.10.
Let $`(u_0,\gamma _0)W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ be globally stable at time $`t=0`$. Then there exists an irreversible quasistatic evolution $`t(u(t),\gamma (t))`$ such that $`(u(0),\gamma (0))=(u_0,\gamma _0)`$.
## 4. Some tools
We introduce a notion of convergence for the functions $`\gamma `$, which is the counterpart of the notion of convergence of sets introduced in . The main property of this convergence is that, if $`u_k`$ converges weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ to some function $`u`$ and $`\varphi ([u_k])\gamma _k`$ $`^{n1}`$-a.e. on $`M`$, then $`\varphi ([u])\gamma `$ $`^{n1}`$-a.e. on $`M`$.
###### Definition 4.1.
Let $`\gamma _k,\gamma L^0(M)^+`$. We say that $`\gamma _k`$ $`\sigma _\phi ^p`$-converges to $`\gamma `$ if the following two conditions are satisfied:
* if $`u_ju`$ weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ and $`\varphi ([u_j])\gamma _{k_j}`$ $`^{n1}`$-a.e. on $`M`$ for some sequence $`k_j\mathrm{}`$, then $`\varphi ([u])\gamma `$ $`^{n1}`$-a.e. on $`M`$;
* there exist a sequence $`u^iW^{1,p}(\mathrm{\Omega }M;^m)`$, with $`sup_i\varphi ([u^i])=\gamma `$ $`^{n1}`$-a.e. on $`M`$, and, for every $`i`$, a sequence $`u_k^iW^{1,p}(\mathrm{\Omega }M;^m)`$, converging to $`u^i`$ weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ as $`k\mathrm{}`$, such that $`\varphi ([u_k^i])\gamma _k`$ $`^{n1}`$-a.e. on $`M`$ for every $`i`$ and $`k`$.
Notice that we do not require any upper bound in $`L^1(M)^+`$ for the functions $`\gamma _k`$.
###### Remark 4.2.
If $`\gamma _k`$ $`\sigma _\phi ^p`$-converges to $`\gamma `$, then in particular there are functions $`u_k^i`$ and $`u^i`$ in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ such that condition (b) in Definition 4.1 holds. We define for every $`k`$
$$\gamma ^i:=\underset{j=1,\mathrm{},i}{sup}\varphi ([u^j])\text{and}\gamma _k^i:=\underset{j=1,\mathrm{},i}{sup}\varphi ([u_k^j]).$$
With this notation it turns out that
$$\gamma =\underset{i\mathrm{}}{lim}\gamma ^i\text{ and }\gamma _k\underset{i}{sup}\gamma _k^i,$$
for every $`k`$.
###### Remark 4.3.
If $`\gamma _k`$ $`\sigma _\phi ^p`$-converges to $`\gamma `$, then
$$\gamma \underset{k\mathrm{}}{lim\; sup}\gamma _k,^{n1}\text{-a.e. on }M,$$
as we can see by modifing the proof of Lemma 4.4 below. Notice that the inequality can be strict, even when $`\gamma _k`$ converges pointwise to a function $`\stackrel{~}{\gamma }`$. As an example, consider $`n=2`$, $`m=1`$, $`p=2`$, $`\mathrm{\Omega }=]2,2[^2`$ and $`M=[0,1]\times \{0\}`$. Let $`\gamma _kL^0(M)^+`$ be defined as follows:
$$\gamma _k(x):=\{\begin{array}{cc}1\hfill & \text{for }x[\frac{i}{k},\frac{i+1}{k}\frac{1}{k^2}[;\hfill \\ & \\ 0\hfill & \text{for }x[\frac{i+1}{k}\frac{1}{k^2},\frac{i+1}{k}[;\hfill \end{array}\text{for }i=0,\mathrm{},k1.$$
It follows from homogenization theory (see , , ) that condition (a) in Definition 4.1 is satisfied with $`\gamma =0`$, hence $`\gamma _k`$ $`\sigma _\phi ^2`$-converges to $`0`$. Furthermore $`\gamma _k`$ converge in measure to $`1`$, so up to a subsequence we have pointwise convergence to $`1=:\stackrel{~}{\gamma }>\gamma `$.
We prove in the following lemma that the $`L^1`$-norm is lower semicontinuous with respect to $`\sigma _\phi ^p`$-convergence.
###### Lemma 4.4.
Let $`\gamma _k`$, $`\gamma L^0(M)^+`$. If $`\gamma _k`$ $`\sigma _\phi ^p`$-converges to $`\gamma `$ then
$$\gamma _{1,M}\underset{k\mathrm{}}{lim\; inf}\gamma _k_{1,M}.$$
(4.1)
###### Proof.
From the hypothesis it follows in particular that there are functions $`u_k^i`$ and $`u^i`$ in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ which satisfy condition (b) in Definition 4.1. With notation from Remark 4.2, let us prove that for every $`i`$
$$\gamma ^i_{1,M}\underset{k\mathrm{}}{lim\; inf}\gamma _k^i_{1,M}.$$
(4.2)
Extracting a subsequence we may assume that $`lim\; inf_k\gamma _k^i_{1,M}`$ is a limit. As $`[u_k^j][u^j]`$ strongly in $`L^p(M;^m)`$ for $`j=1,\mathrm{},i`$, we can extract a further subsequence such that $`[u_k^j][u^j]`$ pointwise $`^{n1}`$-a.e. on $`M`$ for $`j=1,\mathrm{},i`$. By the lower semicontinuity assumption $`(\phi _3)`$ this implies
$$\gamma ^i\underset{k\mathrm{}}{lim\; inf}\gamma _k^i^{n1}\text{-a.e. on }M.$$
By the Fatou lemma we obtain (4.2), which yields
$$\gamma ^i_{1,M}\underset{k\mathrm{}}{lim\; inf}\gamma _k_{1,M}.$$
We then pass to the limit as $`i`$ tends to infinity and obtain (4.1). โ
We now prove a compactness result for the notion of $`\sigma _\phi ^p`$-convergence.
###### Lemma 4.5.
Every sequence in $`L^0(M)^+`$ has a $`\sigma _\phi ^p`$-convergent subsequence.
###### Proof.
Let us denote the $`L^p`$-norm by $`_p`$. Let $`\gamma _kL^0(M)^+`$, let $`w_hL^{\mathrm{}}(\mathrm{\Omega }M;^m)`$ be dense in $`L^p(\mathrm{\Omega }M;^m)`$, and, for every positive integers $`l`$, $`h`$, and $`k`$, let us consider the problem
$$\mathrm{min}\left\{u_p^p+\mathrm{}uw_h_p^p\right\},$$
(4.3)
where the minimum is taken over all functions $`uW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`\varphi ([u])\gamma _k^{n1}`$-a.e. on $`M`$.
To prove that the minimum is achieved, we take a minimizing sequence and we easily obtain that it is bounded in $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Then, up to a subsequence, we can pass to the limit and by using our lower semicontinuity assumption $`(\phi _3)`$ we can prove that the limit function is actually a solution to the minimum problem (4.3). This solution, which is unique by strict convexity, will be denoted by $`u_k^{\mathrm{},h}`$. Notice that this function is bounded in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ uniformly with respect to $`k`$, thus, up to a subsequence, we can pass to the limit in $`k`$ and get that there is a function $`u^{\mathrm{},h}`$ such that $`u_k^{\mathrm{},h}u^{\mathrm{},h}`$ weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Further we define
$$\gamma :=\underset{\mathrm{},h}{sup}\varphi ([u^{\mathrm{},h}])^{n1}\text{-a.e. on }M\text{.}$$
(4.4)
In this way point (b) of Definition 4.1 is automatically satisfied.
We need to prove point (a). To this aim, let $`v_jv`$ weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ be such that $`\varphi ([v_j])\gamma _{k_j}`$ $`^{n1}`$-a.e. on $`M`$ for some sequence $`k_j\mathrm{}`$. We want to prove that $`\varphi ([v])\gamma `$ $`^{n1}`$-a.e. on $`M`$. By density there is a subsequence of $`w_h`$, say $`w_{h_i}`$, which converges strongly to $`v`$ in $`L^p(\mathrm{\Omega }M;^m)`$. Let $`\mathrm{}_i+\mathrm{}`$ be such that $`\mathrm{}_ivw_{h_i}_p^p0`$ as $`i`$ tends to infinity. By the minimality of $`u_{k_j}^{\mathrm{}_i,h_i}`$, we have
$$u_{k_j}^{\mathrm{}_i,h_i}_p^p+\mathrm{}_iu_{k_j}^{\mathrm{}_i,h_i}w_{h_i}_p^pv_j_p^p+\mathrm{}_iv_jw_{h_i}_p^p.$$
Then $`u_{k_j}^{\mathrm{}_i,h_i}`$ is bounded in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ uniformly with respect to $`j`$, and passing to the limit as $`j`$ tends to infinity we get
$$u^{\mathrm{}_i,h_i}_p^p+\mathrm{}_iu^{\mathrm{}_i,h_i}w_{h_i}_p^p\underset{j}{sup}v_j_p^p+\mathrm{}_ivw_{h_i}_p^p.$$
Since $`\mathrm{}_ivw_{h_i}_p^p0`$ as $`i`$ tends to infinity, this inequality ensures that $`u^{\mathrm{}_i,h_i}`$ is bounded in $`L^p(\mathrm{\Omega }M;^m)`$ uniformly with respect to $`i`$, and $`u_{k_j}^{\mathrm{}_i,h_i}w_{h_i}0`$ strongly in $`L^p(\mathrm{\Omega }M;^m)`$. As $`w_{h_i}v`$ strongly in $`L^p(\mathrm{\Omega }M;^m)`$, we deduce that $`u^{\mathrm{}_i,h_i}`$ converges weakly to $`v`$ in $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Then $`[u^{\mathrm{}_i,h_i}]`$ converges strongly to $`[v]`$ in $`L^p(M;^m)`$. Passing to a subsequence, we may also obtain pointwise convergence $`^{n1}`$-a.e. on $`M`$. By (4.4) we have $`\varphi ([u^{\mathrm{}_i,h_i}])\gamma `$ $`^{n1}`$-a.e. on $`M`$, so that the lower semicontinuity assumption $`(\phi _3)`$ yields $`\varphi ([v])\gamma `$ $`^{n1}`$-a.e. on $`M`$, which is precisely the conclusion to point (a) in the definition of $`\sigma _\phi ^p`$-convergence. โ
We shall use the following Helly-type compactness result. We recall that a function $`t\gamma (t)`$ from $`[0,T]`$ into $`L^0(M)^+`$ is said to be increasing if $`\gamma (s)\gamma (t)`$ $`^{n1}`$-a.e. on $`M`$, whenever $`0stT`$.
###### Lemma 4.6.
Let $`t\gamma _k(t)`$ be a sequence of increasing functions from $`[0,T]`$ into $`L^0(M)^+`$. Then there exist a subsequence $`\gamma _{k_j}`$, independent of $`t`$, and an increasing function $`t\gamma (t)`$ from $`[0,T]`$ into $`L^0(M)^+`$, such that $`\gamma _{k_j}(t)`$ $`\sigma _\phi ^p`$-converges to $`\gamma (t)`$ for every $`t[0,T]`$.
###### Proof.
Let $`D`$ be a countable dense subset of $`[0,T]`$ containing $`0`$ and $`T`$. By Lemma 4.5, using a diagonal argument, we can extract a subsequence, still named $`\gamma _k(t)`$, and an increasing function $`t\gamma (t)`$ from $`D`$ into $`L^0(M)^+`$, such that $`\gamma _k(t)`$ $`\sigma _\phi ^p`$-converges to $`\gamma (t)`$ for every $`tD`$.
Let us define
$$\gamma (t+):=\underset{st,sD}{inf}\gamma (s)\text{and}\gamma (t):=\underset{st,sD}{sup}\gamma (s),$$
for every $`t[0,T]`$. It is easy to prove that:
* $`\gamma (t)=\gamma (t)=\gamma (t+)`$ for every $`tD`$;
* $`\gamma (t)\gamma (t+)`$ for every $`t[0,T]`$;
* if $`s<t`$, then $`\gamma (s+)\gamma (t)`$.
Define $`E:=\{t[0,T]:\gamma (t+)=\gamma (t)^{n1}\text{-a.e. in }M\}`$ and $`\gamma (t):=\gamma (t)=\gamma (t+)`$ for every $`tE`$. Note that by (1) $`D`$ is contained in $`E`$ and the definition of $`\gamma (t)`$ agrees with the original one on $`D`$. Then the definition of $`\sigma _\phi ^p`$-convergence and the monotonicity condition imply that $`\gamma _k(t)`$ $`\sigma _\phi ^p`$-converges to $`\gamma (t)`$ for every $`tE`$.
Let us show now that the set $`E^c:=[0,T]E`$ is at most countable. For every pair of positive integers $`i,k`$ we set $`A_{i,k}:=\{t[0,T]:(\gamma (t+)k)(\gamma (t)k)_{1,M}>1/i\}`$, so that we have $`E^c`$ is the union of the sets $`A_{i,k}`$. Therefore it is enough to show that each set $`A_{i,k}`$ is finite. Let $`t_1<\mathrm{}<t_rA_{i,k}`$. Since, by (3), $`(\gamma (t_{j1}+)k)(\gamma (t_j)k)`$ for $`j=2,\mathrm{},r`$, we get
$$\frac{r}{i}\underset{j=1}{\overset{r}{}}(\gamma (t_j+)k)(\gamma (t_j)k)_{1,M}\gamma (t_r+)k_{1,M}k^{n1}(M),$$
so that $`rik^{n1}(M)`$, which implies that $`A_{i,k}`$ is finite. It follows that $`E^c`$ is at most countable, thus we can conclude the proof of the lemma by applying again the compactness Lemma 4.5 for every $`tE^c`$, together with a diagonal argument. โ
The following result plays a crucial role in the proof of point (a) in the Definition 3.4 of quasistatic evolution.
###### Lemma 4.7.
Let $`\gamma _k`$, $`\gamma L^0(M)^+`$. Assume that $`\gamma _k`$ $`\sigma _\phi ^p`$-converges to $`\gamma `$. Then for any $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ with $`\varphi ([v])L^1(M)^+`$ the following inequality holds true:
$$\underset{k\mathrm{}}{lim\; sup}(\varphi ([v])\gamma _k)^+_{1,M}(\varphi ([v])\gamma )^+_{1,M}.$$
(4.5)
###### Proof.
It is not restrictive to assume that the $`lim\; sup`$ is a limit. Let $`u^i`$ and $`u_k^i`$ be the functions considered in point (b) of Definition 4.1. During the proof we shall use the notation introduced in Remark 4.2. As $`\gamma _k^i\gamma _k`$ $`^{n1}`$-a.e. on $`M`$, we have
$$(\varphi ([v])\gamma _k)^+(\varphi ([v])\gamma _k^i)^+,$$
hence
$$\underset{k\mathrm{}}{lim}(\varphi ([v])\gamma _k)^+_{1,M}\underset{k\mathrm{}}{lim\; inf}(\varphi ([v])\gamma _k^i)^+_{1,M}.$$
(4.6)
Passing to a subsequence, we may assume that $`[u_k^i]`$ converges to $`[u^i]`$ $`^{n1}`$-a.e. on $`M`$. By the lower semicontinuity assumption $`(\phi _3)`$ we obtain
$$\gamma ^i\underset{k\mathrm{}}{lim\; inf}\gamma _k^i^{n1}\text{-a.e. on }M,$$
so that Fatou Lemma gives
$$\underset{k\mathrm{}}{lim\; sup}(\varphi ([v])\gamma _k^i)^+_{1,M}(\varphi ([v])\gamma ^i)^+_{1,M},$$
which, together with (4.6), yields
$$\underset{k\mathrm{}}{lim}(\varphi ([v])\gamma _k)^+_{1,M}(\varphi ([v])\gamma ^i)^+_{1,M}.$$
As $`\gamma ^i\gamma `$ $`^{n1}`$-a.e. on $`M`$, inequality (4.5) can be obtained by passing to the limit as $`i\mathrm{}`$. โ
###### Remark 4.8.
The conclusion of Lemma 4.7 does not hold, in general, when $`\gamma _k,\gamma L^{\mathrm{}}(M)^+`$ and $`\gamma _k\gamma `$ weakly\* in $`L^{\mathrm{}}(M)`$. Consider, for instance, the case $`n=2`$, $`m=1`$, $`\mathrm{\Omega }=]4,4[^2`$, $`M=[\pi ,\pi ]\times \{0\}`$, and define $`\gamma _k(x):=1+\mathrm{sin}(kx_1)`$, where $`x_1`$ denotes the first coordinate of $`x`$. Then, $`\gamma _k`$ converges to $`\gamma (x):=1`$ weakly\* in $`L^{\mathrm{}}(M)`$, but (4.5) is not satisfied for $`\varphi ([v])=1`$, since in this case $`(\varphi ([v])\gamma _k)^+_{1,M}=2`$ for every $`k`$, while $`(\varphi ([v])\gamma )^+_{1,M}=0`$.
## 5. The discrete-time problems and proof of the main result
In this section we prove Theorem 3.10 by a discrete-time approximation. We fix a sequence of subdivisions $`(t_k^i)_{0ik}`$ of the interval $`[0,T]`$, with
$`0=t_k^0<t_k^1<\mathrm{}<t_k^{k1}<t_k^k=T,`$ (5.1)
$`\underset{k\mathrm{}}{lim}\underset{1ik}{\mathrm{max}}(t_k^it_k^{i1})=0.`$ (5.2)
For $`i=1,\mathrm{},k`$ we set $`_k^i=(t_k^i)`$, $`\psi _k^i=\psi (t_k^i)`$, $`_k^i=(t_k^i)`$.
For every $`k`$ we define $`u_k^i`$ and $`\gamma _k^i`$ by induction as follows. Let $`(u_0,\gamma _0)`$ be a minimum energy configuration at time $`t=0`$. We set $`(u_k^0,\gamma _k^0):=(u_0,\gamma _0)`$ and define $`(u_k^i,\gamma _k^i)`$ as a solution of the minimum problem
$$\mathrm{min}\left\{_k^i(u,\gamma ):\gamma L^1(M)^+,\gamma \gamma _k^{i1},uAD(\psi _k^i,\gamma )\right\},$$
(5.3)
where the inequality means that $`\gamma \gamma _k^{i1}`$ $`^{n1}`$-a.e. on $`M`$.
###### Remark 5.1.
Consider the minimum problem
$$\mathrm{min}\left\{๐ฒ(u)_k^i,u+\varphi ([u])\gamma _k^{i1}_{1,M}:u=\psi _k^i\text{ on }_0\mathrm{\Omega }\right\},$$
(5.4)
where $`u`$ is assumed to belong to $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Then the following two conditions are equivalent:
* the pair $`(u_k^i,\gamma _k^i)`$ is a solution to (5.3);
* $`u_k^i`$ is a solution to (5.4) and $`\gamma _k^i:=\gamma _k^{i1}\varphi ([u_k^i])`$ $`^{n1}`$-a.e. on $`M`$.
The existence of a solution of (5.3) (or equivalently (5.4)) can be easily obtained by using the direct methods of the calculus of variations. The compactness of a minimizing sequence follows from (2.3) and positiveness of $`\phi `$. The lower semicontinuity follows from $`(W_1)`$, $`(W_2)`$, $`(\phi _3)`$, and from the compactness of the trace operator.
For every $`t[0,T]`$ we define
$$\begin{array}{c}\tau _k(t)=t_k^i,u_k(t)=u_k^i,\gamma _k(t)=\gamma _k^i,\psi _k(t)=\psi (t_k^i),\\ _k(t)=(t_k^i),_k(t)=(t_k^i),\end{array}$$
(5.5)
where $`i`$ is the greatest integer such that $`t_k^it`$. Note that $`u_k(t)=u_k(\tau _k(t))`$, $`\gamma _k(t)=\gamma _k(\tau _k(t))`$, $`\psi _k(t)=\psi (\tau _k(t))`$, $`_k(t)=(\tau _k(t))`$ and $`_k(t)=(\tau _k(t))`$.
###### Remark 5.2.
Since $`\psi _k^iAD(\psi _k^i,\gamma _k^{i1})`$, then by Remark 3.2 we deduce that the $`L^p`$-norms $`u_k^i_p`$ and $`u_k^i_p`$ are bounded uniformly with respect to $`i`$ and $`k`$. Passing to the piecewise constant functions $`tu_k(t)`$ and $`tu_k(t)`$, we have that there exists a positive constant $`C`$ such that
$$u_k(t)_pC\text{and}u_k(t)_pC$$
(5.6)
for every $`k`$ and for every $`t[0,T]`$. Since $`_k(t)(u_k(t),\gamma _k(t))`$ is bounded uniformly with respect to $`k`$, we get also that
$$\gamma _k(t)_{1,M}C,$$
(5.7)
for every $`k`$ and for every $`t[0,T]`$.
We introduce now a sequence of functions which play an important role in our estimates. For a.e. $`t[0,T]`$ we set
$$\theta _k(t):=๐ฒ(u_k(t)),\dot{\psi }(t)_k(t),\dot{\psi }(t)\dot{}(t),u_k(t).$$
(5.8)
In the following lemma we present the main energy estimate for the discrete process.
###### Lemma 5.3.
There exists a sequence $`R_k0`$ such that
$$(\tau _k(t))(u_k(t),\gamma _k(t))(0)(u_0,\gamma _0)+_0^{\tau _k(t)}\theta _k(s)๐s+R_k,$$
for every $`k`$ and for every $`t[0,T]`$.
###### Proof.
We need to prove that there exists a sequence $`R_k0`$ such that
$$_k^i(u_k^i,\gamma _k^i)(0)(u_0,\gamma _0)+_0^{t_k^i}\theta _k(s)๐s+R_k,$$
for any $`k`$ and for any $`i=1,\mathrm{},k`$.
Let us fix $`j`$ and $`k`$ with $`1jk`$. Since $`u_k^{j1}=\psi _k^{j1}`$ on $`_0\mathrm{\Omega }`$, and $`[u_k^{j1}+\psi _k^j\psi _k^{j1}]=[u_k^{j1}]`$ $`^{n1}`$-a.e. on $`M`$, the function $`u_k^{j1}+\psi _k^j\psi _k^{j1}`$ belongs to $`AD(\psi _k^j,\gamma _k^{j1})`$, hence $`_k^j(u_k^j,\gamma _k^j)_k^j(u_k^{j1}+\psi _k^j\psi _k^{j1},\gamma _k^{j1})`$. The proof now can be concluded arguing as in the proof of \[4, Lemma 6.1\]. โ
We are now in a position to prove our main result.
###### Proof of Theorem 3.10.
Let $`(t_k^i)`$, $`0ik`$, be a sequence of subdivisions of the interval $`[0,T]`$ satisfying (5.1) and (5.2). For any $`k`$ consider the pairs $`(u_k^i,\gamma _k^i)`$ inductively defined as solutions of the discrete problems (5.3) for $`i=1,\mathrm{},k`$ with the initial condition $`(u_k^0,\gamma _k^0)=(u_0,\gamma _0)`$. Let $`\tau _k(t)`$, $`u_k(t)`$, $`\gamma _k(t)`$, and $`\psi _k(t)`$ be defined by (5.5) for any $`t[0,T]`$. By Lemma 4.6 there exists a subsequence of $`\gamma _k(t)`$, independent of $`t`$, which $`\sigma _\phi ^p`$-converges to $`\gamma _{\mathrm{}}(t)L^0(M)^+`$, for every $`t[0,T]`$. By (5.7) and Lemma 4.4 we have $`\gamma _{\mathrm{}}(t)L^1(M)^+`$.
Let $`\theta _k(t)`$ be defined by (5.8) for a.e. $`t`$ and let
$$\theta _{\mathrm{}}(t):=\underset{k\mathrm{}}{lim\; sup}\theta _k(t).$$
By (2.4) and (5.6) we deduce that
$$|\theta _k(t)|\alpha _2(C^{p1}+1)\dot{\psi }(t)_p+_k(t)_{}\dot{\psi }(t)_{1,p}+C\dot{}(t)_{},$$
where $`_{}`$ is the norm in the dual space of $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Since the right-hand side of previous formula belongs to $`L^1([0,T])`$, we deduce that $`\theta _{\mathrm{}}`$ belongs to $`L^1([0,T])`$, too, and using the Fatou lemma we get
$$\underset{k\mathrm{}}{lim\; sup}_0^{\tau _k(t)}\theta _k(s)๐s_0^t\theta _{\mathrm{}}(s)๐s.$$
(5.9)
For a.e. $`t[0,T]`$ we can extract a subsequence $`\theta _{k_j}`$ of $`\theta _k`$, depending on $`t`$, such that
$$\theta _{\mathrm{}}(t)=\underset{j\mathrm{}}{lim}\theta _{k_j}(t).$$
By (5.6) the sequence $`u_{k_j}(t)`$ is bounded in $`W^{1,p}(\mathrm{\Omega }M;^m)`$, therefore we can extract a further subsequence, still denoted by $`u_{k_j}(t)`$, which converges weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$ to a function $`u_{\mathrm{}}(t)`$.
Since $`\varphi ([u_{k_j}(t)])\gamma _{k_j}(t)`$ $`^{n1}`$-a.e. on $`M`$, by point (a) in Definition 4.1 we have $`\varphi ([u_{\mathrm{}}(t)])\gamma _{\mathrm{}}(t)`$ $`^{n1}`$-a.e. on $`M`$. On the other hand, as $`u_{k_j}(t)=\psi _{k_j}(t)`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$, we have also $`u_{\mathrm{}}(t)=\psi (t)`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$, so that $`u_{\mathrm{}}(t)AD(\psi (t),\gamma _{\mathrm{}}(t))`$ for every $`t[0,T]`$.
The next step is to prove that the pair $`(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))`$ satisfies property (a) of Definition 3.4. To this aim, let $`\gamma L^1(M)^+`$, $`\gamma \gamma _{\mathrm{}}(t)`$ and $`vAD(\psi (t),\gamma )`$. By the minimality of the incremental solutions $`(u_k(t),\gamma _k(t))`$, we have that $`_k(t)(u_k(t),\gamma _k(t))_k(t)(v_k,\gamma _k(t)\varphi ([v]))`$, where $`v_k:=v+\psi _k(t)\psi (t)`$. Since the functional $`u๐ฒ(u)`$ is weakly lower semicontinuous and strongly continuous, and the function $`t(t)`$ is continuous, we immediately obtain
$`๐ฒ(u_{\mathrm{}}(t))\underset{k\mathrm{}}{lim\; inf}๐ฒ(u_k(t)),๐ฒ(v)=\underset{k\mathrm{}}{lim}๐ฒ(v_k),`$ (5.10)
$`(t),u_{\mathrm{}}(t)=\underset{k\mathrm{}}{lim}_k(t),u_k(t),(t),v=\underset{k\mathrm{}}{lim}_k(t),v_k.`$ (5.11)
So far we have easily obtained that
$`๐ฒ(u_{\mathrm{}}(t))(t),u_{\mathrm{}}(t)`$
$`๐ฒ(v)(t),v+\underset{k\mathrm{}}{lim\; sup}(\varphi ([v])\gamma _k(t))^+_{1,M},`$ (5.12)
where the last term in right-hand side comes from the equality
$$(\gamma \varphi ([v]))\gamma =(\varphi ([v])\gamma )^+,$$
(5.13)
which holds for every $`\gamma L^0(M)^+`$. In order to obtain that the pair $`(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))`$ satisfies point (a) in Definition 3.4 of quasistatic evolution we want to apply Lemma 4.7. To this aim we need to know that $`\varphi ([u_{\mathrm{}}(t)])L^1(M)^+`$. By (5.7) in Remark 5.2 we have that $`\gamma _k(t)_{1,M}`$ is bounded uniformly with respect to $`k`$. As $`u_k(t)`$ belong to $`AD(\psi _k(t),\gamma _k(t))`$, the sequence $`\varphi ([u_k(t)])`$ is bounded in $`L^1(M)^+`$, and by the lower semicontinuity assumption $`(\phi _3)`$ we obtain that $`\varphi ([u_{\mathrm{}}(t)])L^1(M)^+`$ thanks to the Fatou lemma. Then we can apply Lemma 4.7 and we get
$`๐ฒ(u_{\mathrm{}}(t))(t),u_{\mathrm{}}(t)`$
$`๐ฒ(v)(t),v+(\varphi ([v])\gamma _{\mathrm{}}(t))^+_{1,M}.`$ (5.14)
Applying (5.13) to the last term in the right-hand side of (5.14) we conclude that $`(t)(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))(t)(v,\gamma _{\mathrm{}}(t)\varphi ([v]))(t)(v,\gamma )`$ for every $`t[0,T]`$ and point (a) of Definition 3.4 is satisfied.
By the definition of the discrete problems, for every $`k`$ the function $`t\gamma _k(t)`$ is increasing. Passing to the $`\sigma _\phi ^p`$-limit, the same property holds for $`t\gamma _{\mathrm{}}(t)`$, so that point (b) of Definition 3.4 is satisfied.
It remains to prove point (c). For a.e. $`t`$ define
$$\theta (t):=๐ฒ(u_{\mathrm{}}(t)),\dot{\psi }(t)(t),\dot{\psi }(t)\dot{}(t),u_{\mathrm{}}(t).$$
Arguing as in the proof of \[4, Theorem 3.15\] we get
$$\theta _{\mathrm{}}(t)=\theta (t),$$
(5.15)
for a.e. $`t[0,T]`$. This in particular means that the map $`t\theta (t)`$ is measurable. Since we have proved that for every $`t[0,T]`$ the pair $`(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))`$ satisfies points (a) and (b) of Definition 3.4, we are in a position to apply Theorem 3.6 and get
$$(t)(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))(0)(u_0,\gamma _0)_0^t\theta (s)๐s.$$
By (4.1), (5.10), and (5.11) we get
$$(t)(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))\underset{j\mathrm{}}{lim\; inf}_{k_j}(t)(u_{k_j}(t),\gamma _{k_j}(t))\underset{k\mathrm{}}{lim\; sup}_k(t)(u_k(t),\gamma _k(t)).$$
(5.16)
Using Lemma 5.3 and taking (5.9) and (5.15) into account, we obtain
$$\underset{k\mathrm{}}{lim\; sup}_k(t)(u_k(t),\gamma _k(t))(0)(u_0,\gamma _0)+_0^t\theta (s)๐s.$$
(5.17)
By (5.16) and (5.17) we get that
$$(t)(u_{\mathrm{}}(t),\gamma _{\mathrm{}}(t))(0)(u_0,\gamma _0)+_0^t\theta (s)๐s$$
holds true for any $`t[0,T]`$, and this concludes the proof. โ
In the following theorem we prove that for every $`t[0,T]`$ the energy for the discrete-time problems converges to the energy for the continuous-time problem. We emphasize that the theorem is true for any irreversible quasistatic evolution $`t(u(t),\gamma (t))`$ corresponding to a given $`t\gamma (t)`$, not only for the one obtained as limit of the solutions of the discrete-time problems.
###### Theorem 5.4.
For every $`t[0,T]`$ let $`u_k(t)`$ and $`\gamma _k(t)`$ be defined as in the beginning of the proof of Theorem 3.10. Assume that $`\gamma _k(t)`$ $`\sigma _\phi ^p`$-converges to $`\gamma (t)L^1(M)^+`$ for any $`t[0,T]`$. Let $`t(u(t),\gamma (t))`$ be an irreversible quasistatic evolution. For a.e. $`t[0,T]`$ let $`\theta _k(t)`$ be defined as in (5.8), and set
$$\theta (t):=๐ฒ(u(t)),\dot{\psi }(t)(t),\dot{\psi }(t)\dot{}(t),u(t).$$
Then
$`๐ฒ(u(t))(t),u(t)=\underset{k\mathrm{}}{lim}(๐ฒ(u_k(t))_k(t),u_k(t)),`$ (5.18)
$`\gamma (t)_{1,M}=\underset{k\mathrm{}}{lim}\gamma _k(t)_{1,M},`$
for every $`t[0,T]`$. Furthermore
$$\theta _k\theta \text{in }L^1([0,T]),$$
so that there exists a subsequence of $`\theta _k`$ which converges to $`\theta `$ a.e. in $`[0,T]`$.
###### Proof.
For the proof we need to show that
$$\underset{j\mathrm{}}{lim}๐ฒ(u_{k_j}(t))=๐ฒ(u_{\mathrm{}}(t)),$$
(5.19)
for every $`t[0,T]`$, where $`u_{k_j}(t)`$ is the subsequence constructed in the proof of Theorem 3.10, and $`u_{\mathrm{}}(t)`$ is its limit. To this aim, let $`v_j:=u_{\mathrm{}}(t)+\psi _{k_j}(t)\psi (t)`$. By the minimality of the pair $`(u_{k_j}(t),\gamma _{k_j}(t))`$ we obtain that $`_{k_j}(t)(u_{k_j}(t),\gamma _{k_j}(t))_{k_j}(t)(v_j,\gamma _{k_j}(t)\varphi ([u_{\mathrm{}}(t)]))`$, and passing to the limit as $`j`$ goes to infinity, we get by (3.2), (5.10), and (5.11)
$$\begin{array}{cc}& \underset{j\mathrm{}}{lim\; sup}\left[๐ฒ(u_{k_j}(t))_{k_j}(t),u_{k_j}(t)\right]\hfill \\ & \underset{j\mathrm{}}{lim\; sup}\left[๐ฒ(v_j)_{k_j}(t),v_j+(\varphi ([u_{\mathrm{}}(t)])\gamma _{k_j}(t))^+_{1,M}\right]=\hfill \\ & =๐ฒ(u_{\mathrm{}}(t))(t),u_{\mathrm{}}(t)+\underset{j\mathrm{}}{lim\; sup}(\varphi ([u_{\mathrm{}}(t)])\gamma _{k_j}(t))^+_{1,M}.\hfill \end{array}$$
(5.20)
Since $`\gamma _{k_j}(t)`$ $`\sigma _\phi ^p`$-converges to $`\gamma _{\mathrm{}}(t)`$, by Lemma 4.7 we have
$$\underset{j\mathrm{}}{lim\; sup}(\varphi ([u_{\mathrm{}}(t)])\gamma _{k_j}(t))^+_{1,M}0.$$
(5.21)
Taking into account (5.20) and (5.21) we get in particular that
$$\underset{j\mathrm{}}{lim\; sup}๐ฒ(u_{k_j}(t))๐ฒ(u_{\mathrm{}}(t)).$$
(5.22)
This, together with (5.10), gives (5.19).
To conclude the proof it is sufficient to follow the arguments of the proof of \[4, Theorem 8.1\]. โ
The result can be improved under strict convexity assumption.
###### Theorem 5.5.
In addition to the hypotheses of Theorem 5.4, assume that $`\xi W(x,\xi )`$ is strictly convex for a.e. $`x\mathrm{\Omega }M`$ and that $`y\phi (x,y)`$ is convex for $`^{n1}`$-a.e. $`xM`$. Then $`u_k(t)u(t)`$ strongly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$, for every $`t[0,T]`$.
###### Proof.
We observe that for every $`t[0,T]`$ and $`\gamma L^1(M)^+`$ the functional $`v(t)(v,\gamma )`$ is strictly convex on the set of functions $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ with $`v=\psi (t)`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$. Therefore for every $`t`$ there exists a unique function $`uAD(\psi (t),\gamma (t))`$ such that the pair $`(u,\gamma (t))`$ is globally stable at time $`t`$. It follows that $`u(t)`$ coincides with the function $`u_{\mathrm{}}(t)`$ constructed in the proof of Theorem 3.10 and that the whole sequence $`u_k(t)`$ converge to $`u(t)`$ weakly in $`W^{1,p}(\mathrm{\Omega }M;^m)`$. Therefore (5.18) implies that $`๐ฒ(u_k(t))๐ฒ(u(t))`$. Using \[16, Theorem 3\] we deduce that $`u_k(t)u(t)`$ in measure. As
$$|u_k(t)u(t)|^p2^{p1}a_0^1[W(u_k(t))+W(u(t))]+2^{p1}a_0^1b_0,$$
the conclusion follows from the generalized dominated convergence theorem. โ
## 6. Euler conditions
In this section we study the Euler conditions satisfied by globally stable pairs $`(u,\gamma )W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$. Let us fix $`t[0,T]`$ and let $`(u,\gamma )W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ be globally stable at time $`t`$, and let $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ be such that $`v=0`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$. Hence for every $`\epsilon >0`$ the function $`u+\epsilon v`$ belongs to $`AD(\psi (t),\gamma \varphi ([u]+\epsilon [v]))`$, and by the global stability of the pair $`(u,\gamma )`$ at time $`t`$, we have that $`(t)(u,\gamma )(t)(u+\epsilon v,\gamma \varphi ([u]+\epsilon [v]))`$, therefore
$$\underset{\epsilon 0^+}{lim\; inf}\frac{(t)(u+\epsilon v,\gamma \varphi ([u]+\epsilon [v]))(t)(u,\gamma )}{\epsilon }0.$$
(6.1)
The weak formulation of the Euler conditions will be obtained from this inequality. Without loss of generality, we assume that $`(t)`$ is given by (2.5), and we omit the dependence on time. After some standard calculation, one can express (6.1) in the following form
$$\begin{array}{cc}& _{\mathrm{\Omega }M}\left(_\xi W(x,u)H\right):vdx_{\mathrm{\Omega }M}fv๐x_{_1\mathrm{\Omega }}gv๐^{n1}+\hfill \\ & _M\left(g^{}v^{}+g^{}v^{}\right)๐^{n1}+\underset{\epsilon 0^+}{lim\; inf}\frac{(\varphi ([u]+\epsilon [v])\gamma )^+_{1,M}}{\epsilon }0,\hfill \end{array}$$
(6.2)
for any $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`v=0`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$.
To continue our analysis we need now to specify the form of the function $`\phi `$. More precisely, we consider $`\phi :M\times ^m[0,+\mathrm{}]`$ defined by
$$\phi (x,y):=\phi _0(x)+\stackrel{~}{\phi }(x,y)\text{for }y0\text{and}\phi (x,0):=0\text{for all }xM,$$
(6.3)
where $`\phi _0L^1(M)^+`$ and $`\stackrel{~}{\phi }:M\times ^m[0,+\mathrm{}]`$ is a Borel function. We assume that for every $`xM`$ the following properties hold:
* $`\phi (x,y)=0`$ if and only if $`y=0`$;
* the function $`\stackrel{~}{\phi }(x,)`$ belongs to the space $`C^0(^m)C^1(^m\{0\})`$;
* $`\stackrel{~}{\phi }(x,0)=0`$;
* there exists an $`L^{\mathrm{}}`$-function $`\overline{\phi }`$ such that $`|_y\stackrel{~}{\phi }(x,y)|\overline{\phi }(x)`$ for any $`y0`$, where $`_y\stackrel{~}{\phi }(x,y)`$ denotes the vector of the partial derivatives of $`\stackrel{~}{\phi }`$ with respect to $`y`$;
* the limit
$$\stackrel{~}{\psi }(x,y):=\underset{\epsilon 0^+}{lim}_y\stackrel{~}{\phi }(x,\epsilon y)y$$
(6.4)
exists and is finite for any $`y0`$.
###### Remark 6.1.
By using de lโH$`\widehat{\mathrm{o}}`$pital Theorem, one obtain immediately that
$$\stackrel{~}{\psi }(x,y)=\underset{\epsilon 0^+}{lim}\frac{\stackrel{~}{\phi }(x,\epsilon y)}{\epsilon },$$
for any $`xM`$, $`y0`$. It follows from the positiveness of $`\stackrel{~}{\phi }`$ that $`\stackrel{~}{\psi }0`$. Moreover, we get easily that $`\stackrel{~}{\psi }`$ is positively $`1`$-homogeneous with respect to $`y`$, i.e., $`\stackrel{~}{\psi }(x,\lambda y)=\lambda \stackrel{~}{\psi }(x,y)`$, for every $`\lambda >0`$. Furthermore, by (6.4) and (4), we get also
$$|\stackrel{~}{\psi }(x,y)|\overline{\phi }(x)|y|\text{for every }xM\text{ and }y0\text{.}$$
(6.5)
The main result of this section is a theorem which makes explicit the Euler conditions obtained from (6.2) in the case of the function $`\phi `$ specified above. Before stating the theorem, we establish a general result concerning closed linear subspaces of $`L_\mu ^1(\mathrm{\Omega })`$, for an arbitrary Radon measure $`\mu `$ on $`\mathrm{\Omega }`$. We will apply this result to the measure $`\mu =^{n1}\text{ }\text{ }M`$.
The characteristic function of any set $`E`$ is denoted by $`1_E`$, i.e., $`1_E(x)=1`$ if $`xE`$, $`1_E(x)=0`$ otherwise.
###### Lemma 6.2.
Let $`\mu `$ be a Radon measure in $`\mathrm{\Omega }`$ and let $`Y`$ be a closed linear subspace of $`L_\mu ^1(\mathrm{\Omega })`$ with the following properties:
* if $`u,vY`$, then $`uvY`$;
* if $`uY`$ and $`\omega C_c^{\mathrm{}}(\mathrm{\Omega })`$, then $`\omega uY`$.
Then there exists a Borel set $`E\mathrm{\Omega }`$ such that $`Y=\{uL_\mu ^1(\mathrm{\Omega }):u=0\mu \text{-a.e. on }E\}`$.
###### Proof.
We begin by proving that
$$\text{if }uL_\mu ^1(\mathrm{\Omega })\text{ and }|u||v|\text{ for some }vY,\text{ then }uY.$$
(6.6)
Indeed in this case there exists $`\omega L_\mu ^{\mathrm{}}(\mathrm{\Omega })`$ such that $`u=\omega v`$ and there is a sequence $`\omega _kC_c^{\mathrm{}}(\mathrm{\Omega })`$ such that $`\omega _k`$ is bounded in $`L_\mu ^{\mathrm{}}(\mathrm{\Omega })`$ and $`\omega _k\omega `$ $`\mu `$-a.e. on $`\mathrm{\Omega }`$. By (b) we have $`\omega _kvY`$, and by the Lebesgue dominated convergence theorem $`\omega _kv\omega v=u`$ in $`L_\mu ^1(\mathrm{\Omega })`$. Since $`Y`$ is closed, we conclude that $`uY`$.
Now we prove that
$$\text{if }uY\text{ and }t>0,\text{ then }utY\text{ and }(ut)^+Y.$$
(6.7)
As $`|ut||u|`$, we have $`utY`$ by (6.6). Since $`(ut)^+=uut`$, we obtain that $`(ut)^+Y`$.
Next we prove that
$$\text{if }uY\text{ and }t>0,\text{ then }1_{\{u>t\}}Y,$$
(6.8)
where $`\{u>t\}:=\{x\mathrm{\Omega }:u(x)>t\}`$. By (6.7) we deduce that for every $`k>0`$ we have $`k(ut)^+1Y`$. As $`[k(ut)^+]11_{\{u>t\}}`$ pointwise and $`[k(ut)^+]1|u|/t`$, the convergence takes place in $`L_\mu ^1(\mathrm{\Omega })`$ and we conclude that $`1_{\{u>t\}}Y`$.
Let $`(u_k)`$ be a sequence dense in $`Y`$ and let $`E`$ be the intersection of the sets $`\{u_k=0\}`$. It is easy to prove by approximation that $`u=0`$ $`\mu `$-a.e. on $`E`$ for every $`uY`$. Conversely, let $`uL_\mu ^1(\mathrm{\Omega })`$ with $`u=0`$ $`\mu `$-a.e. on $`E`$. For every $`k`$ let
$$A_k:=\{u_1u_2\mathrm{}u_k>1/k\}.$$
By (a) and (6.8) we have $`1_{A_k}Y`$, so that $`(k1_{A_k})u^+`$ and $`(k1_{A_k})u^{}`$ belong to $`Y`$, by (6.6). As $`(k1_{A_k})u^+u^+`$ and $`(k1_{A_k})u^{}u^{}`$ in $`L_\mu ^1(\mathrm{\Omega })`$ we conclude that $`uY`$. โ
###### Lemma 6.3.
Let $`DM`$, let $`Y_D^m`$ be the set of all functions of the form $`[v]`$, with $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ and $`[v]=0`$ $`^{n1}`$-a.e. on $`D`$, and let $`\overline{Y}_D^m`$ be the closure of $`Y_D^m`$ in $`L^1(MM;^m)`$. Then there exists a Borel set $`\stackrel{~}{D}`$ (unique up to $`^{n1}`$-equivalence), containing $`D`$, such that $`\overline{Y}{}_{D}{}^{m}=\{wL^1(MM;^m):w=0^{n1}\text{-a.e. on }\stackrel{~}{D}\}`$.
###### Proof.
Let $`Y_D`$ be the set corresponding to the case $`m=1`$. It is easy to see that $`Y_D^m=(Y_D)^m`$. Therefore it suffices to prove the lemma in the case $`m=1`$.
The conclusion follows from Lemma 6.2 applied to $`\overline{Y}_D`$. It is enough to verify that conditions (a) and (b) are satisfied by $`Y_D`$. Condition (b) is trivial. To prove (a) we consider an open set $`U\mathrm{\Omega }M`$, with $`C^1`$ boundary and $`MU`$, such that $`U`$ lies on the negative side of $`M`$. Given two functions $`u`$ and $`vW^{1,p}(\mathrm{\Omega }M)`$ it is easy to check that $`[u][v]=[u(v\stackrel{~}{v}+\stackrel{~}{u})]`$, where $`\stackrel{~}{u}`$ and $`\stackrel{~}{v}W^{1,p}(\mathrm{\Omega })`$ coincide with $`u`$ and $`v`$ on $`U`$, respectively. โ
In the following theorem we will consider a function $`uW^{1,p}(\mathrm{\Omega }M;^m)`$ such that the divergence of the matrix field $`_\xi W(x,u)H`$ belongs to $`L^q(\mathrm{\Omega }M;^m)`$. It turns out that its normal trace $`(_\xi W(x,u)H)\nu `$ is defined as an element of $`(W^{1\frac{1}{p},p}(_1\mathrm{\Omega };^m))^{}`$. Moreover, we have that the normal traces $`(_\xi W(x,u)H)^{}\nu `$ and $`(_\xi W(x,u)H)^{}\nu `$ (defined on the positive and negative side of $`M`$) are both elements of the space $`(W^{1\frac{1}{p},p}(MM;^m))^{}`$. The duality pairing between $`(W^{1\frac{1}{p},p}(MM;^m))^{}`$ and $`W^{1\frac{1}{p},p}(MM;^m)`$ will be denoted by $`,`$.
###### Theorem 6.4.
Let $`t[0,T]`$ and $`(u,\gamma )W^{1,p}(\mathrm{\Omega }M;^m)\times L^1(M)^+`$ be globally stable at time $`t`$. Assume that $`\phi :M\times ^m[0,+\mathrm{}]`$ is defined as above in (6.3) and it satisfies (1)โ(5). Then
$`\mathrm{div}(_\xi W(x,u)H)=f\text{ on }\mathrm{\Omega }M,`$ (6.9)
$`(_\xi W(x,u)H)\nu =g\text{ on }_1\mathrm{\Omega },`$ (6.10)
$`(_\xi W(x,u)H)^{}\nu +g^{}=(_\xi W(x,u)H)^{}\nu g^{}\text{ on }MM.`$ (6.11)
Let us define
$`A`$ $`:=`$ $`\{xM:0<\varphi ([u])(x)=\gamma (x)\},`$
$`B`$ $`:=`$ $`\{xM:0=\varphi ([u])(x)\text{ and }\gamma (x)=\phi _0(x)\},`$
$`D`$ $`:=`$ $`\{xM:\gamma (x)<\phi _0(x)\},`$
and let $`\stackrel{~}{D}`$ be the set associated with $`D`$ by Lemma 6.3. Then there exists $`hL^{\mathrm{}}(M\stackrel{~}{D};^m)`$ such that
$$(_\xi W(x,u)H)^{}\nu +g^{},[v]=_{M\stackrel{~}{D}}h[v]๐^{n1},$$
(6.12)
for every $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`[v]=0`$ $`^{n1}`$-a.e. on $`D`$. Moreover
* for $`^{n1}`$-a.e. $`xA\stackrel{~}{D}`$ the vector $`h(x)`$ belongs to the segment joining $`0`$ and $`_y\stackrel{~}{\phi }(x,[u](x))`$;
* for $`^{n1}`$-a.e. $`xB\stackrel{~}{D}`$ the vector $`h(x)`$ belongs to the bounded convex set $`K(x):=\{a^m:ay\stackrel{~}{\psi }(x,y),y^m\}`$;
* for $`^{n1}`$-a.e. $`xM(AB\stackrel{~}{D})`$ we have $`h(x)=0`$.
###### Remark 6.5.
It is easy to see that, if $`D`$ is ($`^{n1}`$-equivalent to) a closed set, then $`\stackrel{~}{D}=D`$ (up to $`^{n1}`$-equivalence). A more difficult proof shows that the same result is true if $`D`$ is ($`^{n1}`$-equivalent to) a quasi closed set with respect to $`(1,p)`$-capacity.
It is clear that, if $`\phi _0=0`$, then $`\stackrel{~}{D}=D=\mathrm{}`$.
###### Remark 6.6.
For $`^{n1}`$-a.e. $`xM`$ the vector $`h(x)`$, obtained in Theorem 6.4, represents the cohesive force exerted from the positive lip of the crack on the negative lip. The theorem shows the conditions satisfied by the cohesive force on the different regions of $`M`$ determined by the respective relations between $`\varphi ([u])`$, $`\gamma `$ and $`\phi _0`$.
###### Proof of Theorem 6.4.
Since $`\varphi ([u])\gamma `$ $`^{n1}`$-a.e. on $`M`$, we have $`(\varphi ([u])\gamma )^+=0`$ $`^{n1}`$-a.e. on $`M`$. If $`[v]=0`$ $`^{n1}`$-a.e. on $`M`$, then the $`lim\; inf`$ in (6.2) is actually a limit and it is zero. Therefore (6.9), (6.10), and (6.11) can be obtained from (6.2) by standard argument involving integration by parts and a suitable choice of the test function $`vW^{1,p}(\mathrm{\Omega };^m)`$.
To shorten the notation, we set $`\stackrel{~}{h}:=(_\xi W(x,u)H)^{}\nu +g^{}`$ on $`MM`$. As explained before the statement of the theorem, we have $`\stackrel{~}{h}(W^{1\frac{1}{p},p}(MM);^m))^{}`$. So far, we may rewrite (6.2) as
$$\stackrel{~}{h},[v]+\underset{\epsilon 0^+}{lim\; inf}\frac{(\varphi ([u]+\epsilon [v])\gamma )^+_{1,M}}{\epsilon }0,$$
(6.13)
for any $`vW^{1,p}(\mathrm{\Omega }M;^m)`$ such that $`v=0`$ $`^{n1}`$-a.e. on $`_0\mathrm{\Omega }`$.
Let us extend the definition of $`\stackrel{~}{\psi }`$ by setting $`\stackrel{~}{\psi }(x,0)=0`$ for every $`xM`$. Now we prove that
$$\begin{array}{cc}& \underset{\epsilon 0^+}{lim}\frac{(\varphi ([u]+\epsilon w)\gamma )^+_{1,M}}{\epsilon }=\hfill \\ & =_M\left((_y\stackrel{~}{\phi }(x,[u])w)^+\mathrm{\hspace{0.17em}1}_A+\stackrel{~}{\psi }(x,w)\mathrm{\hspace{0.17em}1}_B\right)๐^{n1},\hfill \end{array}$$
(6.14)
for every $`wL^1(MM;^m)`$ with $`w=0`$ $`^{n1}`$-a.e. on $`D`$. To this aim, it is convenient to split the set $`M`$ into the union of the following two disjoint subsets $`A^{}:=\{xM:[u](x)0\}`$ and $`B^{}:=\{xM:[u](x)=0\}`$.
On $`A^{}`$, as $`\varphi ([u])\gamma `$ $`^{n1}`$-a.e. on $`M`$, we have that
$$\begin{array}{cc}\hfill \frac{(\varphi ([u]+\epsilon w)\gamma )^+}{\epsilon }& \frac{(\varphi ([u]+\epsilon w)\varphi ([u]))^+}{\epsilon }=\frac{(\stackrel{~}{\phi }(x,[u]+\epsilon w)\stackrel{~}{\phi }(x,[u]))^+}{\epsilon }\hfill \\ & (\overline{\phi }(x)w)^+,\hfill \end{array}$$
$`^{n1}`$-a.e. on $`M`$, where we used (6.3), and assumptions (3) and (4). Moreover, we have that
$$\frac{(\varphi ([u]+\epsilon w)\gamma )^+}{\epsilon }(_y\stackrel{~}{\phi }(x,[u])w)^+\mathrm{\hspace{0.17em}1}_A^{n1}\text{-a.e. on }A^{},$$
because $`A=\{0<\varphi ([u])=\gamma \}`$. By the Lebesgue dominated convergence theorem we get
$$_A^{}\frac{(\varphi ([u]+\epsilon w)\gamma )^+}{\epsilon }๐^{n1}_M(_y\stackrel{~}{\phi }(x,[u])w)^+\mathrm{\hspace{0.17em}1}_A๐^{n1},$$
(6.15)
as $`\epsilon 0^+`$, for every $`wL^1(MM;^m)`$.
Let us consider now the integral over $`B^{}`$. If $`wL^1(MM;^m)`$ and $`w=0`$ $`^{n1}`$-a.e. on $`D`$, we have
$$\frac{(\varphi (\epsilon w)\gamma )^+}{\epsilon }=0^{n1}\text{-a.e. on }D,$$
thus we can focus on the set $`B^{}D`$. As $`\gamma \phi _0`$ $`^{n1}`$-a.e. on $`MD`$, for every $`wL^1(MM;^m)`$ with $`w=0`$ $`^{n1}`$-a.e. on $`D`$, we obtain
$$\begin{array}{cc}\hfill \frac{(\varphi (\epsilon w)\gamma )^+}{\epsilon }& \frac{(\varphi (\epsilon w)\phi _0)^+}{\epsilon }=\frac{\stackrel{~}{\phi }(x,\epsilon w)}{\epsilon }\overline{\phi }(x)|w|\hfill \end{array}$$
$`^{n1}`$-a.e. on $`MD`$, where we used (6.3), and assumptions (3) and (4). Moreover, by Remark 6.1 we get that
$$\frac{(\varphi (\epsilon w)\gamma )^+}{\epsilon }(\stackrel{~}{\psi }(x,w))^+\mathrm{\hspace{0.17em}1}_B=\stackrel{~}{\psi }(x,w)1_B^{n1}\text{-a.e. on }B^{},$$
as $`\epsilon 0^+`$, for every $`wL^1(MM;^m)`$ with $`w=0`$ $`^{n1}`$-a.e. on $`D`$. We can apply again the Lebesgue dominated convergence theorem and obtain
$$_B^{}\frac{(\varphi ([u]+\epsilon w)\gamma )^+}{\epsilon }๐^{n1}_M\stackrel{~}{\psi }(x,w)\mathrm{\hspace{0.17em}1}_B๐^{n1},$$
as $`\epsilon 0^+`$, for every $`wL^1(MM;^m)`$ with $`w=0`$ $`^{n1}`$-a.e. on $`D`$. This concludes the proof of (6.14). We note that this equality cannot be true if the condition $`w=0`$ $`^{n1}`$-a.e. on $`D`$ is violated, because in this case
$$\underset{\epsilon 0^+}{lim}\frac{(\varphi (\epsilon w)\gamma )^+_{1,M}}{\epsilon }=\underset{\epsilon 0^+}{lim}\frac{(\phi _0+\stackrel{~}{\phi }(\epsilon w)\gamma )^+_{1,M}}{\epsilon }=+\mathrm{}.$$
Let $`Y_D^m`$ be the space defined in Lemma 6.3. Notice that $`Y_D^mW^{1\frac{1}{p},p}(MM);^m)`$. By (6.13) and (6.14) we have
$$\stackrel{~}{h},w+_M\left[(_y\stackrel{~}{\phi }(x,[u])w)^+\mathrm{\hspace{0.17em}1}_A+\stackrel{~}{\psi }(x,w)\mathrm{\hspace{0.17em}1}_B\right]๐^{n1}0,$$
(6.16)
for any $`wY_D^m`$. In order to localize this inequality, we prove first (6.12). Due to our assumption (4) and to (6.5), if we apply (6.16) to $`w`$ and $`w`$ we deduce that
$$|\stackrel{~}{h},w|\overline{\phi }_{\mathrm{}}w_{1,MD},$$
(6.17)
for every $`wY_D^m`$. It follows that there exists a function $`hL^{\mathrm{}}(MD;^m)`$ such that
$$\stackrel{~}{h},w=_{MD}hw๐^{n1},$$
for every $`wY_D^m`$. This implies that (6.12) is satisfied. By density from (6.16) we obtain
$$_{MD}\left[hw+(_y\stackrel{~}{\phi }(x,[u])w)^+\mathrm{\hspace{0.17em}1}_A+\stackrel{~}{\psi }(x,w)\mathrm{\hspace{0.17em}1}_B\right]๐^{n1}0,$$
(6.18)
for every $`w\overline{Y}_D^m`$. Since by Lemma 6.3 we have $`\overline{Y}{}_{D}{}^{m}=\{wL^1(MM;^m):w=0^{n1}\text{-a.e. on }\stackrel{~}{D}\}`$, we conclude that
$$h(x)y+(_y\stackrel{~}{\phi }(x,[u](x))y)^+\mathrm{\hspace{0.17em}1}_A(x)+\stackrel{~}{\psi }(x,y)\mathrm{\hspace{0.17em}1}_B(x)0,$$
(6.19)
for every $`y^m`$ and for $`^{n1}`$-a.e. $`xM\stackrel{~}{D}`$.
In particular, for $`^{n1}`$-a.e. $`xA\stackrel{~}{D}`$ the equality $`_y\stackrel{~}{\phi }(x,[u](x))y=0`$ implies that $`h(x)y=0`$ (it is enough to use (6.19) with $`y`$ and $`y`$), so that for a given $`xA\stackrel{~}{D}`$ the two vectors $`_y\stackrel{~}{\phi }(x,[u](x))`$ and $`h(x)`$ are parallel, hence there exists $`\lambda (x)`$ such that
$$h(x)=\lambda (x)_y\stackrel{~}{\phi }(x,[u](x))\text{for }^{n1}\text{-a.e. }xA\stackrel{~}{D},$$
(6.20)
and it is easy to verify that $`0\lambda (x)1`$, by using again (6.19). In this way we get condition (a).
On $`B\stackrel{~}{D}`$, from (6.19) we obtain
$$h(x)y+\stackrel{~}{\psi }(x,y)0\text{for }^{n1}\text{-a.e. }xB\stackrel{~}{D},$$
(6.21)
for every $`y^m`$, which is precisely condition (b), by the definition of $`K`$. On the remaining part of $`M\stackrel{~}{D}`$, from (6.19) we get condition (c). This concludes the proof. โ
###### Remark 6.7.
If $`\phi _0(x)>0`$ for $`^{n1}`$-a.e. $`xM`$, and $`(u,\gamma )=(u(t),\gamma (t))`$ for an irreversible quasistatic evolution, then (3.4) implies that the set $`B\stackrel{~}{D}`$ is nonempty only if there exists $`y^m\{0\}`$ such that $`\stackrel{~}{\phi }(x,y)=0`$, for some $`xM`$. This happens, for instance, in the Griffith model, where $`\phi `$ is given by (2.6) with $`a>0`$ and $`b=0`$. In this special case, condition (b) becomes $`h(x)=0`$ $`^{n1}`$-a.e. on $`B\stackrel{~}{D}`$, because $`K(x)=\{0\}`$.
###### Remark 6.8.
If for every $`x`$ the functions $`\xi W(x,\xi )`$ and $`y\phi (x,y)`$ are convex, then for any $`t[0,T]`$ and $`\gamma L^1(M)^+`$, the functional $`u(t)(u,\gamma \phi ([u]))`$ is convex. Therefore, it is possible to prove by standard arguments that conditions (a), (b), and (c) of Theorem 6.4 are equivalent to the inequality
$$_Mhw๐^{n1}+\underset{\epsilon 0^+}{lim}\frac{(\varphi ([u]+\epsilon w)\gamma )^+_{1,M}}{\epsilon }0,$$
for every $`w\overline{Y}_D^m`$. Thus, Euler conditions (6.9), (6.10), (6.11), (a), (b), (c) are not only necessary, but also sufficient to global stability.
We show now an example of a scalar problem, where the Euler conditions of Theorem 6.4 lead to a simplified set of boundary conditions.
###### Example 6.9.
Let $`m=1`$, $`p=2`$, $`W(x,\xi ):=\frac{1}{2}|\xi |^2`$, $`H(t):=0`$, $`g^{}(t)=g^{}(t):=0`$, $`\varphi (y):=|y|`$, which correspond to the energy functional:
$$(t)(u,\gamma ):=\frac{1}{2}_{\mathrm{\Omega }M}|u|^2๐x+_M\gamma ๐^{n1}_{\mathrm{\Omega }M}f(t)u๐x_{_1\mathrm{\Omega }}g(t)u๐^{n1}.$$
Let $`t[0,T]`$ and $`(u,\gamma )W^{1,2}(\mathrm{\Omega }M)\times L^1(M)^+`$ be globally stable at time $`t`$. Then we are in a position to apply Theorem 6.4 and the final part of Remark 6.5, obtaining
$$\{\begin{array}{cc}\mathrm{\Delta }u=f(t)\hfill & \text{on }\mathrm{\Omega }M,\hfill \\ u=\psi (t)\hfill & \text{on }_0\mathrm{\Omega },\hfill \\ \frac{u}{\nu }=g(t)\hfill & \text{on }_1\mathrm{\Omega },\hfill \\ \frac{u}{\nu }=0\hfill & \text{on }M\{0|[u]|<\gamma \},\hfill \\ \left|\frac{u}{\nu }\right|1\text{ and }\frac{u}{\nu }[u]0\hfill & \text{on }M\{|[u]|=\gamma \}.\hfill \end{array}$$
By Remark 6.8 we have also that if $`u`$ solves the previous boundary value problem for a given $`\gamma `$, then the pair $`(u,\gamma )`$ is globally stable at time $`t`$.
## 7. The case of linear elasticity
In this section we show that, with some modifications, it is possible to consider also the case where the uncracked part of the body is linearly elastic, which is excluded by the first inequality in (2.1).
Let $`p=2`$ and $`m=n1`$. We assume now that the bulk energy relative to the displacement $`uW^{1,2}(\mathrm{\Omega }M;^n)`$ has the form of linear elasticity
$$_{\mathrm{\Omega }M}A(x)Eu:Eudx,$$
where $`Eu:=\frac{1}{2}(u+(u)^T)`$ is the symmetric part of the gradient of $`u`$, and $`A`$ satisfies the following properties:
* for every $`x\mathrm{\Omega }`$, $`A(x)`$ is a linear symmetric operator from the space $`๐_{sym}^{n\times n}`$ of symmetric $`n\times n`$ matrices into itself, and the map $`xA(x)`$ is measurable;
* there are two positive constants $`c_0`$ and $`c_1`$ such that
$$c_0\left|\xi \right|^2A(x)\xi :\xi c_1\left|\xi \right|^2$$
(7.1)
for every $`x\mathrm{\Omega }M`$ and $`\xi ๐_{sym}^{n\times n}`$.
For the sake of simplicity in the notation we introduce the $`C^1`$ map $`๐ฌ:L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})`$ defined by
$$๐ฌ(\mathrm{\Psi }):=_{\mathrm{\Omega }M}A(x)\mathrm{\Psi }:\mathrm{\Psi }dx$$
for every $`\mathrm{\Psi }L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})`$, whose differential $`๐ฌ:L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})`$ is given by
$$๐ฌ(\mathrm{\Psi }),\mathrm{\Phi }=2_{\mathrm{\Omega }M}A(x)\mathrm{\Psi }:\mathrm{\Phi }dx,$$
for every $`\mathrm{\Phi }`$, $`\mathrm{\Psi }L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})`$, where $`,`$ denotes now the scalar product in the space $`L^2(\mathrm{\Omega }M;๐_{sym}^{n\times n})`$.
For every $`t[0,T]`$ the total energy of an admissible configuration $`(u,\gamma )W^{1,2}(\mathrm{\Omega }M,^n)\times L^1(M)^+`$ at time $`t`$ is now defined as
$$(t)(u,\gamma ):=๐ฌ(Eu)(t),u+\gamma _{1,M}.$$
Once we have the energy functional, we introduce the notion of global stability as in Definition 3.1.
Since the $`(n1)`$-dimension of $`_0\mathrm{\Omega }`$ is positive, Korn inequality holds (see, e.g., , ): there exists a constant $`C=C(\mathrm{\Omega },_0\mathrm{\Omega })`$ such that
$$u_2CEu_2\text{for all }uW^{1,2}(\mathrm{\Omega };^n)\text{ such that }u=0\text{ on }_0\mathrm{\Omega }\text{.}$$
As an immediate consequence, we get the following Korn-type inequality:
$$u_2CEu_2+(C+1)\psi _2$$
(7.2)
for every $`uW^{1,2}(\mathrm{\Omega }M;^n)`$, and $`\psi W^{1,2}(\mathrm{\Omega };^n)`$ such that $`u=\psi `$ on $`_0\mathrm{\Omega }`$.
Thanks to (7.2), we still have an a priori bound for the displacement $`u`$ as in Remark 3.2.
The definition of irreversible quasistatic evolution of minimum energy configurations is now given replacing $`๐ฒ(u(t)),\dot{\psi }(t)`$ by $`๐ฌ(Eu(t)),E\dot{\psi }(t)`$ in Definition 3.4.
Thanks to the Korn-type inequality (7.2), Theorems 3.7, 3.10, 5.4, and 5.5 (and Remark 3.5) continue to hold, with essentially the same proofs, if we replace $`๐ฒ(u(t))`$ and $`๐ฒ(u(t)),\dot{\psi }(t)`$ by $`๐ฌ(Eu(t))`$ and $`๐ฌ(Eu(t)),E\dot{\psi }(t)`$, respectively, and a similar substitution is done for $`u_k(t)`$.
|
warning/0506/hep-th0506080.html
|
ar5iv
|
text
|
# Gauge thresholds in the presence of oblique magnetic fluxes
## 1 Introduction and Summary
ยฟFrom the very beginning, Type I string theory in the presence of internal magnetic fields has offered a host of interesting effects . From a theoretical point of view, such models are governed by exactly solvable conformal field theories on the worldsheet. The effect of constant abelian field strengths is reflected in the change of boundary conditions for the string coordinates. As a result, perturbative analyses are reliable. From a phenomenologically point of view, magnetized pan-branes or their T-dual branes at angles are the most promising candidate to describe semirealistic string vacua that can capture the essential features of the Standard Model or some of its supersymmetric and/or grand unified extensions .
Turning on internal abelian magnetic fluxes reduces the rank of the CP group to the subgroup commuting with the $`U(1)`$ generators. Chiral fermions may arise in the spectrum and the number of generations, related to the degeneracy of the Landau levels, is a topological number that coincides with the top Chern class of the internal gauge bundle. Since particles interact with a magnetic field according to their helicity, the degeneracy between bosons and fermions is in general removed . However special configurations can preserve some supersymmetry .
Very recently, a new mechanism for moduli stabilization has been proposed based on the use of oblique magnetic fields on non-factorizable tori. Along this line of investigation, in we have described the effect of arbitrary magnetic fields on toroidal compactifications of the type I superstring in various dimensions. In the case of $`๐^6`$, one can attempt the stabilization of all closed string moduli, except dilaton and axion, through the introduction of suitable oblique choices of internal abelian magnetic fluxes while preserving a common $`๐ฉ=1`$ supersymmetry. Unfortunately cancelling all tadpoles both in the R-R and NS-NS sector seems to be harder to achieve than originally proposed in <sup>1</sup><sup>1</sup>1We thank the referee for pointing us out possible problems with tadpole cancellation in the AM model and acknowledge clarifying discussions on this issue with I. Antoniadis and T. Maillard.. In we have also identified the tree level gauge couplings of the surviving Chan-Paton group commuting with the magnetic and thus anomalous $`U(1)`$โs .
In the present paper we would like to extend our analysis to one-loop and compute threshold corrections to the open string gauge couplings. In principle one can play with the rationally quantized values of the internal magnetic fields in order to adjust the thresholds in closely related phenomenologically viable models and make contact with low-energy inputs.
The plan of the paper is as follows. In section 2 we fill in some gaps left open in and write down detailed formulae for the Annulus $`๐`$ and Mรถbius-strip $``$ contributions to the one-loop open string partition function in the presence of internal oblique magnetic fields<sup>2</sup><sup>2</sup>2The torus $`๐ฏ`$ and Klein-bottle $`๐ฆ`$ contributions to the unoriented closed string partition function at one-loop are unaltered, since the terms responsible for lifting the moduli are of order half-loop (disk).. As familiar from the analysis of unoriented open strings stretched between branes with โparallelโ magnetic fields, there are several sectors. Neutral strings connecting branes without magnetic fields preserve $`๐ฉ=4`$ supersymmetry and were described long ago . Singly charged strings connecting neutral branes to magnetized branes can at most preserve $`๐ฉ=2`$ or $`๐ฉ=1`$ supersymmetry in $`D=4`$. Generically supersymmetry is completely broken in these sectors. When the rank of the magnetic flux is not maximal, such as in the $`๐ฉ=2`$ cases, open strings carry generalized zero modes which are combinations of KK momenta and windings determined by the orientation of the magnetic field wrt the fundamental cell of the torus $`๐^6`$. When the rank of the magnetic flux is maximal, such as in $`๐ฉ=1`$ cases, open strings carry discrete multiplicities determined by the index $`I_{ab}`$ of the internal Dirac operator coupled to the magnetic field. In the unoriented case there are also doubly charged strings stretched between magnetized branes and their images under world-sheet parity $`\mathrm{\Omega }`$. Moreover there are dipole strings having their ends on the same (stack of) branes and thus preserving $`๐ฉ=4`$ supersymmetry but carrying โrescaledโ momenta. Finally one has dy-charged strings connecting branes with different magnetic fluxes, generically oblique wrt one another. As shown in , in order to determine the magnetic shifts one has to diagonalize the orthogonal matrix
$$R_{ab}=R_a^{q_a}R_b^{q_b}$$
(1)
where $`q_a,q_b=\pm 1`$ account for the (relative) orientation of the two ends and
$$R(H)=\frac{1H}{1+H},$$
(2)
with $`H_{\stackrel{~}{i}\stackrel{~}{j}}(F)=E_{\stackrel{~}{i}}^iE_{\stackrel{~}{j}}^jF_{ij}`$ the โframeโ components of $`F_{ij}`$. We will mostly concentrate on supersymmetric configurations and derive the detailed open string spectrum. Switching to the transverse closed string channel we check consistency with the boundary state formalism where magnetic shifts show up as phases modulating the reflection coefficients.
In section 5 we pass to consider the effect of turning on an abelian magnetic field in two of the four non-compact directions e.g.
$$F_{\mu \nu }=\delta _{[\mu }^2\delta _{\nu ]}^3fQ$$
(3)
where $`Q`$ is one of the generator of the unbroken Chan-Paton (CP) group. Since the spacetime magnetic โdeformationโ is integrable one can easily write down the relevant contributions: $`๐(f)`$ and $`(f)`$. The closed string spectrum is unaltered to the order at which we work and plays no role in our analysis. Selecting the terms quadratic in $`f`$ (and thus in $`Q`$) and subtracting the IR (in the open string channel) logarithmically divergent terms responsible for their running, we present general formulae for the one-loop threshold corrections to the gauge couplings . After diagonalization of the magnetic rotation matrices, our formulae look very much the same as in the case of โparallelโ magnetic fields which in turn show some similarity with standard formulae for orbifolds . We can thus exploit the available technology in order to write very explicit formulae for the thresholds arising from both $`๐ฉ=2`$ and $`๐ฉ=1`$ sectors<sup>3</sup><sup>3</sup>3$`๐ฉ=4`$ sectors, neither contribute to the (IR) running nor to the thresholds.. In principle the threshold corrections under consideration might be completely determined if the other closed string moduli, except for the overall dilaton dependence, were fixed, in a supersymmetric fashion, by a proper choice of internal magnetic fields following the original proposal of . Unfortunately, we have not been able so far to achieve this goal in a way consistent with tadpole cancellation in the absence of orbifolds or lower dimensional $`\mathrm{\Omega }`$-planes.
In section 6 we conclude with some remarks on dilaton stabilization and a preliminary discussion on the effect of turning on open string Wilson line moduli and their mixing with closed string moduli. We also pay some attention to other low-energy couplings most notably Yukawa couplings.
## 2 Toroidal compactifications with oblique magnetic fluxes
The perturbative spectrum of unoriented strings<sup>4</sup><sup>4</sup>4These are sometimes referred to as โopen descendantsโ, โ(un)orientifoldsโ, type I strings, โฆ They typically but not necessarily require open strings for consistency. is coded in four one-loop amplitudes . Torus $`๐ฏ`$ and Klein-bottle $`๐ฆ`$ represent the contribution of the unoriented closed strings. Annulus $`๐`$ and Mรถbius strip $``$ represent the contributions of unoriented open strings. Our aim in this section is to compute the open string partition function for toroidal compactifications in the presence of oblique magnetic fluxes.
Toroidal compactifications of type I strings without magnetic fluxes were studied long ago . The role of open string Wilson lines in the โadjointโ breaking of the CP group was streamlined. Rank reduction due to a quantized NS-NS antisymmetric tensor background was first pointed out and then further clarified in connection with non-commuting Wilson lines , shift orbifolds and exotic $`\mathrm{\Omega }`$-planes . Special features of rational points were analyzed. Last but not least, the RR emission vertex in the asymmetric superghost picture (-1/2,-3/2) was proposed that involves the RR gauge potential rather than its field strength<sup>5</sup><sup>5</sup>5Though hardly recognized in the overwhelming literature on D-branes, in retrospect this vertex accounts for their RR charge and BPS-ness..
Turning on magnetic fields does not change the one-loop closed string amplitudes $`๐ฏ`$ and $`๐ฆ`$ and thus the closed string spectrum to lowest order (sphere), but does affect the open string spectrum and by open-closed string duality the boundary reflection coefficients. So far only partition functions for cases with โparallelโ fluxes have been explicitly computed, see e.g. . We will momentarily adapt and extend those results to the case of arbitrary magnetic fluxes.
### 2.1 Open string partition function
Let us divide the full set of branes into various stacks $`N_0`$, $`N_1`$, โฆ and turn on constant magnetic fields on each stack except for the first ($`a=0`$) that we leave unmagnetized. The resulting gauge group is $`SO(N_0)\times U(N_1)\times \mathrm{}`$. As shown in , the magnetic $`U(1)`$โs are anomalous and the corresponding photons become massive by eating R-R axions associated to internal (1,1) forms<sup>6</sup><sup>6</sup>6This is a rather petite bouffe for one of the present authorsโ standards.. Henceforth we will focus on the case of $`๐^6`$ for definiteness and suppress the integration measure $`๐t/t`$ as well as the (regulated) contribution of the zero modes of the four non-compact bosonic coordinates $`V_4/(4\pi ^2\alpha ^{}t)^2`$.
### 2.2 Neutral and Dipole strings
The annulus contribution $`๐_{00}`$ from the completely neutral strings is the same as for toroidal compactifications without fluxes
$$๐_{00}=\frac{1}{2}N_0^2๐ฌ(0|\tau _A)\underset{p_{oo}\mathrm{\Lambda }_{KK}}{}\mathrm{exp}(2\pi i\alpha ^{}p_{oo}^2\tau _A)$$
(1)
where $`\tau _A=it/2`$ and
$$๐ฌ(\zeta ^I|\tau )=\frac{1}{2}\underset{\alpha ,\beta }{}c_{\alpha \beta }\frac{\theta [{}_{\beta }{}^{\alpha }](0|\tau )}{\eta ^3(\tau )}\underset{I=1}{\overset{3}{}}\frac{i\theta [{}_{\beta }{}^{\alpha }](\zeta ^I|\tau )}{\theta _1(\zeta ^I|\tau )},$$
(2)
with $`c_{\alpha \beta }=\mathrm{exp}[2\pi i(\alpha +\beta )]`$ implementing the GSO projection. As indicated, only KK momenta $`p_{oo}^{\widehat{i}}=m^iE_i^{\widehat{i}}`$, with $`m^iZ`$ (for singly wrapped branes) and $`E_i^{\widehat{i}}`$ the inverse 6-bein, are allowed. For simplicity, we have set the quantized NS-NS antisymmetric tensor $`B_{ij}`$ and the open string Wilson lines $`A_i^a`$ to zero. We postpone a brief discussion on their effects to the concluding remarks.
The Mรถbius-strip $`\mathrm{\Omega }`$ projection in this sector reads
$$_{00}=\frac{1}{2}N_0๐ฌ(0|\tau _M)\underset{p_{oo}\mathrm{\Lambda }_{KK}}{}\mathrm{exp}(2\pi i\alpha ^{}p_{oo}^2\tau _A)$$
(3)
where $`\tau _M=\tau _A+1/2`$ and, again, only KK momenta are allowed.
Neutral dipole strings starting and ending on the same stack $`a0`$ of branes suffer no magnetic mode shifts and contribute
$$๐_{a\overline{a}}=N_a\overline{N}_a๐ฌ(0|\tau _A)\underset{p_{a\overline{a}}\mathrm{\Lambda }_{a\overline{a}}}{}\mathrm{exp}(2\pi i\alpha ^{}\tau _Ap_{a\overline{a}}^2)$$
(4)
where the lattice sum is over generalized momenta $`p_{a\overline{a}}`$, satisfying $`p_L=R_ap_R`$, which generalizes the condition $`p_L=p_R`$ valid for truly neutral strings, discussed above.
### 2.3 Singly and doubly charged strings
Singly charged strings, connecting unmagnetized branes to magnetized ones, are easy to analyze too. The magnetic shifts read
$$ฯต_{oa}^I=\frac{1}{\pi }\mathrm{arctan}(q_ah_a^I),$$
(5)
where $`h_a^I`$ with $`I=1,2,3`$ are the skew eigenvalues of $`H_{\widehat{i}\widehat{j}}^a=E_{\widehat{i}}^iE_{\widehat{j}}^jF_{ij}^a`$ (frame components!), and turn the supersymmetric โcharacterโ $`๐ฌ(0|\tau _A)`$ into $`๐ฌ(ฯต_{oa}^I\tau _A|\tau _A)`$.
The overall multiplicity, related to the degeneracy of the Landau levels, is given by<sup>7</sup><sup>7</sup>7We assume $`I_{oa}`$ to be positive. A negative $`I_{oa}`$ would imply the presence of massless fermions of opposite chirality in the open string spectrum.
$$I_{oa}=|W_a|V(๐^6)\underset{I}{}q_ah_a^I=\underset{I}{}q_am_a^I$$
(6)
where $`V(๐^6)`$ is the โvolumeโ of $`๐^6`$ in units of $`4\pi ^2\alpha ^{}`$,
$$|W_a|=det\left(\frac{X^i}{\sigma ^\alpha }\right)=\underset{I}{}n_a^I$$
(7)
is the integer wrapping number, and $`m_a^I`$ are the integer magnetic monopole numbers. Dirac quantization indeed constraints the skew eigenvalues of $`(2\pi \alpha ^{})F_{ij}^a`$ (adimensional!) to be given by $`f_a^I=m_a^I/n_a^I`$. If $`๐^6=_I๐_{(I)}^2`$ then $`V(๐^6)=_IV_I`$, with $`V_I`$ the โvolumeโ of $`๐_{(I)}^2`$ and $`f_a^I=V_Ih_a^I`$. In any case, it is easy to prove that
$$I_{oa}=|W_a|V(๐^6)\underset{I}{}\frac{\mathrm{sin}(\pi ฯต_{oa}^I)}{\mathrm{cos}(\pi ฯต_{oa}^I)}=\underset{I}{}\mathrm{sin}(\pi ฯต_{oa}^I)\sqrt{det(๐ข_a+_a)},$$
(8)
where $`๐ข_a`$ and $`_a`$ are the induced worldvolume metric and field strength. This has a clear interpretation in the transverse channel, where it exposes the Born-Infeld (BI) action . The extra product of sinus turns out to cancel a similar factor coming from the $`\theta _1`$โs in the denominator.
If one or more of the $`h_a^I`$ are zero, i.e. $`H^a`$ has not maximal rank, the index vanishes signalling the presence of invariant subtori<sup>8</sup><sup>8</sup>8The label $`u`$ indeed stands for โunmagnetizedโ. $`๐_u^2`$, skewly embedded in $`๐^6`$. The โunmagnetizedโ directions are those fixed under $`R_a`$ and along them the open string can carry generalized momenta simultaneously satisfying $`p_L=R_a^{q_a}p_R`$, at the charged end, and $`p_L=p_R`$, at the neutral end $`a=0`$. Compatibility of the two conditions follows from $`det(R_a^{q_a}1)=0`$.
Generically there are tachyons in these sectors since all susy tend to be broken by the presence of the magnetic flux . However when $`_I(\pm )_Iฯต_{0a}^I=0`$ for some choice of signs the magnetic rotation matrix $`R_a^{q_a}`$ belongs to an $`SU(3)`$ subgroup of $`SO(6)`$ and the sector is at least $`๐ฉ=1`$ supersymmetric . If moreover one of the three mode shifts is zero, let us say $`ฯต_{0a}^u=0`$ for $`u=3`$, then $`ฯต_{0a}^1=\pm ฯต_{0a}^2`$ and the sector preserves $`๐ฉ=2`$ susy. Thanks to some Jacobi theta function identities the annulus amplitudes vanish in both cases and read
$$๐_{oa}^{๐ฉ=1}=N_0N_aI_{oa}๐ฌ(ฯต_{oa}^I\tau _A|\tau _A)$$
(9)
or
$$๐_{oa}^{๐ฉ=2}=N_0N_a\mathrm{\Lambda }_{oa}^uI_{oa}^{}๐ฌ(ฯต_{oa}^I\tau _A|\tau _A),$$
(10)
where $`\mathrm{\Lambda }_{oa}^u`$ denotes the lattice sum in the unmagnetized complex direction, and
$$I_{oa}^{}=\underset{Iu}{}q_am_a^I$$
(11)
is the reduced index that counts the degenaracy of the Landau levels in the (four) transverse magnetized directions.
As mentioned above, the lattice $`\mathrm{\Lambda }_{ao}^u`$ consists in those generalized momenta that satisfy $`p_L=R_a^{q_a}p_R`$ and $`p_L=p_R`$. For generic choices of magnetic fluxes and thus $`R_a`$ the resulting momenta are neither pure KK nor pure windings but rather mixtures of the two. The โunmagnetizedโ directions satisfy $`R_au=u`$ or, equivalently, $`H_au=0`$.
Doubly charged strings connecting magnetic branes with their $`\mathrm{\Omega }`$ images carry CP multiplicity $`N_a^2`$ or $`\overline{N}_a^2`$, depending on the sign of $`q_a=q_b`$, suffer doubled magnetic mode shifts $`ฯต_{aa}^I=2ฯต_{0a}^I`$ and appear with rescaled degeneracy of the Landau levels
$$I_{aa}=\underset{I}{}2q_am_a^I=8I_{oa},I_{aa}^{}=\underset{Iu}{}2q_am_a^I=4I_{oa}^{}.$$
(12)
Moreover, this sector receives a Mรถbius strip contribution.
For supersymmetric configurations, one has
$$๐_{aa}^{๐ฉ=1}=\frac{N_a^2}{2}8I_{oa}๐ฌ(2ฯต_{0a}^I\tau _A|\tau _A)$$
(13)
or
$$๐_{aa}^{๐ฉ=2}=\frac{N_a^2}{2}4I_{oa}^{}\mathrm{\Lambda }_{oa}^u(\tau _A)๐ฌ(2ฯต_{oa}^I\tau _A|\tau _A)$$
(14)
where as indicated $`\mathrm{\Lambda }_{aa}^u=\mathrm{\Lambda }_{oa}^u`$ is the same as in the singly charged sector<sup>9</sup><sup>9</sup>9When $`ฯต_{oa}^I=1/2`$ for some $`I`$ a separate treatment is required.. The Mรถbius strip reads
$$_{aa}^{๐ฉ=1}=\frac{N_a}{2}\widehat{I}_{aa}\widehat{๐ฌ}(2ฯต_{0a}^I\tau _A|\tau _M)$$
(15)
or
$$_{aa}^{๐ฉ=2}=\frac{N_a}{2}\widehat{I}_a^{}\widehat{\mathrm{\Lambda }}_{oa}^u(\tau _A)\widehat{๐ฌ}(2ฯต_{oa}^I\tau _A|\tau _M)$$
(16)
where a priori $`8I_{oa}\widehat{I}_{aa}+8I_{oa}`$ and $`4I_{oa}\widehat{I}_{aa}^{}+4I_{oa}`$, both with jumps of 2 units, allow for all possible (anti)symmetrizations under $`\mathrm{\Omega }`$ . Although in the simple toroidal models we explicitly consider $`\widehat{I}_{aa}=I_{aa}=8I_{oa}`$ and $`\widehat{I}_{aa}^{}=I_a^{}=4I_{oa}`$, turning on a non vanishing $`B`$ and/or (discrete) Wilson lines may change the situation .
### 2.4 Dy-charged strings
We are now ready to discuss the last and most subtle case of open strings strecthed between branes with oblique magnetic fields. As obvious we will recover the simpler case of parallel magnetic fields as a limit. As shown in , in order to compute the magnetic shifts one has to diagonalize the orthogonal matrix $`R_{ab}=R_a^{q_a}R_b^{q_b}`$ (in the frame basis!). For $`๐^6`$, the eigenvalues come in (three) complex conjugate pairs $`\rho _{ab}^{I\pm }=\mathrm{exp}(\pm i2\beta _{ab}^I)`$. The magnetic shifts are then given in general by
$$ฯต_{ab}^I=\frac{2}{2\pi }\beta _{ab}^I.$$
(17)
If $`R_aR_b=R_bR_a`$, the simpler abelian composition rule $`ฯต_{ab}^I=ฯต_{ao}^I\pm ฯต_{ob}^I`$ applies. Only the annulus contributes to these sectors and, depending on the amount of supersymmetry preserved, reads<sup>10</sup><sup>10</sup>10We try to consistently use $`F`$ to denote the 2-form in the coordinate basis and $`H`$ to denote the antisymmetric matrix in the frame basis.
$$๐_{ab}^{๐ฉ=1}=N_a\overline{N}_bI_{ab}๐ฌ(ฯต_{ab}^I\tau _A|\tau _A)$$
(18)
for $`๐ฉ=1`$ sectors, where
$$I_{ab}=C_3(_a^{q_a}_b^{q_b})=\frac{W_aW_b}{3!(2\pi )^3}_{๐^6}(q_aF_a+q_bF_b)^3,$$
(19)
and
$$๐_{ab}^{๐ฉ=2}=N_a\overline{N}_bI_{ab}^{}\mathrm{\Lambda }_{ab}^u(it)๐ฌ(ฯต_{oa}^I\tau _A|\tau _A)$$
(20)
for $`๐ฉ=2`$ sectors, where
$$I_{ab}^{}=C_2^{}(_a^{q_a}_b^{q_b})=\frac{W_aW_b}{2!(2\pi )^2}_{๐_{}^4}(q_aF_a+q_bF_b)^2,$$
(21)
with $`๐_{}^4`$ denoting the effectively magnetized subtorus, comprising the coordinates for which $`(q_aF_a+q_bF_b)_{ij}X^j0`$. Along the complementary unmagnetized torus $`๐_u^2`$, open strings carry zero modes that contribute to the lattice $`\mathrm{\Lambda }_{ab}^u`$. This consists in those generalized momenta that satisfy $`p_L=R_a^{q_a}p_R`$ and $`p_L=R_b^{q_b}p_R`$, which are compatible with one another since $`R_{ab}=R_a^{q_a}R_b^{q_b}`$ has unit eigenvalues as a consequence of $`(q_aH_a+q_bH_b)`$ having zero eigenvalues. For generic choices of magnetic fluxes and thus $`R_a`$ and $`R_b`$ the resulting momenta are neither pure KK nor pure windings but rather mixtures of the two. At first sight, there seems to be some ambiguity in the definition of the โunmagnetizedโ directions in these sectors. Indeed $`R_aR_bu=u`$ implies $`R_bR_av=v`$ for $`v=R_bu`$ and also $`(H_a+H_b)z=0`$ for $`z=(1+H_b)^1u`$. However the three possible choices ($`u`$, $`v`$ or $`z`$) are equivalent in that they yield the same results for the masses and multiplicities of the open string states. We will check this statement by means of T-duality in section 4.
The above expressions clearly encompass $`๐_{\overline{a}b}=๐_{a\overline{b}}`$ and $`๐_{\overline{a}\overline{b}}=๐_{ab}`$, upon properly choosing the signs $`q_a`$ and $`q_b`$.
Eqs. (19) and (21) indicate that the degeneracy of the Landau levels in each sector is given by the relevant Chern class of the internal tensor gauge bundle, which in turn coincides with the index of the Dirac operator coupled to the combined magnetic fields. For the purpose of checking consistency with the transverse closed string channel and accounting for the emergence of the BI and Wess-Zumino (WZ) terms, it is crucial to observe that
$`I_{ab}`$ $`=`$ $`V(๐^6){\displaystyle \underset{I}{}}n_a^In_b^I{\displaystyle \frac{\mathrm{sin}(\pi ฯต_{ab}^I)}{\mathrm{cos}(\pi ฯต_a^I)\mathrm{cos}(\pi ฯต_b^I)}}`$ (22)
$`=`$ $`V(๐^6)W_aW_b\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}{\displaystyle \underset{I=1}{\overset{3}{}}}\mathrm{sin}(\pi ฯต_{ab}^I)`$
and
$`I_{ab}^{}\mathrm{\Lambda }_{ab}^{(u)}`$ $`=`$ $`V(๐^6){\displaystyle \underset{Iu}{}}n_a^In_b^I{\displaystyle \frac{\mathrm{sin}(\pi ฯต_{ab}^I)}{\mathrm{cos}(\pi ฯต_a^I)\mathrm{cos}(\pi ฯต_b^I)}}`$ (23)
$`=`$ $`V(๐^6)W_aW_b\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}{\displaystyle \underset{Iu}{}}\mathrm{sin}(\pi ฯต_{ab}^I)\stackrel{~}{\mathrm{\Lambda }}_{ab}^{(u)}.`$
To this end, using $`h_a^I=\mathrm{tan}(\pi ฯต_a^I)`$ and elementary trigonometry, one first expresses the BI action in the form
$$W_a\sqrt{det(1+q_aH_a)}=\underset{I}{}n_a^I\sqrt{(1+(h_a^I)^2)}=\underset{I}{}\frac{n_a^I}{\mathrm{cos}(\pi ฯต_a^I)}.$$
(24)
In the case of parallel magnetic fluxes, one then has
$$\mathrm{sin}(\pi ฯต_{ab}^I)=\mathrm{sin}(\pi ฯต_a^I+\pi ฯต_b^I)=\frac{(q_ah_a^I+q_bh_b^I)}{\sqrt{1+(h_a^I)^2}\sqrt{1+(h_b^I)^2}}$$
(25)
and the denominator can be used to cancel the BI factors and to get precisely (22).
In the case of arbitrary magnetic fluxes, one has to work a little harder . The product of sines $`_I\mathrm{sin}(\pi ฯต_{ab}^I)`$ can be related to the characteristic polynomial of $`R_{ab}`$, $`P(\lambda )=det(R_{ab}\lambda I)`$, with $`\lambda =1`$<sup>11</sup><sup>11</sup>11Some of its remarkable properties have been discussed in .. Plugging in (22) the trigonometric formula
$$\underset{Iu}{}\mathrm{sin}(\pi ฯต_{ab}^I)=\sqrt{\stackrel{}{det}\left(\frac{R_{ab}1}{2}\right)}=\frac{\sqrt{\stackrel{}{det}(q_aH_a+q_bH_b)}}{\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}},$$
(26)
proven in the Appendix, the BI terms in the denominator cancel and the index reads
$`I_{ab}`$ $`=`$ $`V(๐^6)\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}\sqrt{{\displaystyle \frac{1}{2^6}}P(1)}`$ (27)
$`=`$ $`V(๐^6)\sqrt{det(q_aH_a+q_bH_b)}=V(๐^6)\mathrm{Pfaff}(q_aH_a+q_bH_b)`$
$`=`$ $`{\displaystyle \frac{1}{23!}}{\displaystyle _{๐^6}}(q_aF_a+q_bF_b)^3`$
up to signs.
When $`\mathrm{Pfaff}(q_aH_a+q_bH_b)=0`$ i.e. when $`det(q_aH_a+q_bH_b)=0`$ then $`det(R_a^{q_a}R_b^{q_b}1)=0`$ one has unmagnetized directions along which open strings carry zero modes, i.e. mixtures of KK momenta and windings. We have already observed that the various kernels are isomorphic. One eventually finds
$`I_{ab}^{}`$ $`=`$ $`V(๐_{}^4)\sqrt{det(1+q_aH_a)^{}}\sqrt{det(1+q_bH_b)^{}}\sqrt{{\displaystyle \frac{1}{2^4}}P^{}(1)}`$ (28)
$`=`$ $`V(๐_{}^4)\sqrt{det(q_aH_a+q_bH_b)^{}}=V(๐_{}^4)\mathrm{Pfaff}(q_aH_a+q_bH_b)^{}`$
$`=`$ $`{\displaystyle \frac{1}{22!}}{\displaystyle _{๐_{}^4}}(q_aF_a+q_bF_b)^2`$
as expected.
## 3 Channel duality and tadpoles
In order to check the validity of our derivation of the open string spectrum, encoded in the direct (loop) open string channel, we would now like to compute the resulting transverse (tree level) closed string channel. For consistency one expects to find a boundary-to-boundary amplitude of the form
$$\stackrel{~}{๐}=\underset{a,b}{}N_aN_bB_a|\mathrm{exp}(\pi \mathrm{}_{cl})|B_b,$$
(1)
where the presence of $`_{cl}`$ means that only states in the closed string spectrum are allowed to be exchanged.
The superstring boundary state $`|B(F)`$ in the presence of an arbitrary (electro-)magnetic field was constructed long ago and reconsidered more recently . It consists of various ingredients and obviously depends on the choice of boundary conditions for the worldsheet supercurrent, i.e. for the worldsheet fermions and superghosts. The ghost contribution is independent from the magnetic flux and we will not display it for simplicity. Indeed, since all electric components vanish in our case, $`F_{io}=0`$, we can choose a light-cone gauge and work with the eight transverse coordinates only, $`i,j=2,\mathrm{}9`$, and forget about (super)ghosts altogether.
The contribution of the bosonic coordinates
$$|B_a^{(X)}=\sqrt{det(๐ข_a+_a)}\mathrm{exp}(\underset{n>o}{}\stackrel{~}{a}_n^iR_{ij}(F_a)a_n^j)|O_a$$
(2)
Here we are back to the coordinate basis, where $`R_{ij}`$ is not an orthogonal matrix! One switches from one to the other by means of $`E_{\widehat{i}}^i`$ ad its inverse $`E_i^{\widehat{i}}`$. Taking into account the obvious generalization associated to multiple wrapping and the presence of a (flat) non-trivial induced metric. Notice the presence of the BI action that generalizes the overall volume contribution of the CM position when $`F_a0`$. The bosonic zero-mode contribution is implicit in $`|O_a`$ and deserves a special treatment. It consists in a sum over all $`p_L=R_ap_R`$. In compact cases<sup>12</sup><sup>12</sup>12For non-compact directions $`p_L=p_R`$ and, even in the presence of magnetic fields, this results in the familiar Neumann condition of โno momentum flowโ through the boundary. This observation turns out to be relevant for our later purposes of computing thresholds., this results in an infinite but discrete number of choices, e.g. windings for $`F_a=0`$ or generalization thereof for $`F_a0`$.
The contribution of the fermionic coordinates is notoriously much subtler. In the NS-NS sector, there are no fermionic zero-modes, since the modes are half-integers and one has
$$|B_a,\eta _{NSNS}^{(\psi )}=\mathrm{exp}(i\eta \underset{n1/2}{}\stackrel{~}{\psi }_n^iR_{ij}(F_a)\psi _n^j)|\eta $$
(3)
where the $`\eta =\pm `$ stands for possible GSO projections and the light-cone gauge roughly speaking corresponds to the choice of the canonical (left-right symmetric) superghost picture $`q=\stackrel{~}{q}=1`$.
In the R-R sector, fermions admit zero-modes, whose contribution replaces the BI action with the WZ coupling
$$|B_a,\eta _{RR}^{(\psi )}=\frac{1}{\sqrt{det(๐ข_a+_a)}}\mathrm{exp}(i\eta \underset{n>0}{}\stackrel{~}{\psi }_n^iR_{ij}(F_a)\psi _n^j)|O_a,\eta $$
(4)
where
$$|O_a,\eta =U_{A\stackrel{~}{B}}(F_a)|A,\stackrel{~}{B}$$
(5)
with
$$U_{A\stackrel{~}{B}}(F_a)=\left[\mathrm{AExp}\left(\frac{1}{2}F_{ij}^a\mathrm{\Gamma }^{ij}\right)\right]_{A\stackrel{~}{B}},$$
(6)
where the notation $`\mathrm{AExp}`$ implies that one has to antisymmetrize the vector indices of the $`\mathrm{\Gamma }`$ matrices i.e.
$$\mathrm{AExp}\left(\frac{1}{2}F_{ij}^a\mathrm{\Gamma }^{ij}\right)=1\frac{1}{2}F_{ij}^a\mathrm{\Gamma }^{ij}+\frac{1}{8}F_{ij}^aF_{kl}^a\mathrm{\Gamma }^{[ij}\mathrm{\Gamma }^{kl]}+\mathrm{}.$$
(7)
The full boundary state<sup>13</sup><sup>13</sup>13A similar analysis allows one to construct crosscap states. We refrain from doing so here since there are no issues at stake for the Klein bottle $`\stackrel{~}{๐ฆ}`$ or Mรถbius strip $`\stackrel{~}{}`$. then reads
$$|B=\frac{1}{2}\underset{a}{}N_a(|B_a,+_{NSNS}|B_a,_{NSNS}+|B_a,+_{RR}+|B_a,_{RR}).$$
(8)
Let us now consider for definiteness the amplitude
$$\stackrel{~}{๐}_{ab}=B_a|\mathrm{exp}(\pi \mathrm{}_{cl})|B_b.$$
(9)
for $`ab`$ with $`[R_a,R_b]0`$. Since $`_{cl}=L_o+\stackrel{~}{L}_oc/12`$ is a (transverse) Lorentz scalar, $`L_o=\alpha ^{}p_L^2/4+_{n>o}[na_na_n+\psi _n\psi _n]`$ one can perform a simultaneous rotation of all $`a_n^i`$โs (both annihilation $`a_{n>o}^i`$ and creation $`a_{n<o}^i`$ modes) by say $`R_b`$: $`\widehat{a}_n^i=R_b^i{}_{j}{}^{}a_{n}^{j}`$ that leaves $`L_o`$ invariant and preserves the canonical commutation rules. The net result is to transfer the effect of the rotation on the other boundary state $`|B_a`$ that, once written in terms of $`\widehat{a}_n^i`$ and $`\stackrel{~}{a}_n^i`$, depends on the combined rotation $`R_{ab}=R_aR_b^1`$. Thence everything, except for the zero-modes and overall BI or WZ actions, goes through in the same way as for closed string bouncing between an unmagnetized brane and a magnetized one. In particular mode shifts in the direct channel give rise to phases in the transverse channel. Relying on the expressions for $`I_{ab}`$ and $`I_{ab}^{}`$ that have been derived at the end of the previous (sub)section, one can also reproduce the expected BI action or WZ couplings.
### 3.1 UV Divergences
We are thus ready to address the question of UV divergences in the presence of oblique fluxes and their cancellation. As it is well known, they are associated to diagrams (tadpoles) of massless particles (dis)appearing from (into) the vacuum . In particular tadpoles of RR massless fields belonging to closed sectors with non-vanishing Witten index are responsible for chiral anomalies in the low energy effective theory . On the other, hand NS-NS tadpoles are less dangerous in principle <sup>14</sup><sup>14</sup>14For a derivation of the dilaton tadpole in the type I superstring see . They simply signal the instability of the chosen vacuum configuration .
Since R-R and NS-NS fields couple to pan-branes according to the WZ and BI actions , respectively, turning on internal open string fluxes induces lower dimensional R-R charges and NS-NS tensions . Quite remarkably, some of these can be negative for special choices of fluxes on $`๐^6`$ preserving at most $`๐ฉ=1`$ supersymmetry and correspond to stable bound states not at threshold . As a consequence, one may try to satisfy the consistency constraints without adding lower dimensional D-branes and/or $`\mathrm{\Omega }`$-planes . For $`๐^4`$ and/or for $`๐ฉ=2`$ supersymmetric configurations on $`๐^6`$, instead, BPS bound states are necessarily at threshold. Yet, even in these cases, one can play with the non-polynomial BI action and derive supersymmetric configurations associated to non-linear instantons .
In order to expose potential massless tadpoles at genus 1/2 from the disk and projective plane, we start by performing an S modular trasformation $`\tau _A=\frac{it}{2}\frac{1}{\tau }_A=i\mathrm{}`$ on the Annulus amplitude $`๐`$ and get
$`\stackrel{~}{๐}_{ab}`$ $`=`$ $`{\displaystyle \frac{N_aN_b|W_a||W_b|}{2^5}}{\displaystyle ๐\mathrm{}\sqrt{det(G_{ab})}\stackrel{~}{\mathrm{\Lambda }}_{ab}(i\mathrm{})\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}}`$ (10)
$`{\displaystyle \underset{\alpha \beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{2}}\left({\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](0|i\mathrm{})}{\eta ^3(i\mathrm{})}}\right)^{1+u}{\displaystyle \underset{I=1}{\overset{3u}{}}}{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](iฯต_{ab}^I|i\mathrm{})}{\theta [{}_{\frac{1}{2}}{}^{\frac{1}{2}}](iฯต_{ab}^I|i\mathrm{})}}2\mathrm{sin}(\pi ฯต_{ab}^I),`$
where we use the modular properties of Jacobi theta functions and open/closed duality for the index. The zero mode contribution can be dealt with by means of a Poisson resummation
$$\stackrel{~}{\mathrm{\Lambda }}_{ab}(i\mathrm{})=\underset{w_{ab}}{}\mathrm{exp}\left(\frac{2\pi \mathrm{}w_{ab}^iG_{ij}w_{ab}^j}{4\alpha ^{}}\right)$$
(11)
Modular transformation properties of $`\theta `$ and other functions are listed in an Appendix.
Similarly, for the Mรถbius strip $``$, the relevant modular trasformation is $`P=TST^2S`$, that acts on the modular parameter according to $`\tau _M=\frac{it+1}{2}\frac{1}{2}+\frac{i}{2t}=\frac{1}{2}+i\mathrm{}`$ so that $`\mathrm{}=2t`$, and yields
$`\stackrel{~}{}_a`$ $`=`$ $`2N_a|W_a|{\displaystyle ๐\mathrm{}\sqrt{det(G_{aa})}\stackrel{~}{\mathrm{\Lambda }}_{aa}(i\mathrm{})\underset{\alpha \beta }{}\frac{c_{\alpha \beta }}{2}\left(\frac{\theta [{}_{\beta }{}^{\alpha }](0|i\mathrm{})}{\eta ^3(i\mathrm{})}\right)^{1+u}\underset{I=1}{\overset{3u}{}}\frac{\theta [{}_{\beta }{}^{\alpha }](iฯต_{aa}^I/2|i\mathrm{})}{\theta [{}_{\frac{1}{2}}{}^{\frac{1}{2}}](iฯต_{aa}^I/2|i\mathrm{})}}`$ (12)
$`\sqrt{det(1+q_aH_a)}{\displaystyle \underset{I=1}{\overset{3u}{}}}2\mathrm{sin}(\pi ฯต_{aa}^I/2).`$
At this order the unoriented closed string spectrum is unaffected by the internal magnetic field and the Klein bottle amplitude $`๐ฆ`$ gives rise to
$$\stackrel{~}{๐ฆ}=2^5๐\mathrm{}\sqrt{det(G_6)}\mathrm{\Lambda }_6(i\mathrm{})\underset{\alpha \beta }{}\frac{c_{\alpha \beta }}{2}\left(\frac{\theta [{}_{\beta }{}^{\alpha }](0|i\mathrm{})}{\eta ^3(i\mathrm{})}\right)^4,$$
(13)
after an S modular transformation.
In the RR sector, the (closed string) IR limit $`\mathrm{}\mathrm{}`$ is dominated by the exchange of massless states and yields
$`\stackrel{~}{๐}_{m=0}^{RR}`$ $``$ $`2^58_s{\displaystyle \underset{ab}{}}{\displaystyle \underset{q_aq_b}{}}N_a^{q_a}W_aN_b^{q_b}W_b\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}{\displaystyle \underset{I=1}{\overset{3}{}}}\mathrm{cos}(\pi ฯต_{ab}^I)`$
$`\stackrel{~}{}_{m=0}^{RR}`$ $``$ $`28_s{\displaystyle \underset{aq_a}{}}N_a^{q_a}W_a`$
$`\stackrel{~}{๐ฆ}_{m=0}^{RR}`$ $``$ $`2^5\mathrm{\hspace{0.17em}8}_s`$ (14)
The deceiving simplicity of $`\stackrel{~}{}_{m=0}`$ is to be ascribed to the often used identity
$$\sqrt{det(1+q_aH_a)}=1/\underset{I=1}{\overset{3}{}}\mathrm{cos}(\pi ฯต_{a0}^I)=1/\underset{I=1}{\overset{3}{}}\mathrm{cos}(\pi ฯต_{aa}^I/2)$$
(15)
At this point it is useful to take advantage of the spinorial representation of the rotation matrix $`R(F)`$, introduced above,
$$U(R)=\frac{1}{\sqrt{det(1+qH)}}\mathrm{AExp}\left(\frac{q}{2}\gamma ^{\widehat{i}}\gamma ^{\widehat{j}}H_{\widehat{i}\widehat{j}}\right)$$
(16)
in order to recognize that
$$\underset{I=1}{\overset{3}{}}2\mathrm{cos}(\pi ฯต_{ab}^I)=Tr_s[U(R_{ab})]=Tr_s[U(R_a)U(R_b)]$$
or, equivalently,
$$\underset{I}{}2\mathrm{cos}(\pi ฯต_{ab}^I)=\sqrt{det\left(R_{ab}+1\right)}=\frac{2^3\sqrt{det(1+q_aH_aq_bH_b)}}{\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}}$$
(17)
In this case after a sum over the possible orientations, $`q=\pm 1`$, only wedge products with even numbers of $`H`$ survive. Tracing over the spinor indices of $`\gamma `$ matrices, the above expression factorizes in a sum of squares that are easily interpreted in terms of the total R-R charge of D9-branes, $`\mathrm{\Omega }`$9-plane and of the individual R-R charges of the lower dimensional objects induced by the fluxes. Observing that the series in $`\mathrm{AExp}`$ actually truncates at order $`d/2=3`$ in our case and that
$$๐ฌ_a^{\widehat{i}\widehat{j}}=\frac{1}{2^3}ฯต^{\widehat{i}\widehat{j}\widehat{i}_1\widehat{j}_1\widehat{i}_2\widehat{j}_2}H_{\widehat{i}_1\widehat{j}_1}^aH_{\widehat{i}_2\widehat{j}_2}^a$$
accounts for the induced D5-brane charge of the magnetized D9-branes, one eventually finds the complete R-R tadpole condition
$`\stackrel{~}{๐}_{m=0}^{RR}+\stackrel{~}{๐ฆ}_{m=0}^{RR}+\stackrel{~}{}_{m=0}^{RR}=\left({\displaystyle \underset{a}{}}2N_aW_a\mathrm{\hspace{0.17em}32}\right)^2+{\displaystyle \underset{\widehat{i}\widehat{j}}{}}{\displaystyle \underset{ab}{}}N_aW_aN_bW_b๐ฌ_a^{\widehat{i}\widehat{j}}๐ฌ_b^{\widehat{i}\widehat{j}}`$ (18)
$`=`$ $`\left({\displaystyle \underset{a}{}}2N_aW_a\mathrm{\hspace{0.17em}32}\right)^2+{\displaystyle \underset{\widehat{i}\widehat{j}}{}}\left({\displaystyle \underset{a}{}}N_aW_a๐ฌ_a^{\widehat{i}\widehat{j}}\right)^2=0`$
This consistency condition, derived here using CFT techniques i.e. channel duality, coincides with the one based on the analysis of the BI and WZ actions or on anomaly cancellation arguments. Unfortunately, due to subtleties with the choice of the wrapping numbers, the AM model does not satisfy these consistency requirement even if the spectrum is not chiral. We have not been able so far to find consistent variants of the AM model, although to the best of our knowledge no โno-goโ theorem prevents their existence. One possible way out would be to include magnetized $`\overline{D}9`$-branes, preserving the same $`๐ฉ=1`$ susy and corresponding to $`W_a<0`$.
By similar means one can study the NS-NS sector. Taking the limit $`\mathrm{}\mathrm{}`$ one finds
$`\stackrel{~}{๐}_{m=0}^{NSNS}`$ $``$ $`2^5{\displaystyle \underset{ab}{}}{\displaystyle \underset{q_aq_b}{}}N_a^{q_a}|W_a|N_b^{q_b}|W_b|\sqrt{det(1+q_aH_a)}\sqrt{det(1+q_bH_b)}`$
$`\times [2+{\displaystyle \underset{I=1}{\overset{3}{}}}2\mathrm{cos}(2\pi ฯต_{ab}^I)]`$
$`\stackrel{~}{}_{m=0}^{NSNS}`$ $``$ $`2{\displaystyle \underset{aq_a}{}}8N_a^{q_a}|W_a|\sqrt{det(1+q_aH_a)}[2+{\displaystyle \underset{I=1}{\overset{3}{}}}2\mathrm{cos}(2\pi ฯต_{ao}^I)]`$
$`\stackrel{~}{๐ฆ}_{m=0}^{NSNS}`$ $``$ $`2^5\mathrm{\hspace{0.17em}8}_v.`$ (19)
Using
$$2\underset{I}{}\mathrm{cos}(2\pi ฯต_{ab}^I)=\underset{I}{}(e^{2i\pi ฯต_{ab}^I}+e^{2i\pi ฯต_{ab}^I})=Tr_v(R_{ab})=Tr_v(R_aR_b),$$
(20)
proven in the appendix yields the complete massless NS-NS tadpole condition
$`\stackrel{~}{๐}_{m=0}^{NSNS}+\stackrel{~}{๐ฆ}_{m=0}^{NSNS}+\stackrel{~}{}_{m=0}^{NSNS}`$ (21)
$`=`$ $`\left({\displaystyle \underset{a}{}}2N_a|W_a|\sqrt{det(1+q_aH_a)}\mathrm{\hspace{0.17em}32}\right)^2+{\displaystyle \underset{\widehat{i}\widehat{j}}{}}{\displaystyle \underset{ab}{}}N_a^{q_a}|W_a|N_b|W_b|๐ฏ_a^{\widehat{i}\widehat{j}}๐ฏ_b^{\widehat{i}\widehat{j}}`$
$`=`$ $`\left({\displaystyle \underset{a}{}}N_a|W_a|\sqrt{det(1+q_aH_a)}\mathrm{\hspace{0.17em}32}\right)^2+{\displaystyle \underset{\widehat{i}\widehat{j}}{}}\left({\displaystyle \underset{a}{}}N_a|W_a|๐ฏ_a^{\widehat{i}\widehat{j}}\right)^2=0`$
The overall tension of the bound state of magnetized D9-branes is positive, being the positive branch of a square root. As a result, the vanishing of the dilaton tadpole, despite the presence of the negative contribution from the tension of the $`\mathrm{\Omega }`$9-plane, seems hard to achieve with non-trivial fluxes compatibly with RR-tadpole cancellation. The remaining massless tadpoles indicate the presence of induced lower dimensional tensions of both signs, which in turn are derivatives of the potential generated by the BI couplings
$$๐ฏ_{\widehat{i}\widehat{j}}^a=E_{\widehat{i}}^iE_{\widehat{j}}^j\frac{๐ฑ_a}{G^{ij}}.$$
(22)
Indeed it is easy to prove that
$$E_{\widehat{i}}^iE_{\widehat{j}}^j\frac{๐ฑ_a}{G^{ij}}=\frac{1}{2}(R_{\widehat{i}\widehat{j}}^a+R_{\widehat{j}\widehat{i}}^a)=\frac{1}{2}\underset{q_a=\pm 1}{}R_{\widehat{i}\widehat{j}}^{aq_a}.$$
(23)
Let us stress once again that configurations of this kind are not bound states at threshold since their tension is the modulus of an algebraic sum rather than the arithmetic sum of moduli (positive numbers).
## 4 T-duality
An alternative way to understand the geometry behind the dy-charged string sectors relies on T-duality that transforms pairs of obliquely magnetized D9-branes into pairs of intersecting unmagnetized D6-branes. Obviously the required T-duality depends on the pair of branes under consideration and for generic oblique fluxes it is impossible to T-dualize the complete set of magnetized D9-branes into a set of (neutral) intersecting D6-branes. Yet, one can proceed pair by pair. The procedure is particularly rewarding for $`๐ฉ=2`$ sectors where the computation of $`I_{ab}^{}`$ and the determination of the โunmagnetizedโ directions $`u_{ab}`$ and thus of the generalized KK momenta $`p_{ab}`$ carried by open strings is rather subtle if not ambiguous to some extent.
For definiteness let us consider two examples that illustrate the general procedure: the sectors 5-8 and 5-4 of the AM model<sup>15</sup><sup>15</sup>15We use these subsectors only for illustrative purposes. .
In the first 5-8 case (in units of $`1/4\pi ^2\alpha ^{}`$)
$$F_5=dx^1dx^3dy^1dy^3F_8=dx^1dy^1+dx^2dy^2dx^3dy^3$$
(1)
in the coordinate system where $`x^i=x^i+2\pi k^i\sqrt{\alpha ^{}}`$, same for $`y^i`$. Barring the $`(x^2,y^2)`$ subtorus where any T-duality does the job, there are two possible T-duality trasformations of the remaining $`๐^4`$: $`T_{y^1}T_{x^3}`$ and $`T_{x^1}T_{y^3}`$. Let us choose the first and combine it with $`T_{y^2}`$ for definiteness. Neglecting the common non-compact spacetime dimensions, the problems is reduced to considering a $`๐^6`$ with two intersecting but unmagnetized D3-branes spanning the worldvolumes
$`D3_5(a,b,c)=a(E_{x^1}+q_5E_{\stackrel{~}{x}^3})+b(E_{y^3}q_5E_{\stackrel{~}{y}^1})+cE_{x^2}`$ (2)
$`D3_8(d,e,f)=d(E_{x^1}+q_8E_{\stackrel{~}{y}^1})+e(E_{y^3}q_8E_{\stackrel{~}{x}^3})+f(E_{x^2}q_8E_{\stackrel{~}{y}^2})`$ (3)
where $`a,b,c,d,e,f`$ are real parameters subject to periodic identifications ($`a_ia_i+1`$) and $`E_{x^i}`$ are orthogonal (since the metric is diagonal at the susy point) vectors along the (T-dualized) directions, normalized to
$`|E_{x^1}|=r_1^2=2^{3/2}|E_{\stackrel{~}{y}^1}|=\stackrel{~}{r}_1^2=2^{3/2}`$ (4)
$`|E_{x^2}|=r_2^2=2^{1/2}|E_{\stackrel{~}{y}^2}|=\stackrel{~}{r}_2^2=2^{1/2}`$ (5)
$`|E_{\stackrel{~}{x}^3}|=\stackrel{~}{r}_3^2=2^{1/2}|E_{y^3}|=r_3^2=2^{1/2}.`$ (6)
Along $`\stackrel{~}{๐}_{(2)}^2`$ $`D3_5`$ and $`D3_8`$ intersect once at an angle $`\beta _{58}^{(2)}=\mathrm{arctan}(\sqrt{2})`$. Let us focus on the remaining $`\stackrel{~}{๐}^4`$. We expect to find two orthogonal directions along which the 5-8 strings carry momentum and winding. KK momentum $`P_{58}`$ lies along the common longitudinal direction
$$P_{58}=\kappa \frac{E_{x^1}+q_8E_{\stackrel{~}{y}^1}q_5q_8E_{y^3}+q_5E_{\stackrel{~}{x}^3}}{|E_{x^1}+q_8E_{\stackrel{~}{y}^1}q_5q_8E_{y^3}+q_5E_{\stackrel{~}{x}^3}|^2},$$
(7)
that stretches once along the fundamental cell and has length $`|P_{58}|^2=\kappa ^2\sqrt{8}/15`$ with $`\kappa `$ an arbitrary integer. The allowed winding $`W_{58}`$ is aligned along the unique direction which is orthogonal to both branes
$$W_{58}=\nu (E_{x^1}8q_8E_{\stackrel{~}{y}^1}2q_5q_8E_{y^3}4q_5E_{\stackrel{~}{x}^3}).$$
(8)
It winds $`1+8+2+4(41)=12`$ times around the fundamental cell of $`T^4`$ and has length $`|W_{58}|^2=\nu ^215\sqrt{8}`$. The minimal allowed value of $`\nu `$ is $`1/15`$.
We are left with the magnetized plane $`\mathrm{\Pi }_M`$ spanned by the two the vectors
$`V_5=E_{x^1}2q_8E_{\stackrel{~}{y}^1}+2q_5q_8E_{y^3}+q_5E_{\stackrel{~}{x}^3}`$ (9)
$`V_8=2E_{x^1}+2q_8E_{\stackrel{~}{y}^1}+3q_5q_8E_{y^3}3q_5E_{\stackrel{~}{x}^3},`$ (10)
that lie along the worldline of the projections of $`D3_5`$ and $`D3_8`$ in the two-plane orthogonal to $`๐_{(2)}^2`$, $`P_{58}`$ and $`W_{58}`$. The two vectors are such that $`|V_5|^2=15/\sqrt{2}`$, $`|V_8|^2=45/\sqrt{2}`$ and $`V_5V_8=15/\sqrt{2}`$ and thus form an angle $`\beta _{58}^{(1)}=\mathrm{arccos}(1/\sqrt{3})=\mathrm{arctan}(\sqrt{2})`$! This being the same as the one in the $`T_{(2)}^2`$ confirms that $`๐ฉ=2`$ susy is preserved in this sector.
In order to compute $`I_{58}^{(1^{})}`$ one can T-dualize back and use open / closed string duality that yields
$$I_{ab}^{}L_PL_P^{}=V(T^{tot})\underset{Iu}{}\mathrm{sin}(\pi ฯต_{ab}^I)\sqrt{det(1+H_a)}\sqrt{det(1+H_b)}$$
(11)
Plugging numbers $`V(T^4)=4`$, $`\mathrm{sin}(\pi ฯต_{58})=\sqrt{2/3}`$, $`L_P=\sqrt{15/\sqrt{8}}`$, $`\sqrt{det(1+q_5H_5)}=5/4`$, $`\sqrt{det(1+q_8H_8^{})}=3\sqrt{3}/4`$, $`L_P^{}=15/L_W=\sqrt{15/\sqrt{8}}`$ one gets
$$I_{58}^{(1^{})}=1.$$
(12)
Let us now turn our attention on the 4-5 sector of the AM model
$$F_5=dx^1dx^3dy^1dy^3F_4=dx^2dx^3dy^2dy^3$$
(13)
A possible choice for the T-duality transformation is $`T_{y^1}T_{y^2}T_{x^3}`$ that yields
$`D3_4=a_4(E_{x^2}+q_4E_{\stackrel{~}{x}^3})+b_4(E_{y^3}q_4E_{\stackrel{~}{y}^2})+c_4E_{x^1}`$ (14)
$`D3_5=a_5(E_{x^1}+q_5E_{\stackrel{~}{x}^3})+b_5(E_{y^3}q_5E_{\stackrel{~}{y}^1})+c_5E_{x^2}`$ (15)
The allowed KK momenta lie along the common longitudinal direction
$$P_{45}=\kappa \frac{q_4E_{x^1}+q_5E_{x^2}+q_4q_5E_{\stackrel{~}{x}^3}}{|q_4E_{x^1}+q_5E_{x^2}+q_4q_5E_{\stackrel{~}{x}^3}|^2}$$
(16)
that stretches once ($`1+1+1(31)=1`$) along the fundamental cell of $`T^6`$ and has length $`|P_{45}|^2=\kappa ^2/3\sqrt{2}`$ with $`\kappa `$ an integer. The allowed windings stretch along the unique direction orthogonal to both branes
$$W_{45}=\nu (4q_5E_{\stackrel{~}{y}^1}+q_4E_{\stackrel{~}{y}^2}+E_{y^3})$$
(17)
that winds $`4+1+1(31)=4`$ times the fundamental cell, the minimal allowed value of $`\nu `$ is $`\nu =1/6`$ as can be seen geometrically or by requiring $`|W_{45}|_{min}^2=|P_{45}|_{min}^2`$ for the minimal non-vanishing zero-modes.
The magnetized 4-plane $`\mathrm{\Pi }_M`$ is spanned by the worldvolumes of the two D2-branes
$`D2_4^{}=g_4(E_{x^1}2q_4q_5E_{x^2}2q_5E_{\stackrel{~}{x}^3})+h_4(E_{y^3}q_4E_{\stackrel{~}{y}^2})=g_4\widehat{E}_{g_4}+h_4\widehat{E}_{h_4}`$ (18)
$`D2_5^{}=g_5(E_{x^1}5q_4q_5E_{x^2}+q_5E_{\stackrel{~}{x}^3})+h_5(E_{y^3}q_5E_{\stackrel{~}{y}^1})=g_5\widehat{E}_{g_5}+h_5\widehat{E}_{h_5}`$ (19)
obtained neglecting the common longitudinal direction $`P_{45}`$ i.e. taking the orthogonal complements to $`P_{45}`$ of $`D3_4`$ and $`D3_5`$. The hypervolumes spanned by $`\mathrm{\Pi }_M`$, $`D2_4^{}`$, $`D2_5^{}`$ are given by
$$\widehat{V}=\sqrt{det(\widehat{E}_A\widehat{E}_B)}$$
(20)
where $`\widehat{E}_A=\{\widehat{E}_{g_4},\widehat{E}_{h_4},\widehat{E}_{g_5},\widehat{E}_{h_5}\}`$ for $`\mathrm{\Pi }_M`$, while $`\widehat{E}_a=\{\widehat{E}_{g_4},\widehat{E}_{h_4}\}`$ for $`D2_4^{}`$ and $`\widehat{E}_b=\{\widehat{E}_{g_5},\widehat{E}_{h_5}\}`$ for $`D2_5^{}`$. The relevant scalar products are
$`|\widehat{E}_{g_4}|^2=6\sqrt{2}|\widehat{E}_{h_4}|^2=2\sqrt{2}|\widehat{E}_{g_5}|^2=15\sqrt{2}|\widehat{E}_{h_5}|^2={\displaystyle \frac{5}{4}}\sqrt{2}`$ (21)
$`\widehat{E}_{g_4}\widehat{E}_{g_5}=6\sqrt{2}\widehat{E}_{h_4}\widehat{E}_{h_5}=\sqrt{2}\widehat{E}_{g_4}\widehat{E}_{h_4}=\widehat{E}_{g_4}\widehat{E}_{h_5}=\widehat{E}_{g_5}\widehat{E}_{h_4}=\widehat{E}_{g_5}\widehat{E}_{h_5}=0`$
So that
$$V(\mathrm{\Pi }_M)=18A(D2_4^{})=2\sqrt{6}A(D2_4^{})=\frac{5}{2}\sqrt{6}$$
(22)
Then the intersection angles are such that
$$\underset{I}{}\mathrm{sin}(\pi ฯต_{ab}^I)=V(\mathrm{\Pi }_M)/A(Dp_a)A(Dp_b)$$
(23)
In our 4-5 case, since $`\mathrm{sin}(\pi ฯต_{45}^1)=\mathrm{sin}(\pi ฯต_{45}^2)`$ for susy reasons, one has
$$\mathrm{sin}(\pi ฯต_{45})=\sqrt{18/30}=\sqrt{3/5}$$
(24)
that means
$$\mathrm{tan}(\pi ฯต_{45})=\sqrt{3/2}$$
(25)
that coincides with the result of diagonalizing $`R_{45}`$! In order to compute the index $`I_{45}^{}`$ and confirm that the minimal winding is indeed $`1/6`$ it is very convenient to factorize the problem in the two subtori $`๐^6=๐_X^3๐_Y^3`$. On the first subtorus $`๐_X^3=๐_{x^1x^2\stackrel{~}{x}^3}^3`$, we have two D2-branes intersecting along the common direction $`P_{45}`$ at an angle such that $`\mathrm{cos}(\beta _{ab}^X)=\sqrt{2/5}`$. The fig.(1) displays the the orientation of $`P_{45}`$ wrt the cell. As already stated $`P_{45}`$ winds only once within the fundamental cell. In the second sub-torus $`๐_X^3=๐_{\stackrel{~}{y}^1\stackrel{~}{y}^2y^3}^3`$ the two D1-branes interesect only at the origin and span a plane orthogonal to $`W_{45}`$, as fig.(2) shows.
The distance between two such planes, i.e. intersections, along the direction of $`W_{45}`$ is $`|W_{45}|/6`$ hence $`\nu _{min}=1/6`$. Thus $`I_{45}^{}=1`$, which is nicely consistent with open - closed channel duality after putting numbers $`V(T^6)=\sqrt{8}`$, $`\mathrm{sin}(\beta )=\sqrt{3/5}`$, $`\sqrt{det(1+H_4)}=2`$ and $`\sqrt{det(1+H_5)}=5/4`$, $`L_P=\sqrt{3\sqrt{2}}`$ and $`L_P^{}=6/L_W=\sqrt{6/\sqrt{2}}`$.
## 5 Gauge Coupling Thresholds
Once the perturbative open string spectrum is known, computing some of the low-energy effective couplings is quite straightforward. Tree level gauge couplings were determined in and turned out to be given by
$$\frac{4\pi }{g_a^2}=e^\mathrm{\Phi }\sqrt{det(๐ข_a+_a)}=e^\mathrm{\Phi }|W_a|\sqrt{det(G+F_a)}$$
(1)
Moduli dependence is hidden inside the induced metric $`๐ข_a`$. The induced internal magnetic field $`_a=F_{ij}^a_\alpha X^i_\beta X^j`$ satisfies the standard Dirac quantization condition. At supersymmetric points, the expression simplifies and indeed coincides with the WZ term due to the identity of tension and charge for the magnetized brane (configuration). Moreover, in principle, for proper choices of the oblique magnetic fields all closed string moduli, except for the complexified dilaton, could be frozen. The ratios of the couplings would then be completely determined. More precisely, we are ignoring possible mixings of the open string Wilson line moduli, i.e. we are setting their VEVโs to zero for the time being.
Although a consistent variant of the AM model has not yet been found, our aim is to extend the tree level analysis \[mbt\] to one-loop and derive the running of the gauge couplings as well as their threshold corrections. This is a preliminary step towards the study of gauge coupling unification in models with oblique magnetic fluxes or closely related (orbifold) models that might be phenomenologically more appealing but still solvable.
We will follow the strategy pioneered by and successfully applied to type I orbifolds in , to generic type I vacuum configurations in and to intersecting brane models in , based on the background field method.
As hinted at in the introduction, the method consists in applying an abelian, constant and small magnetic field in some spacetime directions, computing the effect of such an integrable deformation and then extracting the desired (quadratic) term in the one-loop effective action.
Only open strings that have at least one end on the spacetime magnetized brane will sense the presence of the magnetic field and can a priori contribute to the renormalization of the corresponding gauge coupling. In principle, one should consider dipole strings, preserving $`๐ฉ=4`$, as well as singly- and doubly-charged strings, preserving $`๐ฉ=1`$ or $`๐ฉ=2`$. However, similar to what happens in simpler cases with parallel magnetic fields or untwisted sector of orbifolds, $`๐ฉ=4`$ sectors neither contribute to the running nor to the thresholds, while $`๐ฉ=1`$ or $`๐ฉ=2`$ sectors contribute both to the running and to the thresholds. Massless open string states contribute to the logarithmic running, and we will retrieve the field-theory $`\beta `$-function coefficients studying the IR limit of the relevant one loop amplitudes. The contribution to the thresholds from $`๐ฉ=2`$ sectors is particularly simple since the gauge coupling is 1/2 BPS-saturated , only the zero-modes coded in the โmagneticallyโ deformed internal lattice sum will survive but no string oscillator modes . The contribution to the thresholds from $`๐ฉ=1`$ sectors is slightly more involved since the gauge coupling is not BPS-saturated in this case . As obvious from the discussion of the spectrum in Section 2 there are no lattice sums in these sectors, but magnetically shifted string oscillator modes can and do contribute. Luckily we will recast our anlysis along the lines of where the modular integral for the case of intersecting branes were computed and finally expressed in terms of $`\mathrm{\Gamma }`$ functions. Once again the moduli dependence hidden in the lattice sums or the magnetic shifts (T-dual to angles) can be fixed for particular choices of the internal oblique fluxes.
### 5.1 General analysis
For definiteness, we turn on an abelian magnetic field in spacetime directions 2 and 3, viz.
$$F_{\mu \nu }=\delta _{[\mu }^2\delta _{\nu ]}^3fQ$$
(2)
where $`Q`$ is one of the generator of the unbroken CP group, normalized so that $`Tr_N(Q)=0`$ and $`Tr_N(Q^2)=1/2`$. Depending on the embedding of $`Q`$ in the CP group one can find different behaviours. We will mostly focus on the case in which $`Q`$ is a generator of a non-abelian and thus non-anomalous factor.
As for internal fluxes, the spacetime magnetic deformation is integrable. Amplitudes on surfaces with no boundaries, such as torus and the Klein Bottle are insensible to the external field. Annulus and Mรถbius strip do couple to the external field and the connected generating functional depends on $`f`$. The main effects of turning on $`f`$ are the magnetic shifts $`ฯต_{ab}^Q`$ of the transverse spacetime modes and the degeneracy $`I_{ab}^Q`$ of Landau levels for the string modes in the $`[23]`$ plane. Both are related to the charge $`Q`$ of the open string according to<sup>16</sup><sup>16</sup>16To be pedantic for the annulus $`Q_a`$ actually means $`Q_{(a)}1_{(b)}`$ and $`Q_a+Q_b`$ actually means $`Q_{(a)}1_{(b)}+1_{(a)}Q_{(b)}`$.
$$ฯต_{ab}^Q=\frac{1}{\pi }[\mathrm{arctan}(Q_af)+\mathrm{arctan}(Q_bf)]$$
(3)
for $`๐_{ab}(f)`$ with $`I_{ab}^Q=(Q_a+Q_b)f/2\pi `$, and
$$ฯต_{aa}^Q=\frac{2}{\pi }\mathrm{arctan}(Q_af)$$
(4)
for $`_{aa}(f)`$ with $`\widehat{I}_{aa}^Q=2Q_af/2\pi `$.
Expanding the Annulus and the Mรถbius amplitudes up to second order in $`f`$ one gets the one-loop gauge threshold for the group $`Q`$ belongs to
$$\mathrm{\Delta }_Q=\frac{dt}{t}(๐_Q^{\prime \prime }(0)+_Q^{\prime \prime }(0))=\frac{dt}{4t}B_Q(t)$$
(5)
that implicitly depends on the moduli fields through the dependence on the latter of the masses of the unoriented open string states running in the loop. We use $`Q`$ to label the (factor) group we are computing the threshold of.
The only allowed momenta are along the light-cone directions $`0`$ and $`1`$. Analytic continuation and volume regularization thus yield
$$\frac{V_{LC}dp^+dp^{}}{(2\pi )^2}\mathrm{exp}(\pi \alpha ^{}tp^+p^{})=\frac{V_{LC}}{(2\pi )^2\alpha ^{}t}.$$
(6)
One can easily write down the contribution of singly as well as doubly charged unoriented strings<sup>17</sup><sup>17</sup>17Sums over $`a`$ or $`b`$ include branes as well as their images under $`\mathrm{\Omega }`$, which in our conventions sends $`q`$ into $`q`$.:
$`๐_Q^{๐ฉ=1}(f)`$ $`=`$ $`{\displaystyle \underset{a,b}{}}I_{ab}{\displaystyle \underset{\alpha ,\beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{2}}tr_{N_a\times N_b}\left[{\displaystyle \frac{(Q_a+Q_b)f}{(2\pi )^3\alpha ^{}t}}{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](ฯต_{ab}^Q\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^Q\tau _A|\tau _A)}}\right]{\displaystyle \underset{I}{}}{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](ฯต_{ab}^I\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^I\tau _A|\tau _A)}}`$
$`_Q^{๐ฉ=1}(f)`$ $`=`$ $`{\displaystyle \underset{a}{}}\widehat{I}_{aa}{\displaystyle \underset{\alpha \beta }{}}c_{\alpha \beta }Tr_{N_a}\left[{\displaystyle \frac{Q_af}{(2\pi )^3\alpha t}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^Q\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^Q\tau _A|\tau _M)}}\right]{\displaystyle \underset{I}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^I\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^I\tau _A|\tau _M)}})`$ (7)
where $`\tau _A=it/2`$, $`\tau _M=\tau _A+1/2`$, and, denoting as usual by $`u`$ the internal unmagnetized direction, when present,
$`๐_Q^{๐ฉ=2}(f)`$ $`=`$ $`i{\displaystyle \underset{a,b}{}}I_{ab}^{}\mathrm{\Lambda }_{ab}^u(\tau _A){\displaystyle \underset{\alpha ,\beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{2}}tr_{N_a\times N_b}\left[{\displaystyle \frac{(Q_a+Q_b)f}{(2\pi )^3\alpha ^{}t}}{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](ฯต_{ab}^Q\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^Q\tau _A|\tau _A)}}\right]`$ (8)
$`{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](0|\tau _A)}{\eta ^3(\tau _A)}}{\displaystyle \underset{Iu}{}}{\displaystyle \frac{\theta [{}_{\beta }{}^{\alpha }](ฯต_{ab}^I\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^I\tau _A|\tau _A)}}`$
$`_Q^{๐ฉ=2}(f)`$ $`=`$ $`i{\displaystyle \underset{a}{}}\widehat{I}_{aa}^{}\mathrm{\Lambda }_{aa}^u(\tau _A){\displaystyle \underset{\alpha \beta }{}}c_{\alpha \beta }N_aTr_{N_a}\left[{\displaystyle \frac{Q_af}{(2\pi )^3\alpha t}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^Q\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^Q\tau _A|\tau _M)}}\right]{\displaystyle \frac{\theta [_\beta ^\alpha ](0|\tau _M)}{\eta ^3(\tau _M)}}`$ (9)
$`{\displaystyle \underset{Iu}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^I\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^I\tau _A|\tau _M)}}`$
$`๐ฉ=4`$ sectors, such as neutral and dipole strings with opposite $`Q`$ charge at their ends do not contribute to the thresholds since their modes are not shifted and they simply receive an overall factor reflecting the โmagnetic deformationโ of the lattice sum. Similarly, the Mรถbius strip does not contribute if $`Q`$ is part of $`SO(N_0)`$, associated to โunmagnetizedโ branes if present, so that the two ends have opposite charge. Henceforth we set $`\alpha ^{}=1/2`$ for convenience.
### 5.2 $`๐ฉ=1`$ sectors
In $`๐ฉ=1`$ supersymmetric sectors, expanding to quadratic order one gets
$`๐_Q^{๐ฉ=1}(f)`$ $`=`$ $`i{\displaystyle \underset{a,b}{}}I_{ab}{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{f}{2\pi }}\right)^2{\displaystyle \underset{\alpha \beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{8\pi ^2}}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle \frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0)}{\eta ^3}}{\displaystyle \underset{I}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{ab}^I\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^I\tau _A|\tau _A)}})+\mathrm{}`$
$`_Q^{๐ฉ=1}(f)`$ $`=`$ $`i{\displaystyle \underset{a}{}}I_{aa}{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{f}{2\pi }}\right)^2{\displaystyle \underset{\alpha \beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{2\pi ^2}}Tr_{N_a}(Q_a^2){\displaystyle \frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0|\tau _M)}{\eta ^3(\tau _M)}}{\displaystyle \underset{I}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^I\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^I\tau _A|\tau _M)}}+\mathrm{}`$ (10)
Summing over spin structures and using the generalized Jacobi $`\theta `$ function identity
$$\underset{\alpha \beta }{}c_{\alpha \beta }\frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0)}{\eta ^3}\underset{I}{}\frac{\theta [_\beta ^\alpha ](ฯต^I\tau )}{\theta _1(ฯต^I\tau )}=2\pi \underset{I}{}\frac{\theta _1^{}(ฯต_{ab}^I\tau )}{\theta _1(ฯต^I\tau )}$$
(11)
give
$`B_Q^{๐ฉ=1}(t)`$ $`=`$ $`{\displaystyle \frac{i}{\pi }}{\displaystyle \underset{a,b}{}}I_{ab}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle \underset{I}{}}{\displaystyle \frac{\theta _1^{}(ฯต_{ab}^I\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^I\tau _A|\tau _A)}}`$
$`\widehat{B}_Q^{๐ฉ=1}(t)`$ $`=`$ $`{\displaystyle \frac{i}{\pi }}{\displaystyle \underset{a}{}}I_{aa}Tr_{N_a}(2Q_a^2){\displaystyle \underset{I}{}}{\displaystyle \frac{\theta _1^{}(ฯต_{aa}^I\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^I\tau _A|\tau _M)}}`$ (12)
At this point it is easy to extract the $`\beta `$-function coefficents from the IR limit of (5.2)
$`b_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{a,b}{}}I_{ab}Tr_{N_a\times N_b}(Q_a+Q_b)^2`$
$`\widehat{b}_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{a}{}}I_{aa}Tr_{N_a}(2Q_a)^2`$ (13)
Since all vectors belong to $`๐ฉ=4`$ multiplets, $`\beta `$-function are positive, i.e. all non-abelian couplings grow in the UV.
In order to perform the integral and compute $`\mathrm{\Delta }_Q`$ we switch to the transverse channel and end up with the following expressions
$`\mathrm{\Delta }_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle \underset{a,b}{}}I_{ab}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle \underset{I}{}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\theta _1^{}(ฯต_{ab}^I|i\mathrm{})}{\theta _1(ฯต_{ab}^I|i\mathrm{})}}๐\mathrm{}`$
$`\widehat{\mathrm{\Delta }}_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{a}{}}Tr_{N_a}(2Q_a)^2I_{aa}{\displaystyle \underset{I}{}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\theta _1^{}(ฯต_{0a}^I|i\mathrm{}+1/2)}{\theta _1(ฯต_{0a}^I|i\mathrm{}+1/2)}}๐\mathrm{}`$ (14)
Series expansion
$$\frac{\theta _1^{}(ฯต|\tau )}{\theta _1(ฯต|\tau )}=\pi \mathrm{cot}(\pi ฯต)+2\underset{k=1}{\overset{\mathrm{}}{}}\zeta (2k)ฯต^k(E_{2k}(\tau )1),$$
(15)
where $`\zeta (2k)=(2\pi )^{2k}|B_{2k}|/(2k)!`$ and $`E_{2k}(\tau )`$ is an Eisenstein series with modular weight $`2k`$, expose the potentially divergent terms
$`\delta _Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \underset{a,b}{}}{\displaystyle \frac{I_{ab}}{4}}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle \underset{I}{}}\mathrm{cot}(\pi ฯต_{ab}^I){\displaystyle _0^{\mathrm{}}}๐\mathrm{}`$
$`\widehat{\delta }_Q^{๐ฉ=1}`$ $`=`$ $`2{\displaystyle \underset{a}{}}\widehat{I}_{aa}Tr_{N_a}(Q_a^2){\displaystyle \underset{I}{}}\mathrm{cot}(\pi ฯต_{0a}^I){\displaystyle _0^{\mathrm{}}}๐\mathrm{}`$ (16)
that eventually cancel thanks to (NS-NS) tadpole cancellation, for the non-anomalous $`Q`$, with $`Tr(Q)=0`$. The latter condition has to be used in order to dispose of the divergent terms with $`f`$ insertions in two different boundaries. Divergences from insertions on the same boundary cancel between annulus and Mรถbius strip thanks to tadpole cancellation.
The finite terms boil down to integrals of the form
$$_0^{\mathrm{}}๐\mathrm{}\underset{k=1}{\overset{\mathrm{}}{}}2\zeta (2k)ฯต^k(E_{2k}(i\mathrm{})1)=\pi \mathrm{log}\left[\frac{\mathrm{\Gamma }(1ฯต)}{\mathrm{\Gamma }(1+ฯต)}\right]+2\pi ฯต\gamma _E$$
(17)
$$_0^{\mathrm{}}๐\mathrm{}\underset{k}{}2\zeta (2k)ฯต^k(E_{2k}(i\mathrm{}+1/2)1)=\pi \mathrm{log}\left[\frac{\mathrm{\Gamma }(12ฯต)}{\mathrm{\Gamma }(1+2ฯต)}\right]+2\pi ฯต\gamma _E$$
(18)
Actually the last contributions, linear in $`ฯต`$, drop after summing over the three internal directions in supersymmetric cases.
Summing the various contributions one finally gets
$`\mathrm{\Delta }_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \underset{a,b}{}}{\displaystyle \frac{I_{ab}}{2}}Tr_{N_a\times N_b}(Q_a^2+Q_b^2){\displaystyle \underset{I}{}}\mathrm{log}\left[{\displaystyle \frac{\mathrm{\Gamma }(1ฯต_{ab}^I)}{\mathrm{\Gamma }(1+ฯต_{ab}^I)}}\right]`$
$`\widehat{\mathrm{\Delta }}_Q^{๐ฉ=1}`$ $`=`$ $`{\displaystyle \underset{a}{}}I_{aa}Tr_{N_a}(2Q_a)^2{\displaystyle \underset{I}{}}\mathrm{log}\left[{\displaystyle \frac{\mathrm{\Gamma }(1ฯต_{aa}^I)}{\mathrm{\Gamma }(1+ฯต_{aa}^I)}}\right],`$ (19)
where $`ฯต_{aa}^I=2ฯต_{ao}^I`$.
### 5.3 $`๐ฉ=2`$ sectors
Thresholds corrections from $`๐ฉ=2`$ sectors are much easier to compute since they correspond to BPS saturated couplings. Indeed, for $`๐ฉ=2`$ supersymmetric sectors, the terms quadratic in $`f`$ read
$`๐_Q^{๐ฉ=2}(f)`$ $`=`$ $`{\displaystyle \underset{a,b}{}}I_{ab}^{}\mathrm{\Lambda }_{ab}^u(\tau _A){\displaystyle \frac{1}{2}}\left({\displaystyle \frac{f}{2\pi }}\right)^2{\displaystyle \underset{\alpha \beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{8\pi ^2}}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle \frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0|\tau _A)}{\eta ^3(\tau _A)}}{\displaystyle \frac{\theta [_\beta ^\alpha ](0|\tau _A)}{\eta ^3(\tau _A)}}`$
$`{\displaystyle \underset{Iu}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{ab}^I\tau _A|\tau _A)}{\theta _1(ฯต_{ab}^I\tau _A|\tau _A)}}+\mathrm{}`$
$`_Q^{๐ฉ=2}(f)`$ $`=`$ $`{\displaystyle \underset{a}{}}I_{aa}^{}\mathrm{\Lambda }_{aa}^u(\tau _A){\displaystyle \frac{1}{2}}\left({\displaystyle \frac{f}{2\pi }}\right)^2{\displaystyle \underset{\alpha \beta }{}}{\displaystyle \frac{c_{\alpha \beta }}{8\pi ^2}}Tr_{N_a}(2Q_a)^2{\displaystyle \frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0|\tau _M)}{\eta ^3(\tau _M)}}`$ (20)
$`{\displaystyle \frac{\theta [_\beta ^\alpha ](0|\tau _M)}{\eta ^3(\tau _M)}}{\displaystyle \underset{Iu}{}}{\displaystyle \frac{\theta [_\beta ^\alpha ](ฯต_{aa}^I\tau _A|\tau _M)}{\theta _1(ฯต_{aa}^I\tau _A|\tau _M)}})+\mathrm{}`$
and the Jacobi $`\theta `$ function identity
$$\underset{\alpha \beta }{}c_{\alpha \beta }\frac{\theta ^{\prime \prime }[_\beta ^\alpha ](0)}{\eta ^3}\frac{\theta [_\beta ^\alpha ](0)}{\eta ^3}\underset{Iu}{}\frac{\theta [_\beta ^\alpha ](\zeta ^I|\tau )}{\theta _1(\zeta ^I|\tau )}=4\pi ^2,$$
(21)
valid for $`_I\zeta ^I=0`$, imply that only the lattice sum over 1/2 BPS states contributes.
Manipulations similar to the above yield the following results for $`\beta `$-function coefficents in $`๐ฉ=2`$ sectors,
$`b_Q^{๐ฉ=2}`$ $`=`$ $`{\displaystyle \underset{a,b}{}}I_{ab}^{}Tr_{N_a\times N_b}(Q_a+Q_b)^2`$
$`\widehat{b}_Q^{๐ฉ=2}`$ $`=`$ $`{\displaystyle \underset{a}{}}\widehat{I}_{aa}^{}N_aTr_{N_a}(2Q_a)^2`$ (22)
Since all vectors belong to $`๐ฉ=4`$ multiplets, $`\beta `$-functions are positive, i.e. all non-abelian couplings grow in the UV.
For the thresholds one has
$`\mathrm{\Delta }_Q^{๐ฉ=2}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{a,b}{}}I_{ab}^{}Tr_{N_a\times N_b}(Q_a+Q_b)^2{\displaystyle _0^{\mathrm{}}}\mathrm{\Lambda }_{ab}^u(it){\displaystyle \frac{dt}{t}}`$
$`\mathrm{\Delta }_Q^{๐ฉ=2}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{a}{}}I_{aa}^{}Tr_{N_a}(2Q_a)^2{\displaystyle _0^{\mathrm{}}}\mathrm{\Lambda }_{aa}^u(it){\displaystyle \frac{dt}{t}}`$ (23)
The integrals of the โregulatedโ lattice sums can be performed with the aid of the formula
$$_0^{\mathrm{}}\frac{dt}{t}\underset{(k_1,k_2)(0,0)}{}\mathrm{exp}(\pi \mathrm{}|k_1+Uk_2|^2/V_2U_2)=\gamma _E\mathrm{log}[4\pi V_2U_2|\eta (U)|^4]$$
(24)
where $`V_2`$ is the volume and $`U`$ the complex structure of the unmagnetized torus. Inserting into (24) one gets
$`\mathrm{\Delta }_Q^{๐ฉ=2}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{a,b}{}}I_{ab}^{}Tr_{N_a\times N_b}(Q_a+Q_b)^2[\mathrm{ln}(V_2U_2|\eta (U)|^44\pi )\gamma _E]`$
$`\mathrm{\Delta }_Q^{๐ฉ=2}`$ $`=`$ $`2{\displaystyle \underset{a}{}}I_{aa}^{}Tr_{N_a}(Q_a^2)[\mathrm{ln}(4V_2U_2|\eta (U)|^44\pi )\gamma _E]`$ (25)
For proper choices of the internal oblique fluxes all closed string moduli are fixed, modulo mixing with massless open string states, and the above formulae give simply numbers as we will momentarily see.
We would now like to comment on the effect of large extra dimensions on the coupling costants running. Setting $`U_1=0`$ for simplicity and expanding the logarithm in the last expression yields
$$\mathrm{ln}(4\pi V_2U_2|\eta (U)|^4)=\mathrm{ln}(4\pi VU_2)\frac{\pi r_2}{3r_1}+4\underset{n}{}\mathrm{ln}(1e^{2\pi \frac{r_2}{r_1}n})$$
(26)
where $`r_2r_1`$ are the radii of $`T_u^2`$ so that $`U_2=r_2/r_1`$. One can envisage two different situations
* $`r_2r_1`$, when the radii are fixed at the string scale, and corrections do not seem to affect the usual logarithmic behavior.
* $`r_2>>r_1`$. Power corrections are dominant, this is the scenario already described in , where power corrections are induced by the running in the loop of bulk particle, with KK towers organized in $`๐ฉ=2`$ multiplets. Power law behavior can be exploited to lower the unification scale. This is achieved if one of the unmagnetized eigenvectors points toward a large extra dimension.
## 6 Outlook
In the present paper we have derived explicit formulae for the one-loop contributions to type I string compactifications on tori with arbitrary magnetic fluxes . We have checked consistency with the transverse channel and identified the correct tadpole conditions. Further insights in the geometry of these vacuum configurations has been gained by means of T-duality . We have then turned our attention to the one-loop threshold corrections to the non-abelian gauge couplings and derived very compact expressions thereof, relying on similar analyses for type I orbifolds and intersecting brane models . Although unrealistic in many respects, toroidal models of this kind may be used as building blocks or rather starting points for type I orbifolds and other solvable (supersymmetric) compactifications. In particular the emergence of induced lower dimensional R-R charges and NS-NS tensions of both signs plays a crucial role in solving some long standing puzzles .
Given the high level of control one has on this class of models, one can restrict oneโs attention onto those that resemble as closely as possible the Standard Model or some of its supersymmetric or grand unified generalizations. In principle, the magnetic fields can be tuned so as to produce the desired gauge group and fermionic content, and achieve gauge coupling unification. With more effort one can try to generate the correct pattern of Yukawa couplings and trigger supersymmetry breaking in a controllable way .
Stabilizing dilaton and axion may require switching to a T-dual description in terms of D3-branes that allows the introduction of closed string 3-form fluxes. Yet the same goal may be achieved by means of non-perturbative effects such as D5-brane and D-string instantons. In any case, at present the possibility that all moduli be stabilized by perturbative effects remains a challenge. The presence of dilaton tadpoles at different orders in perturbation theory may help achieving this goal.
Moreover, open string Wilson line moduli, especially those charged under the anomalous $`U(1)`$โs, can mix with closed string moduli, due to their contribution to D-terms<sup>18</sup><sup>18</sup>18We thank L. Ibanez and the referee of for vigorously pointing this out to us.. This complicates the analysis, that has been so far performed at the origin of the open string moduli space. In this respect, it is reassuring to observe that scalars in vector multiplets can be lifted by orbifold projections and in any case they can be treated exactly along the lines of . In order to set the stage for the discussion of the lifting of scalars in chiral multiplets one should compute the superpotential, i.e. the Yukawa couplings . For charged open string states, the relevant amplitudes involve mutually non-abelian twists in general and the perspective of computing them is daunting<sup>19</sup><sup>19</sup>19As suggested by C. Bachas and E. Kiritsis it may prove convenient to extract the coupling from factorization of a four-point amplitude.. Yet it may be worth proving.
## Acknowledgements
We would like to thank P. Anastasopoulos, C. Angelantonj, I. Antoniadis, S. Ferrara, F. Fucito, A. Lionetto, J. F. Morales Morera, G. Pradisi, M. Prisco, A. Sagnotti, and Ya. Stanev for useful discussions. During completion of this work M.B. was visiting the String Theory group at Ecole Polytechnique, E. Kiritsis and his colleagues are warmly acknowledged for their kind hospitality. This work was supported in part by INFN, by the MIUR-COFIN contract 2003-023852, by the EU contracts MRTN-CT-2004-503369 and MRTN-CT-2004-512194, by the INTAS contract 03-516346 and by the NATO grant PST.CLG.978785.
## Appendix A: Some useful formulae
$`{\displaystyle \underset{Iu}{}}\mathrm{sin}(\pi ฯต_{ab}^I)`$ $`=`$ $`{\displaystyle \underset{Iu}{}}{\displaystyle \frac{1}{2i}}(e^{i\pi ฯต_{ab}^I}e^{i\pi ฯต_{ab}^I})`$ (1)
$`=`$ $`\sqrt{{\displaystyle \underset{Iu}{}}e^{i\pi ฯต_{ab}^I}\left({\displaystyle \frac{1}{2i}}\right)(e^{i2\pi ฯต_{ab}^I}1)e^{i\pi ฯต_{ab}^I}\left({\displaystyle \frac{1}{2i}}\right)(e^{i2\pi ฯต_{ab}^I}1)}`$
$`=`$ $`\sqrt{\stackrel{}{det}\left({\displaystyle \frac{R_{ab}1}{2}}\right)}={\displaystyle \frac{\sqrt{\stackrel{}{det}(H_a+H_b)}}{\sqrt{det(1+H_a)}\sqrt{det(1+H_b)}}}`$
$`{\displaystyle \underset{Iu}{}}\mathrm{cos}(\pi ฯต_{ab}^I)`$ $`=`$ $`{\displaystyle \underset{Iu}{}}{\displaystyle \frac{1}{2}}(e^{i\pi ฯต_{ab}^I}+e^{i\pi ฯต_{ab}^I})`$ (2)
$`=`$ $`\sqrt{{\displaystyle \underset{Iu}{}}e^{i\pi ฯต_{ab}^I}\left({\displaystyle \frac{1}{2}}\right)(e^{i2\pi ฯต_{ab}^I}+1)e^{i\pi ฯต_{ab}^I}\left({\displaystyle \frac{1}{2}}\right)(e^{i2\pi ฯต_{ab}^I}+1)}`$
$`=`$ $`\sqrt{det\left({\displaystyle \frac{R_{ab}+1}{2}}\right)}={\displaystyle \frac{\sqrt{det(1+H_aH_b)}}{\sqrt{det(1+H_a)}\sqrt{det(1+H_b)}}}`$
$`=`$ $`Tr(U_{ab})=Tr(U_aU_b)`$
where $`U_a=\frac{\mathrm{AExp}(\frac{1}{2}F_{a,ij}\mathrm{\Gamma }^{ij})}{\sqrt{det(1+H_a)}}`$. As a corollary
$$\underset{Iu}{}2\mathrm{cos}(\pi ฯต_a^I)=\frac{1}{\sqrt{det(1+H_a)}}=Tr_s(U_a)$$
(3)
$$2\underset{Iu}{}\mathrm{cos}(2\pi ฯต_{ab}^I)=\underset{Iu}{}(e^{2i\pi ฯต_{ab}^I}+e^{2i\pi ฯต_{ab}^I})=Tr_v(R_{ab})=Tr_v(R_aR_b)$$
(4)
## Appendix B: Theta functions
### Definitions
In order to fix notations, we report in this appendix the Jacobi $`\theta `$-functions, we used throughout the paper. Let $`q=e^{2\pi i\tau }`$ they are defined as guassian sums
$$\theta [{}_{\beta }{}^{\alpha }](z|\tau )=\underset{n}{}q^{\frac{1}{2}(n\alpha )^2}e^{2\pi i(z\beta )(n\alpha )}$$
(5)
where $`\alpha \beta R`$.
Equivalently, for particular values of characteristics, such as $`\alpha \beta =0\text{ , }\frac{1}{2}`$ they are given also in terms of infinite product as follows
$`\theta [{}_{\frac{1}{2}}{}^{\frac{1}{2}}](z|\tau )=\theta _1(z|\tau )`$ $`=`$ $`2q^{\frac{1}{8}}\mathrm{sin}(\pi z){\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1e^{2\pi iz}q^m)(1e^{2\pi iz}q^m)`$
$`\theta [{}_{0}{}^{\frac{1}{2}}](z|\tau )=\theta _2(z|\tau )`$ $`=`$ $`2q^{\frac{1}{8}}\mathrm{cos}(\pi z){\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1+e^{2\pi iz}q^m)(1+e^{2\pi iz}q^m)`$
$`\theta [{}_{0}{}^{0}](z|\tau )=\theta _3(z|\tau )`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1+e^{2\pi iz}q^{m\frac{1}{2}})(1+e^{2\pi iz}q^{m\frac{1}{2}})`$
$`\theta [{}_{\frac{1}{2}}{}^{0}](z|\tau )=\theta _4(z|\tau )`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}(1q^m)(1e^{2\pi iz}q^{m\frac{1}{2}})(1e^{2\pi iz}q^{m\frac{1}{2}})`$ (6)
The Dedekind function $`\eta `$ is defined as
$$\eta (\tau )=q^{\frac{1}{24}}\underset{n=1}{\overset{\mathrm{}}{}}(1q^n)$$
(7)
The Eisenstein series are
$$E_r=\underset{m0}{\overset{\mathrm{}}{}}\underset{n0}{\overset{\mathrm{}}{}}\frac{1}{(m+n\tau )^r}$$
(8)
with $`r>2`$. Moreover they can be expressed as polynomial of elliptic funtions
$$E_{2k}(\tau )=1+\frac{(2\pi i)^{2k}}{(2k1)!\zeta (2k)}\underset{n=1}{\overset{\mathrm{}}{}}\sigma _{2k1}(n)e^{2\pi in\tau }$$
(9)
where $`\zeta (2k)`$ is Riemann zeta function and $`\sigma _{2k1}(n)`$ is the divisor function
$$\sigma _{2k1}(n)=\underset{d|n}{}d^{2k1}.$$
(10)
### Modular Transformations
Under T and S modular trasformations on the arguments the functions, given above, have pecular properties:
$`\theta [{}_{\beta }{}^{\alpha }](z|\tau +1)`$ $`=`$ $`e^{i\pi \alpha (\alpha 1)}\theta [{}_{\beta +\alpha \frac{1}{2}}{}^{\alpha }](z|\tau )`$
$`\eta (\tau +1)`$ $`=`$ $`e^{\frac{i\pi }{12}}\eta (\tau )`$
$`E_{2k}(\tau +1)`$ $`=`$ $`E_{2k}(\tau )`$
$`\theta [{}_{\beta }{}^{\alpha }]({\displaystyle \frac{z}{\tau }}|{\displaystyle \frac{1}{\tau }})`$ $`=`$ $`(i\tau )^{\frac{1}{2}}e^{2i\pi \alpha \beta +i\pi z^2/\tau }\theta [{}_{\alpha }{}^{\beta }](z|\tau )`$
$`\eta ({\displaystyle \frac{1}{\tau }})`$ $`=`$ $`(i\tau )^{\frac{1}{2}}\eta (\tau )`$
$`E_{2k}({\displaystyle \frac{1}{\tau }})`$ $`=`$ $`\tau ^2kE_{2k}(\tau )`$ (11)
The modular transformation P on the Jacobi functions is more involved as it consists in a sequence of T and S transformation ($`P=TST^2S`$). on the modular parameter $`\tau _M=\frac{1}{2}+\frac{it}{2}`$
$`\theta [{}_{\beta }{}^{\alpha }]({\displaystyle \frac{z}{it}}|{\displaystyle \frac{1}{2}}+{\displaystyle \frac{i}{2t}})`$ $`=`$ $`e^{i\pi \alpha (\alpha 1)2\pi i(\alpha +\beta 1/2)^2+2\pi z^2/t}\sqrt{it}\theta [{}_{1/2\alpha \beta }{}^{\alpha +2\beta 2}](z|{\displaystyle \frac{1}{2}}+{\displaystyle \frac{it}{2}})`$
$`\eta ({\displaystyle \frac{1}{2}}+{\displaystyle \frac{i}{2t}})`$ $`=`$ $`e^{i\pi /4}\sqrt{it}\eta ({\displaystyle \frac{1}{2}}+{\displaystyle \frac{it}{2}})`$ (12)
|
warning/0506/hep-ph0506158.html
|
ar5iv
|
text
|
# References
SINP/TNP/05-06
TIFR/TH/05-17
hep-ph/0506158
NLO-QCD Corrections to Dilepton Production in the Randall-Sundrum Model
Prakash Mathews<sup>a</sup> <sup>*</sup><sup>*</sup>*prakash.mathews@saha.ac.in, V. Ravindran<sup>b</sup> ravindra@mri.ernet.in, K. Sridhar<sup>c</sup> sridhar@theory.tifr.res.in
a) Theory Group, Saha Institute of Nuclear Physics, 1/AF Bidhan Nagar,
Kolkata 700 064, India.
b) Harish-Chandra Research Institute, Chhatnag Road, Jhunsi, Allahabad, India.
c) Department of Theoretical Physics, Tata Institute of Fundamental Research,
Homi Bhabha Road, Mumbai 400 005, India.
ABSTRACT
The dilepton production process at hadron colliders in the Randall-Sundrum (RS) model is studied at next-to-leading order in QCD. The NLO-QCD corrections have been computed for the virtual graviton exchange process in the RS model, in addition to the usual $`\gamma ,Z`$-mediated processes of standard Drell-Yan. $`K`$-factors for the cross-sections at the LHC and Tevatron for differential in the invariant mass, $`Q`$, and the rapidity, $`Y`$, of the lepton pair are presented. We find the $`K`$-factors are large over substantial regions of the phase space.
In brane-world models, the four dimensional universe is a dynamical hypersurface: a $`D_3`$-brane (or 3-brane) existing in a higher dimensional spacetime. In many such models, the Standard Model (SM) fields are localized on the brane and only gravity can propagate in the bulk. The scale of quantum gravity can be lowered down from the Planck scale to the TeV scale in these models making it exciting for high-energy physics not only because these suggest fresh perspectives to the solution of the hierarchy problem but also because these models throw open the possibility of the discovery of new physics at energies accessible to collider experiments. In addition, these models provide new frameworks for gauge symmetry and supersymmetry breaking and suggest theoretical approaches to the cosmological constant problem and dark-matter problem.
The simplest model seeking to address the gauge hierarchy problem was the the ADD model proposed by Arkani-Hamed, Dimopoulos and Dvali , where, starting from a higher dimensional theory, an effective four-dimensional theory at a scale $`M_S\mathrm{TeV}`$ is obtained. This is done by compactifying the extra dimensions to magnitudes which are large compared to the Planck length .
The main problem that one faces within the ADD model is the reappearance of disparate scales $`viz.`$, the string scale and the inverse of the compactification radius. It was an attempt to avoid this problem that led to the formulation of the Randall-Sundrum (RS) model . In the RS model the single extra dimension $`\varphi `$ is compactified on a $`๐^1/๐^2`$ orbifold with a radius $`R_c`$ which is somewhat larger than the Planck length. Two 3-branes, the Planck brane and the TeV brane, are located at the orbifold fixed points $`\varphi =0,\pi `$, with the SM fields localised on the TeV brane. The five-dimensional metric, which is non-factorisable or $`warped`$ is of the form
$$ds^2=e^{๐ฆR_c\varphi }\eta _{\mu \nu }dx^\mu dx^\nu +R_c^2d\varphi ^2.$$
(1)
The exponential warp factor $`e^{๐ฆR_c\varphi }`$ serves as a conformal factor for fields localised on the brane. Thus the huge ratio $`\frac{M_P}{M_{EW}}10^{15}`$ can be generated by the exponent $`๐ฆR_c`$ which needs to be only of $`๐ช(10)`$ thereby providing a way of avoiding the hierarchy problem. There remains the problems of stabilising $`R_c`$ against quantum fluctuations but this can be done by introducing an extra scalar field in the bulk .
The tower of massive Kaluza-Klein (KK) excitations of the graviton, $`h_{\mu \nu }^{(\stackrel{}{n})}`$, interact with the SM particles by:
$`_{int}`$ $``$ $`{\displaystyle \frac{1}{M_P}}T^{\mu \nu }(x)h_{\mu \nu }^{(0)}(x){\displaystyle \frac{e^{\pi ๐ฆR_c}}{M_P}}{\displaystyle \underset{1}{\overset{\mathrm{}}{}}}T^{\mu \nu }(x)h_{\mu \nu }^{(n)}(x).`$ (2)
$`T^{\mu \nu }`$ is the symmetric energy-momentum tensor for the particles on the 3-brane. The masses of the $`h_{\mu \nu }^{(\stackrel{}{n})}`$ are given by
$`M_n`$ $`=`$ $`x_n๐ฆe^{\pi ๐ฆR_c},`$ (3)
where the $`x_n`$ are the zeros of the Bessel function $`J_1(x)`$. The zero-mode couples weakly and decouples but the couplings of the massive RS gravitons are enhanced by the exponential $`e^{\pi ๐ฆR_c}`$ leading to interactions of electroweak strength. Consequently, except for the overall warp factor in the RS case, the Feynman rules in the RS model are the same as those for the ADD case .
The basic parameters of the RS model are
$`m_0`$ $`=`$ $`๐ฆe^{\pi ๐ฆR_c},`$
$`c_0`$ $`=`$ $`๐ฆ/M_P,`$ (4)
where $`m_0`$ is a scale of the dimension of mass and sets the scale for the masses of the KK excitations, and $`c_0`$ is an effective coupling. The interaction of massive KK gravitons with the SM fields can be written as
$$_{int}\frac{c_0}{m_0}\underset{n}{\overset{\mathrm{}}{}}T^{\mu \nu }(x)h_{\mu \nu }^{(n)}(x).$$
(5)
Since $`๐ฆ`$ is related to the curvature of the fifth dimension we need to restrict it to small enough values to avoid effects of strong curvature. On the other hand $`๐ฆ`$ should not be too small compared to $`M_P`$ because that would reintroduce a hierarchy. These considerations suggest $`0.01c_00.1`$. For our analysis we choose to work with the RS parameters $`c_0`$ and $`M_1`$ the first excited mode of the graviton rather then $`m_0`$.
The decoupling of the graviton zero-mode and the existence of a mass gap in the spectrum of KK gravitons imply that it is only the resonant production and decay of the heavier KK modes or the virtual effects of the KK modes that one can hope to detect in collider experiments. The phenomenology of resonant production of the KK excitations and the virtual effects have already been studied in processes like dilepton production , diphoton production , $`t\overline{t}`$ production at hadron colliders , $`\tau `$-production at a linear collider and pair production of KK modes in $`e^+e^{}`$ and hadron hadron colliders . The sensitivity of the CMS experiment to the resonant production of RS graviton KK modes has been studied for electron pair production . Recently Dร has reported the first direct search for RS graviton KK modes using dielectron, dimuon, and diphoton events .
In an earlier work we had presented NLO-QCD corrections for $`e^+e^{}`$ hadrons and dilepton pair production at hadron colliders in the ADD model. These results for the dilepton pair production case are extended to the RS model, in this paper. We note that it is the same virtual graviton exchange process that contributes to dilepton production in both the ADD and RS models. The leading order process being the same, the QCD corrections are also not model-dependent. However, as explained above, the differences between the two models arise because of the difference in the summation over the tower of KK gravitons and also in the overall factors. Consequently, the relative weight of the subprocess cross-section due to graviton exchange vis-a-vis the SM subprocess will be different in the two models. This results in different $`K`$-factors in the ADD and RS models and the dependence of the $`K`$-factors on the kinematic variables are also different. In this letter, we present the results for dilepton production at the LHC and Tevatron in the RS model.
The process we are interested in is where two hadrons $`P_1,P_2`$ scatter and give rise to leptonic final states, say $`\mu ^+,\mu ^{}`$
$`P_1(p_1)+P_2(p_2)\mu ^+(l_1)+\mu ^{}(l_2)+X(P_X),`$ (6)
where $`p_1,p_2`$ are the momenta of incoming hadrons $`P_1`$ and $`P_2`$ respectively and $`\mu ^{},\mu ^+`$ are the outgoing leptons which have the momenta $`l_1,l_2`$. The final inclusive hadronic state is denoted by $`X`$ and carries the momentum $`P_X`$. The hadronic cross section can be expressed in terms of partonic cross sections convoluted with appropriate parton distribution functions as follows
$`2S{\displaystyle \frac{d\sigma ^{P_1P_2}}{dQ^2}}(\tau ,Q^2)`$ $`=`$ $`{\displaystyle \underset{ab=q,\overline{q},g}{}}{\displaystyle _0^1}๐x_1{\displaystyle _0^1}๐x_2f_a^{P_1}(x_1)f_b^{P_2}(x_2)`$ (7)
$`\times {\displaystyle _0^1}dz\mathrm{\hspace{0.17em}\hspace{0.17em}2}\widehat{s}{\displaystyle \frac{d\widehat{\sigma }^{ab}}{dQ^2}}(z,Q^2)\delta (\tau zx_1x_2).`$
The scaling variables are defined by $`k_1=x_1p_1,k_2=x_2p_2`$ where $`k_1,k_2`$ are the momenta of incoming partons.
$`(p_1+p_2)^2`$ $``$ $`S,(k_1+k_2)^2\widehat{s},(l_1+l_2)^2=q.qQ^2,`$
$`\tau `$ $`=`$ $`{\displaystyle \frac{Q^2}{S}},z={\displaystyle \frac{Q^2}{\widehat{s}}},\tau =x_1x_2z.`$ (8)
The partonic cross section for the process $`a(k_1)+b(k_2)j(q)+{\displaystyle \underset{i}{\overset{m}{}}}X_i(p_i)`$ is given by
$`2\widehat{s}{\displaystyle \frac{d\widehat{\sigma }^{ab}}{dQ^2}}`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{jj^{}=\gamma ,Z,G}{}}{\displaystyle ๐PS_{m+1}|M^{abjj^{}}|^2P_j(q)P_j^{}^{}(q)^{jj^{}l^+l^{}}(q)}.`$ (9)
In the above equation, the sum over Lorentz indices between matrix element squared and the propagators is implicit through a symbol โdot productโ. The $`m+1`$ body phase space is defined as
$`{\displaystyle ๐PS_{m+1}}`$ $`=`$ $`{\displaystyle \underset{i}{\overset{m}{}}\left(\frac{d^np_i}{(2\pi )^n}2\pi \delta ^+(p_i^2)\right)\frac{d^nq}{(2\pi )^n}2\pi \delta ^+(q^2Q^2)}`$ (10)
$`\times (2\pi )^n\delta ^{(n)}(k_1+k_2+q+{\displaystyle \underset{i}{\overset{m}{}}}p_i),`$
where $`n`$ is the space-time dimension. The propagators are
$`P_\gamma (q)`$ $`=`$ $`{\displaystyle \frac{i}{Q^2}}g_{\mu \nu },`$ (11)
$`P_Z(q)`$ $`=`$ $`{\displaystyle \frac{i}{(Q^2M_Z^2iM_Z\mathrm{\Gamma }_Z)}}g_{\mu \nu },`$ (12)
$`P_G(q)`$ $`=`$ $`๐(Q^2)B_{\mu \nu \lambda \rho }(q),`$ (13)
where
$`B_{\mu \nu \rho \sigma }(q)`$ $`=`$ $`\left(g_{\mu \rho }{\displaystyle \frac{q_\mu q_\rho }{M_{n}^{}{}_{}{}^{2}}}\right)\left(g_{\nu \sigma }{\displaystyle \frac{q_\nu q_\sigma }{M_{n}^{}{}_{}{}^{2}}}\right)+\left(g_{\mu \sigma }{\displaystyle \frac{q_\mu q_\sigma }{M_{n}^{}{}_{}{}^{2}}}\right)\left(g_{\nu \rho }{\displaystyle \frac{q_\nu q_\rho }{M_{n}^{}{}_{}{}^{2}}}\right)`$ (14)
$`{\displaystyle \frac{2}{n1}}\left(g_{\mu \nu }{\displaystyle \frac{q_\mu q_\nu }{M_{n}^{}{}_{}{}^{2}}}\right)\left(g_{\rho \sigma }{\displaystyle \frac{q_\rho q_\sigma }{M_{n}^{}{}_{}{}^{2}}}\right).`$
The function $`๐(Q^2)`$ in the graviton propagator Eq. (13), results from summing over the KK modes, given by
$`๐(Q^2)`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{Q^2M_n^2+iM_n\mathrm{\Gamma }_n}}{\displaystyle \frac{\lambda }{m_0^2}},`$ (15)
where $`M_n`$ are the masses of the individual resonances and the $`\mathrm{\Gamma }_n`$ are the corresponding widths. The graviton widths are obtained by calculating their decays into final states involving SM particles. $`\lambda `$ is defined as
$`\lambda (x_s)`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{x_s^2x_n^2i\frac{\mathrm{\Gamma }_n}{m_0}x_n}{x_s^2x_n^2+\frac{\mathrm{\Gamma }_n}{m_0}x_n}},`$ (16)
where $`x_s=Q/m_0`$. We have to sum over all the resonances to get the value of $`\lambda (x_s)`$. This is done numerically and for a given value of $`x_s`$, we retain all resonances which contribute with a significance greater than one per mil, and treat the remaining KK modes as virtual particles (in which case the sum can be done analytically).
We now present the distributions in the invariant lepton pair mass, $`Q`$, and the rapidity of the lepton pair, $`Y`$ at the LHC ($`\sqrt{S}=14`$ TeV) and Tevatron ($`\sqrt{S}=1.96`$ TeV). From these distributions the effects of the NLO-QCD corrections can be clearly discerned. For the parton density sets we adopt in leading order (LO) MRST 2001 LO ($`\mathrm{\Lambda }=0.1670\mathrm{GeV}`$) and in next-to-leading order the MRST 2001 NLO ($`\mathrm{\Lambda }=0.2390`$ GeV). For LHC we choose the kinematic ranges $`300\mathrm{GeV}<Q<3000\mathrm{GeV}`$ and $`|Y|<2.2`$ at $`Q=1.5`$ TeV. For Tevatron $`300\mathrm{GeV}<Q<1000`$ GeV and $`|Y|<0.9`$ at $`Q=300`$ GeV. The renormalisation scale is taken to be same as the factorization scale $`\mu _F`$ and $`\mu _F`$ si chosen to be $`\mu _F=Q`$.
The cross-section $`d\sigma /dQ`$ as a function of $`Q`$ to NLO is presented in Fig. 1a for LHC. For the figure, we have chosen the representative values of the RS model parameters: $`M_1=1.5`$ TeV the first RS resonance mass and the coupling constant $`c_0=0.01`$. The width of the resonance is related to $`c_0`$ and hence a smaller $`c_0`$ corresponds to a narrow resonance. The subsequent resonance are determined by $`m_0`$ and $`x_n`$. To LO the dilepton case has been presented in . To see the effect of the NLO effect we study the $`K`$-factor for the $`Q`$ and $`Y`$ distribution.
The $`K`$-factor for the invariant lepton pair mass distribution defined by
$`K^I=\left[{\displaystyle \frac{d\sigma _{LO}^I(Q)}{dQ}}\right]^1\left[{\displaystyle \frac{d\sigma _{NLO}^I(Q)}{dQ}}\right],`$ (17)
where $`I=SM`$, $`I=SM+GR`$ for both SM and gravity combined and $`I=GR`$ for only gravity. It is possible to define $`K^{GR}`$ for the invariant lepton pair mass distribution, as there is no interference with SM . The results are presented in Fig. 1b. The parameters chosen are the same as in Fig. 1a. In order to understand the behaviour of $`K`$-factor of the model involving both SM and gravity, it is useful to express it as
$`K^{(SM+GR)}(Q)={\displaystyle \frac{K^{SM}+K^{GR}K^{(0)}}{1+K^{(0)}}},`$ (18)
where we have introduced a quantity $`K^{(0)}`$, defined as the ratio of the LO distribution of gravity to SM, given by
$`K^{(0)}(Q)=\left[{\displaystyle \frac{d\sigma _{LO}^{SM}(Q)}{dQ}}\right]^1\left[{\displaystyle \frac{d\sigma _{LO}^{GR}(Q)}{dQ}}\right].`$ (19)
The behaviour of $`K^{(0)}(Q)`$ is governed by competing couplings constants of SM and gravity and the parton fluxes. In the RS case the gravity contribution is significant in the resonance region, (see Fig.1a). In the off resonance region the $`K`$-factor is hence purely $`K^{SM}`$. In the resonance region where the gravity effect dominates the $`K^{(SM+GR)}`$ factor shifts to the $`K^{GR}`$ value (see Fig. 1b). This behaviour of the $`K`$-factor of the RS case is very distinct from the corresponding case we presented in the ADD case . To incorporate the NLO effects for an appropriate distribution one needs to take into account the behaviour of the $`K`$-factor accordingly. For $`M_1=300`$ GeV the $`K`$-factor is about $`1.5`$ in the resonance region. This is due to the fact that at loq $`Q`$ ($`Q=300`$ GeV) the gluon flux becomes dominant at Tevatron. The behaviour of $`K^{GR}`$ is the same as in the ADD case .
In Fig. 1c, we have plotted the scale variations of the $`Q`$ distribution for both LO and NLO cross sections. We define $`R^I`$ for the invariant lepton mass distribution as
$`R_{LO}^I`$ $`=`$ $`\left[{\displaystyle \frac{d\sigma _{LO}^I(Q,\mu =\mu _0)}{dQ}}\right]^1\left[{\displaystyle \frac{d\sigma _{LO}^I(Q,\mu )}{dQ}}\right],`$
$`R_{NLO}^I`$ $`=`$ $`\left[{\displaystyle \frac{d\sigma _{NLO}^I(Q,\mu =\mu _0)}{dQ}}\right]^1\left[{\displaystyle \frac{d\sigma _{NLO}^I(Q,\mu )}{dQ}}\right],`$ (20)
where $`\mu _0`$ is a fixed scale which is chosen to be $`\mu _0=1.5`$ TeV for LHC. As can be seen from the figure, the inclusion of the NLO corrections stabilises the cross-section with respect to the scale $`\mu `$. Here we have chosen $`\mu _0=1.5`$ TeV, ie. the first resonance region. The scale variation is driven by the gravity part as its the dominant contribution.
In Fig. 2a the double differential cross section $`d^2\sigma /dQdY`$ is displayed for rapidity region $`|Y|2.2`$ for a $`Q`$ value of 1.5 TeV. To plot this distribution the $`Q`$ value is chosen such that it lies at the first resonance, where the gravity effect dominates. Hence the dominant contribution is purely gravity. The RS model parameters remain the same as before. The $`K`$ factor as a function of $`Y`$ is plotted for a choice of $`Q`$ in the resonance region where the dominant contribution to $`K^{(SM+GR)}`$ factor comes from the gravity part Fig. 2b. The $`R`$ ratio using the $`Y`$ distribution is plotted in Fig. 2c for the central region of rapidity and for a $`Q`$ value of 1.5 TeV. In this region the scale variation is also dictated by the gravity contribution.
The corresponding analysis for the Tevatron is done for the $`Q`$ range $`300<Q<1000`$ GeV and for the RS parameter $`M_1=300`$ GeV and the coupling $`c_0=0.01`$. At low $`Q`$ the gravity effects of the RS model is dominant in the resonance and off the resonance region the effect is negligible. As $`Q`$ increases the effect of gravity starts to become comparable to the SM contribution as is seen towards the third resonance in Fig. 3a. In Fig. 3b we have plotted the $`K`$-factor for $`Q`$ distribution at the Tevatron. Using Eq. (18) we can understand the behaviour of $`K^{SM+GR}`$. As expected the behaviour of $`K^{GR}`$ is same as the ADD case . The double differential cross section $`d^2\sigma /dQdY`$ for $`Q=300`$ GeV is plotted as a function of rapidity $`Y`$. In the resonance region the dominant contribution is from the RS. In contrast for ADD only at large $`Q`$ the gravity effects became comparable to the SM at Tevatron. The corresponding $`K`$-factor is plotted in Fig. 4b. Scale variation for the $`Q`$ and double differential $`dQdY`$ is given in Fig. 3c and Fig. 4c respectively.
In a recent analysis by Dร , the LO cross section was scaled by a constant $`K`$-factor of 1.34 to account for the NLO effect for the RS case. This does not yield a realistic picture as can be seen from Fig. 3b. In the RS case due to the resonant production, the $`K`$-factor is very different from the ADD case reported earlier .
In summary, we have presented the results for the cross-section for dilepton production in the Randall-Sundrum model at the LHC and Tevatron. The large incident gluon flux at the LHC makes the NLO QCD corrections very important. Moreover, when the NLO corrections are taken into account the cross-sections are stabilised with respect to scale variations. In order to derive robust bounds on the RS model at the LHC using the dilepton production process, the inclusion of the NLO QCD corrections in the cross-section is crucial.
Acknowledgments: PM thanks S. Moretti for useful discussion. VR would like to thank Prof. W. L. van Neerven for discussion. The work of PM and KS is part of a project (IFCPAR Project No. 2904-2) on โBrane-World Phenomenologyโ supported by the Indo-French Centre for the Promotion of Advanced Research, New Delhi, India. PM would also like to thank IPPP, Durham for warm hospitality where part of this work was done. PM and VR thank S. Raychaudhuri for providing the code that evaluates the RS KK mode sum in the propagator.
Figure Caption
Figure 1. (a) The cross section is plotted as a function of invariant mass $`Q`$ of the lepton pair for $`M_1=1.5`$ TeV at LHC. (b) The corresponding $`K`$-factor for $`Q`$ distribution SM, gravity and SM plus gravity. (c) Scale variation of the cross section at LO and NLO as defined in Eq. (19) for $`Q=1.5`$ TeV.
Figure 2. (a) The double differential cross section $`d^2\sigma /dQdY`$ is plotted as a function of rapidity $`Y`$ for $`Q=1.5`$ TeV at LHC. (b) The K-factor for the distribution in (a) is plotted for the rapidity range. (c) The scale variation of the ratio R is plotted as a function of $`\mu /\mu _0`$ for $`Y=0`$.
Figure 3. (a) The cross section is plotted as a function of invariant mass $`Q`$ of the lepton pair for $`M_1=300`$ GeV and $`c_0=0.01`$ at the Tevatron. (b) The $`K`$-factor for $`Q`$ distribution for the same RS parameters in (a) is plotted. (c) The variation of the cross section with respect to the scale.
Figure 4. (a) The double differential cross section $`d^2\sigma /dQdY`$ is plotted as a function of rapidity $`Y`$ for $`Q=300`$ GeV at the Tevatron for the RS parameters $`M_1=300`$ GeV and $`c_0=0.01`$. (b) The $`K`$-factor for the distribution in (a) is plotted for the rapidity range. (c) The scale variation of the ratio R is plotted as a function of $`\mu /\mu _0`$ for the central rapidity region $`Y=0`$.
|
warning/0506/math0506084.html
|
ar5iv
|
text
|
# Chern Classes of the Moduli Stack of Curves
## 1. Introduction
Let $`g`$ and $`n`$ be non-negative integers such that $`n>22g`$. We denote by $`\overline{}_{g,n}`$ the Deligne-Mumford stack of stable genus $`g`$ curves with $`n`$ marked points. More generally, if $`P`$ is a set with $`n`$ elements, it will be technically convenient to work with $`\overline{}_{g,P}`$, i.e., the stack of genus $`g`$ stable curves whose marked points are labelled by $`P`$. The natural projection $`\phi `$ from the stack $`\overline{}_{g,P}`$ to the coarse moduli space $`\overline{M}_{g,P}`$ induces an isomorphism $`\phi _{}`$ at the level of Chow rings - hereafter we shall only deal with rational coefficients. Following , we denote by $`R^{}(\overline{M}_{g,P})`$ the tautological ring of $`\overline{M}_{g,P}`$. By abuse of notation, we shall denote the image of $`R^{}(\overline{M}_{g,P})`$ under $`(\phi _{})^1`$ by the same symbol.
Tautological classes have been intensely studied in the last few years: see for a synoptic survey on the most recent developments in this area. In particular, it is not at all clear which classes lie in the tautological ring. In fact, constructions of tautological or non-tautological classes can be very diverse in nature: combinatorial - as conjectured by Kontsevich and proved in, e.g., \- or purely algebro-geometric (e.g., ). As shown by Mumford in , the Grothendieck-Riemman-Roch Theorem allows one to express Chow classes in terms of tautological ones. Incidentally, this is done for the canonical class of $`\overline{}_{g,P}`$ in . That calculation can be rephrased in terms of stacks. For foundational material on stacks we refer the reader to and, especially for Chern classes, to .
In the present paper, we extend Mumfordโs work and calculate all Chern classes of the (smooth) moduli stack of curves, i.e., of the tangent bundle $`๐ฏ_{\overline{}_{g,P}}`$ to $`\overline{}_{g,P}`$. In spite of their geometric significance, these classes have not been hitherto computed. As above, we apply the Grothendieck-Riemann-Roch Theorem. In fact, we refine combinatorial arguments, and we manage to get explicit formulas. This shows in particular that such classes are tautological, thus yielding new elements in $`R^{}(\overline{M}_{g,P})`$.
Our presentation is rather concise since most of the theoretic material has been explored by several authors. We briefly recall the needed preliminaries in Section 2 and prove the main result in Section 3. Finally, we give some examples and compare our formulas with previous results.
Throughout, we shall work over the field of complex numbers.
## 2. Preliminaries
Let $`\overline{}_{g,P}`$ be the moduli stack of $`P`$-pointed stable curves of genus $`g`$. As usual, $`\pi :๐\overline{}_{g,P}`$ will denote the universal curve with sections $`\sigma _p`$ for $`pP`$. Set, further, $`D:=_{pP}\sigma _p_{}(1)`$. The universal cotangent classes on $`\overline{}_{g,P}`$ are defined as $`\psi _p=c_1(\sigma _p^{}(\omega _\pi ))`$ for $`pP`$, where $`\omega _\pi `$ is the relative dualizing sheaf of $`\pi `$. The collection of all moduli stacks $`\overline{}_{g,P}`$ is equipped with some natural morphisms, namely:
(1)
$$\xi _G:\underset{vV}{}\overline{}_{g(v),l(v)}\overline{}_{g,P},$$
where $`G`$ is a stable graph - see, e.g., . For example, it is well known that the boundary $`\overline{}_{g,P}`$ can be described in terms of the following morphisms:
(2)
$$\xi _{irr}:\overline{}_{g1,P\{q_1,q_2\}}\overline{}_{g,P},$$
(3)
$$\xi _{h,A}:\overline{}_{h,A\{r_1\}}\times \overline{}_{gh,A^c\{r_2\}}\overline{}_{g,P},$$
where $`0hg`$, $`AP`$, and both $`2h1+|A|`$ and $`2(gh)1+|A^c|`$ are positive. Finally, let $`\delta `$ be the boundary class defined as
(4)
$$\delta =\frac{1}{2}\xi _{irr}(1)+\frac{1}{2}\underset{h}{}\underset{AP}{}\xi _{h,A}(1).$$
Let $`K=c_1\left(\omega _\pi (D)\right)`$. Following , the Mumford classes on $`\overline{}_{g,P}`$ are defined as $`\kappa _m=\pi _{}(K^{m+1})`$. For $`P=\mathrm{}`$ their analogue was first introduced by Mumford in . Another generalization of Mumfordโs $`\kappa _m`$โs to the case of $`P`$-pointed curves is given by the classes $`\stackrel{~}{\kappa }_m=\pi _{}(c_1(\omega _\pi )^{m+1})`$. As proved in , the following relationship holds:
(5)
$$\kappa _m=\stackrel{~}{\kappa }_m+\underset{pP}{}\psi _p^m.$$
The Hodge bundle $`๐ผ`$ on $`\overline{}_{g,P}`$ is defined as $`\pi _{}\omega _\pi `$. From and we have
$$ch(๐ผ)=g+\frac{1}{2}\underset{m1}{}\frac{B_{2m}}{(2m)!}\{\stackrel{~}{\kappa }_{2m1}+$$
$$\xi _{irr}(\psi _{q_1}^{2m2}\psi _{q_1}^{2m3}\psi _{q_2}+\mathrm{}+\psi _{q_2}^{2m2})+$$
$$\underset{h=0}{\overset{g}{}}\underset{AP}{}\xi _{G_{h,A}}(\psi _{r_1}^{2m2}1\psi _{r_1}^{2m3}\psi _{r_2}+\mathrm{}+1\psi _{r_2}^{2m2})\},$$
where $`B_{2m}`$ are the Bernoulli numbers<sup>1</sup><sup>1</sup>1We recall that Bernoulli numbers are defined as follows:
$$\frac{x}{e^x1}=1\frac{1}{2}x+\underset{m1}{}\frac{B_{2m}}{(2m)!}x^{2m}.$$
. Note in particular that
(6)
$$ch_1(๐ผ)=c_1(๐ผ):=\lambda =\frac{1}{12}\left(\kappa _1\underset{pP}{}\psi _p\right)+\delta .$$
As first shown by Mumford in , the Grothendieck-Riemann-Roch Theorem (the G-R-R Theorem for short) can be applied to the universal curve $`\pi :๐\overline{}_{g,P}`$. Alternatively, one can use the G-R-R Theorem stated in . For the sake of completeness, we report the statement below.
###### Theorem 1.
(G-R-R Theorem) Let $``$ be a locally free sheaf on $`๐`$. Then
$$ch(\pi _!)=\pi _{}\left(ch()Td^{}(\mathrm{\Omega }_\pi )\right),$$
where $`\mathrm{\Omega }_\pi `$ is the sheaf of relative Kรคhler differentials.
In Proposition 1, $`ch()`$ and $`Td()`$ denote the Chern character and the dual Todd class of $``$, respectively. Some formulas for these classes can be found, for instance, in . Here, we just remark two basic facts. First, notice that
(7)
$$ch(^{})=rk()+\underset{j1}{}(1)^jch_j().$$
Second, let $`\mu =(1^{m_1}2^{m_2}\mathrm{}i^{m_i}\mathrm{})`$ be a partition of weight $`j`$, where $`j`$ is a positive integer. Define $`ch_\mu ()`$ to be the product $`ch_1^{m_1}()ch_2^{m_2}()\mathrm{}ch_i^{m_i}()\mathrm{}`$ As proved in , (2.14โ), the following holds:
(8)
$$c_j()=\underset{\mu j}{}(1)^{jl(\mu )}\underset{r1}{}\frac{((r1)!)^{m_r}}{m_r!}ch_\mu (),j1,$$
where the sum ranges over all partitions $`\mu `$ of $`j`$, and $`l(\mu )`$ is the length of $`\mu `$.
Finally, we recall some properties of $`Z`$, the singular locus of $`\pi `$. For more details the reader is referred to . $`Z`$ is a closed substack of codimension $`2`$ in $`๐`$. Moreover, there exists a double รจtale covering $`\epsilon :\stackrel{~}{Z}Z`$ obtained from the choice of the branches incident at the nodes corresponding to points in $`Z`$. Let $`\iota :\stackrel{~}{Z}๐`$ be the natural composition. Denote by $``$ and $`^{}`$ the line bundles corresponding to the cotangent directions along the branches. Then $`\epsilon ^{}(๐ฉ_Z)=^{}`$, where $`๐ฉ_Z`$ is the normal bundle of $`Z`$ in $`๐`$. Moreover, note that $`\pi \iota `$ maps $`\stackrel{~}{Z}`$ onto $`\overline{}_{g,P}`$. In other words, we have
(9)
$$\pi \iota =\xi _{irr}+\underset{h}{}\underset{\begin{array}{c}AP\end{array}}{}\xi _{h,A},$$
where both $`2h1+|A|`$ and $`2(gh)1+|A^c|`$ are positive.
## 3. The Chern Character of $`\overline{}_{g,P}`$
In this section we apply the G-R-R Theorem to the sheaf $`\mathrm{\Omega }_\pi (D)\omega _\pi `$. By standard duality theorems and deformation theory , $`\pi _{}\left(\mathrm{\Omega }_\pi (D)\omega _\pi \right)`$ is the cotangent bundle on $`\overline{}_{g,P}`$. In the sequel, we closely follow , p. 302 ff.
We recall that
(10)
$$ch\left(\mathrm{\Omega }_\pi (D)\right)=ch\left(\omega _\pi (D)\right)ch\left(๐ช_Z\right),$$
and
(11)
$$Td^{}(\mathrm{\Omega }_\pi )=Td^{}(\omega _\pi )\left(Td^{}(๐ช_Z)\right)^1,$$
where $`๐ช_Z`$ is viewed as a sheaf on $`๐`$. For the purpose of what follows, we need to determine the power series $`ch(๐ช_Z)Td^{}(๐ช_Z)^1`$. Since $`Td^{}(๐ช_Z)^1`$ is a polynomial in the Chern characters $`ch_k(๐ช_Z)`$, the G-R-R Theorem applied to $`\iota `$ yields a universal power series $`\mathrm{\Theta }`$. For all $`\nu :YX`$, an inclusion of a smooth codimension two subvariety in a smooth variety, the series $`\mathrm{\Theta }`$ satisfies the following identity:
(12)
$$ch(๐ช_Y)Td^{}(๐ช_Y)^1=\nu _{}\left(\mathrm{\Theta }(c_1\left(๐ฉ_Y\right),c_2\left(๐ฉ_Y\right))\right).$$
To compute $`\mathrm{\Theta }`$, take $`Y=D_1D_2`$. Then the exact sequence
$$0๐ช_X(D_1D_2)๐ช_X(D_1)๐ช_X(D_2)๐ช_X๐ช_Y0$$
yields
(13)
$$ch(๐ช_Y)=(1e^{D_1})(1e^{D_2}).$$
By , p. 303, we have
(14)
$$Td^{}(๐ช_Y)^1=\frac{D_1D_2}{D_1+D_2}\frac{1e^{D_1D_2}}{(1e^{D_1})(1e^{D_2})}.$$
Therefore, we get
(15)
$$ch(๐ช_Y)Td^{}(๐ช_Y)^1=D_1D_2\underset{j1}{}(1)^{j1}\frac{(D_1+D_2)^{j1}}{j!}.$$
Thus, the following holds:
(16)
$$\mathrm{\Theta }(D_1+D_2,D_1D_2)=\underset{j1}{}(1)^{j1}\frac{(D_1+D_2)^{j1}}{j!}.$$
As noted in , (13) can be applied to the morphism $`\iota :\stackrel{~}{Z}๐`$ as well. In this case, we need to replace the Chern classes of $`๐ฉ_Z`$ with those of $`\epsilon ^{}๐ฉ_Z`$.
We now state the main result of this section.
###### Theorem 2.
The Chern character of the cotangent bundle on $`\overline{}_{g,P}`$ is given by
$`ch\left(๐ฏ_{\overline{}_{g,P}}^{}\right)`$ $`=`$ $`{\displaystyle \underset{j1}{}}{\displaystyle \frac{\kappa _{j1}}{j!}}+{\displaystyle \frac{1}{2}}{\displaystyle \underset{t1}{}}{\displaystyle \frac{\kappa _t}{t!}}{\displaystyle \underset{m3}{}}a_m\kappa _{m1}+`$
$`ch(๐ผ)1{\displaystyle \frac{1}{2}}\xi _{irr}\left(\mathrm{\Xi }^{(1)}\right){\displaystyle \frac{1}{2}}{\displaystyle \underset{h=0}{\overset{g}{}}}{\displaystyle \underset{AP}{}}\xi _{G_{h,A}}\left(\mathrm{\Xi }^{(2)}\right),`$
where
(18)
$$a_m=\underset{h=1}{\overset{\frac{m1}{2}}{}}\frac{B_{2h}}{(2h)!(m2h)!},$$
(19)
$$\mathrm{\Xi }^{(1)}=\underset{k1}{}(1)^{k1}\frac{\left(\psi _{q_1}+\psi _{q_2}\right)^{k1}}{k!}$$
and
(20)
$$\mathrm{\Xi }^{(2)}=\underset{k1}{}(1)^{k1}\frac{\left(\psi _{r_1}1+1\psi _{r_2}\right)^{k1}}{k!}$$
Proof. By (10) and (11), we have
$`ch\left(\left(\mathrm{\Omega }_\pi (D)\omega _\pi \right)Td^{}(\mathrm{\Omega }_\pi )\right)`$ $`=`$ $`ch(\omega _\pi )Td^{}(\omega _\pi )Td^{}(๐ช_Z)^1ch(\omega _\pi (D))`$
$``$ $`ch(\omega _\pi )Td^{}(\omega _\pi )Td^{}(๐ช_Z)^1ch(๐ช_Z).`$
As proved in , the first Chern class of $`\omega _\pi (D)`$ is equal to $`\psi _q`$. Here $`\psi _q`$ denotes the universal cotangent class on $`\overline{}_{g,P\{q\}}`$ corresponding to the marked point $`q`$.
Since $`๐ช_Z`$ is supported on $`Z`$ and the marked points are non-singular, we get
$`ch(\omega _\pi )Td^{}(\omega _\pi )Td^{}(๐ช_Z)^1ch(\omega _\pi (D))`$ $`=`$ $`ch(\omega _\pi )Td^{}(\omega _\pi )ch(\omega _\pi (D))`$
$`+`$ $`Td^{}(๐ช_Z)^11,`$
and
(23)
$$ch(\omega _\pi )Td^{}(\omega _\pi )Td^{}(๐ช_Z)^1ch(๐ช_Z)=Td^{}(๐ช_Z)^1ch(๐ช_Z).$$
In particular, it is easy to check that (3) is equal to
(24)
$$ch(\omega _\pi )Td^{}(\omega _\pi )\left[ch(\omega _\pi (D))1\right]+ch(\omega _\pi )Td^{}(\mathrm{\Omega }_\pi ).$$
By definition, we have
(25)
$$\begin{array}{c}ch(\omega _\pi (D))=e^{\psi _q},\\ \\ ch(\omega _\pi )Td^{}(\omega _\pi )=e^{(\psi _qD)}\frac{\psi _qD}{e^{(\psi _qD)}1}=\frac{D\psi _q}{e^{D\psi _q}1}.\end{array}$$
We recall that the formal expansion of the second power series in (25) is given by
(26)
$$1+\frac{1}{2}(\psi _qD)+\underset{j1}{}\frac{B_{2j}}{(2j)!}(\psi _q+D)^{2j},$$
where $`B_{2j}`$ is the $`(2j)`$-th Bernoulli number. Since $`\psi _qD=0`$, the first term of (24) is given by
(27)
$$\underset{j1}{}\frac{\psi _q^j}{j!}+\frac{1}{2}\underset{t1}{}\frac{\psi _q^{t+1}}{t!}\underset{m3}{}a_m\psi _q^m,$$
where $`a_m`$ is defined in (18).
Let us apply $`\pi _{}`$ to (3). The contribution in (24) yields
(28)
$$\underset{j1}{}\frac{\kappa _{j1}}{j!}+\frac{1}{2}\underset{t1}{}\frac{\kappa _t}{t!}\underset{m3}{}a_m\kappa _{m1}+ch(๐ผ)1.$$
On the other hand, the contribution in (23) is given by
(29)
$$\frac{1}{2}(\pi \iota )_{}\left(\mathrm{\Theta }(c_1\left(\epsilon ^{}๐ฉ_Z\right),c_2\left(\epsilon ^{}๐ฉ_Z\right))\right).$$
By (9), this is equivalent to
$$\frac{1}{2}\xi _{irr}\left(\mathrm{\Xi }^{(1)}\right)+\frac{1}{2}\underset{h=0}{\overset{g}{}}\underset{AP}{}\xi _{G_{h,A}}\left(\mathrm{\Xi }^{(2)}\right),$$
where $`\mathrm{\Xi }^{(1)}`$ and $`\mathrm{\Xi }^{(2)}`$ are defined in (19) and (20), respectively.
$`\mathrm{}`$
###### Remark 3.
By Theorem 2 and (6), we get
$$ch_0\left(๐ฏ_{\overline{}_{g,P}}^{}\right)=rk(๐ฏ_{\overline{}_{g,P}}^{})=3g3+n,$$
$$ch_1\left(๐ฏ_{\overline{}_{g,P}}^{}\right)=K_{\overline{}_{g,P}}=13\lambda +\underset{pP}{}\psi _p2\delta ,$$
where $`K_{\overline{}_{g,P}}`$ is the canonical class of the stack $`\overline{}_{g,P}`$ and $`\delta `$ is defined in (4). These formulas agree with previous known results: see, e.g., , .
###### Remark 4.
Note that Theorem 2 and Formula (7) give the Chern character of $`๐ฏ_{\overline{}_{g,P}}`$.
By Formula (8), we obtain an expression for the Chern classes of $`\overline{}_{g,P}`$.
###### Corollary 5.
For $`j1`$, we have
(30)
$$c_j(\overline{}_{g,P})=\underset{\mu j}{}(1)^{jl(\mu )}\underset{r1}{}\frac{((r1)!)^{m_r}}{m_r!}ch_\mu (\overline{}_{g,P}).$$
$`\mathrm{}`$
Example. We give some examples for low $`j`$โs. In higher degrees, one can use John Stembridgeโs symmetric function package SF for maple . For the sake of simplicity, we denote $`_p\psi _p`$ by $`\psi `$. As noted in Remark 3, we have
$$c_1(\overline{}_{g,P})=13\lambda \psi +2\delta .$$
From Corollary 5, we get
$$ch_2(\overline{}_{g,P})=\frac{\kappa _2}{3}+\frac{1}{4}\xi _{irr}\left(\psi _{q_1}+\psi _{q_2}\right)+\frac{1}{4}\underset{h,A}{}\xi _{h,A}\left(\psi _{r_1}1+1\psi _{r_2}\right).$$
hence
$`c_2(\overline{}_{g,P})`$ $`=`$ $`{\displaystyle \frac{1}{2}}(13\lambda \psi +2\delta )^2{\displaystyle \frac{1}{3}}\kappa _2{\displaystyle \frac{1}{4}}\xi _{irr}\left(\psi _{q_1}+\psi _{q_2}\right)`$
$`{\displaystyle \frac{1}{4}}{\displaystyle \underset{h,A}{}}\xi _{h,A}\left(\psi _{r_1}1+1\psi _{r_2}\right).`$
Finally, the degree $`3`$ Chern character is equal to
$`ch_3(\overline{}_{g,P})`$ $`=`$ $`{\displaystyle \frac{\kappa _3}{12}}ch_3(๐ผ)+{\displaystyle \frac{1}{12}}\xi _{irr}\left(\psi _{q_1}^2+\psi _{q_1}\psi _{q_2}+\psi _{q_2}^2\right)+`$
$`{\displaystyle \frac{1}{12}}{\displaystyle \underset{h,A}{}}\xi _{h,A}\left(\psi _{r_1}^21+\left(\psi _{r_1}\psi _{r_2}\right)+1\psi _{r_2}^2\right).`$
Thus, we get
$`c_3(\overline{}_{g,P})`$ $`=`$ $`[+13\lambda +\psi 2\delta ][{\displaystyle \frac{1}{3}}\kappa _2+{\displaystyle \frac{1}{4}}\xi _{irr}(\psi _{q_1}+\psi _{q_2})`$
$`+{\displaystyle \frac{1}{4}}{\displaystyle \underset{h,A}{}}\xi _{h,A}(\psi _{r_1}1+1\psi _{r_2})]`$
$`{\displaystyle \frac{1}{6}}\left[13\lambda \psi +2\delta \right]^3+2ch_3(\overline{}_{g,P}).`$
As a result of Theorem 2 and the definition of tautological classes, we get new elements in the tautological ring of $`\overline{}_{g,P}`$. Precisely, the following holds.
###### Corollary 6.
The Chern classes of $`\overline{}_{g,P}`$ are tautological.
$`\mathrm{}`$
|
warning/0506/astro-ph0506328.html
|
ar5iv
|
text
|
# 1 Abstract
## 1 Abstract
We confirm a 23 day periodicity during a large flare in 1997 for X-ray data of X-ray satellite RXTE all sky monitor (ASM), two TeV gamma ray data from Utah Seven Telescope and HEGRA, with a Fourier analysis. We found the three results to be the same with a newly estimated error. We confirm the presence of a frequency dependent power (1/f noise) in a frequency-power diagram. Further, we calculated a chance probability of the occurrence of the 23 day periodicity by considering the 1/f noise and obtained a chance probability 4.88 $`\times 10^3`$ for the HEGRA data:this is more significant than the previous result by an order. We also obtained an indentical periodicity with another kind of timing analysisโepoch folding method for the ASM data and HEGRA data. We strongly suggest an existence of the periodicity. We divided the HEGRA data into two data sets, analyzed them with a Fourier method, and found an unstableness of the periodicity with a 3.4 sigma significance. We also analyzed an energy spectra of the X-ray data of a RXTE proportional counter array and we found that a combination of three physical parametersโ a magnetic field, a Lorentz factor, and a beaming factorโis related to the periodicity.
## 2 Introduction
Gamma ray emission from active galactic nuclei (AGN) has been measured by detectors in an orbit and by detectors on the ground. The EGRET detector on the gamma ray satellite CGRO was sensitive to GeV gamma rays and detected 90 Blazars (e.g., ). Air Cherenkov detectors on the ground are sensitive to TeV energies. They detected the ten BlazarsโMkn 421 (z = 0.030), Mkn 501 (z = 0.034), PKS 2155-304 (z = 0.116), 1ES1959+650 (z = 0.048), 1ES 2344+514 (z = 0.044), H1426+428(z = 0.129), PKS2005-489 (z = 0.071), H2356-309 (z = 0.165), 1ES1218+304 (z = 0.182) and 1ES1101-232 (z = 0.186). The energy spectra of Blazars has two components. One component extends from the radio to the X-ray band, while the other is in the gamma ray range. Low energy photons are interpreted to be a result of synchrotron emission by accelerated high energy electrons, and high energy photons are interpreted to be resulted of inverse Compton scattering of the synchrotron emission by high energy electrons. There are two types of Blazars, the BL Lac type and the QSO type. The BL Lac is of two types: high frequency BL Lac (HBL) and low frequency BL Lac (LBL), where the names correspond to the peak frequency of the synchrotron emission. The peak frequency of the synchrotron emission and that of the inverse Compton emission occur in the radio band and Xray band for LBL and in the X-ray band and TeV band for HBL.
The TeV gamma ray flux of HBLs Mkn421 and Mkn501 are usually lower than that of the Crab nebula. In 1997, the flux of Mkn501 increased up to 10 Crab over 3 months. Two Cherenkov detectorsโUtah Seven Telescope (Utah TA) and HEGRAโand X-ray satellite RXTE all sky monitor (ASM) observed Mkn501 simultaneously during this flare. Hayashida et al. studied the Utah TA data during this flare with a Fourier analysis and suggested two periodicitiesโ13 day and 23 day. Kranich studied both the HEGRA and ASM data during this flare with a Fourier analysis and obtained a 22.5 day periodicity with a chance probability of 0.028 for the HEGRA data and a 22.5 day periodicity with a chance probability of 0.047 for the ASM data. These results weakly suggest the 23 day periodicity. There is some problem in these results. The frequency-power diagram shows frequency dependent power (1/f noise). This 1/f noise in the frequency-power diagram is well known in AGN and a Blackhole candidate binary in X-ray band. The origin of the noise is unknown. Hayashida et al. do not consider an effect of the 1/f noise on both a significance and an error of the periodicity. Kranich does not consider an effect of the 1/f noise on an error of the periodicity, but he considers an effect of the 1/f noise on a significance of the periodicity. However, he uses an unreliable model of the 1/f noise for deducing a chance probability. He fitted a raw power spectra to obtain a model of the 1/f noise without discussing model reliability. With Poisson statistics, the power in a raw power spectra has a 100% error because it follows a $`\chi ^2`$ distribution of 2 degrees of freedom. In the case of Poisson statistics and the 1/f noise, this error is less than 100%:however, power still has a large error.
In this study, we study these three data sets with a Fourier analysis and also use another kind of a timing analysis in order to increase the reliability of the periodicity. We binned a raw power spectra in order to obtain a power spectra that is statistically reasonable, obtain the best model of the 1/f noise, and estimate a chance probability by considering the reliability of the 1/f noise model. We obtain a lower chance probability than Kranich by an order. We obtain an error of the periodicity by considering an effect of the 1/f noise and found that the three results are same with a newly estimated error. We study the stability of the periodicity and analyze an energy spectra of the X-ray satellite RXTE proportional counter array (PCA) during this flare , in order to set a limit on an origin of the periodicity.
## 3 data
### 3.1 timing analysis
We obtained the HEGRA data in 1997 as a form of (a MJD, flux, an error of flux). There are four kinds of dataโCT1 (no moon), CT1(moon), CT2, and CTsys corresponding to different data acquisition conditions or detectors. We use summed data for a timing analysis and show a lightcurve of the HEGRA data in MJD 50545-50661 in figure 1. We obtained the Utah TA data in 1997 as the form of (a MJD, flux, an error of flux) for a timing analysis and show a lightcurve of the Utah TA data in figure 2. We obtained the ASM data of a 90 s dwell in the form of (a MJD, a rate, an error of a rate) for a timing analysis. The error of a rate for the ASM data includes a systematic error of 3% derived using a lightcurve of Crab. We show a lightcurve of the ASM data from 1996 to 2000 in figure 3. We use a span MJD 50545-50661 for both the HEGRA and ASM data, which is the same as in Kranich . Further we use a span MJD 50520-50665 for the Utah TA data. We show a lightcurve of the ASM data in MJD 50545-50661 in figure 4. We also use a span MJD 50300-50900 for the ASM data because the ASM data is plentiful and we require more than 10 cycles for increasing a reliability of a periodicity.
### 3.2 spectral analysis
We obtain the PCA raw dataโstandard 2 data that is suitable for an energy spectral analysisโand use a span MJD 50300-50900. We use only the data of the top layer in the PCA, which has a low background. Using a standard tool FTOOLS 4.2, we selected the data with normal conditionsโan elevation greater than 10 degrees, neglecting time of the SAA passage and time less than 30 minuites of the SAA passage, offset $`0.02`$, and electron0 $`0.1`$. We use a background model, L7 model, which is generally used for a faint source. We obtain one energy spectra for each continuous observation for 10 ksec. The total number of energy spectra is 56. We calibrate a systematic error of the PCA data with an energy spectra of Crab. We fitted the energy spectra of Crab with an absorbed power law of the form dN/dE = $`e^{N_\mathrm{H}\sigma (E)}E^\alpha `$ and obtain a statistical disagreement $`\chi ^2`$/d.o.f$`=`$26.35. Here, $`\sigma (E)`$ is a crosssection of a photoabsorption, and $`\alpha `$ is a photon index. We use a column density of neutral hydrogen $`N_\mathrm{H}=32.70\times 10^{20}`$ cm<sup>-2</sup>, which is given by EINLINE. When we add a 1% systematic error to the energy spectra of Crab and fitted this with an abosorbed power law, we obtain a statistical agreement $`\chi ^2`$/d.o.f = 1.25. The energy spectra of Crab should be fitted with an absorbed power law. We need 1 % systematic error in order to fit the energy spectra of Crab with an absorbed power law. We found that there is 1 % systematic error in PCA data itself. Therefore, we add 1% systematic error to an energy spectra of Mkn501. We fitted an energy spectra of Mkn501 with an absorbed power law of a form dN/dE$`=e^{N_\mathrm{H}\sigma (E)}E^\alpha `$. We use $`N_\mathrm{H}=`$ 2.08 $`\times 10^{20}`$ cm<sup>-2</sup>, which is given by EINLINE. We obtain a data set in the form of (a MJD, a photon index, an error of a photon index).
## 4 timing analysis
### 4.1 periodicity
When we use a normal Fourier analysis for an unevenly spaced data set, leakage of the power to neighbouring frequencies is a problem. A window function is normally applied to each data point. We instead use a Fourier analysis in which a weighted power is applied . The formula is as follows.
$$P(w)=\frac{1}{2\sigma ^2}[\frac{[_j(h_j\overline{h})\mathrm{cos}w(t_j\tau )]^2}{_j\mathrm{cos}^2w(t_j\tau )}+\frac{[_j(h_j\overline{h})\mathrm{sin}w(t_j\tau )]^2}{_j\mathrm{sin}^2w(t_j\tau )}]$$
(1)
$$\mathrm{tan}(2w\tau )=\frac{_j\mathrm{sin2}wt_j}{_j\mathrm{cos2}wt_j}$$
(2)
Here, $`\overline{h}`$ is an average of a count rate, $`\sigma `$ is variance of data, ($`t_i,h_i`$) are the $`i`$th observation time and rate respectively, and $`\tau `$ is an input that does not vary with a time offset. This formula is the same with a least square analysis . Kranich used the same method. We calculate a power spectra with 100$`\times N/2`$ frequencies in order to compensate a gap in a frequency and increase the reliability of a power spectra because a power has an error of about 100% as previously mentioned in the introduction. $`N`$ is the number of data and $`N/2`$ is the number of independent frequencies. We show the power spectra for three data sets in figure 5. We obtain a maximum power as a periodicity. We obtain a 22.5 day periodicity (5$`\times 10^7`$Hz) for the HEGRA and ASM data. We obtain a 23.6 day periodicity as the second largest peak for the Utah TA data. We obtain a peak around second harmonics (1$`\times 10^6`$ Hz) in power spectra of Utah TA and ASM data. We do not obtain a peak around secound harmonics in power spectra of HEGRA data. We also obtain a 23.6 day periodicity for the ASM data in MJD 50300-50900. We discuss an error of these periodicities at a later stage.
We also attempt to another timing analysisโepoch folding methodโin order to increase the reliability of the periodicity for the HEGRA and ASM data. We make a folded lightcurve with a period $`P`$ from 1 day to 116 days in steps of 0.5 day, and calculate $`\chi ^2(P)`$=$`\mathrm{\Sigma }(h_ih_0)^2/\sigma _i`$ and $`\chi ^2(P)`$/d.o.f. for each folded lightcurve. Here, $`h_i`$ is a rate of the ith data points, $`h_0`$ is an average count rate, and $`\sigma _i`$ is an error of the ith data points in a folded lightcurve. We obtain a maximum peak in $`\chi ^2(P)`$/d.o.f vs a period $`P`$ as a periodicity. The lightcurve is composed of multiple frequencies. The epoch folding analysis is sensitive to the sum of multiple frequencies although a Fourier analysis is sensitive to a monochromatic frequency. We show $`\chi ^2(P)`$/d.o.f diagrams in figure 6. A 22.5 day periodicity is present in the ASM data. A 22.5 day periodicity is present as the second largest peak in the HEGRA data. We found an identical periodicity with an epoch folding method. We show the phase diagrams for a 22.5 day periodicity of the HEGRA and ASM data in figure 7. We confirm that the phase diagrams are broad.
A power spectra has noise equal to 1/f. When only Poission statistics exist, we can estimate a chance probability of a detected periodicity with exp($`z`$). Here, $`z`$ is a power. When there are Poisson statistics and 1/f noise, the chance probability does not follow $`\mathrm{exp}(z)`$. We have to calculate a chance probability by making simulated data sets that show the 1/f noise in a power spectra. First, we must obtain a model of the 1/f noise for each data set. We calculate a power spectra with N/2 frequencies for a fourier analysis in order to obtain independent data points. We make a binned power spectra in order to obtain power spectra that is statistically reasonable as shown in figure 8. We confirm the 1/f noise for three data sets. We fit these power spectra with a model of a form, power $`P(f)=1+\alpha \times f^\beta `$. Here, $`f`$ is a frequency and $`\alpha ,\beta `$ are constants. $`P(f)=`$1 indicates Poisson statistics. We note an area at $`10^5`$ Hz and $`10^7`$ Hz as A. We obtain (A,$`\beta `$) instead of ($`\alpha ,\beta `$) because $`\alpha `$ and $`\beta `$ are strongly coupled and have a large error. We show a fitted power spectra with a model in figure 8 and fitted parameters in table 1. We remove the data point at 5$`\times 10^7`$ Hz as a periodicity when we fit the HEGRA data although this data point is included in figure 8. Second, we create 1000 simulated data sets in the form of (a MJD, a rate) by taking an inverse Fourier transform of this model ($`f,P(f)`$. We input two parameters (mod, $`\beta `$โ) to generate the simulated data. Here, a mod is a ratio parameter between the Poisson statistics and 1/f noise. We use an actual time history for the simulation data. We analyze the 1000 simulation data sets with a Fourier method, take an average of the 1000 powers in each frequency, and make one binned power spectra. We then fit this power spectra with a model of a form, power $`P(f)=1+\alpha \times f^\beta `$, and obtain ($`A^{},\beta `$โ). $`\beta `$โ and $`\beta `$โ are not always the same. We consider two extreme conditions (A$`+\mathrm{\Delta }`$A, $`\beta \mathrm{\Delta }\beta `$)(case 1) and (A$`+\mathrm{\Delta }`$A, $`\beta +\mathrm{\Delta }\beta `$)(case 2), using which we obtain a low significance for the periodicity. Here, $`\mathrm{\Delta }`$ is a 1 sigma statistical error. We tune (mod, $`\beta `$โ) so that ($`A^{},\beta `$โ) matches (A$`+\mathrm{\Delta }`$A, $`\beta \mathrm{\Delta }\beta `$). We show simulated binned power spectra with adjusted parameters in figure 9 for the HEGRA data, figure 10 for the Utah TA data, and figure 11 for the ASM data. Third, we create 10000 or 1000 simulation data sets with adjusted parameter values (mod, $`\beta `$โ), analyze simulated data with a Fourier method, and search for a maximum power $`P_{max}`$ in each simulated power spectra. We show the distribution of maximum power $`P_{max}`$ in figure 9 for the HEGRA data, figure 10 for the Utah TA data, and figure 11 for the ASM data. We calculate chance probability $`P_{ch}=N(P_{max}P_{obs})/N_0`$. Here, $`P_{obs}`$ is the power for the observed periodicity, and $`N_0`$ is the number of simulation data. We show $`N(P_{max}P_{obs})`$, $`N_0`$ in table 1. We perform the same analysis for another condition (A$`+\mathrm{\Delta }`$A, $`\beta +\mathrm{\Delta }\beta `$). We obtain lower chance probability for the two conditions as a chance probability of the periodicity. We obtain a chance probability of 4.88$`\times 10^3`$ for the HEGRA data, 0.981 for the Utah TA, and 0.200 for the ASM data, as shown in table 1. The 22.5 day periodicity seems to be significant in the observed binned power spectra of HEGRA data as figure 8. However, two fitting parameters $`A,\beta `$ of the 1/f noise have large errors as shown in table 1. When we consider an extreme condition of case 1, 2, the simulated binned power spectra in these condition have large power as shown in figure 9. Therefore, we obtain low chnace probability 4.88$`\times 10^3`$ for HEGRA data. For the epoch folding method, we do not obtain a significance of the periodicity because of complex treatment of huge data sets.
We estimate an error of the periodicity for a Fourier analysis. The chance probability does not follow exp($`z`$) because there is an 1/f noise. Here, $`z`$ is a power. Therefore, we deduce an error of the periodicity with the simulation data. We make the simulation data of Gaussain of (R, $`\sigma `$) in a form of (a MJD, a rate). Here, R is an observed count rate, and $`\sigma `$ is a variance of data. The error of a count rate is only a Poisson fluctuation. In order to include a fluctuation of the 1/f noise, we use a variance of the data. We use an actual time history for the simulation data. We create 1000 simulation data sets, make a Fourier analysis, search for a maximum peak as a periodicity in each power spectra and obtain 1000 periodicities. We show the distributions of a periodicity for three data sets in figure 12. We fitted each distribution of periodicities with a Gaussian and obtain a 1 sigma statistical error of the periodicity. We found that the three periodicities are the same with a 1.3 sigma significance and found that a width of the periodicity is as narrow as $`\mathrm{\Delta }P/P0.01`$. In the case of the epoch folding method, we do not obtain an error of the periodicity because of complex treatment of huge data sets.
### 4.2 stability of periodicity
We study stability of the periodicity in order to set a limit on the origin of the periodicity. We divided statistically good dataโHEGRA dataโinto two and analyzed these data with a Fourier method, as shown in table 2. We estimated an error of a periodicity for a Fourier analysis, as described in the previous section:we show this in figure 13. We found that the periodicity is unstable with a 3.4 sigma significance.
## 5 Change of energy spectra
We create a phase diagram of a photon index for the 23.6 day periodicity, that is, a periodicity for the ASM data in MJD 50300-50900, as shown in figure 14. A clear relation between a photon index and a phase can be observed. X-rays have been considered as synchrotron emission by a synchrotron self Compton model. There is an energy spectral change by a synchrotron cooling:however, the time scale is about $`10^3`$. Therefore, the observed relation is related to another physics. A photon index $`\alpha `$ changes from 1.8 to 2.4. The index in an energy spectra $`\nu F_\nu `$(erg s<sup>-1</sup>) vs. $`\nu `$ (Hz) is $`\alpha +2`$. There is an index change from $`0.4`$ to 0.2. We obtain a photon index 2.345$`\pm `$0.015 at a phase of 0.0 and found a negative index with (2.345-2.000)/0.015=23 sigma significance. We also obtain a photon index 1.874$`\pm `$0.0084 at a phase of 0.8 and found a positive index with (2.000-1.874)/0.0084 = 15 sigma significance. The PCA is sensitive to energy from 3 keV to 20 keV. A negative index means that a peak energy of the synchrotron emission is below 3 keV, and the positive index means that a peak energy of the synchrotron emission is above 20 keV. Therefore, a change in the index from a negative value to a positive value means that the peak energy of the synchrotron emission moves from a low energy to a high energy. The peak energy of the synchrotron emission is written as $`E_p=10^6\gamma ^2B\delta /(1+z)`$. Here, $`\gamma `$ is a Lorentz factor of accelerated electrons, $`B`$ is a magnetic field, $`\delta `$ is a beaming factor of jet, and $`z`$ is a redshift. We found that a combination of these parameters is related to the periodicity and changes during this flare by $`\mathrm{\Delta }(\gamma ^2B\delta )/(\gamma ^2B\delta )6`$.
## 6 Conclusion
We analyze three data setsโHEGRA, Utah TA, and ASMโin 1997 for Mkn501 with a Fourier method and confirm a 23 day periodicity for these three data sets. We found that the three results are the same with a newly estimated error. We confirm a 1/f noise in a frequency-power diagram and obtain a chance probability of 4.88$`\times 10^3`$ for a periodicity in the HEGRA data by considering a 1/f noise: this is more significant than the previous result by an order. We obtain the 23 day periodicity with more than 10 cycles for the ASM data. We also obtain the same periodicity with an epoch folding method for the HEGRA and ASM data. Therefore, we strongly suggest the existence of the 23 day periodicity. We confirm that the phase diagrams of the HEGRA and ASM data are broad. We found that a width of the periodicity is as narrow as $`\mathrm{\Delta }P/P0.01`$. We divided the HEGRA data into two, analyzed these data with a Fourier method, and found that a periodicity is unstable with a 3.4 sigma significance. We analyzed an X-ray energy spectra of Mkn501 during this flare and found that a combination of three physical parametersโa magnetic field, a Lorentz factor, and a beaming factorโis related to the periodicity and is changed during this flare by $`\mathrm{\Delta }(\gamma ^2B\delta )/(\gamma ^2B\delta )6`$.
We thank both Dr.Mitsuda, Dr.Dotani and Dr. Teshima for an useful advice and discussion across the complete study and analysis. We thank Dr.Kranich for giving us the HEGRA data set. We thank Dr.Yamaoka for the RXTE PCA energy spectra.
|
warning/0506/hep-lat0506027.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
From a theoretical viewpoint the ascent of โoverlapโ fermions , i.e. fermions which at zero quark mass satisfy the Ginsparg Wilson (GW) relation ($`\rho `$ is a parameter that will be specified later)
$$\gamma _5D+D\widehat{\gamma }_5=0,\widehat{\gamma }_5=\gamma _5(1\frac{1}{\rho }D)$$
(1)
and thus realize a lattice version of the continuum chiral symmetry
$$\delta \psi =\widehat{\gamma }_5\psi ,\delta \overline{\psi }=\overline{\psi }\gamma _5$$
(2)
together with an index theorem , represents a major breakthrough in the field of non-perturbative studies of QCD. We know how to discretize fermions in a way that preserves the relevant symmetries: $`(i)`$ gauge invariance, $`(ii)`$ flavor symmetry, and $`(iii`$) chiral invariance. Unfortunately, from a practical viewpoint the usefulness of this concept is limited by the fact that the overlap tends to be one to two orders of magnitude more expensive, in terms of CPU time, than a standard Wilson Dirac operator.
In this paper we study a variant of the overlap operator which makes use of a UV filtered Wilson kernel. Here, the โfilteringโ refers to replacing the original (โthinโ) links of the gauge configuration in the standard definition of the Wilson kernel by โthickโ links obtained through APE or HYP smearing. This is a legal change of discretization as long as one keeps the iteration level and smearing parameters fixed all the way down to the continuum, since the โthickโ links transform under a local gauge transformation in the same way as the โthinโ links; it should be seen as a modification of the operator and not of the gauge background. Such filtering has been used in the context of staggered quarks, where it has been found to reduce UV fluctuations, in particular taste changing interactions due to highly virtual gluons . In Ref. filtered staggered quarks were compared against overlap quarks (where the filtered version was merely considered for completeness), and it was observed that a single filtering step may speed up the forward application of the overlap operator $`D_{\mathrm{ov}}`$ on a source vector by a factor 2-4, depending on the gauge background. This was seen to come through a reduction of the degree of the Chebychev polynomial needed to approximate the inverse square root or sign function in the definition of the massless overlap
$$aD_{\mathrm{ov}}=\rho \left[1+D_{\mathrm{W},\rho }(D_{\mathrm{W},\rho }^{}D_{\mathrm{W},\rho })^{1/2}\right]=\rho \left[1+\gamma _5\mathrm{sign}(a\gamma _5D_{\mathrm{W},\rho })\right]$$
(3)
with $`D_{\mathrm{W},\rho }=D_\mathrm{W}\rho /a`$ the Wilson operator at negative mass $`\rho /a`$. However, what matters in view of most phenomenological applications is the performance of the massive operator (bare quark mass $`m`$)
$$D_{\mathrm{ov},m}=(1\frac{am}{2\rho })D_{\mathrm{ov}}+m$$
(4)
in the process of calculating a given physical observable to a pre-defined accuracy. In other words the total CPU time spent depends on:
1. The number of forward applications of the shifted Wilson operator $`D_{\mathrm{W},\rho }`$ (or, generally speaking, of the kernel) needed to construct the massless overlap operator (3).
2. The number of iterations spent on inverting the so-constructed massive operator (4) for a given renormalized quark mass (or a given $`M_\pi ^2`$).
3. The number of gauge backgrounds needed to reach a pre-defined statistical accuracy of the desired observable at a given lattice spacing $`a`$.
4. The lattice spacing needed to enter the scaling window.
The main emphasis of this paper will be on point 1; in particular we attempt to give an understanding of the observed speedup in terms of the spectral properties of the underlying hermitean (shifted) Wilson operator $`H_\mathrm{W}=\gamma _5D_{\mathrm{W},\rho }`$. At first sight it might seem that point 2 does not need to be considered at all. At fixed bare mass $`m`$ and fixed $`\rho `$ the filtered and the unfiltered overlap do not differ on this point, since the number of forward applications of $`D_{\mathrm{ov},m}`$ to get a column of the inverse depends only on its condition number, and that is $`2\rho /m`$ for either variety. As we shall see, the optimum $`\rho `$ (w.r.t. locality) gets reduced through filtering whereas $`Z_m=Z_S^1=Z_P^1`$ increases and this means that in the filtered case one has to use a smaller bare mass to work at a fixed physical $`m^{\mathrm{ren}}=Z_mm`$. These two aspects tend to compensate, and as a result there is little net effect on point 2 from filtering. Whether in points 3 and 4 filtering brings further savings is not clear, but we plan to address this issue in the future.
Let us try to obtain a first understanding of the effect of filtering in terms of the spectrum of the underlying (non-hermitean) Wilson operator. We are going to compute all eigenvalues, and to avoid spending too much CPU time on this illustration, we shall do this in 2D, but it is clear that the conceptual issue is โ mutatis mutandis โ the same as in 4D. In $`d`$ dimensions the Wilson Dirac operator has $`d+1`$ branches, and the respective flavor multiplicities are
$$\left(\genfrac{}{}{0pt}{}{d}{0}\right),\left(\genfrac{}{}{0pt}{}{d}{1}\right),\mathrm{},\left(\genfrac{}{}{0pt}{}{d}{d}\right).$$
(5)
Thus in 2D the Wilson operator has 3 branches with multiplicities 1,2,1, while in 4D it has 5 branches with multiplicities 1,4,6,4,1, respectively. Fig. 1 shows the complete spectrum of $`D_\mathrm{W}`$ with the hopping parameter fixed at its tree-level critical value, $`\kappa =0.25`$, on 10 configurations of size $`16^2`$ at $`\beta =3.2`$ in the quenched Schwinger model. Besides the โthinโ link operator also its UV filtered descendent is shown. In terms of the kernel spectrum the filtering is seen to have the following effects:
1. The two/four โbelliesโ are depleted โ in particular exactly real modes which cannot be assigned in a unique way to one of the three/five branches are severely suppressed.
2. The horizontal scatter of any of the three/five branches diminishes.
3. The additive mass renormalization of the physical (leftmost) branch is substantially reduced.
If one were to ignore the kernel non-normality (we shall come back to this point), the spectrum of $`D_\mathrm{W}`$ could be linked, on a mode-by-mode basis, to the one of $`D_{\mathrm{W},\rho }^{}D_{\mathrm{W},\rho }`$, $`H_\mathrm{W}=\gamma _5D_{\mathrm{W},\rho }`$ and $`D_{\mathrm{ov}}`$. Then the first observation above (the reluctance of the filtered eigenvalues to show up near the projection point $`\rho `$) simply means that the effect of filtering on the spectrum of $`H_\mathrm{W}`$ is to deplete the vicinity of the origin by pushing the eigenvalues further towards the ends of the interval $`[2d+1,2d1]`$. In spite of the caveat mentioned, the thinning effect that (any kind of) smearing has on the spectrum of $`H_\mathrm{W}`$ near zero is indeed the reason for the speedup in point 1 above. A bigger interval $`[0,ฯต^2[`$ or $`]ฯต,ฯต[`$ that does not need to be covered by the polynomial/rational approximation to the $`1/\sqrt{.}`$ or $`\mathrm{sign}(.)`$ function translates into a lower degree and thus into fewer forward applications of the kernel operator.
In the remainder of this article we shall address the spectral properties of $`H_\mathrm{W}`$ in more detail (Sec. 2), and show that (a reasonable amount of) filtering does not degrade the locality properties of $`D_{\mathrm{ov}}`$, but rather makes the overlap operator *more local* (Sec. 3). We continue with an explicit demonstration that the kernel non-normality gets reduced by filtering (Sec. 4). We add some observations relevant to phenomenological applications of the filtered overlap; in particular $`Z_A`$ is shown to be much closer to the tree-level value 1 than for the unfiltered variety (Sec. 5). We rate this as a sign that perturbation theory might work far better for the filtered overlap. We make an attempt to compare our simple filtering recipe against other approaches (Sec. 6). Finally, the appendix contains spectral data which suggest that the spectral density of $`H_\mathrm{W}`$ at the origin is non-zero for any $`\beta `$ and any filtering level.
We shall use pure gauge backgrounds and set the scale through the Sommer parameter $`r_0`$ . We choose the Wilson gauge action, and since $`r_0(\beta )`$ is known it is easy to select $`\beta `$ values such that the resulting lattices are matched, i.e. have fixed spatial size $`L1.5\mathrm{fm}`$, with the resolution varying by a factor 3 from the coarsest to the finest lattice โ see Tab. 1 for details. Henceforth we set $`a=1`$.
## 2 Speedup and kernel spectrum
In a quenched simulation the overhead, in terms of CPU time, of overlap versus Wilson quarks comes in the first place from the polynomial or rational approximation to the $`1/\sqrt{.}`$ or $`\mathrm{sign}(.)`$ function in (3). Let us assume<sup>1</sup><sup>1</sup>1In fact, these values are rather close to the actual situation at $`\beta 6.0`$, after projecting out the lowest 10-15 eigenvectors. that the lowest eigenvalue of the unfiltered $`|H_\mathrm{W}|`$ is $`0.14`$ while the highest eigenvalue takes the free field value 7. This leads to the task to construct a polynomial/rational approximation of the inverse square root over the range $`[ฯต^2,1]`$ with $`ฯต=0.02`$ the inverse condition number of $`|H_\mathrm{W}|`$. Modest filtering will lift the lowest eigenvalue to something like $`0.49`$, while the largest eigenvalue is almost invariant. Then the task is to construct the approximation over the range $`[\stackrel{~}{ฯต}^2,1]`$ with $`\stackrel{~}{ฯต}=0.07`$ the filtered inverse condition number. The lower bound increasing from $`0.0004`$ to $`0.0049`$ means that one gets away with a smaller overall polynomial degree or an increased minimum root of the denominator polynomial. Therefore, the filtered overlap requires fewer forward applications of $`D_{\mathrm{W},\rho }^{}D_{\mathrm{W},\rho }`$ and this is how the savings on CPU time in point 1 above come about. In the remainder of this section we will elaborate on this statement, replace the fictitious numbers by actual figures from real simulations and see that the conclusion remains unchanged.
Fig. 2 shows, as an illustration, the 15 lowest eigenvalues of $`|H_\mathrm{W}|`$ on 25 configurations at $`\beta =6.0`$, without filtering and after 1,3 steps of APE or HYP smoothing. The filtering increases the upper end of the band of eigenvalues shown. In fact, just this upper end matters in terms of CPU time, since in practice one projects out the lowest few modes and constructs the function in (3) over the relevant spectral range of $`|H_\mathrm{W}|`$ on the subspace orthogonal to these modes. Hence the sequence of the 15th eigenvalue represents the relevant quantity, if 14 modes are treated exactly, and this band gets lifted by filtering. Evidently, a single APE step is less efficient than a single HYP step, and adding two more steps lifts the 15th eigenvalue further, but the lifting factor is no more as large as it was in the first step. Here and below we use the parameters $`\alpha _{\mathrm{APE}}=0.5`$, $`\alpha _{\mathrm{HYP}}=(0.75,0.6,0.3)`$ (for details of the $`SU(3)`$ projection see e.g. or the appendix of ) and, unless stated otherwise, $`\rho =1`$.
Fig. 3 shows the mean and the standard deviation of the 15 lowest eigenvalues of $`|H_\mathrm{W}|`$, with our standard filtering options (none, 1 APE, 3 APE, 1 HYP, 3 HYP). In this logarithmic representation it is easy to see that (apart from the coarsest lattice which represents a special case discussed in App. A) all 15 eigenvalues get lifted, at a given coupling, by virtually the same factor. Specifically, the 15th<sup>2</sup><sup>2</sup>2This number needs to be scaled with the physical box volume; working, for any given $`\beta `$, in a $`(2.0\mathrm{fm})^4`$ box instead of $`(1.5\mathrm{fm})^4`$, our statement would most likely be adequate for the 47th mode. eigenvalue gets multiplied by $`\lambda _{1\mathrm{HYP}}/\lambda _{\mathrm{none}}=4.8,4.2,3.2`$ at $`\beta =5.84,6.00,6.26`$. Thus the lifting effect that filtering has on the โbulkโ part of the $`|H_\mathrm{W}|`$ spectrum diminishes somewhat towards the continuum, but for accessible couplings it remains substantial. Details of the ensemble average of the 15th eigenvalue are collected in Tab. 2. The second observation is that the bands become flatter at large $`\beta `$, hence the onset of the โbulkโ becomes a less ambiguous concept at weaker coupling. Had we chosen the 10th or 20th mode to define the โbulk edgeโ instead of the 15th, this would cause a small change at $`\beta =6.26`$, but it would make a substantial difference at the smallest $`\beta `$ shown.
A point of theoretical interest is whether the low-lying eigenvalues of the (shifted) hermitean Wilson operator $`H_\mathrm{W}`$ are correlated, between different smearing levels, just as the low-lying eigenvalues of the final $`D_{\mathrm{ov}}`$ were found to be correlated for large enough $`\beta `$ . Fig. 4 shows that this is almost true โ the eigenvalues correlate if they are sufficiently large in absolute magnitude, but the correlation weakens closer to the origin. Here, a technical issue comes along. Ideally, one would pair the eigenvalues by considering a smooth interpolation between the two filtering recipes. Changes in topology (as seen by the overlap operator) would then be evident as stray points in quadrants 2 or 4. However, since we just know the eigenvalues shown we decided to pair them starting from 0. Now there are no points in quadrants 2 or 4 by definition and changes in topology manifest themselves through a reduced correlation of the few lowest eigenvalues in absolute magnitude. Such topology changes are expected to occur with an $`O(a^2)`$ re-definition of the overlap operator, e.g. by changing the filtering or $`\rho `$ .
A similar conclusion is drawn from the flow of eigenvalues $`H_\mathrm{W}`$ as shown in Fig. 5 for one $`16^4`$ configuration. One effect of filtering is to stretch the whole scenery in the vertical direction (note the vertical scale). Filtering also shifts the entire eigenvalue flow to the left which is consistent with the reduction of the additive mass renormalization of the kernel operator as discussed in the introduction. Note that there is, from a conceptual viewpoint, no reason to prefer one filtering level over any other one; what we see is just a manifestation of the $`O(a^2)`$ ambiguity of the overlap operator .
To assess the CPU time needed for the massless overlap, the behavior of the โbulk edgeโ of the $`|H_\mathrm{W}|`$ spectrum is one ingredient. What really matters is the condition number, thus we need to study the largest eigenvalue, too. From the naive discussion around Fig. 1 in the introduction one expects that filtering barely affects the largest eigenvalue of $`|H_\mathrm{W}|`$. It turns out that this is indeed true, for instance at $`\beta =6.0`$ a single HYP filtering step lifts it from $`6.55(1)`$ to $`6.88(1)`$. Hence, filtering has an overall beneficial effect on the condition number as illustrated in Fig. 6. Without projection the condition number fluctuates wildly and occasionally it may increase through filtering (i.e. the lowest eigenvalue decreases, cf. Fig. 2) but after projecting 14 eigenmodes this never occurs. The bottom line is that the combination of filtering and projection reduces the condition number much more vigorously than either one alone could do. Average condition numbers after projecting out 14 eigenmodes are collected in Tab. 3 (regarding the first entry, cf. App. A). As a side remark we note that the horizontal increase to the left explains why in a fixed physical volume simulating unfiltered overlap quarks on a coarse lattice is not so much cheaper than on a fine one; for the filtered version this penalty is reduced.
We have also studied the condition number of $`|H_\mathrm{W}|`$ as a function of the parameter $`\rho `$. With and without filtering the minimum is rather shallow and at a $`\rho `$ value above $`1`$. Since in the free case
$$ฯต=\{\begin{array}{cc}\rho /(8\rho )\hfill & \mathrm{for}\mathrm{\hspace{0.33em}0}<\rho 1\hfill \\ (2\rho )/(8\rho )\hfill & \mathrm{for}\mathrm{\hspace{0.33em}1}\rho <2\hfill \end{array}$$
(6)
we expect that larger $`\beta `$ values will further drive the minimum location towards $`\rho =1`$.
The last step is to convert the reduced condition number, brought by the filtering, of $`|H_\mathrm{W}|`$ on the subspace orthogonal to the lowest 14 modes into a lower degree of the polynomial/rational approximation of the $`1/\sqrt{.}`$ function in (3) and thus into actual savings of CPU time in step 1 of the introduction.
The precise speedup factor depends on the implementation of the massless overlap operator (3). For definiteness let us consider the approximation of the inverse square root over the range $`[ฯต^2,1]`$ through Chebychev polynomials . Fig. 7 shows on the l.h.s. for a few inverse condition numbers $`ฯต`$ of $`|H_\mathrm{W}|`$ the well known exponential fall-off pattern of the truncation error of the Chebychev approximation versus the number of applications of $`H_\mathrm{W}^2=D_{\mathrm{W},\rho }^{}D_{\mathrm{W},\rho }`$. What matters for our purpose is the dependence of the polynomial degree required to reach a fixed minimax accuracy โ say $`\delta =10^8`$ over the full approximation range โ on $`ฯต`$. As is evident from the r.h.s. of that figure, the relation
$$\mathrm{degree}ฯต^1$$
(7)
holds in good approximation. Thus, from (7) and a look at Tab. 3 one predicts that at $`\beta =6.0`$ and $`\rho =1`$ a single HYP step will speed up the construction of the overlap (on average) by a factor $`53.2/13.3=4.00`$, and this is in good agreement with what we find in actual runs (see Fig. 8). On a coarser lattice this factor would be somewhat larger ($`4.56`$ at $`\beta =5.84`$) while on a finer lattice it tends to decrease ($`3.04`$ at $`\beta =6.26`$), but it certainly remains substantial at all accessible couplings.
To approximate the inverse square root or sign function over the relevant range, two main strategies are found in the literature. Polynomial and rational representations have been tried. We have concentrated on the Chebychev variant, since this one is efficient and easy to implement. It goes without saying that the lifting effect on the bulk of the $`|\gamma _5D_{\mathrm{kern}}|`$ eigenvalues translates into similar savings on CPU time in step 1 of the introduction, if another representation is used. For instance, in the rational approach it is the increase of the smallest zero of the denominator polynomial that lets one get away with fewer iterations in the inner multishift CG.
For five dimensional variants of the overlap operator , in particular the domain wall formulation, the computational gain comes from the reduction of the extent of the fifth dimension needed to reach a given residual mass. What we would like to stress here is simply that our proposal to replace the โthinโ links by โthickโ links is generically useful for any kind of overlap variant.
## 3 Locality
It has been shown that the overlap operator cannot be ultralocal, as opposed to the Wilson operator where $`D_\mathrm{W}(x,y)=0`$ for $`xy_1>1`$. To guarantee the universality of the underlying field theory and hence to obtain the correct continuum limit it is sufficient to have an operator with
$$D_{\mathrm{ov}}(x,y)\mathrm{exp}(\nu xy)\mathrm{for}xy1$$
(8)
where the localization $`\nu `$ is of the order of the cut-off, i.e. $`\nu =O(1)`$ \[in lattice units\]. In practice, for a given lattice spacing the condition (8) gives an upper bound on any physical mass that one can extract, and it is therefore crucial to have an operator as local as possible, i.e. with a maximal $`\nu `$. In it has been demonstrated that the standard overlap operator indeed obeys (8). It is clear that their proof goes through for our filtered variant, but it is open in which way the localization $`\nu `$ is influenced. Naively, one might think that the locality will deteriorate, since the original links entering the covariant derivative of the filtered kernel spread over a larger volume. As first observed by Kovacs , the filtered overlap turns out to be even more local than the standard one and this is achieved without tuning $`\rho `$.
In Fig. 9 we plot the localization of $`D_{\mathrm{ov}}`$ at $`\beta =6.0`$ with two projection parameters ($`\rho =1.0,1.4`$) and two filtering options (none, 1 HYP). The ordinate is the maximum over the 2-norm of $`D_{\mathrm{ov}}\eta `$ at $`x`$ with $`\eta `$ a normalized $`\delta `$-peak source vector at the point $`y`$ in the lattice, the abscissa is the โtaxi driverโ distance $`d_1=xy_1`$ to the location of the $`\delta `$-peak, i.e. we plot the function
$$f(d_1)=\mathrm{sup}\{||(D_{\mathrm{ov}}\eta )(x)||_2|||xy||_1=d_1\}$$
(9)
versus $`d_1`$, as first studied in . Comparing the two unfiltered operators (black/dark diamonds and crosses) one finds their result reproduced that (at this $`\beta `$) adjusting $`\rho `$ to a value around $`1.4`$ lets $`f(d_1)`$ fall off steeper than with the value $`1.0`$ which is the canonical choice in view of the spectrum of the Wilson operator sufficiently close to the continuum (cf. Fig. 1). The interesting observation is that a single HYP step together with $`\rho =1.0`$ (red/light squares) results in an even steeper descent than the unfiltered version with $`\rho =1.4`$ (which was chosen to nearly optimize the locality of the unfiltered operator). The last curve shown (red/light pluses) indicates that one should not attempt to combine the filtering with a $`\rho `$ value that would be optimal for the unfiltered operator.
An obvious question is whether filtering remains useful on fine lattices. Fig. 10 shows the fall-off at four couplings with no smearing, 1 APE and 1 HYP step, with $`\rho =1`$ fixed. On the coarsest lattice smearing alters the locality just modestly, on the two intermediate ones ($`\beta =5.84,6.00`$) the locality gets substantially improved, with HYP doing a better job than APE. On the finest lattice, the improvement is still sizable, but there is almost no difference among the two filtering recipes. At this coupling further smearing steps would then diminish the locality. The localization measured with the definition (29) \[which we use for technical reasons discussed below\] is summarized in Tab. 4.
There is a loose connection between the localization of $`D_{\mathrm{ov}}`$ and the spectrum of $`H_\mathrm{W}`$, for instance
$$D_{\mathrm{ov}}(x,y)\mathrm{const}\times \mathrm{exp}(\frac{\theta }{2}xy_1)$$
(10)
is a bound found in , where $`||.||`$ is the matrix norm in Dirac and color space. The exponent $`\theta /2`$ in (10) is defined via the largest and smallest eigenvalue of $`D_{\mathrm{W},\rho }^{}D_{\mathrm{W},\rho }`$ through
$$\mathrm{cosh}(\theta )=\frac{\lambda _{\mathrm{max}}/\lambda _{\mathrm{min}}+1}{\lambda _{\mathrm{max}}/\lambda _{\mathrm{min}}1}=\frac{1+ฯต^2}{1ฯต^2}$$
(11)
where we like to express the r.h.s. in terms of the inverse condition number $`ฯต`$ of $`|H_\mathrm{W}|`$. Expanding either side to first order one obtains the simple relation (after getting rid of the unphysical $`\theta <0`$ solution)
$$\frac{\theta }{2}=ฯต+O(ฯต^2).$$
(12)
As already mentioned in the exponent $`\theta /2`$, defined via the spectral properties of the underlying $`|H_\mathrm{W}|`$, is a rather bad estimate for the actual localization $`\nu `$. The situation is not much better for the filtered variety, as a brief comparison of our Tabs. 3 and 4 reveals<sup>3</sup><sup>3</sup>3Tab. 3 contains the condition number on the subspace orthogonal to the 14 lowest modes, while $`ฯต`$ in (10, 12) refers to the full operator. For two reasons we propose to re-interpret (12) as a prediction for the locality of $`D_{\mathrm{ov}}`$ with $`ฯต`$ the ratio of the lower to the upper end of the *bulk* of eigenvalues of $`|H_\mathrm{W}|`$. A practical hint is that the unprojected condition number fluctuates wildly (see Fig. 6), whereas the localization is rather stable for all configurations in an ensemble. Furthermore, in it is shown that an isolated near-zero mode of $`|H_\mathrm{W}|`$ does normally not affect the locality of $`D_{\mathrm{ov}}`$. Of course, one cannot repeat that argument indefinitely, but still a test whether a modified $`ฯต`$ helps is interesting.. Though quantitatively unsuccessful, this connection still gives a qualitative hint that the overlap operator with a filtered Wilson kernel might enjoy better localization properties due to the reduced condition number of $`H_W`$. There are more detailed bounds in the literature , but it seems fair to say that a quantitative understanding of the localization of $`D_{\mathrm{ov}}`$ in terms of the spectral properties of $`H_W`$ is a challenge.
The localization $`\nu `$ as a function of the projection parameter $`\rho `$ is presented in Fig. 11. For $`\beta =5.84,6.00`$ the optimum parameter for the unfiltered operator is around $`\rho =1.6,1.4`$, respectively. For the 1 HYP operator the localization at $`\rho =1.0`$ does not fall short of the maximal one by a large amount; this is why we restrict much of our investigation with a filtered $`D_{\mathrm{ov}}`$ to the case $`\rho =1.0`$. Still, the figure suggests that an optimal $`\rho `$ for the 1 HYP filtered operator may be *smaller* than 1, and it decreases with increasing $`\beta `$; at $`\beta =5.84`$ we find $`\rho _{\mathrm{opt}}^{1\mathrm{HYP}}1.0`$ and at $`\beta =6.00`$ we find $`\rho _{\mathrm{opt}}^{1\mathrm{HYP}}0.8`$.
After dealing with some technical issues to make sure that an $`48^4`$ lattice is large enough (see App. B), we have studied $`\nu `$ defined via (29) as function of $`\rho `$ in the free case; the result is shown in Fig. 12. The pattern observed in Fig. 11 should thus not come as a surprise, filtering simply drives the locality properties of the overlap operator towards the free field case. In fact, Fig. 12 offers a simple explanation why it is so difficult to predict the localization $`\nu `$ from spectral properties of the underlying $`|H_\mathrm{W}|`$ operator โ in the free case the inverse condition number (6) in the range $`0<\rho <1`$ is monotonic, while $`\nu `$ has a non-trivial extremum at $`\rho _{\mathrm{opt}}^{\mathrm{free}}0.54`$.
## 4 Kernel non-normality
An operator $`A`$ is called normal, if it commutes with its adjoint
$$[A,A^{}]=0$$
(13)
which implies that its left and right eigenbasis coincide. Normality has special implications for lattice Dirac operators. For a normal Dirac operator $`D=_k\lambda _k|kk|`$ which, in addition, is $`\gamma _5`$-hermitean
$$\gamma _5D\gamma _5=D^{}$$
(14)
we immediately obtain
$$D^{}=\underset{k}{}\lambda _k^{}|kk|=\underset{k}{}\lambda _k\gamma _5|kk|\gamma _5$$
(15)
and this implies that eigenmodes with real $`\lambda _k`$ are chiral (or may be linearly combined to chiral modes in case of degeneracies). Furthermore, for such a $`D`$ the eigenvectors of the hermitean Dirac operator
$$H=\gamma _5D=\underset{k}{}\lambda _k\gamma _5|kk|$$
(16)
are given by $`\sqrt{\lambda _k^{}}|k\pm \sqrt{\lambda _k}\gamma _5|k`$ with the corresponding eigenvalues $`\pm |\lambda _k|`$.
The continuum Dirac operator is normal, and so are the naive and staggered discretizations (but the latter two yield more than one flavor in the continuum limit). The GW relation (1) together with $`\gamma _5`$-hermiticity (14) also implies normality of the operator, hence $`D_{\mathrm{ov}}`$ is normal. In fact, the overlap construction can be described as extracting the unique unitary part of $`D_{\mathrm{kern}}/\rho `$ , and for a normal kernel it reduces to a simple radial projection of the $`D_{\mathrm{kern}}/\rho `$ eigenvalues onto the unit circle.
The shifted Wilson operator, which we use as a kernel, is not normal. Some consequences of this non-normality have been explored in other contexts . Here, it suffices to point out that the relations between the eigenmodes of the overlap operator, its kernel and the hermitean Dirac operator are not as simple as above for the case of a normal operator. Typically, an eigenvector of the (hermitean) kernel will mix into every mode of the overlap operator, which we expect to have a detrimental effect on the efficiency of overlap construction algorithms. Thus, a practically relevant question is whether UV filtering can reduce the amount of non-normality of the overlap kernel.
To quantify the non-normality of $`D_\mathrm{W}`$ we measure the 2-norm of the commutator; technically
$$[D_{\mathrm{W},1},D_{\mathrm{W},1}^{}]|\eta $$
(17)
is averaged over a number of normalized random vectors $`|\eta `$. In Fig. 13 the commutator (17) is shown for all $`\beta `$ and smearing levels (since this is not a physical observable, we use lattice units). Evidently, any kind of filtering reduces it โ the filtered kernel is thus closer to normality and has left- and right-eigenvectors that are better aligned than for the unfiltered version. Whether โsmartโ overlap construction algorithms can be written which exploit this property is an open question.
## 5 Physics perspectives
To explore the physics potential of filtered overlap quarks a quenched spectroscopy study would be highly desirable. Physical results should reproduce โ after a continuum extrapolation โ results in the traditional โthin linkโ formulation. It would be interesting to see whether the speedup in point 1 of the introduction gets enhanced in points 2-4; in particular if scaling and/or asymptotic scaling set in earlier, this would make a real difference. Unfortunately, a detailed scaling study requires substantial computational resources, but as a first step in this direction we want to investigate the renormalization of the axial-vector current with filtered overlap quarks.
We follow the method of , where one starts from the usual (chirally rotated) densities
$`P(x)`$ $`=`$ $`\psi _1(x)\gamma _5[(1{\displaystyle \frac{1}{2\rho }}D_{\mathrm{ov}})\psi _2](x)`$ (18)
$`A_\mu (x)`$ $`=`$ $`\psi _1(x)\gamma _\mu \gamma _5[(1{\displaystyle \frac{1}{2\rho }}D_{\mathrm{ov}})\psi _2](x)`$ (19)
with $`\psi _1\psi _2`$ (flavor non-singlet) and defines the correlators \[$`x=(๐ฑ,t)`$\]
$`G_{PP}(t)`$ $`=`$ $`{\displaystyle \underset{๐ฑ}{}}P(๐ฑ,t)P^c(\mathrm{๐},0)`$ (20)
$`G_{AP}(t)`$ $`=`$ $`{\displaystyle \underset{๐ฑ}{}}\overline{}_4A_4(๐ฑ,t)P^c(\mathrm{๐},0)`$ (21)
where $`\overline{}_4`$ is the symmetric derivative in the time direction and $`P^c`$ is the conjugate of (18), i.e. with the flavor indices $`12`$ interchanged. With these correlators at hand one forms the ratio
$$\rho (t,m_1,m_2)=\frac{G_{\overline{}AP}(t)}{G_{PP}(t)}$$
(22)
where the second and third argument indicate that the spinors $`\psi _1`$ and $`\psi _2`$ in the densities (18, 19) are solutions to the massive operators $`D_{\mathrm{ov},m_1}`$ and $`D_{\mathrm{ov},m_2}`$, respectively. On account of the axial Ward identity (AWI) the ratio $`\rho `$ should be constant in time, and for light enough quarks (22) tends indeed to plateau rather nicely (see e.g. Fig. 1 in ). In a slightly sloppy but transparent notation the plateau value is $`\rho (m_1,m_2)`$. This quantity will โ to the extent to which the AWI is respected at finite lattice spacing โ only depend on the sum<sup>4</sup><sup>4</sup>4In principle, we might use the covariant conserved current for overlap quarks (see and the 2nd work in ) with the โthinโ links replaced by โthickโ links. Then the last term on the r.h.s. of (23, 24) would be absent, and the AWI would be an exact identity. However, there is a practical problem with APE or HYP filtering, due to the $`SU(3)`$ projection involved. The solution via stout/EXP links is in exact analogy to the dynamical case discussed in the last section. of the quark masses, and thus defines the $`m^{\mathrm{AWI}}`$ quark masses
$$\rho (m_1,m_2)=\rho (m_1+m_2)+O(a^2)=m_1^{\mathrm{AWI}}+m_2^{\mathrm{AWI}}+O(a^2).$$
(23)
The actual data for our $`Z_A`$ determination for quenched filtered and unfiltered overlap quarks are generated with couplings and geometries as given in the last line of Tab. 1. We restrict ourselves to the canonical choice $`\rho =1`$. We plot $`\rho (m_1,m_2)`$ versus $`m_1+m_2`$ for various quark mass combinations and filtering levels in Fig. 14 for $`\beta =5.66`$ and in Fig. 15 for $`\beta =5.84,6.00`$, respectively. They form one universal band, i.e. different $`m_1`$ and $`m_2`$ combinations with a fixed sum $`m_1+m_2`$ always give the same $`\rho (m_1+m_2)`$ \[within errors\]. Furthermore, the relationship is in good approximation linear, but there is an anomaly without filtering at our strongest coupling (Fig. 14). Here, the slope is *negative*, and this supports the view established in App. A that with $`\beta =5.66`$ and $`\rho =1.0`$ the projection point is โinโ or โto the leftโ of the physical branch of the underlying Wilson operator, and we effectively operate in the โzero fermionโ sector. Also at $`\beta =5.84`$ the unfiltered plateau was not very pronounced either, resulting in a large systematic uncertainty beyond the statistical error quoted below. We use the ansatz
$$\rho (m_1+m_2)=\mathrm{const}+\frac{1}{Z_A}(m_1+m_2)+\mathrm{const}(m_1+m_2)^2$$
(24)
and see whether we obtain acceptable fits and whether the first constant is consistent with zero. It turns out that this is the case, and the associate $`Z_A`$ values are summarized in Tab. 5.
It is interesting to discuss both the general pattern of these $`Z_A`$ values and the relation to 1-loop perturbation theory. Evidently, at fixed $`\beta `$ and $`\rho `$ the filtered $`Z_A`$ is much closer to the tree-level value 1. We recover the relative strength ordering of Sect. 2, i.e. one APE step is less efficient than 3 APE or a single HYP step, but the latter is topped by 3 HYP steps. At $`\beta =6.0`$ we compare to $`\rho =1.4`$ which is the the standard choice for the โthin linkโ overlap. Without filtering, $`Z_A^{\mathrm{none}}(\rho =1.4)1.554`$ is about half of $`Z_A^{\mathrm{none}}(\rho =1.0)3.145`$, and this means that the choice $`\rho =1.4`$ is not just near-optimal w.r.t. locality, but also beneficial to tame (one particular) renormalization. Once the filtering recipe is specified, $`Z_A`$ seems to be monotonic in $`6/\beta =g_0^2`$, as expected from perturbation theory. In the unfiltered case the 1-loop value is included in the last line of Tab. 5 for comparison. Assuming that in perturbation theory $`1<Z_A^{1\mathrm{HYP}}<Z_A^{\mathrm{none}}`$ holds for $`\rho =1`$, one may compare the deviation of the unfiltered $`\beta =6.0`$ operator $`3.1451.264=1.881`$ to $`1.1531=0.153`$ which then amounts to an upper bound in the 1 HYP case. Evidently, the discrepancy is dramatically reduced, which in view of the perturbative results in , should not come as a surprise. To get a slightly more quantitative view, we consider it useful to fit our data without filtering at $`\rho =1`$ to a Pade-type ansatz of the form
$$Z_A^{\mathrm{none}}=\frac{1+c_1x+c_2x^2}{1+(c_11.585648)x}$$
(25)
with $`x=1/\beta `$, where the perturbative knowledge (cf. caption of Tab. 5) is built-in as a constraint. In the same spirit a Pade ansatz for any of the filtered operators reads
$$Z_A^{1\mathrm{H}\mathrm{Y}\mathrm{P}/\mathrm{\hspace{0.17em}3}\mathrm{HYP}}=\frac{1+c_1x+c_2x^2}{1+c_3x}$$
(26)
with โ as of now โ no constraint on $`c_1c_3`$ yet. There is a problem with the functional forms (25, 26), since our data sets contain 2 and 3 entries, respectively, and there is zero degree of freedom. Still, for an illustration such a โfitโ might be worth while, and the result is shown in Fig. 16. With sufficient data the curves would contain two pieces of information. The asymptotic slope for $`x0`$ would predict the perturbative 1-loop coefficients for $`Z_A^{1\mathrm{HYP}}`$, $`Z_A^{3\mathrm{HYP}}`$. And the pole in (25, 26), i.e. the values $`c_11.585648`$ or $`c_3`$, respectively, would predict the coupling where the perturbative description breaks down. Hence, if the curves in Fig. 16 are indicative at all, it seems that filtering renders the perturbative 1-loop coefficient of $`Z_A`$ much smaller, but the perturbative range gets barely enhanced.
## 6 Discussion
In this paper we have studied the massless overlap operator constructed from a filtered Wilson kernel where the original โthinโ links were replaced by โthickโ links which behave in the same manner under local gauge transformations. This is a legal change of the fermion discretization as long as one particular filtering recipe \[e.g. 1 HYP step with $`\alpha _{\mathrm{HYP}}=(0.75,0.6,0.3)`$\] is maintained at all couplings. It amounts to an $`O(a^2)`$ re-definition of $`D_{\mathrm{ov}}`$ at fixed $`\rho `$, as does a change of $`\rho `$ at fixed filtering level.
Our key observations are the following. First, the onset of the โbulkโ part of the spectrum of the underlying shifted hermitean Wilson operator $`H_\mathrm{W}=\gamma _5(D_\mathrm{W}\rho )`$ gets lifted. This leads to an increased inverse condition number $`ฯต`$ (after projection typically by a factor 2-4 through a single HYP step) and the latter reflects itself in a reduction (by the same factor) of the polynomial degree (and thus the number of forward applications of $`H_\mathrm{W}^2`$) needed to construct the inverse square root over the relevant range. What is the precise impact on CPU requirements to invert the massive operator is a topic for future research. Second, at standard couplings the filtered massless overlap is โ even with the untuned canonical choice $`\rho =1`$ โ better localized than the unfiltered version with an optimally tuned $`\rho `$ could ever be. Our finding is backed by the observation that in the free case the optimum $`\rho `$ (w.r.t. locality) is around $`0.54`$ and thus substantially smaller than the typical $`\rho 1.4`$ used in the past. Our third observation is that the filtered kernel is much closer to being a normal operator. In other words the left- and right-eigenvectors of $`D_\mathrm{W}`$ are better aligned with higher filtering level, and in this respect the effect of the โthickโ links is the same as a shift much closer towards the continuum under which the overlap construction (3) tends to be a simple radial projection of the $`D_\mathrm{W}`$ eigenvalues. Finally, our fourth observation is that the renormalization constant of the axial-vector current is much closer to 1 with filtering than without. We rate this as a sign that lattice perturbation theory for the filtered overlap might work much better than for the unfiltered variety. If this is indeed so, and if it goes through for 4-fermion operators, it is likely to be the most important consequence of our work, since it offers the perspective of considerably reduced theoretical uncertainties in electroweak precision studies.
Let us finally discuss a variety of proposals in the literature that are similar in spirit to the one put forth in the present paper.
There is a top-level version deriving from โparametrized fixed-point fermionsโ. The idea behind this approach pursued by Hasenfratz and Niedermayer is that true fixed-point fermions would satisfy the GW relation exactly , but a practical implementation is always ultralocal. Hence, sticking such an ansatz into the overlap formula (3) yields fermions with exact chiral symmetry and otherwise properties that are at least as good (but typically better) than the version with a plain Wilson kernel .
Bietenholz has considered a variety of actions, originally based on RG concepts . The idea was that an action with a spectrum close to the GW circle could be iteratively improved in its chiral properties. Over time the focus has shifted towards using the overlap formula (3) to have exact chiral symmetry, but it is clear that the kernel of his โhypercubic overlapโ benefits from a larger inverse condition number $`ฯต`$ of $`|\gamma _5D_{\mathrm{kern}}|`$ just as we do.
Gattringer and collaborators construct a โchirally improvedโ Dirac operator that involves the full Dirac Clifford algebra with links restricted to the hypercube. The coefficients are adjusted such that (for a given coupling) the violation of the GW relation is minimized . The problem is the same as in the Bietenholz approach: a single forward application with such a kernel is so expensive that the improvement, if it is not โperfectโ, does not really pay off.
DeGrand has considered โ both perturbatively and non-perturbatively โ Wilson and clover action varieties that involve smeared gauge links . Based on this experience he went on to construct a โvariant overlapโ which starts from a kernel with only scalar/vector terms and smoothed links, and is thus sufficiently cheap as to allow for sticking it into the overlap formula .
The closest to what we do is found in the work of Kovacs . He uses a โfat-link cloverโ overlap in which all links are smeared, together with the tree-level value $`c_{\mathrm{SW}}=1`$. As far as we know, he was the first author to notice that such a filtered kernel allows for the untuned choice $`\rho =1`$, and still the resulting overlap shows good localization properties.
A related approach has been pursued by the Adelaide group . Their โfat link irrelevant cloverโ overlap quarks are built from a clover action in which only the irrelevant pieces (i.e. the Wilson and the Sheikoleslami-Wohlert terms) use smeared links, but not the covariant derivative. They found a similar speedup factor in the construction of the overlap operator (cf. โstep 1โ in the introduction) and tied it to the reduced spectral density of $`|\gamma _5D_{\mathrm{kern}}|`$ near the origin.
Finally, โoverlapโ quarks with smeared gauge links have been used by several lattice collaborations. RBC has found that the residual mass of domain-wall fermions at fixed $`N_5`$ gets reduced , though they miss out an important ingredient, the projection to $`SU(3)`$. UKQCD has used overlap valence quarks with 3-fold HYP smeared links on staggered sea as supplied by the MILC collaboration, finding a surprisingly good signal on as few as 10 configurations . Similarly, LHP and NPLQCD have used filtered domain-wall valence quarks on staggered sea to compute the pion form factor and the $`I=2`$ $`\pi \pi `$-scattering length , respectively.
There is another idea that should not be confused with filtering. Using an improved gauge action has been found to reduce $`\rho _{|H_\mathrm{W}|}(0)`$ by up to an order of magnitude . There is, however, an important practical difference to the filtering concept, which is a modification of the fermion action. As already discussed in , a better choice of the gauge action improves, in the first place, the very low end of the $`|H_\mathrm{W}|`$ eigenvalue distribution. After projecting out the lowest $`O(15)`$ eigenvectors (which nowadays is a standard thing to do ) much of the advantage is lost (in Fig. 3 of the lifting factor diminishes to the right). By contrast, filtering lifts the complete low-energy end of the $`|H_\mathrm{W}|`$ eigenvalues (in our Fig. 3 one finds an almost-universal lifting factor) and the usefulness of filtering is not vitiated by the projection. Still, it might be interesting to see whether the two ideas can be fruitfully combined.
An extension of the filtering concept to full QCD is straightforward, albeit hampered by a technical problem. These days, most dynamical fermion simulations are set up with a HMC algorithm, and the latter requires the fermion action to be differentiable w.r.t. the gauge links. The kernel of our filtered overlap quarks is differentiable w.r.t. the โthickโ links, but not w.r.t the elements of the original set, due to the projection involved in the APE or HYP procedure. A convenient way out is offered by the stout/EXP links introduced in , involving a differentiable mapping between the โthickโ and โthinโ links. In pure gauge observables the usefulness of this smearing recipe was found to be restricted to small parameter values , and one may fear that this feature persists in stout/EXP overlap quarks, since in perturbation theory they are equivalent to APE filtered overlap fermions with $`\alpha _{\mathrm{APE}}=1/(16\alpha _{\mathrm{EXP}})`$ . Thus, due to the pole at $`\alpha _{\mathrm{EXP}}=1/6`$ we expect them to have a โnarrow therapeutic rangeโ in parameter space, but it is clear that there is no conceptual issue in simulating full QCD with filtered overlap quarks beyond the difficulties met in the unfiltered case .
To summarize, our suggestion is to use the overlap recipe (3) with an unimproved ($`c_{\mathrm{SW}}=0`$) Wilson kernel in which all links are replaced by some smeared descendents of the actual gauge background. We recommend to stay with a moderate amount of link โfatteningโ, e.g. with a single step of standard HYP smearing . The projection parameter $`\rho `$ may be fixed at its canonical value 1, and in this sense the filtered overlap involves *less tuning* than the unfiltered version<sup>5</sup><sup>5</sup>5Of course, there are parameters in the filtering recipe, but our results show that they hardly matter. Thus filtering allows one to trade a parameter that needs to be tuned for parameters on which the lattice data show very little sensitivity.. An important restriction is that the choice of iteration level and smearing parameter must be the same for all couplings considered in a scaling study. This is one point on which our proposal differs from some of the attempts reviewed above which involve coefficients (e.g. in the extended $`\gamma `$-algebra) that are adjusted โby handโ to yield a GW-type spectrum at one standard value of the gauge coupling. The other difference is that our kernel remains cheap and still requires fewer $`D_{\mathrm{kern}}^{}D_{\mathrm{kern}}`$ forward applications. From a practical viewpoint, a clear advantage is the ease of implementation of the โfiltered overlapโ โ everyone with a running overlap code has it (in disguise).
### Acknowledgments
It is a pleasure to thank Ferenc Niedermayer and Tom DeGrand for useful conversation or correspondence. This paper was supported by the Swiss NSF.
## App. A: Cumulative eigenvalue distributions in 4D and 2D
In this appendix we discuss what can be learned from the cumulative eigenvalue distribution (CED). We consider both the eigenvalues in 4D generated for the main part of this paper and data from dedicated runs in the quenched Schwinger model (QED with massless fermions in 2D) to elucidate the effects that filtering and changing $`\beta `$ have on the spectral density of the hermitean Wilson operator $`H_\mathrm{W}=\gamma _5D_{\mathrm{W},1}`$.
Fig. 17 presents the cumulative eigenvalue distribution (CED) of the 15 smallest eigenvalues of $`|H_\mathrm{W}|`$ on the ensembles discussed before. We show it both in standard form and in double logarithmic form, and the scale on the ordinate follows from the requirement that it would extend up to 1, if all eigenvalues were calculated (cf. Fig. 18 below). For the two intermediate couplings ($`\beta =5.84,6.00`$) we see the expected linear rise of the CED near the origin, which soon gets complemented by a higher order piece. The coefficient of the linear part is a measure for the spectral density of the hermitean Wilson operator at the origin, $`\rho _{|H_\mathrm{W}|}(0)`$. That density being non-zero means that there is a finite probability to encounter arbitrarily small eigenvalues. The main effect of smearing is to reduce this spectral density, as is evident from the double logarithmic plots โ here the initial slope 1 piece gets shifted downwards, and this corresponds to a smaller coefficient in front of the linear piece in the standard representation. Our data at $`\beta =6.26`$ are of lesser quality โ here we definitely cannot identify a linearly dominated regime. The situation is far more favorable in Fig. 18 where quenched Schwinger model data are shown. Apart from the statistics, the main difference is that all eigenvalues (extending up to $`3`$ in 2D) are included. The higher the filtering level or $`\beta `$, the more pronounced is the โjumpโ in the CED at $`\lambda 1`$. Note that with a chiral kernel *all* eigenvalues of $`|\gamma _5D_{\mathrm{kern}}|`$ would be there, i.e. the CED would be a step function at $`\lambda =1`$. Finally, to come back to Fig. 17, the situation at the strongest coupling ($`\beta =5.66`$) is different, since here the linear piece in the unfiltered CED is *not* larger than in the filtered versions. This is, because our choice $`\rho =1`$ lets us โlooseโ the fermion โ at this coupling our projection point is somewhere โinโ the physical branch or โto the leftโ of it, while for the filtered version $`\rho =1`$ is still appropriate. One might avoid such a situation by choosing a larger $`\rho `$ with the unfiltered kernel, but an even safer option might be to refrain from simulating unfiltered overlap quarks on such coarse lattices. It looks like this is a situation where the filtered overlap may help a lot, since it allows simulations on coarser lattices than the unfiltered operator, but in order to really be useful such simulations should be in the scaling regime (and not just in the right universality class), and this is, of course, not yet clear.
The spectral properties of $`H_\mathrm{W}`$ play a role in the context of the physical interpretation of the Aoki phase . The latter is a conjectured phase, originally specific to $`N_f=2`$ active Wilson fermions at negative mass, in which after switching off an external trigger term
$$S_{\mathrm{source}}=\pm h\overline{\psi }\gamma _5\sigma _3\psi $$
(27)
parity and flavor break spontaneously and a condensate ($`\mathrm{const}0`$)
$$\underset{h0^\pm }{lim}\overline{\psi }\gamma _5\sigma _3\psi =\pm \mathrm{const}$$
(28)
forms. Good numerical evidence for a non-zero condensate (28) in the (dynamical) 2-flavor case for an appropriate choice of the negative mass $`\rho (\beta )`$ is found in . Ref. argues that in the massless limit of the continuum theory a condensate of the form (28) is simply an axial rotation of the usual (flavor diagonal) condensate and thus breaks neither parity nor flavor. They relate the spectral density of $`D_\mathrm{W}`$ to that of $`H_\mathrm{W}`$ and argue that the absence of a gap (around the origin) of the latter is indicative of chiral symmetry breaking and that $`\rho _{|H_\mathrm{W}|}(0)>0`$ if and only if (28) is non-zero. This was later elucidated to be a continuum argument , which โ in view of our Sec. 4 โ might be an important point.
The next issue is whether there is an Aoki phase in the quenched theory with 2 valence (but 0 sea) flavors . The simplest expectation is that qualitatively the picture with the 5 Aoki โfingersโ goes through, though the phase boundary is somewhat shifted w.r.t. the $`N_f=2`$ case.
Our 4D data in Fig. 17 clearly show the suppression of $`\rho _{|H_\mathrm{W}|}(0)`$ as one approaches the continuum, but we cannot see any sign that this distribution would vanish at some โcriticalโ coupling. Given the uniform pattern in the figures (apart from the scale on the $`y`$-axis they seem qualitatively similar), it seems more likely to us that $`\rho _{|H_\mathrm{W}|}(0)`$ will stay non-zero for arbitrary couplings.
To test this view, we analyze the quenched Schwinger model where high statistics can be reached. The couplings and geometries are chosen such as to have a fixed physical volume, with a box size about 5 times larger than the Compton wavelength of the lightest degree of freedom in the chiral limit of the $`N_f=1`$ theory. A survey of the parameters is given in Tab. 6 and for technical details we refer to .
Fig. 19 provides an overview over the complete $`|H_\mathrm{W}|`$ eigenvalue distribution; one sees a โpeakโ at $`\lambda =1`$ forming that gets more pronounced with higher $`\beta `$ and higher filtering level. This โpeakโ corresponds to the โjumpโ at $`\lambda =1`$ in the CED of $`|H_\mathrm{W}|`$ in Fig. 18.
Fig. 20 presents the distribution of the lowest eigenvalue of $`|H_\mathrm{W}|`$. At low $`\beta `$ this distribution accumulates at zero, at intermediate values of the coupling there is a horizontal band of eigenvalues connecting down to zero, and at the largest $`\beta `$ there are just scattered eigenvalues. Evidently, one cannot draw a final conclusion whether these scattered eigenvalues really make up for a non-zero $`\rho (0)`$, but it seems worth while to study this band in the region of $`\beta `$ values where it is clearly visible and see whether changing $`\beta `$ implies some structure, or whether it just stays flat, regardless of $`\beta `$.
Fig. 21 presents the CED in the area of interest, the very-low $`\lambda `$ region. At $`\beta =5.0,7.2`$ we show the data from the high-statistics run with 10 eigenvalues per configuration (bottom line of Tab. 6), but we checked that the results are consistent with what we get from the runs where all eigenvalues were determined. At a given coupling filtering clearly reduces $`\rho _{|H_\mathrm{W}|}(0)`$. The overall impression is that changing $`\beta `$ merely rescales the $`y`$-axis, in striking analogy with what we have seen in 4D (Fig. 17). If this is indeed true, the natural conclusion is that $`\rho _{|H_\mathrm{W}|}(0)>0`$ at any finite $`\beta `$ in the quenched theory.
Fig. 22 contains a summary of our determinations of the spectral densities $`\rho _{|H_\mathrm{W}|}(0)`$ for various $`\beta `$ and filtering levels, extracted from the initial slopes in the CED shown in Fig. 21. It looks like eventually the density decreases exponentially in $`\beta `$ and changing the filtering level amounts to an overall rescaling factor which is, in good approximation, independent of $`\beta `$. Obviously, this is just numerical evidence, but the message seems to be as clear as one can possibly hope for from a numerical experiment. Note that for practical reasons we cannot take the infinite volume limit, but given our physical box size we expect finite volume effects to be exponentially small.
We remind the reader that (both in 2D and in 4D) we were working at fixed negative mass $`m_0=1`$ and pushed towards the continuum line. In other words, we were trying to stay as far outside the Aoki phase as one can, if one wants to be in the supercritical region with one overlap fermion. Of course, our data do not exclude the existence of a critical $`\beta `$, but they favor the view that there is no $`\beta _{\mathrm{crit}}`$ that makes the $`|H_{\mathrm{W},1}|`$ spectral density strictly zero and therefore we conjecture that $`\rho _{|H_\mathrm{W}|}(0)>0`$ throughout the supercritical region. Still, we do not see why this would create a problem for the localization of the overlap operator, since the two seem not one-to-one inversely connected.
Finally, there is a simple argument that $`\rho _{|H_\mathrm{W}|}(0)>0`$ holds in all quenched or unquenched theories with a massive overlap determinant at all couplings. Acquire infinite statistics at $`\beta =0,N_f=0`$. When integrating over the full configuration space the spectral density is certainly non-zero. Results for the case of interest at finite $`\beta `$ and maybe finite $`N_f`$ can be obtained through reweighting. As long as one can guarantee that there is no configuration where the reweighting factor vanishes, the spectral density will be modified, but it cannot be made strictly zero. This holds true in the quenched case and in the dynamical theory with an overlap determinant ($`m0,2\rho `$), but it would not be true with Wilson fermions at a negative mass.
## App. B: Overlap operator locality in the free case
In this appendix we collect some technical points to make sure that a numerical investigation of the localization $`\nu `$ versus $`\rho `$ as shown in Fig. 12 for a $`48^4`$ lattice is not overwhelmed with finite size effects.
Considering $`f(d_1)`$ as given in (9) on a finite lattice one encounters a technical problem that is evident in Fig. 23. The free field case is far from showing rotational symmetry and the supremum function (9) has a couple of initial bumps (in particular at $`d_1=4,8,12`$), and this means that one needs to go to sufficiently large distances to measure the slope in a logarithmic representation. On the other hand, the choice to measure the distance in the 1-norm leads to rather large finite size effects for $`d_1>L`$, in particular the region near the maximal distance $`d_1=2L`$ is heavily contaminated. Therefore, we tried
$$\nu =\frac{1}{2}\mathrm{log}(f(L1)/f(L+1))$$
(29)
as a technical definition of the localization $`\nu `$ in (8). The comparison between the $`32^4`$ and $`48^4`$ geometries shows that our choice to evaluate the logarithmic derivative at $`d_1=L`$ produces rather consistent $`\nu `$ values, and we take this as a sign that they cannot be far from the asymptotic exponent. For the (two-digit-precision) projection parameter that we find to be optimal w.r.t. locality in the free case, $`\rho _{\mathrm{opt}}^{\mathrm{free}}0.54`$, the correlator is explicitly shown to be steeper than in the $`\rho =1`$ case.
|
warning/0506/nucl-th0506038.html
|
ar5iv
|
text
|
# References
Comparison of the extended linear sigma model and chiral perturbation theory
W.P. Alvarez<sup>(a,b)</sup>E-mail:alvarezwilson@hotmail.com, K. Kubodera<sup>(b)</sup>E-mail:kubodera@sc.edu, F. Myhrer<sup>(b)</sup><sup>ยง</sup><sup>ยง</sup>ยงE-mail:myhrer@sc.edu,
<sup>(a)</sup> Facultad de Ingenerรญa, Ciencias Fรญsicas y Matemรกticas, Universidad Central del Ecuador, Quito,Ecuador
<sup>(b)</sup>Department of Physics and Astronomy, University of South Carolina, Columbia,
SC 29208, USA
The pion-nucleon scattering amplitudes are calculated in tree approximation with the use of the extended linear sigma model (ELSM) as well as heavy baryon chiral perturbation theory (HB$`\chi `$PT), and the non-relativistic forms of the ELSM results are compared with those of HB$`\chi `$PT. We find that the amplitudes obtained in ELSM do not agree with those derived from the more fundamental effective approach, HB$`\chi `$PT.
The linear sigma model , which provides an illuminating example of spontaneous chiral symmetry breaking in strong interactions, has been studied extensively in the literature. Some of the consequences of this model, however, are known to be in conflict with observation. Notably, the isoscalar pion-nucleon ($`\pi N`$) scattering length predicted by the model is larger than the experimental value by an order of magnitude. Furthermore, the model predicts the axial coupling constant $`g_A`$ to be unity, whereas empirically $`g_A1.26`$. Despite the known limitations of the model, its simplicity has invited many authors to use it for exploring the consequences of chiral symmetry in nuclear physics; see e.g. Ref.. Nauenberg and Bjorken and Lee introduced an extended linear sigma model (ELSM) by adding a pair of extra terms (which jointly preserve chiral symmetry) to the original linear sigma model lagrangian. An important feature of ELSM is that $`g_A`$ is no longer restricted to be unity. Furthermore, via chiral rotations of the fields, ELSM leads to the non-linear chiral lagrangian of Weinberg (with $`g_A1`$) in the limit of an infinitely massive scalar field. Recently ELSM has been used to investigate the $`g_A`$ dependence of the $`\pi N`$ scattering lengths and the $`\pi N`$ sigma term $`\mathrm{\Sigma }_N`$ . It has been found in Ref. that ELSM can reproduce the very small experimental value of the $`\pi N`$ isoscalar scattering length, $`a_{\pi N}^{(+)}`$, and furthermore the same model can reproduce the large empirical value of the $`\pi N`$ sigma term, $`\mathrm{\Sigma }_N`$, without invoking any $`\overline{s}s`$ component of the nucleon.
Meanwhile, low-energy hadronic physics can be described by an effective field theory (EFT) of QCD known as โchiral perturbation theoryโ ($`\chi `$PT) . The $`\chi `$PT Lagrangian, $`_{\chi PT}`$, reflects the symmetries and the pattern of symmetry breaking of the underlying QCD. $`_{\chi PT}`$ is expanded in powers of $`Q/\mathrm{\Lambda }_\chi 1`$ where $`Q`$ denotes the typical four-momentum of the process in question or the pion mass, $`m_\pi `$, which represents the small explicit chiral symmetry breaking scale; $`\mathrm{\Lambda }_\chi 4\pi f_\pi `$ 1 GeV, is the chiral scale. The parameters appearing in $`_{\chi PT}`$, called the low-energy constants (LECโs), effectively subsume the high-energy physics that has been integrated out. These LECโs could in principle be determined from the underlying theory, but in practice they are fixed phenomenologically from experimental data. Once the LECโs are determined, $`_{\chi PT}`$ represents a complete and hence model-independent Lagrangian up to a specified chiral order. Furthermore, starting from $`_\chi `$, one can develop, for the amplitude of a given process, a well-defined perturbation scheme by organizing the relevant Feynman diagrams according to powers in $`Q/\mathrm{\Lambda }_\chi `$. If all the Feynman diagrams up to a given power, $`\nu `$, in $`Q/\mathrm{\Lambda }_\chi `$ are taken into account, then the results are model-independent up to this order, with the contributions of higher order terms suppressed by an extra power of $`Q/\mathrm{\Lambda }_\chi `$. A problem one encounters in extending $`\chi `$PT to the nucleon sector is that, as the nucleon mass $`m_\text{N}`$ is comparable to the cut-off scale $`\mathrm{\Lambda }_\chi `$, a straightforward application of expansion in $`Q/\mathrm{\Lambda }`$ becomes difficult . This difficulty can be circumvented by employing heavy-baryon chiral perturbation theory (HB$`\chi `$PT) , which essentially consists in shifting the reference point of the nucleon energy from 0 to $`m_\text{N}`$ and in integrating out the small component of the nucleon field as well as the anti-nucleonic degrees of freedom. An effective Lagrangian in HB$`\chi `$PT therefore involves as explicit degrees of freedom the pions and the large components of the redefined nucleon field. The expansion parameters in HB$`\chi `$PT are $`Q/\mathrm{\Lambda }_\chi `$, $`m_\pi /\mathrm{\Lambda }_\chi `$ and $`Q/m_\text{N}`$. Since $`m_\text{N}\mathrm{\Lambda }_\chi `$, it is convenient to combine chiral and heavy-baryon expansions and introduce the chiral index $`\overline{\nu }`$ defined by $`\overline{\nu }=d+(n/2)2`$. Here $`n`$ is the number of fermion lines that participate in a given vertex, and $`d`$ is the number of derivatives (with $`m_\pi `$ counted as one derivative). A similar power counting scheme can also be introduced for Feynman diagrams as well . HB$`\chi `$PT has been used with great success to the one-nucleon sector, see, e.g., Ref. .
We therefore consider it informative to compare the predictions of ELSM with those of HB$`\chi `$PT. As an example of this comparison, we consider here the tree-level $`\pi N`$ scattering amplitudes calculated in ELSM and HB$`\chi `$PT to lowest order corrections in $`Q/\mathrm{\Lambda }_\chi `$.
The lagrangian of the extended sigma model (ELSM) consists of the standard linear sigma model lagrangian plus two pion-nucleon interaction terms with a common coupling constant proportional to $`(g_A1)`$. The additional terms are a vector- and a pseudo-vector coupling term . Thus the lagrangian of ELSM reads
$``$ $`=`$ $`\overline{\psi }i\psi g\overline{\psi }\left[\sigma +i\gamma _5\stackrel{}{\pi }\stackrel{}{\tau }\right]\psi +{\displaystyle \frac{1}{2}}\left[(_\mu \sigma )^2+(_\mu \stackrel{}{\pi })^2\right]`$ (1)
$`+{\displaystyle \frac{1}{2}}\mu _0^2\left[\sigma ^2+\stackrel{}{\pi }^2\right]{\displaystyle \frac{\lambda }{4}}\left[\sigma ^2+\stackrel{}{\pi }^2\right]^2+_{\chi sb}`$
$`+\left({\displaystyle \frac{g_A1}{f_\pi ^2}}\right)\left[\left(\overline{\psi }\gamma _\mu {\displaystyle \frac{\stackrel{}{\tau }}{2}}\psi \right)\left(\stackrel{}{\pi }\times ^\mu \stackrel{}{\pi }\right)+\left(\overline{\psi }\gamma _\mu \gamma _5{\displaystyle \frac{\stackrel{}{\tau }}{2}}\psi \right)\left(\sigma ^\mu \stackrel{}{\pi }\stackrel{}{\pi }^\mu \sigma \right)\right],`$
where the parameters $`\lambda `$ and $`\mu _0`$ are assumed to be real and positive. The last line proportional to $`(g_A1)`$ represents the additional $`\pi N`$ coupling terms introduced in . As for the explicit chiral symmetry breaking term $`_{\chi sb}`$, we consider three terms (see e.g., Refs.):
$`_{\chi sb}`$ $`=`$ $`\epsilon _1\sigma \epsilon _2\stackrel{}{\pi }^2\epsilon _3\overline{\psi }\psi `$ (2)
The first term is the โstandardโ chiral symmetry breaking term in the linear sigma model, while the second term arises naturally in $`\chi `$PT. The third term proportional to $`\epsilon _3`$ was discussed in, e.g. , and we remark that a term proportional to $`\overline{\psi }\psi `$ appears in $`\chi `$PT with a coefficient proportional to $`m_\pi ^2c_1`$, where $`c_1`$ is a low-energy constant in $`\chi `$PT .
As usual, we redefine the scalar field relative to its vacuum expectation value, $`<\sigma >_0=f_\pi `$, and introduce the new scalar field $`s`$ defined by $`s=\sigma f_\pi `$. The requirement that the energy is minimum for $`<\sigma >_0=f_\pi `$ gives the following relation $`\mu _0^2\lambda f_\pi ^2=\epsilon _1/f_\pi `$ . The pion mass is found to be
$`m_\pi ^2=\epsilon _1/f_\pi +\mathrm{\hspace{0.33em}2}\epsilon _2.`$ (3)
In what follows we evaluate the $`\pi N`$ scattering amplitude using the lagrangian in Eq.(1) properly modified to account for the redefinition of the scalar field, $`\sigma `$ $``$ $`s`$, explained above.
The $`\pi N`$ scattering $`T`$-matrix is conventionally written as
$`T_{\alpha \beta }`$ $`=`$ $`T^{(+)}\delta _{\alpha \beta }+T^{()}{\displaystyle \frac{1}{2}}[\tau _\alpha ,\tau _\beta ]`$ (4)
where $`\alpha `$ and $`\beta `$ are the initial and final pion isospin indices, respectively, and $`T^{(\pm )}`$ are defined as
$`T^{(\pm )}=A^{(\pm )}+B^{(\pm )}{\displaystyle \frac{1}{2}}\gamma _\mu (k_1^\mu +k_2^\mu )`$ (5)
Here $`k_1`$ and $`k_2`$ are the incoming and outgoing pion momenta, respectively, in the center-of-mass system. It is understood that, in order to obtain the scattering amplitude, $`T_{\alpha \beta }`$ should be sandwiched between the relevant Dirac spinors (which however are suppressed in Eq.(4)). To compare the $`\pi N`$ amplitudes evaluated in ELSM with the ones obtained in HB$`\chi `$PT, we have to treat the nucleons in ELSM as heavy, non-relativistic fields of mass $`m_N`$. Therefore, the ELSM amplitudes, $`A^{(\pm )}`$ and $`B^{(\pm )}`$, in Eq.(5) and the Dirac spinors describing the initial and final nucleons in ELSM need to be expanded in powers of $`1/m_N1/M`$. The corresponding non-relativistic $`\pi N`$ scattering amplitudes, $`g^{(\pm )}`$ and $`h^{(\pm )}`$, are customarily defined by
$`\stackrel{~}{T}_{\alpha \beta }`$ $`=`$ $`[g^{(+)}+i\stackrel{}{\sigma }(\stackrel{}{k}_1\times \stackrel{}{k}_2)h^{(+)}]\delta _{\alpha \beta }+[g^{()}+i\stackrel{}{\sigma }(\stackrel{}{k}_1\times \stackrel{}{k}_2)h^{()}]{\displaystyle \frac{1}{2}}[\tau _\alpha ,\tau _\beta ]].`$ (6)
It is understood here that $`\stackrel{~}{T}_{\alpha ,\beta }`$ is to be sandwiched between the initial and final nucleon Pauli spinors and iso-spinors to yield the scattering amplitude. The amplitudes, $`g^{(\pm )}`$ and $`h^{(\pm )}`$, calculated in ELSM and HB$`\chi `$PT are denoted by $`g_{ESM}^{(\pm )}`$, $`h_{ESM}^{(\pm )}`$, $`g_{\chi PT}^{(\pm )}`$ and $`h_{\chi PT}^{(\pm )}`$, respectively. Comparison between $`g_{ESM}^{(\pm )}`$ and $`g_{\chi PT}^{(\pm )}`$ and between $`h_{ESM}^{(\pm )}`$ and $`h_{\chi PT}^{(\pm )}`$ is our main concern in what follows.
In Ref. the elastic $`\pi N`$ scattering amplitude in ELSM was calculated in the tree approximation and the expressions for the four amplitudes, $`A^{(+)}`$, $`A^{()}`$, $`B^{(+)}`$ and $`B^{()}`$, are given in Eqs.(17a-d) of Ref..<sup>1</sup><sup>1</sup>1We remark that the overall sign of the amplitude $`B^{()}`$, Eq.(17d) in , should be changed. The corresponding $`g_{ESM}^{(\pm )}`$ and $`h_{ESM}^{(\pm )}`$ amplitudes were derived in Ref.. The $`\pi N`$ scattering amplitude in HB$`\chi `$PT was evaluated by, e.g., Meissner et al.. For our present purposes, we only need the tree approximation amplitudes. The amplitudes, $`g_{\chi PT}^{(\pm )}`$ and $`h_{\chi PT}^{(\pm )}`$, were rederived in Ref. and it has been confirmed that in the tree approximation the results agree with those of Ref..
As mentioned earlier, Weinbergโs non-linear sigma model can be derived from ELSM in the limit of $`m_\sigma \mathrm{}`$. Therefore, to facilitate comparison with the HB$`\chi `$PT expressions, the amplitudes obtained in ELSM are further simplified by assuming that $`m_\sigma `$ is heavy compared to the pion mass and energy and expanding the amplitudes in powers of $`1/m_\sigma `$. We assume that $`m_\sigma `$ $`M`$ $`\mathrm{\Lambda }_\chi 1`$ GeV, whereas $`m_\pi ,\omega `$ and $`\sqrt{t}`$ are of order $`Q\mathrm{\Lambda }_\chi `$. We also assume that the chiral symmetry breaking parameters $`\epsilon _i`$ are of order $`Q^2`$. We restrict our comparison to the lowest powers of $`Q`$ in each amplitude, and we use the fact that the LECs, $`c_i`$ ($`i=1,2,3`$), in HB$`\chi `$PT are of the natural order of magnitude ($`c_i\mathrm{\Lambda }_\chi 1`$).
In comparing the amplitudes obtained in the two approaches under consideration, we find it convenient to introduce the following decompositions:
$`g_{ESM}^{(\pm )}=\stackrel{~}{g}^{(\pm )}+\delta g_{ESM}^{(\pm )},g_{\chi PT}^{(\pm )}=\stackrel{~}{g}^{(\pm )}+\delta g_{\chi PT}^{(\pm )},`$ (7)
$`h_{ESM}^{(\pm )}=\stackrel{~}{h}^{(\pm )}+\delta h_{ESM}^{(\pm )},h_{\chi PT}^{(\pm )}=\stackrel{~}{h}^{(\pm )}+\delta h_{\chi PT}^{(\pm )}.`$ (8)
In the above, $`\stackrel{~}{g}^{(\pm )}`$ represents the part that has a common analytic expression between $`g_{ESM}^{(\pm )}`$ and $`g_{\chi PT}^{(\pm )}`$ , whereas $`\delta g_{ESM}^{(\pm )}`$ and $`\delta g_{\chi PT}^{(\pm )}`$ represent the parts that do not have common analytic expressions. Similarly for $`\stackrel{~}{h}^{(\pm )}`$. The terms common between ELSM and HB$`\chi `$PT are given by
$`\stackrel{~}{g}^{(+)}`$ $`=`$ $`\left({\displaystyle \frac{g_A^2}{f_\pi ^2}}\right)\left({\displaystyle \frac{2\omega ^2m_\pi ^2\omega ^4+(\stackrel{}{k}_1\stackrel{}{k}_2)^2}{4M\omega ^2}}\right)`$
$`\stackrel{~}{g}^{()}`$ $`=`$ $`\left({\displaystyle \frac{\omega }{2f_\pi ^2}}\right)\left({\displaystyle \frac{g_A^2}{f_\pi ^2}}\right){\displaystyle \frac{\stackrel{}{k}_1\stackrel{}{k}_2}{2\omega }}+{\displaystyle \frac{1}{f_\pi ^2}}\left({\displaystyle \frac{\omega ^4m_\pi ^2\omega ^2+\omega ^2(\stackrel{}{k}_1\stackrel{}{k}_2)}{4M\omega ^2}}\right)`$
$`+`$ $`\left({\displaystyle \frac{g_A^2}{f_\pi ^2}}\right){\displaystyle \frac{1}{4M\omega ^2}}\left[2\omega ^4+2\omega ^2m_\pi ^2m_\pi ^2(\stackrel{}{k}_1\stackrel{}{k}_2)\omega ^2(\stackrel{}{k}_1\stackrel{}{k}_2)+(\stackrel{}{k}_1\stackrel{}{k}_2)^2\right]`$
$`\stackrel{~}{h}^{(+)}`$ $`=`$ $`{\displaystyle \frac{g_A^2}{2\omega f_\pi ^2}}\left({\displaystyle \frac{g_A^2}{f_\pi ^2}}\right)\left({\displaystyle \frac{\omega ^2+m_\pi ^2\stackrel{}{k}_1\stackrel{}{k}_2}{4M\omega ^2}}\right)`$
$`\stackrel{~}{h}^{()}`$ $`=`$ $`{\displaystyle \frac{1}{f_\pi ^2}}\left({\displaystyle \frac{\omega ^2g_A^2(\stackrel{}{k}_1\stackrel{}{k}_2)}{4M\omega ^2}}\right)`$ (9)
As for $`\delta g^{(\pm )}`$ and $`\delta h^{(\pm )}`$, our ELSM calculation leads to the following results:
$`\delta g_{ESM}^{(+)}`$ $`=`$ $`{\displaystyle \frac{M}{f_\pi ^2}}\left\{{\displaystyle \frac{\left[tm_\pi ^2+2\epsilon _2\right]}{m_\sigma ^2}}{\displaystyle \frac{\epsilon _3}{M}}+๐ช(M^2,m_\sigma ^2)\right\}`$ (10)
$`\delta g_{ESM}^{()}`$ $`=`$ $`{\displaystyle \frac{g_A}{f_\pi ^2}}\left\{{\displaystyle \frac{\epsilon _3\left(t2m_\pi ^2\right)}{2M\omega }}+\mathrm{}\right\}`$ (11)
$`\delta h_{ESM}^{(+)}`$ $`=`$ $`{\displaystyle \frac{1}{f_\pi ^2}}\left({\displaystyle \frac{tm_\pi ^2+2\epsilon _2}{4Mm_\sigma ^2}}{\displaystyle \frac{g_A\epsilon _3}{M\omega }}\right)+\mathrm{}`$ (12)
$`\delta h_{ESM}^{()}`$ $`=`$ $`0.`$ (13)
Meanwhile, an HB$`\chi `$PT calculation (in tree approximation) gives, to the order $`Q^2/\mathrm{\Lambda }_\chi ^2`$, the following results.
$`\delta g_{\chi PT}^{(+)}`$ $`=`$ $`\left(c_3\left[t2m_\pi ^2\right]+4m_\pi ^2c_12\omega ^2c_2\right){\displaystyle \frac{1}{f_\pi ^2}}`$ (14)
$`\delta g_{\chi PT}^{()}`$ $`=`$ $`0`$ (15)
$`\delta h_{\chi PT}^{(+)}`$ $`=`$ $`0`$ (16)
$`\delta h_{\chi PT}^{()}`$ $`=`$ $`{\displaystyle \frac{c_4}{f_\pi ^2}},`$ (17)
where $`t`$ = $`(k_1k_2)^2`$ is the four-momentum transfer .
We now discuss to what extent the sigma model (ELSM in our case) simulates the effective field theory (here HB$`\chi `$PT). We first look at the results for $`g^{(+)}`$. It is informative to examine what values ELSM gives to the LECs, $`c_i`$ ($`i=1,\mathrm{},4`$), that appear in HB$`\chi `$PT. To this end, let us impose the requirement $`\delta g_{ESM}^{(+)}=\delta g_{\chi PT}^{(+)}`$. Comparison of the momentum transfer ($`t`$) and energy ($`\omega `$) dependences in Eq.(10) and Eq.(14) leads us to identify
$`c_3{\displaystyle \frac{M}{m_\sigma ^2}},c_1{\displaystyle \frac{M}{m_\sigma ^2}}\left({\displaystyle \frac{m_\pi ^2+2\epsilon _2}{4m_\pi ^2}}\right){\displaystyle \frac{\epsilon _3}{4m_\pi ^2}}\mathrm{and}c_2=0.`$ (18)
The result $`c_2=0`$ means that ELSM fails to generate the energy-dependence of $`g^{(+)}`$ required by HB$`\chi `$PT. With the use of the sigma-meson mass scale, $`m_\sigma M1`$GeV, we find $`c_3c_11`$GeV<sup>-1</sup>. These results are not inconsistent with those found in Ref. based on the resonance saturation assumption. There it was shown that the $`\mathrm{\Delta }`$-resonance gives a major contribution to $`c_2`$ and $`c_3`$, whereas the empirical value of $`c_1`$ can be explained by a scalar resonance contribution in the two-pion channel. Furthermore, this scalar-meson resonance was found to give a $``$30% contribution to $`c_3`$ . These features are compatible with our finding that ELSM, which contains no $`\mathrm{\Delta }`$-field, leads to $`c_2=0`$ and to the value of $`c_3`$ that is significantly smaller than the empirically determined value. As for $`g^{()}`$, we notice that $`\delta g_{ESM}^{()}`$ in Eq.(11) is of order $`๐ช(Q^3)`$, i.e., this amplitude has no terms of order $`๐ช(Q^2)`$. This feature is consistent with Eq.(15).
Regarding the โspin-flipโ amplitudes $`h^{(\pm )}`$, we note that $`h^{(\pm )}`$ in Eq.(6) are accompanied by a factor of $`๐ช(Q^2)`$. This means that, to the chiral order under consideration, the comparison of the ELSM and HB$`\chi `$PT results for $`h^{(\pm )}`$ should be limited to the $`๐ช(1)`$ terms. Comparison between $`\delta h_{ESM}^{()}`$ in Eq.(13) and $`\delta h_{\chi PT}^{()}`$ in Eq.(17) leads to the conclusion that $`c_4=0`$. This implies that ELSM cannot generate the iso-vector, spin-dependent term in the $`\pi N`$ scattering amplitude predicted by HB$`\chi `$PT. Again we refer to Ref. , where it is shown that the empirical value of $`c_4`$ can be explained, within the resonance saturation assumption, by dominant contributions from the $`\mathrm{\Delta }`$-resonance and the $`\rho `$-meson. Since none of these hadrons are included in ELSM, it should come as no surprise that $`c_4=0`$ in ELSM. As for $`\delta h^{(+)}`$, Eq.(12) indicates that $`\delta h_{ESM}^{(+)}`$ is of $`๐ช(Q)`$, i.e., it has no contribution of $`๐ช(1)`$. This feature is consistent with the fact that HB$`\chi `$PT generates no $`\delta h^{(+)}`$ amplitude of chiral orders lower than $`Q^2`$ \[see Eq.(16)\].
We remark en passant that, if we take the limit $`m_\sigma \mathrm{}`$ and require $`\epsilon _3=0`$, then ELSM leads to $`c_1=c_2=c_3=c_4=0`$, and ELSM and HB$`\chi `$PT give identical tree approximation $`\pi N`$ scattering amplitudes, $`g_{ESM}^{(\pm )}=g_{\chi PT}^{(\pm )}=\stackrel{~}{g}^{(\pm )}`$, and $`h_{ESM}^{(\pm )}=h_{\chi PT}^{(\pm )}=\stackrel{~}{h}^{(\pm )}`$. This is however a very special case.
The above comparison indicates that, in general, the ELSM fails to reproduce some of the $`\pi N`$ scattering amplitude properties, (e.g., the energy dependence) that are predicted by HB$`\chi `$PT. It is well known that $`\mathrm{\Delta }`$ degrees of freedom play a role in describing $`\pi N`$ scattering even at very low energies. In HB$`\chi `$PT, although only the pion and nucleon are explicit degrees of freedom, the effects of the $`\mathrm{\Delta }`$-resonance are subsumed in the LECs, $`c_2,c_3`$ and $`c_4`$. By contrast, the $`\mathrm{\Delta }`$-field is normally not included in sigma models. This difference seems to be the main cause of the failure for ELSM to reproduce the HB$`\chi `$PT results. A lesson we learn from the present study is that, although the linear sigma model (either in its original version or in the form of ELSM) is often used as a convenient tool for exploring consequences of chiral symmetry in nuclear physics, conclusions obtained in such studies should be taken with caution.
W.P.A thanks T.-S. Park for many useful suggestions. We are grateful to the referee for his/her extremely useful comments and for drawing our attention to the results given in Ref.. This work is supported in part by the US NSF Grant No. PHY-0140214.
|
warning/0506/physics0506130.html
|
ar5iv
|
text
|
# H-theorem for classical matter around a black hole
## Acknowledgments
P. N. is supported by the Ministero dellโIstruzione, dellโUniversitร e della Ricerca (MIUR) via the Programma PRIN 2004: โMetodi matematici delle teorie cineticheโ.
|
warning/0506/hep-th0506219.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Recently, a very interesting relation between four-dimensional $`๐ฉ=2`$ supersymmetric BPS black holes and topological strings has been proposed . This correspondence relates the black hole free energy in type IIB string theory, compactified on a CalabiโYau threefold, to the topological and anti-topological string amplitudes on this same manifold, according to
$$_{BH}=_{top}+\overline{}_{top}.$$
(1)
In this relation, the complex structure moduli of the CalabiโYau are fixed in terms of the black hole charges by certain attractor equations.
Topological strings are often related to matrix models. It is well known that the topological theory on the conifold is perturbatively equivalent to $`c=1`$ bosonic non-critical string theory at self-dual radius, and hence to a matrix model . In fact, there are quite general correspondences between matrix models and topological strings on non-compact CalabiโYaus based on ground ring considerations .
Usually, the advocated correspondence relates the matrix model free energy $`_{MM}`$ directly to the topological string amplitude $`_{top}`$. However, as is by now well known, the exponential of $`_{top}`$ should not be viewed as a partition function, but rather as a wave function . Thus, it seems unnatural to relate $`_{MM}`$ directly to $`_{top}`$. If $`_{MM}`$ is a true free energy one should rather make the identification<sup>1</sup><sup>1</sup>1Note that this does not mean that the identification of the $`c=1`$ free energy with the topological string amplitude is wrong, since in that case one divides the natural (and real) free energy by two so to make it agree with 2d space-time calculations. However, the result is only true for a particular choice of polarization of the phase space $`H^3(CY)`$.
$$_{MM}=_{top}+\overline{}_{top},$$
(2)
thereby directly relating the free energy of the matrix model to the one for the black hole. We will see many indications in this paper that this is the right way to think of the relation between matrix models and topological strings. In particular, as we will show in an explicit example in section 3, the black hole free energy resulting from a deformed conifold with complex deformation parameter is precisely given by the free energy of the 0A matrix model at the self-dual radius. From the construction we present in section 2, the generalization to $`n`$ times the self-dual radius, and its interpretation in terms of $`n`$ conifolds, is straightforward<sup>2</sup><sup>2</sup>2This will also serve to clarify some of the results in , where the issue of associating specific radii to a certain number of conifolds was discussed.. The underlying property of the 0A model which makes this interpretation natural is its holomorphic factorization, as we will discuss in section 2.
Note that the identification of the 0A model at self-dual radius with a single conifold is different from what one naively may expect from ground ring relations. The reason for this is exactly the holomorphic factorization, which forces us to look at a manifold which in a sense is also the โholomorphic square rootโ of the one given by the defining relation of the matrix model ground ring. By studying the 0A and 0B matrix models at various radii, and investigating their relation to the $`c=1`$ matrix model, we show how these new geometries are constructed.
The procedure also gives new insights on the genus expansions of the matrix model free energy. In particular there will be a crucial mixing between genus 0 and genus 1 terms. The result of the mixing is that the genus 1 term in the expansion of the free energy will have the numerical coefficient $`1/12`$ for both the holomorphic and the anti-holomorphic term. As explained in , and later pointed out in , this is what is required for string theory to resolve the singularity associated to the shrinking of a cycle in the geometry. Again, this result is different from what one may expect from a naive ground ring analysis.
We then turn to the correspondence between matrix models and black holes. As has been argued recently , the relations mentioned above allow for a description of the black hole entropy<sup>3</sup><sup>3</sup>3There has been some discussion recently on the question whether the quantity calculated in should really be called an entropy, or rather an index. We will use the term โentropyโ throughout this paper, but the reader should be aware that this term is not to be taken too literally. in terms of the free energy of a matrix model. However, there are some important issues that need to be clarified in the proposed correspondence. Most importantly, in non-compact CalabiโYaus are considered as internal spaces for string compactification. When we view these as local models for compact CalabiโYaus, we would like to think of all parameters of the noncompact CalabiโYau as moduli. Hence, the manifold must have at least one A-cycle at infinity, and it is not straightforward to obtain the dependence of the black hole entropy on all the charges. In the present letter we remedy this deficiency by treating compact CalabiโYaus. It may sound strange that a matrix model can say something about string theory on a compact CalabiโYau. The reason this happens here is that one can find special charge configurations which, through the attractor equations of , result in a singular compactification space only at the horizon. This is enough to allow one to calculate the black hole free energy from the matrix model.
Thus, our internal space is a truly compact manifold which only near the black hole horizon โdecompactifiesโ into a conifold-like geometry. Note that this decompactification only involves quantities that are expressed in terms of the complex structure moduli. In particular, it seems perfectly possible to keep the Kรคhler volume of the Calabi-Yau finite throughout space-time. Since it is this volume which appears in the four-dimensional Newtonโs constant, we can really speak of four-dimensional gravity with nonzero coupling constant (and hence, for instance, of true black holes) in this context.
The paper is organized as follows. In section 2 we introduce the matrix models and describe the new geometrical interpretation of the 0A matrix model. We explain the crucial mixing of genus 0 and genus 1 terms that takes place at multiples of the self-dual radius, and its implications for which geometry one should consider. We also work out the case of fractional radii. In section 3 we discuss the correspondence between black holes and matrix models. We describe how the attractor equations can โdecompactifyโ a compact internal space on the black hole horizon. As an explicit example, we study a CalabiโYau with a conifold point. We match variables, Legendre transforms, and large classical terms on the matrix model and black hole sides. Finally, we summarize and discuss our results.
While this manuscript was being prepared for submission, we received the interesting paper , which discusses the 0A and 0B matrix models in a lot of detail. Also in that paper, the holomorphic factorization of the matrix model partition functions plays an important role.
## 2 The geometry of the 0A matrix model
This section describes the geometrical interpretation we propose for the 0A matrix model. Let us begin by explaining the matrix model nomenclature we use. In the eigenvalue description, the $`c=1`$ matrix model describes free fermions in an inverted harmonic oscillator potential, with the Fermi sea filled on one side of the potential. This model is nonperturbatively unstable due to tunnelling. The 0A matrix model<sup>4</sup><sup>4</sup>4The particular model we study was introduced and further studied in -, and is also known in the literature as the โdeformed matrix modelโ. It can be shown by integrating out eigenvalue phases that the gauged and holomorphic matrix model that is often used to describe the two-dimensional type 0A string theory is equivalent to this Hermitean matrix model. differs from this model by a term $`M/x^2`$ which is added to the potential. This deformation effectively removes one side of the potential, thus creating a stable model with one Fermi sea. One could also consider the undeformed matrix model with both sides of the potential filled. This corresponds to the 0B matrix model, which also is non-perturbatively stable. The 0A and 0B matrix models were constructed in , and their relations at different radii were discussed in detail in , to which we refer for further reading.
We will mainly be interested in matrix model free energies. Unless stated otherwise, the free energies are given in the grand canonical ensemble. We use the notation
$$_{MM}=2\pi \beta RF_{MM}=\mathrm{ln}๐ต_{MM},$$
(3)
where $`MM`$ can stand for โ$`c=1`$โ, โ0Aโ or โ0Bโ. $`๐ต_{MM}`$ and $`F_{MM}`$ are the usual partition function and free energy of the matrix model. In the case of the $`c=1`$ matrix model, we have
$$_{c=1}(\mu ,R)=\mathrm{Re}f(i\mu ,R)$$
(4)
with
$$f(i\mu ,R)=\underset{n,m=0}{}\mathrm{ln}\left(\frac{2n+1}{2}+\frac{2m+1}{2R}+i\mu \right).$$
(5)
The genus expansion becomes
$$_{c=1}(\mu ,R)=\frac{R}{2}\mu ^2\mathrm{ln}(\mu )\frac{1}{24}\left(R+\frac{1}{R}\right)\mathrm{ln}(\mu )+\mathrm{}$$
(6)
Throughout the paper, we use units where $`\alpha ^{}=1/2`$ on the 0A and 0B sides and $`\alpha ^{}=1`$ on the $`c=1`$ side. This means that the various self-dual radii are $`R_{SD}^A=\frac{1}{2},R_{SD}^B=1`$ and $`R_{SD}^{c=1}=1`$, respectively.
The expansion for the 0A free energy is given by
$`_{0A}(\mu ,R,q)`$ $`=`$ $`2\text{Re}\left[f({\displaystyle \frac{q+i\mu }{2}},2R)\right]`$ (7)
$`=`$ $`2\text{Re}[R(q/2+i\mu /2)^2\mathrm{ln}(q/2+i\mu /2)`$
$`{\displaystyle \frac{1}{24}}(2R+{\displaystyle \frac{1}{2R}})\mathrm{ln}(q/2+i\mu /2)+\mathrm{}]`$
where $`q`$ is related to the coefficient of the deformation term in the potential as $`q^2=M+1/4`$. In the corresponding two-dimensional string theory, $`q`$ is the net amount of D0-brane charge in the background. The above formula explicitly displays the holomorphic factorization mentioned in the introduction. In a very precise sense, the 0A partition function is the holomorphic square of the โcomplexifiedโ $`c=1`$ partition function.
Let us explain some of the perhaps strange-looking factors of $`i`$ in the above formulae. In the literature one usually encounters expansions in the parameters $`\mu `$ or $`\mu +iq`$. However, as emphasized in , it is really the sign in front of $`\mu `$ that changes when taking the complex conjugate. This may seem like an academic point since all the signs are going to be squared away anyway. However, it will be important when we construct the geometries, and natural later on when we are matching variables. Moreover, it *will* matter for the nonperturbative part of the theory .
### 2.1 The 0A matrix model at self-dual radius
By going to a double scaling limit, it has been shown that the free energy of the $`c=1`$ matrix model at self-dual radius is identical to the topological string amplitude on the conifold . The two terms in Eq. (6) then correspond to the genus 0 and genus 1 terms of the topological string. From Eq. (7) we also see that at self-dual radius, the free energy of the 0A matrix model is identical to the sum of the topological and anti-topological amplitudes of the conifold. We have the relation
$$_{0A}(\mu ,R_{SD}^A,q)=2\text{Re}\left[f(\frac{q+i\mu }{2},R_{SD}^{c=1})\right]=2\text{Re}\left[_{top}\left(\frac{q+i\mu }{2}\right)\right].$$
(8)
Thus, we relate the 0A matrix model at self-dual radius to the conifold, with the equation
$$uv+(\mu iq)=st.$$
(9)
Note that the 0A theory has enough real parameters to describe one complex modulus. Only in the above expression and similar ones that follow, in order to make contact with existing literature, we use the โconventionalโ notation where $`\mu `$ is the real part of the parameter and $`q`$ the imaginary part. Note that we can do this without loss of generality, since we can always for example rescale $`u`$ and $`s`$ by a factor of $`i`$.
Of course, since we take the real part of $`f`$, we could just as well have written it as a function of $`qi\mu `$, leading to a conifold of the form
$$uv+(\mu +iq)=st.$$
(10)
Now, we can see how these two manifolds are related to the usual ground ring geometry. It has been proposed that the ground ring equation for the 0A model at self-dual radius including both its deformations is
$$(uv+\mu )^2=stq^2.$$
(11)
We can rewrite this as
$$(uv+\mu iq)(uv+\mu +iq)=st.$$
(12)
It is useful to view this geometry as a fibration over the $`uv`$-plane. The fiber $`st=`$ const is a cylinder, except over the loci $`uv+(\mu \pm iq)=0`$, where it is the intersection of two complex planes in a single point. As is well-known (see the appendix of for a pictorial explanation) the complex structure moduli of the manifold are related to A-cycles which are localized near these loci, and B-cycles which start there and run off to infinity. Since for $`q`$ large these loci are far away from each other, the geometry of Eq. (12) then effectively reduces to two independent copies of the deformed conifolds we mentioned above. This is the intuitive reason why in the right polarization and perturbatively, the topological amplitude on the ground ring geometry and the sum of topological and anti-topological conifold amplitudes give the same result.
Since the 0A model is well defined beyond its perturbative expansion , it would clearly be interesting to further explore also its nonperturbative aspects at special radii.
### 2.2 The 0A matrix model at other radii
It is natural to ask whether the 0A matrix model at other radii corresponds to topological string theories on other singular geometries. It was argued in that the type II string compactified on a compact CalabiโYau with shrinking cycles is only non-singular if the corresponding topological string amplitude has a one-loop term coefficient of $`k/12,k`$. Eq. (7) shows that, apart from at self-dual radius, the 0A matrix model does not satisfy this requirement. Hence, it seems that we are in big trouble if we want to identify the general matrix models we consider with topological strings โ in particular if we would like these topological strings to live on double scaling limits of compact manifolds, as we will in the next section.
However, because of the double scaling limit the parameters in the model will be of the same order of magnitude as Planckโs constant, and it is not immediately obvious anymore that one can match the expressions on the matrix model and the topological string side genus by genus. In particular, it might be the case that the genus 0 term on the matrix model side contributes to the genus 1 term on the topological string side. Similar types of genus mixing have previously been considered in . Below we argue that this is the correct way of viewing the matrix model free energy.
As a motivation we show that, in order for the 0A expression to reduce to 0B as $`M0`$, we need to make such a reinterpretation of terms. If we plug in $`q=\frac{1}{2}`$ in Eq. (7) we get an expression
$$\begin{array}{c}\hfill _{0A}(\mu ,R,\frac{1}{2})=2\text{Re}\left[\frac{R}{4}(\frac{1}{4}\mu ^2)\mathrm{ln}(\mu +i\frac{1}{2})\frac{1}{24}(2R+\frac{1}{2R})\mathrm{ln}(\mu +i\frac{1}{2})+\mathrm{}\right],\end{array}$$
(13)
where we have skipped imaginary and analytic terms. We see that we cannot match the genus 0 (1) term in this expression to the genus 0 (1) term of the 0B free energy directly. For example, the self-dual radius of the 0B model is 1, while in this expression it appears to be $`1/2`$. However, if we move the $`\frac{R}{16}\mathrm{ln}(\mu +iq)`$ from the genus 0 term to the genus 1 term in Eq. (13), we indeed get the expression for half the 0B free energy<sup>5</sup><sup>5</sup>5We only get half since in the 0A theory, half of the states have to be removed .. Note that we also get the correct self-dual radius for 0B by making this shift.
We now turn to the reinterpretation of terms suitable for describing topological strings on CalabiโYau manifolds. To this end we use the formula of Gopakumar and Vafa for $`_{c=1}(R_{SD}^{c=1}/n)`$:
$$_{c=1}(\mu ,\frac{R_{SD}^{c=1}}{n})=\underset{k=(n1)/2}{\overset{(n1)/2}{}}_{c=1}(\frac{\mu ik}{n},R_{SD}^{c=1}).$$
(14)
Using Eq. (7) and going to $`n`$ times the self-dual radius, this can be recast into a formula for $`_{0A}`$:
$$_{0A}(\mu ,nR_{SD}^A,q)=2\text{Re}\left[\underset{k=(n1)/2}{\overset{(n1)/2}{}}f(\frac{(q+\frac{2k}{n}+i\mu )}{2},R_{SD}^{c=1})\right].$$
(15)
The 0A free energy at $`n`$ times the self-dual radius is thus given as two times the real part of the sum of $`n`$ $`c=1`$ free energies at self-dual radius. Since the coefficient in front of the 1-loop term of each $`f`$ is $`1/12`$, this immediately shows that we have succeeded in rearranging the terms so that they make sense from a type II string theory point of view. It also means, by the result of Ghoshal and Vafa , that it computes a sum of $`2\text{Re}_{top}`$ on $`n`$ conifolds.
We are now in a position to say something about the geometrical interpretation of the 0A matrix model at $`n`$ times the self-dual radius. It should correspond to a certain double scaling limit of the topological theory on a CalabiโYau with $`n`$ three-cycles that can shrink at different loci in moduli space. Call the distance to these loci $`t^k`$, $`k=\frac{n1}{2},\mathrm{},\frac{n1}{2}`$. Then the limit described by the 0A matrix model is $`t^k0`$ and $`g_{top}0`$ with
$$\frac{t^k}{g_{top}}=\left(q+\frac{2k}{n}\right)+i\mu $$
(16)
kept fixed<sup>6</sup><sup>6</sup>6Eq. (16) is the correct equation if the parameters $`t^k`$ are chosen so that all $`(t^k)^2\mathrm{ln}t^k`$ terms in $`_{top}`$ appear with the same coefficient, and up to an overall normalization.. Thus the CalabiโYau must allow for all cycles to shrink simultaneously. Note also that the matrix model, having only two parameters, describes a very special limit of this geometry.
There are of course many geometries satisfying these properties, but some have a more natural interpretation than others. As an example, let us work out the geometry in more detail for the case $`n=2`$. At twice the self-dual radius, the free energy is (recall that $`R_{SD}^{c=1}=2R_{SD}^A=1`$)
$`_{0A}(\mu ,1,q)`$ $`=2\text{Re}\left[f({\displaystyle \frac{q\frac{1}{2}+i\mu }{2}},1)+f({\displaystyle \frac{q+\frac{1}{2}+i\mu }{2}},1)\right]`$
$`=2\text{Re}\left[_{top}\left({\displaystyle \frac{q\frac{1}{2}+i\mu }{2}}\right)+_{top}\left({\displaystyle \frac{q+\frac{1}{2}+i\mu }{2}}\right)\right].`$ (17)
We see that there are two loci in parameter space where the corresponding CalabiโYau should have conifold singularities. Note that, again, this is half the number one would expect by a naive ground ring analysis. Following the arguments in , such a manifold can be created by modding out the conifold $`st=uv+(\mu iq)`$ by a $`_2`$, changing variables from $`(s^2,t^2)`$ to $`(s,t)`$, and deforming the resulting $`A_2`$-singularity by $`\pm \frac{1}{2}`$, leading to
$$\begin{array}{cc}\hfill st& =(uv+\mu i\left(q1/2\right))(uv+\mu i\left(q+1/2\right))\hfill \\ & =(uv+\mu )^22iq(uv+\mu )M.\hfill \end{array}$$
(18)
Just as at the end of the previous section, this procedure boils down to simply multiplying the equations for the loci of the single conifolds.
However, perturbatively we can rewrite Eq. (17) in several other ways, such as
$`_{0A}(\mu ,1,q)`$ $`=2\text{Re}\left[_{top}\left({\displaystyle \frac{q\frac{1}{2}+i\mu }{2}}\right)+_{top}\left({\displaystyle \frac{q+\frac{1}{2}i\mu }{2}}\right)\right]`$ (19)
$`=2\text{Re}\left[_{top}\left({\displaystyle \frac{q\frac{1}{2}+i\mu }{2}}\right)+_{top}\left({\displaystyle \frac{\left(q+\frac{1}{2}\right)+i\mu }{2}}\right)\right]+\text{non-pert.}`$
where in the first expression we have used $`\text{Re}[f(z)]=\text{Re}[f(\overline{z})]`$. The second expression can be easily verified by examining the genus expansion. The detailed form of the non-perturbative contribution was given in for the case of $`q=0`$. We might then conclude that the resulting manifold is given by
$`st`$ $`=\left(uv+\mu i\left(q1/2\right)\right)\left(uv+\mu +i\left(q+1/2\right)\right)`$
$`=\left(uv+\mu \right)^2+M+i\left(uv+\mu \right).`$ (20)
We claim that this latter manifold is the more natural one corresponding to the ground state of the Fermi sea. (By general ground ring arguments , one can argue that extra terms proportional to $`uv`$ and $`1`$ should correspond to excitations of the Fermi sea.) The reason for this is that we can now make contact with the higher genus analysis described in , which is equivalent to the Kodaira-Spencer description of the topological string . See also , where similar techniques are used. To do this we need to consider a superpotential given by
$$W=\frac{M}{D^2}+\left(\frac{\mu }{D}X+t_{2k}D^{2k1}\right)^2,$$
(21)
where $`[D,X]=i`$. For the free energy we need not consider the perturbations and we can put all $`t_{2k}=0`$. Hence, commuting everything to the right, and putting $`X=0`$, we find
$$W=\frac{M}{D^2}+\frac{\mu ^2}{D^2}i\frac{\mu }{D^2}.$$
(22)
We see that this matches the structure of the expression in Eq. (20) for $`uv=0`$ and up to an irrelevant complex conjugation. For further details on how to perform the higher genus calculations in this framework, see . It would of course be very interesting to see if the manifolds that are natural from the Kodaira-Spencer point of view also allow for a more natural embedding into truly compact CalabiโYaus.
For completeness, let us work out the case for fractional radii $`R^A=R_{SD}^A/n`$. In this case, the 0A free energy should be written
$`_{0A}(\mu ,{\displaystyle \frac{R_{SD}^A}{n}},q)`$ $`=2\text{Re}\left[f({\displaystyle \frac{q+i\mu }{2}},{\displaystyle \frac{R_{SD}^{c=1}}{n}})\right]`$
$`=2\text{Re}\left[{\displaystyle \underset{k=(n1)/2}{\overset{(n1)/2}{}}}f({\displaystyle \frac{q+2k+i\mu }{2n}},R_{SD}^{c=1})\right].`$ (23)
In terms of the CalabiโYau, a natural interpretation of this sum, which contains $`n`$ terms, is that the total charge $`q`$ and potential $`\mu `$ is associated to the $`n`$ conifolds in a specific way<sup>7</sup><sup>7</sup>7Running slightly ahead of the black hole part of our story, let us make the following interesting observation. For the $`l^{}`$th conifold, say, the attractor equations , fix the complex structure moduli (including the imaginary part ) at the horizon to
$$CX^{l+1}=\frac{1}{n}(2l+1+q+i\mu )1,$$
(24) where $`0ln1`$. It is interesting to compare this with the energy eigenvalues of the 0A matrix model :
$$E^l=i(2l+1+q+i\mu ).$$
(25) Given our identifications, and the fact that the topological partition function is peaked at the attractor value , a relation between the attractor fixed point values and the energy eigenvalues of the 0A matrix model is not unexpected..
## 3 Black hole entropy and compact CalabiโYaus
We now turn to the relation between matrix models and black hole entropy. In Ref. , the relation $`S_{BH}=F_{MM}/T_{MM}`$ is derived. $`F_{MM}`$ is in the canonical ensemble and $`T_{MM}=(2\pi R_{MM})^1`$ is the matrix model temperature.
Matrix models are usually related to topological strings on non-compact CalabiโYaus. In the case of a compact CalabiโYau the number of independent three-cycles is $`b^3=2(h^{(2,1)}+1)`$, and these are naturally divided into symplectic pairs of A- and B-cycles. The complex structure moduli space then has dimension $`h^{(2,1)}`$. It can be parameterized by the periods of the holomorphic $`(3,0)`$-form on $`h^{2,1}`$ A-cycles, or more invariantly by considering the periods on all $`(h^{2,1}+1)`$ A-cycles as projective coordinates. For more details, the reader is referred to for a review on the special geometry of CalabiโYaus.
We would like to think of the non-compact manifolds as local models for compact Calabi-Yaus, and of the periods of the $`n`$ A-cycles in these geometries as $`n`$ true complex structure moduli. This means there has to be at least one extra A-cycle โat infinityโ. In the formula $`S_{BH}=F_{MM}/T_{MM}`$, the left hand side is a function of the black hole charges. Since the number of electromagnetic charges of the four-dimensional black hole equals the total number of three-cycles, it is important to take the extra A-cycle(s) into account when obtaining the dependence of the black hole entropy on the charges in this framework.
In this section we derive an explicit correspondence between the 0A matrix model at self-dual radius and a four-dimensional half-BPS $`๐ฉ=2`$ black hole on a compact internal space with a conifold point. This involves finding charge configurations that, through the attractor equations , fix the moduli to a conifold point at the horizon. Let us however stress again that this โdecompactificationโ only takes place near the black hole horizon, and only for quantities that are sensitive to the complex structure moduli โ the Newtonโs constant being the most notable exception. We will identify variables, thermodynamical ensembles and double scaling limits on both sides, thus tying up one end left loose in Ref. . We also verify that the Legendre transforms, taking us from $`_{BH}`$ to $`S_{BH}`$ on the black hole side and from the grand canonical to the canonical ensemble on the matrix model side, coincide as proposed in Ref. . Finally, we explore the large classical contributions present on both sides.
For a review on compactification in the black hole context, see Ref. . In , two different ways to deal with the truly noncompact case by introducing cutoffs were discussed.
### 3.1 Charges and decompactification
Consider for simplicity a CalabiโYau $``$ with just one complex structure modulus. For example, one could think of $``$ as the mirror quintic, which has been thoroughly studied in . Extending the treatment to the general case is straightforward. We choose a symplectic basis $`A^I,B_I`$, $`I=0,1`$ of $`H_3(,)`$, which is such that the period of $`A^1`$ shrinks to zero at the conifold point. Let $`X^I`$ and $`F_I`$ be the periods of the holomorphic three-form on $`A^I`$ and $`B_I`$. Each pair of cycles leads to a four-dimensional gauge field, and hence to an electric and a magnetic charge. Our objective is to express the entropy of the black hole as a function of its electromagnetic charges $`q_I`$ and $`p^I`$.
Recall that the entropy is given by
$$S_{BH}(q_I,p^I)=_{BH}(\varphi ^I,p^I)\varphi ^J\frac{}{\varphi ^J}_{BH}(\varphi ^I,p^I),$$
(26)
where $`\varphi ^I`$ are the chemical potentials conjugate to $`q_I`$. $`_{BH}(\varphi ^I,p^I)`$ can be obtained from the topological partition function $`_{top}`$ on the CalabiโYau as
$$_{BH}(\varphi ^I,p^I)=2\text{Re}_{top}(t,g_{top}).$$
(27)
Here $`t=X^1/X^0`$ is a parameter on moduli space, and $`g_{top}`$ is the topological string coupling constant. The correspondence holds if $`_{top}`$ is evaluated at
$$t=\frac{p^1+i\varphi ^1/\pi }{p^0+i\varphi ^0/\pi },\text{ }g_{top}=\frac{\pm 4\pi i}{p^0+i\varphi ^0/\pi }.$$
(28)
We now want to compute $`_{BH}`$ using matrix model technology. To this end we use Eq. (7), and the fact that $`_{c=1}`$ equals the topological partition function on the conifold . To be more specific, in the double scaling limit $`t0`$, $`g_{top}0`$ at constant $`\mu _{top}t/g_{top}`$, the partition function is given by
$$_{c=1}(\mu =\mu _{top})=\mathrm{Re}_{top}(i\mu _{top}).$$
(29)
The equality (29) is valid only after appropriately fixing the gauge on the topological string theory side, and up to large classical terms, on which we will comment in a moment. Thus, Eqs. (7), (27) and (29) give $`_{BH}=_{0A}`$, upon identifying variables. This is done by computing the $`(q+i\mu )^2\mathrm{ln}[(q+i\mu )/\beta ]`$ contribution to the zero order term of $`_{0A}`$, and its counterpart $`(p^1+i\varphi ^1/\pi )^2\mathrm{ln}[(p^1+i\varphi ^1/\pi )/(p^0+i\varphi ^0/\pi )]`$ in $`_{BH}`$. Doing this carefully, using e.g. Eq. (2.16) of Ref. , yields the identification
$`\mu `$ $`\varphi ^1/\pi ,`$
$`q`$ $`p^1,`$
$`\beta `$ $`p^0+i\varphi ^0/\pi .`$ (30)
Note that $`\beta `$ is to be considered as an independent variable. It appears only in the the genus 0 and 1 contributions, exactly as $`p^0+i\varphi ^0/\pi `$. Let us stress that $`\mu `$ is identified with $`\varphi ^1`$, and not with $`p^1`$.
With the correspondence (3.1) we have $`_{BH}=_{0A}`$, and the fact that $`\mu \varphi ^1`$ shows that the Legendre transforms on both sides indeed match<sup>8</sup><sup>8</sup>8On the black hole side the transform really contains a $`\varphi ^0_{BH}/\varphi ^0`$ term which is not present on the matrix model side. Since $`\beta `$ only appears in the genus 0 and 1 terms, this contribution only contains non-universal terms. However, since we will be interested also in the non-universal terms, we choose the black hole charges in such a way that $`\varphi ^0=0`$ in what follows.. Thus we have explicitly verified the conclusion that $`S_{BH}=F_{0A}/T_{SD}`$ (Eq. (1.2) of Ref. ), where $`F_{0A}`$ is the canonical free energy of the 0A matrix model, and $`T_{SD}=1/\pi `$ is the self-dual temperature. To be precise, in this ensemble the identification of $`q`$ and $`\beta `$ is as in (3.1), and instead of $`\mu `$ we now have $`N`$, which is the number of fermions measured from the top of the $`x^2`$ part of the potential. This variable is to be identified with the electric charge $`q_1`$ of the black hole, and has expectation value $`N=\frac{1}{\pi }_{0A}/\mu `$. Having made the connection (3.1) we need to identify the double scaling limit on the black hole side. Indeed, Eq. (29) is only valid in that limit, and thus it is only in this limit that $`_{0A}`$ correctly computes the entropy. Eq. (28) shows that the appropriate limit is $`p^0+i\varphi ^0/\pi \mathrm{}`$ while $`p^1+i\varphi ^1/\pi `$ remains constant<sup>9</sup><sup>9</sup>9Note that this limit coincides exactly with the usual matrix model double scaling limit -.. Hence $`p^1`$ remains constant, and using the attractor equations, it is straightforward to show that at least two of $`p^0,q_0`$ and $`q_1`$ go to infinity. The attractor equations also give the following condition on these three charges:
$$p^0\text{Im}(F_0\overline{F}_1)+q_0\text{Im}(F_1\overline{X}^0)+q_1\text{Im}(X^0\overline{F}_0)=0,$$
(31)
where all periods are evaluated at $`t=0`$. When the charges satisfy these requirements, the CalabiโYau will be effectively non-compact at the black hole horizon.
### 3.2 Classical terms
Next, let us consider the large classical terms appearing in the black hole entropy. Computing the genus 0 contribution gives
$$_{BH}=2\text{Re}\left[\frac{\pi i}{4}A_1(CX^0)^2+\frac{\pi i}{2}A_2(CX^0)(CX^1)+\frac{1}{2}(\frac{CX^1}{2})^2\mathrm{ln}\frac{CX^1}{CX^0}+\mathrm{}\right],$$
(32)
where terms that are vanishing or finite and regular in the double scaling limit have been omitted. Here, $`CX^Ip^I+i\varphi ^I/\pi `$, and $`A_i`$ are numerical constants depending on the CalabiโYau. Explicitly $`A_1=(F_0/X^0)|_{t=0}`$ and $`A_2=(F_1/X^0)|_{t=0}`$. Note that the first two terms become large in the double scaling limit.
In principle, there are large classical contributions also to the matrix model free energy. Regularizing the potential of the $`c=1`$ matrix model $`Vx^2`$ according to
$$V\left(x\right)=\frac{x^2}{\alpha ^{}}+Ax^4,$$
(33)
gives, up to numerical factors, a grand canonical matrix model free energy of the form
$$_{c=1}(\mu ,\beta )\frac{1}{A^2}\beta ^2\frac{1}{A}\mu \beta \mu ^2\mathrm{ln}\frac{\mu }{\beta }+\mathrm{}$$
(34)
We see that since we can identify $`CX^0\beta `$, $`CX^1i\mu `$, the structure of this expression is in complete accordance with (32). For the example of the mirror quintic, we also checked that the sign of the leading term is the same in both equations. To precisely match the two undetermined coefficients in (32), one would need to consider a potential which is regularized by two coefficients, such as $`Vx^2+Ax^4+Bx^6`$.
## 4 Conclusions and outlook
In this paper, we have studied the correspondence between matrix models, topological strings and four-dimensional $`๐ฉ=2`$ half-BPS black holes. We have in particular studied the relations between these systems for the case of the 0A matrix model at multiples of its self-dual radius. When relating the matrix models to topological strings, we have argued that it is more natural to match the matrix model free energy to $`2\text{Re}_{top}`$ than to $`_{top}`$. Consequently, the geometry naturally associated to the matrix model at self-dual radius is the deformed conifold. This conifold can be viewed as the โholomorphic square rootโ of the manifold that follows from the ground ring equations. At multiples of the self-dual radius we again find such a holomorphicโanti-holomorphic factorization, and the matrix model should be associated with certain non-compact CalabiโYau manifolds with $`n`$ three-cycles that can shrink to zero volume. We found that it is plausible that such local geometries can be embedded in compact CalabiโYaus.
We noted that the matrix model free energies and the topological partition functions need not match each other genus by genus. In particular, a mixing of genus 0 and genus 1 terms will occur. This ensures that the coefficients in front of the genus 1 terms on the topological string side are always $`1/12`$ for both the holomorphic and the anti-holomorphic contributions, as required for the resolution of compact singular spaces by string theory.
Using the recently conjectured correspondence between topological strings and black holes in type IIB string compactification, we were able to directly relate the matrix model free energy to the one for the black hole. An important new point here was that the theory is compactified on a compact CalabiโYau, that through the attractor equations develops a singularity only at the black hole horizon. This allowed us to get a matrix model description of the black hole, in spite of the compactness of the CalabiโYau.
The relation was calculated explicitly for an internal space with a single conifold point. The variables matched perfectly and we saw that the Legendre transforms between the canonical and grand canonical ensemble on the matrix model side, and between $`S_{BH}`$ and $`_{BH}`$ on the black hole side were identical. It was also shown that the large classical terms in $`_{BH}`$ and $`_{MM}`$ can be matched in form by regulating the matrix model potential.
There has been much interest in the consequences of the topological string / black hole relation recently, and it seems that many more interesting results in this direction lie ahead. Let us mention some lines of further investigation related to the results of this paper. First of all, it would be extremely interesting to understand the nonperturbative corrections on the different sides of the story better. Many of the models that we have mentioned are perturbatively equivalent, but have nonperturbative differences. Studying these better, as well as their relations to black holes, may give us some intuition about the correct nonperturbative completion of topological string theory, and about the question of how unique such a completion is. A closer nonperturbative study may also lead to relations with the baby universes of . In this respect, the sums over different conifolds we have mentioned are also suggestive.
On a more technical level, the notion of the โholomorphic square rootโ of the ground ring geometry needs to be made more precise. On a case-by-case basis, the correct geometries are not hard to guess, but it seems that by using Kodaira-Spencer theory a more rigorous definition should also be possible.
Another interesting point to work out further would be the actual embedding of the local models into compact CalabiโYaus. Ultimately, this leads to the intriguing mathematical question of which local CalabiโYau manifolds allow an embedding into compact CalabiโYaus. Already in the two-moduli case this seems to be a very nontrivial issue.
Finally, we repeat an open question that was mentioned in : could we completely skip the topological string step and directly relate the matrix models to the black holes? Since the black holes have an $`AdS_2\times S^2`$ near-horizon region, one would expect the supergravity theory to be equivalent to a $`CFT_1`$ on the boundary of $`AdS_2`$. It seems natural to relate the two factors of the matrix model partition function to the two boundaries of this space. It would be interesting to make such a holographic description precise.
## Acknowledgments
UD is a Royal Swedish Academy of Sciences Research Fellow supported by a grant from the Knut and Alice Wallenberg Foundation. The work was supported by the Swedish Research Council (VR) and the Royal Swedish Academy of Science.
|
warning/0506/hep-th0506039.html
|
ar5iv
|
text
|
# On the stability of field-theoretical regularizations of negative tension branes
## I Introduction
Scalar fields propagating in extra dimensions have been widely investigated. A suitable combination of bulk and brane potentials can provide a nontrivial profile for a bulk field, which can serve many purposes. For instance, it can be employed for stabilizing the size of the extra space, as in the GoldbergerโWise gw mechanism; alternatively, it can be used to localize matter in a narrow region in the extra space, as originally done by Rubakov and Shaposhnikov rush . Although gauge fields are not localized in Ref. rush , this mechanism can be considered as a field theory regularization of a deltaโlike lower dimensional object, as for instance a brane of string theory. The field theory description can have interesting phenomenological consequences: for instance, it allows to localize fields of different families at slightly different positions in the bulk, so to explain the fermion mass hierarchy of the standard model ahsc ; if the scalar field can vary over cosmological times, it can allow for a greater baryon and CP violation at early times mpst , and overcome the obstacles for baryogenesis that characterize several models of extra dimensions sacha .
In these examples, the bulk scalar has standard kinetic terms. A few works discuss the possibility of more general kinetic terms. There are at least two reasons to be interested in this possibility. One is related to the idea of selfโtuning nemanja . In extra dimensions, a brane tension can give rise to a warping of the extra space, without inducing a $`4d`$ expansion rs1 . This typically requires a fineโtuning between the energies in the bulk and on the brane. However, one can hope that the fineโtuning can be avoided in some special cases. Ref. nilles presents a general theorem against this possibility, which however relies on the presence of bulk fields with standard kinetic terms. Mechanisms of self-tuning with fields with nonstandard kinetic terms were advanced for instance in Refs. kim ; holdom . In particular, Ref. holdom makes use of a scalar field with a quartic derivative term.
A second motivation is the attempt to resolve a negative tension brane in field theory. The kink solution of Ref. rush has positive energy density, and thus it can be regarded as a resolved positive tension brane. Ref. koka makes use of a scalar field with the โwrongโ sign for the kinetic term, and finds a kink profile with negative energy density. More in general, Ref. maxim shows that any attempt to resolve a negative tension brane through a bulk scalar necessary requires that this field is a ghost. In principle, one may hope that a ghost could also serve the purpose of selfโtuning, since the negative kinetic term may absorb a brane tension to restore $`4d`$ flatness. <sup>1</sup><sup>1</sup>1We thank Nemanja Kaloper for drawing our attention to this possibility.
Clearly, these proposals call for a discussion of their stability. For instance, the set-up of Ref. koka is clearly unstable, due to the presence of the ghost field. However, the situation is more subtle for the main model discussed in Ref. maxim . In this model, the kinetic term of the scalar field changes sign along the bulk. The profile of the field provides a resolution of both a positive and a negative brane, placed at two different locations in the bulk (where the kinetic term has the โcorrectโ and the โwrongโ sign, respectively). It is not obvious a priori which of these two regions controls the stability of the background. A calculation maxim , based on an effective $`4d`$ potential, suggests that the model has no tachyons. However, as also acknowledged in Ref. maxim , the stability against ghosts requires a more accurate calculation which was beyond the aims of that work. More in general, we find very interesting to discuss the stability of models with a kinetic term which is not sign definite. This is the main purpose of the present investigation.
The stability of a given background requires a careful general relativity treatment. The fluctuations of the scalar field source and mix with the (scalar) fluctuations of the geometry. Such calculations are standard in $`4d`$ cosmology, for what concerns the generation of inhomogeneities and growth of structures. This framework has been extended to extra dimensional models, both for discussing the stability, and the coupling of bulk fluctuations to brane fields (with a focus on accelerator phenomenology). A general formalism can be found in Ref. brandenberger . The application to radion phenomenology was first done in Ref. csaba , and then in several other works. However, the exact identification of the physical excitations, with the bulk scalarโgeometry mixing fully taken into account was performed only recently kmp .
We generalize the computation of Ref. kmp , valid for a bulk field $`\varphi `$ with standard kinetic terms, to the case of a kinetic term of the form $`K\left(\varphi \right)\left(\varphi \right)^2`$, where $`K`$ is an arbitrary function of the bulk scalar. Although we refer to Ref. kmp for some of the details, we try to keep the present discussion selfโcontained. We identify the exact $`4d`$ physical modes of the system. Both the kinetic and mass term for the various modes are obtained by decomposing the original action and integrating along the compact coordinate. We explicitly show that the portion of the bulk integral where $`K>0`$ gives a positive contribution to the kinetic coefficient of each mode, while the bulk region with $`K<0`$ gives a negative contribution. Whether a mode is or is not a ghost then depends on where it is mostly localized. <sup>2</sup><sup>2</sup>2It is worth remarking that all these calculations are semiclassical, based on linear quantum fluctuations on a given classical background. In a path integral formulation, one has to include any type of fluctuations. In particular, one is force to consider fluctuations with arbitrarily negative kinetic energy. However, in the absence of a rigorous path integral formulation, where gravity is also included, it is worth investigating whether the model is stable or if problems arise already at the semiclassical level.
The linearized computation becomes questionable around the points where $`K`$ vanishes. At the technical level, one can expect singular terms in the equations for the perturbations. This is not obvious a priori. For instance, Ref. finelli discusses a somewhat similar problem in scalar field $`4d`$ cosmology. In that case, the equation for the metric perturbation is singular when the kinetic term for the background inflaton vanishes, $`\dot{\varphi }=0`$. However, the equation for the MukhanovโSasaki musa variable, which properly identifies the physical mode of the system is regular. Also in the present case, one may hope that, due to the mixing with gravity, the kinetic term for the proper physical excitations may be regular where $`K=0`$. The computation shows that this is not the case.
We find that the singularity is mild enough that one can find normalizable modes in both the regions $`K0`$ and $`K0`$. However, it is also strong enough that these regions are not in communication; at the technical level, it is not possible to obtain junction conditions which relate the profile of a mode across the points where $`K=0`$. We show this explicitly for the model of Ref. maxim ; however, we also show that this a very general conclusion, irrespectively of the form of $`K`$. The safest interpretation is probably that the theory for the perturbations is ill defined due to the vanishing of the kinetic function, and that modifications (for instance, the introduction of higher derivative terms) are necessary to have a better defined quantum field theory. To confirm the separation of the two regions, we attempt to regularize the singularity through a cutโoff, and to investigate the behavior of the solutions as the cut-off is removed. Doing so, we find that the limiting solutions either vanish where $`K`$ is positive or where it is negative. This regularization explicitly shows that the eigenvalue problem which determines the bulk profile of the modes effectively splits into two eigenvalue problems, characterized by a different mass spectrum. Hence, a physical mode can only live in one of the two regions; the modes which have support where $`K<0`$ are ghosts, and preclude the stability of the background.
The plan of the paper is the following. In Section II we present the general computation for the perturbations, for a generic kinetic function $`K`$. In Section III we apply it to the main model of Ref. maxim . In Section IV we discuss how our findings generalize beyond this application.
## II General formalism
We start from the action for a scalar field $`\varphi `$ plus gravity,
$$S_{\mathrm{bulk}}=_0^{z_0}d^5x\sqrt{g}\left[\frac{M^3}{2}R\frac{K\left(\varphi \right)}{2}\left(\varphi \right)^2V\left(\varphi \right)\right],$$
(1)
which is defined on a compact and periodic extra dimension $`z`$. For the moment, we assume that there are no branes present, so that the calculation is simpler. We discuss below how the result changes when boundary branes are present. We note the presence of a nonstandard kinetic term, where we allow for an arbitrary function $`K\left(\varphi \right)`$. The standard case corresponds to $`K=1`$. More in general, $`\varphi `$ is a ghost whenever $`K<0`$. Finally, we note that a possible bulk cosmological constant is implicitly included in (1), by simply shifting the zero point energy of $`\varphi `$.
We are interested in background solutions which only depend on $`z`$, and with the factorizable geometry
$$ds^2=A\left(z\right)^2\left[dz^2+\eta _{\mu \nu }dx^\mu dx^\nu \right].$$
(2)
(notice $`z`$ is a conformal coordinate; also, notice we have chosen the mostly positive signature for the Minkowsky metric). It is straightforward to canonically normalize the scalar field, through the relation
$$\phi \sqrt{K}๐\varphi ,$$
(3)
so that the background Einstein equations are
$`6M^3{\displaystyle \frac{A^2}{A^2}}`$ $`=`$ $`{\displaystyle \frac{K}{2}}\varphi ^2A^2V,`$
$`{\displaystyle \frac{A^{\prime \prime }}{A}}`$ $`=`$ $`2{\displaystyle \frac{A^2}{A^2}}{\displaystyle \frac{K}{3M^3}}\varphi ^2.`$ (4)
These two equations can be combined to give the equation of motion for $`\varphi `$,
$$\varphi ^{\prime \prime }+3\frac{A^{}}{A}\varphi ^{}+\frac{K^{}}{2K}\varphi ^2\frac{A^2}{K}V^{}=0.$$
(5)
Prime on $`\varphi `$ or $`A`$ denotes differentiation with respect to $`z`$, while $`K^{}dK/d\varphi `$, and analogously for $`V`$. Although (3) is formally defined only where $`K`$ is positive, it is immediate to verify that eqs. (4)โ(5) hold in general.
Periodicity conditions supplement these equations, and allow to determine the background solution. As we mentioned, we are interested in the stability of the background against scalar perturbations. We introduce the perturbations of the bulk scalar, which we denote as $`\delta \varphi (x,z)`$. In addition, one can reduce the system of metric perturbations to a unique mode $`\mathrm{\Phi }(x,z)`$, which, in the $`5d`$ longitudinal gauge, appears as
$$ds^2=A\left(z\right)^2\left[\left(1+2\mathrm{\Phi }\right)dz^2+\left(1\mathrm{\Phi }\right)\eta _{\mu \nu }dx^\mu dx^\nu \right]$$
(6)
If $`K`$ is positive definite, one expects that also the calculation for the perturbations can be readily obtained from the standard one, upon the redefinition (3). However, as we mentioned in the Introduction, the situation is more delicate if $`K`$ can change sign. As in the standard case, the two perturbations $`\delta \varphi `$ and $`\mathrm{\Phi }`$ are related by constraint Einstein equations, and only one linear combination of them is dynamical. Among different possible dynamical variables, one is particularly convenient for the diagonalization of the action; it is the $`5d`$ generalization of the MukhanovโSasaki musa variable $`v`$, introduced for the study of scalar perturbations in $`4d`$ cosmology. In the present case, combining the analysis of kmp with the redefinition (3), we find
$$vZ\left(\frac{\mathrm{\Phi }}{2}\frac{A^{}}{A\varphi ^{}}\delta \varphi \right),$$
(7)
where
$$Z\sqrt{|K|}\frac{A^{5/2}\varphi ^{}}{A^{}}.$$
(8)
A lengthy but straightforward computation confirms that the action for the perturbations (obtained by expanding at second order in the perturbations the starting action (1)) acquires a particularly simple form in terms of the mode $`v`$,
$$S^{\left(2\right)}=\frac{1}{2}d^5x\sigma \left(K\right)v\left[\mathrm{}+\frac{d^2}{dz^2}\frac{Z^{\prime \prime }}{Z}\right]v,$$
(9)
where $`\sigma \left(K\right)`$ denotes the sign of $`K`$. It is worth noting that the use of the generalized MukhanovโSasaki variable automatically โrescales awayโ the function $`K`$ from the kinetic term, up to its sign. However, the potential problems with vanishing $`K`$ are โencodedโ in the fact that $`Z`$ vanishes for $`K=0`$, so that, unless there are cancellations with $`Z^{\prime \prime }`$, the effective potential for the perturbations is divergent. We already discussed this problem in the Introduction; in the next Section we will discuss how this problem can be dealt with in a specific example. Here, we simply proceed with the computation, by decomposing the $`5d`$ variable $`v`$ into KK modes,
$$v(x,z)=\underset{n}{}\stackrel{~}{v}_n\left(z\right)Q_n\left(x\right).$$
(10)
The modes $`Q_n`$ represent quantum fields in the $`4d`$ description of the model, while $`\stackrel{~}{v}_n`$ are the corresponding wave functions in the bulk. They are determined by separating the equation of motion for $`v`$ which follows from (9),
$$\left(\frac{d^2}{dz^2}\frac{Z^{\prime \prime }}{Z}+m_n^2\right)\stackrel{~}{v}_n=0.$$
(11)
This eigenvalue equation, together with the periodicity condition along the extra dimension, determines the bulk profiles $`\stackrel{~}{v}_n`$, as well as the spectrum of the theory, $`\left\{m_n^2\right\}`$. The exact profile found here is then employed to compute whether a mode is a ghost or not, as we show now.
As remarked in the Introduction, we are interested in the kinetic terms of the $`4d`$ modes. Due to the hermiticity of (9), different modes are orthogonal, and (9) separates into the sum of decoupled quadratic actions for each mode,
$$S^{\left(2\right)}=\underset{n}{}S_n^{\left(2\right)}=\frac{1}{2}\underset{n}{}C_nd^4xQ_n\left[\mathrm{}m_n^2\right]Q_n,$$
(12)
where (comparing with (9)) it is immediate to see that the coefficients $`C_n`$ are given by
$$C_n=๐z\sigma \left(K\right)\stackrel{~}{v}_n^2.$$
(13)
A further rescaling $`\widehat{Q}_nQ_n/\sqrt{|C_n|}`$ renders the mode canonically normalized. However, the sign of $`C_n`$ is not rescaled away by this redefinition, and it thus controls whether the mode $`\widehat{Q}_n`$ is a ghost or not. As we see, each coefficient $`C_n`$ gets positive contributions where $`K>0`$, and negative contributions where $`K<0`$. So, the nature of a mode is determined by whether its wave function is mostly localized where $`K>0`$, or $`K<0`$.
For completeness, we conclude this section by giving the result of the computation when also boundary terms are present. More specifically, the extra coordinate is assumed to lie on the compact interval $`0<z<z_0`$, delimited by two orbifold ($`Z_2`$ symmetric) boundary branes. We assume that the brane contains a $`\varphi `$dependent potential term,
$$S_{\mathrm{brane},i}=d^4x\sqrt{\gamma _i}\left\{2M^3\left[K\right]+U\left(\varphi \right)\right\}_i,$$
(14)
where $`\gamma _i`$ is the induced metric at the brane location, while $`\left[K\right]`$ denotes the jump of the trace of the extrinsic curvature across the brane. In principle, kinetic terms for $`\varphi `$ could be also present at the boundaries, and they would have the effect of modifying the kinetic term for the $`4d`$ modes. However, we will not consider them here.
The branes enforce boundary conditions, which replace the periodicity conditions considered so far. For the background, we have
$$U_i=\mathrm{\hspace{0.17em}6}M^3\frac{A^{}}{A^2}|_i,U_i^{}=\pm \frac{2K\varphi ^{}}{A}|_i,$$
(15)
where the upper/lower sign refer to the brane at $`z=0/z_0`$, respectively. For the perturbations we get instead
$`\mathrm{\Phi }^{}+2{\displaystyle \frac{A^{}}{A}}\mathrm{\Phi }{\displaystyle \frac{2K\varphi ^{}}{3M^3}}\delta \varphi =0`$
$`\delta \varphi ^{}\varphi ^{}\mathrm{\Phi }{\displaystyle \frac{A}{2K}}U^{\prime \prime }\delta \varphi +{\displaystyle \frac{K^{}\varphi ^{}}{K}}\delta \varphi =0.`$ (16)
These new equations, together with the eigenvalue equation (11), allow to determine the spectrum and the bulk profiles of the modes. A simple extension of the calculation of kmp shows that the coefficients $`C_n`$ acquires a boundary contribution
$$C_n=๐z\sigma \left(K\right)\stackrel{~}{v}_n^2\frac{3M^3A^4}{4A^{}}\stackrel{~}{\mathrm{\Phi }}_n^2|_0^{z_0},$$
(17)
where we have defined $`\mathrm{\Phi }=_n\stackrel{~}{\mathrm{\Phi }}_nQ_n`$, in strict analogy to (10). The boundary values of $`\stackrel{~}{\mathrm{\Phi }}_n`$ can be obtained from the ones of $`\stackrel{~}{v}_n`$ through the definition (7) and the boundary conditions (16).
## III Application
Let us discuss how the above formalism applies to the model of Ref. maxim . The model is characterized by the bulk action (1), with
$`K\left(\varphi \right)`$ $`=`$ $`3M^3A_0{\displaystyle \frac{\varphi /\varphi _0}{\varphi _0^2\varphi ^2}},`$
$`V\left(\varphi \right)`$ $`=`$ $`6M^3w^2A_0^2\left\{1{\displaystyle \frac{1}{4A_0}}{\displaystyle \frac{\varphi }{\varphi _0}}{\displaystyle \frac{\varphi ^2}{\varphi _0^2}}\right\},`$ (18)
extending on a periodic interval ($`M^3`$ is the five dimensional Planck mass while $`A_0,w,\varphi _0`$ are some constants). The background solution is particularly simple and elegant. In normal coordinates, defined as
$$ds^2=dy^2+A\left(y\right)\eta ^{\mu \nu }dx_\mu dx_\nu ,$$
(19)
one has
$`\varphi \left(y\right)`$ $`=`$ $`\varphi _0\mathrm{cos}\left(wy\right),`$
$`A\left(y\right)`$ $`=`$ $`\mathrm{exp}\left[A_0\mathrm{cos}\left(wy\right)\right].`$ (20)
The geometry is characterized by a nontrivial periodic evolution of the warp factor $`A`$. The sourcing bulk scalar is also periodic in the bulk, and the scalar density $`\varphi ^2`$ is mostly localized at $`y=0,\pi /w`$, where $`A`$ has its extrema. This model represents an attempt to regularize a brane/antiโbrane system. This is shown in fig. 1, where we compare the background (20) with the RandallโSundrum background rs1 . In the present case, the maximum (minimum) of $`A`$ is due to a delocalized field rather than to a positive (negative) tension brane. As shown in maxim , a bulk scalar field with the standard sign for the kinetic term cannot give rise to a minimum of $`A`$, irrespectively of its potential. Indeed, we observe from the definitions of $`K(\varphi )`$ and $`\varphi (y)`$ that $`K`$ is positive in the interval $`0wy<\pi /2`$ close to the maximum of $`A(y)`$, whereas it is negative in the region $`\pi /2<wy\pi `$ near the minimum of the warp factor where $`\varphi `$ regularizes the negative tension brane.
Let us now compute the scalar perturbations around the background (20). It is convenient to rewrite eq. (11) in normal coordinates. In terms of the rescaled variable $`\psi A^{1/2}\stackrel{~}{v}`$ (for shortening the notation, we omit the subscript $`n`$) one finds
$$\frac{d^2\psi }{dy^2}+\mathrm{e}^{2A_0\mathrm{cos}\left(wy\right)}m^2\psi V_{\mathrm{eff}}\psi =0,$$
(21)
where
$`V_{\mathrm{eff}}{\displaystyle \frac{w^2}{4}}[\mathrm{cot}^2\left(wy\right)+{\displaystyle \frac{8}{\mathrm{sin}^2\left(wy\right)}}{\displaystyle \frac{4}{\mathrm{sin}^2\left(2wy\right)}}+`$
$`+{\displaystyle \frac{8A_0}{\mathrm{cos}\left(wy\right)}}+16A_0^2\mathrm{sin}^2\left(wy\right)].`$ (22)
It is always possible to rescale $`y`$ such that $`w=1`$. We do so from now on. The effective potential is $`Z_2`$ symmetric around $`y=0`$ (โregularizedโ positive brane) and $`y=\pi `$ (โregularizedโ negative brane), where it diverges to $`+\mathrm{}`$ (see fig. 2). In addition, it is unbounded from below where the kinetic function $`K`$ vanishes. Despite of the singularities, we will now show that (21) admits solutions which are finite everywhere. However, there is an intrinsic ambiguity in matching the solutions across the singularities of $`V_{\mathrm{eff}}`$. To see this, let us first solve (21) for $`A_0=0`$, and then for small $`A_0`$ (although the model (18) is trivial for $`A_0=0`$, the formal problem (21) is still defined).
For $`A_0=0`$, the potential is symmetric also around $`y=\pi /2`$; in the interval $`0y\pi /2`$, we approximate it as
$`V_{\mathrm{eff}}\{\begin{array}{c}\frac{2}{y^2}+\frac{1}{6},0y1,\\ \\ \frac{1}{4\left(y\frac{\pi }{2}\right)^2}+\frac{5}{3},1y\frac{\pi }{2}.\end{array}`$ (26)
This approximated form reproduces the potential where it diverges (up to terms which vanish as $`y0,\pi /2`$), and it is very close to the exact potential everywhere. We verified that the numerical solutions to the exact problem (22) are very well approximated by the analytic ones of (26). The โmatching pointโ $`y=1`$ has been chosen by comparison with the exact form of the potential, and we verified that the solutions do not change significantly if the matching point is slightly moved away from $`1`$.
Equation (21), with the approximated potential (26), is solved by
$`\psi `$ $`=`$ $`C_1\left[\mathrm{cos}\left(\alpha y\right){\displaystyle \frac{\mathrm{sin}\left(\alpha y\right)}{\alpha y}}\right]`$
$`+C_2\left[\mathrm{sin}\left(\alpha y\right)+{\displaystyle \frac{\mathrm{cos}\left(\alpha y\right)}{\alpha y}}\right],0y1,`$
$`\psi `$ $`=`$ $`C_3\sqrt{\xi }J_0\left(\beta \xi \right)+C_4\sqrt{\xi }Y_0\left(\beta \xi \right),1y{\displaystyle \frac{\pi }{2}},`$
where in the last line we have defined $`\xi \pi /2y`$, and where
$$\alpha \sqrt{m^2\frac{1}{6}},\beta \sqrt{m^2\frac{5}{3}}.$$
(28)
Since the potential is symmetric around $`\pi /2`$, in the range $`\pi /2y\pi `$ the solution can be written as (III), upon the substitution $`y\pi y,C_iD_i`$.
Let us now identify the unknown quantities, and see whether we can determine them by boundary conditions (continuity of $`\psi `$ and $`\psi ^{}`$ at $`y=1,\pi /2,\pi 1`$). In the first interval, the mode proportional to $`C_2`$ diverges at the origin. This immediately sets $`C_2=0`$. Analogously, regularity at $`\pi `$ sets $`D_2=0`$. We are thus left with $`7`$ unknowns: the six coefficients $`C_{1,3,4}`$, $`D_{1,3,4}`$, and the eigenvalue $`m^2`$. One coefficient cannot be determined, since the overall normalization of the mode cannot be obtained from the linearized calculation we are performing (the normalization is fixed as explained in the previous section). Hence, we can fix here $`C_1=1`$.
We are thus left with $`6`$ unknowns, and with the request of continuity of $`\psi `$ and its derivative at the three locations $`y=1,\pi /2,\pi 1`$. The modes are regular at $`y=1,\pi 1`$, so we have four nontrivial boundary conditions there. As can be expected, the problems arise for the matching at $`y=\pi /2`$. At this point, the mode (III) vanishes, while its derivative diverges,
$$\psi C_3\sqrt{\xi }+\frac{2C_4}{\pi }\sqrt{\xi }\mathrm{ln}\xi ,\mathrm{as}\xi =\frac{\pi }{2}y0.$$
(29)
(and analogously for $`y`$ approaching $`\pi /2`$ from the right). Hence, the two boundary conditions at $`\pi /2`$ are absent, and a global solution cannot be determined. This confirms the expectation that the linearized problem is ill posed, due to the vanishing of the kinetic function $`K`$.
One can expect that higher order terms can be relevant where $`K`$ vanishes, providing a cutโoff where the effective potential is unbounded. In the following we regularize the potential by hand with a cutโoff at $`y\pi /2`$ (which allows to solve the linearized problem). More precisely, we take the potential to be constant (matching the value from (26)), in the interval $`\pi /2ฯตy\pi /2+ฯต`$; we then study the behavior of the solution as $`ฯต0`$. Fortunately, this procedure leads to a well defined and normalizable solution. This limiting solution is the one which could have been easily guessed from elementary quantum mechanics. The coefficients of the modes which are more divergent as $`y\pi /2`$, namely $`C_4`$ and $`D_4`$, vanish. The first eigenvalues are approximately given by (in units of $`w^2`$)
$$m^26,\mathrm{\hspace{0.17em}20},\mathrm{\hspace{0.17em}42},\mathrm{\hspace{0.17em}72},\mathrm{}$$
(30)
As always for symmetric potentials, each eigenvalue $`m^2`$ admits two degenerate solutions, one symmetric and one antisymmetric around $`\pi /2`$. We confirmed these solutions through a fully numerical calculation <sup>3</sup><sup>3</sup>3The numerical calculation is a boundary value problem, and it can be solved with a shooting method. See kmp for details. using the exact potential (22). Although modes with negative $`m^2`$ are in principle a possibility, the numerical analysis does not reveal indications for their existence.
The more relevant case of $`A_00`$ can be studied analogously. The additional terms in the potential can be also approximated by simple polynomials, and analytic solutions can be obtained in terms of new special functions. Alternatively, the problem can be studied numerically for any given value of the cutโoff, and one can verify that also in this case limiting solutions are reached when $`ฯต`$ is sent to zero. The results are quite interesting, since โ for the reasons we will now argue โ the limit $`A_00`$ is not continuous.
For small $`A_0`$ the eigenvalues occur in finely split pairs. This is due to the fact that the $`A_0`$ part of the potential (22) is not symmetric around $`\pi /2`$. As $`A_00`$, the values (30) are recovered; however, the limiting solutions are not any longer symmetric or antisymmetric around $`\pi /2`$. For any non-vanishing $`A_0`$ (not necessarily small), the limiting solutions split in two groups: one characterized by modes which are nonvanishing only in the interval $`0y\pi /2`$ (group I), and one by modes which are nonvanishing in the complementary interval $`\pi /2y\pi `$ (group II). This can be easily explained. As we have seen, for $`ฯต=0`$ the two halves of the space are not in communication. Hence, the eigenvalue problem (21) effectively decouples in two different problems. Unless the potential is symmetric in the two halves, the two eigenvalue problems admit different eigenvalues. It is then impossible for an eigenmode to have support in both halves. The decoupling is visible in the $`ฯต0`$ limit, as shown for a particular case in fig. 3.
Let us finally return to the initial question, whether a $`4d`$ mode is a ghost or not. Solutions in group I have support in the interval $`0y\pi /2`$, where $`K`$ is positive. The normalization coefficient $`C_n`$ for these modes, see eq. (13), is positive, and hence they are well behaving $`4d`$ scalars. On the contrary, the modes which have support in the other half of the bulk, where $`K`$ is negative, are ghosts. Although we have investigated only a limited set of values for $`A_0`$, this second class of models has always been present for all the values we have considered (in a comparable amounts to the modes in group I). We therefore conclude that the model (18) is unstable already at the semiclassical level.
## IV Discussion
Our main motivation was to study explicitly how the vanishing of the kinetic function $`K`$ affects the system of perturbations. Quite in general, one can expect singularities in correspondence to the zeros of $`K`$. We found that, for the specific model considered in Ref. maxim , the singularities are mild enough so that the regions where $`K0`$ and $`K0`$ admit normalizable modes for the perturbations; however, we also saw that the singularity is strong enough so that these regions do not communicate, and the modes have support only in one of them. Modes which have support where $`K`$ is negative are ghosts, and preclude the stability of the background. We now argue that this effect is quite general, irrespectively of the detailed form of $`K`$. To see this, assume that $`K`$ vanishes at some given point $`y_{}`$ in the bulk. In general, we can expand,
$$K(yy_{})^\alpha \xi ^\alpha ,\alpha >0$$
(31)
at small $`\xi `$. Using the background eqs. (4) and (5) to determine the functional form of $`\varphi ^{}`$ and $`A^{}`$ where $`K`$ vanishes, we find that, in a neighborhood of $`y_{}`$,
$$Z\left|\frac{KV^{}}{VK^{}}\right|^{1/2}\left|\frac{KV/\xi }{VK/\xi }\right|^{1/2}.$$
(32)
If the potential $`V`$ is regular and nonvanishing at $`y_{}`$, we then find
$$Z\xi ^{1/2}V_{\mathrm{eff}}\frac{Z^{\prime \prime }}{Z}\xi ^2.$$
(33)
Recalling the expansion (26), we see that the model (18) is not exception to the general rule, and that the degree of divergency of the effective potential is typically $`2`$. Hence, the problems and the instability found for this model are expected to be a general issue, whenever the kinetic function is allowed to vanish.
As manifest in (33), a possible โcureโ to the instability could be to arrange $`V`$ to vanish precisely where $`K`$ also does. This is a rather trivial solution, which however is likely to postpone the problem at orders higher than quadratic. More in general, one may hope that higher orders in $`\varphi `$ may allow the theory to have a better behavior where the quadratic term vanishes.
Acknowledgements
We thank Nemanja Kaloper, Stefan Groot Nibbelink and Maxim Pospelov for very useful discussions. This work has been supported in part by the Department of Energy under contract DE-FG02-94ER40823 at the University of Minnesota.
|
warning/0506/math0506559.html
|
ar5iv
|
text
|
# Some congruences for traces of singular moduli
## Abstract.
We address a question posed by Ono \[7, Problem 7.30\], prove a general result for powers of an arbitrary prime, and provide an explanation for the appearance of higher congruence moduli for certain small primes. One of our results overlaps but does not coincide with a recent result of Jenkins . This result essentially coincides with a recent result of Edixhoven , and we hope that the comparison of the methods, which are entirely different, may reveal a connection between the $`p`$-adic geometry and the arithmetic of half-integral weight Hecke operators.
Partially supported by NSF grant DMS-0501225
1. Introduction and discussion of the results
Throughout the paper $`D`$ and $`d`$ denote positive and non-negative integers, respectively, which satisfy the congruences
$$D0,1mod4\text{and}d0,3mod4.$$
We denote by $`\chi _d=\left(\frac{d}{}\right)`$ and $`\chi _D=\left(\frac{D}{}\right)`$ the quadratic Dirichlet characters associated with the imaginary and real quadratic fields $`(\sqrt{d})`$ and $`(\sqrt{D})`$ correspondingly. The character $`\chi _d`$ (resp. $`\chi _D`$) is primitive if $`d`$ (resp. $`D`$) is a fundamental discriminant.
Zagier considered in the nearly holomorphic modular forms
$$f_d=q^d+\underset{D>0}{}A(D,d)q^D$$
of weight $`1/2`$ and
$$g_D=q^D+\underset{d0}{}B(D,d)q^d.$$
of weight $`3/2`$. (Here and in the following $`q=\mathrm{exp}(2\pi i\tau )`$ with $`\mathrm{}(\tau )>0`$.) The explicit recursive construction of $`f_d`$ and $`g_D`$, provided by Zagier, guarantees that $`A(D,d),B(D,d)`$.
For an integer $`m1`$ the Hecke operators $`T(m)`$ act on these forms and conserve the integrality of the Fourier coefficients. Following we denote by $`A_m(D,d)`$ and $`B_m(D,d)`$ the coefficient of $`q^D`$ in $`f_d|_{\frac{1}{2}}T(m)`$ and the coefficient of $`q^D`$ in $`g_D|_{\frac{3}{2}}T(m)`$, respectively. Zagier proved that
(1)
$$A_m(D,d)=B_m(D,d),$$
and this common value divided by $`\sqrt{D}`$ is the (twisted if $`D>1`$) trace of a certain modular function. This interpretation in terms of traces of singular moduli is a primary source of motivation for the investigation of these numbers.
Ahlgren and Ono studied the arithmetic of traces of singular moduli in , and, in particular, proved the congruences
(2)
$$A_m(1,p^2d)0modp$$
if the prime $`p`$ splits in $`(\sqrt{d})`$ and $`pm`$. In this connection Ono posed a question \[7, Problem 7.30\] whether there are natural generalizations of (2) modulo arbitrary powers of $`p`$. Numerical evidence indicates that if $`\chi _d(p)=\chi _D(p)`$, then
(3)
$$A_m(D,p^{2n}d)0modp^n$$
with the maximum congruence moduli which exceeds $`p^n`$ for $`p11`$.
The question splits into two parts: to find similar congruences which hold for powers of an arbitrary prime and to find series of stronger congruences for special primes.
We firstly comment on the former part of the question. Recently Edixhoven used the interpretation of the numbers $`A_m(1,p^{2n}d)`$ as traces of singular moduli and the local moduli theory of ordinary elliptic curves in positive characterstic and obtained the following result.
###### Theorem (Edixhoven).
If $`D=1`$ and $`\chi _d(p)=1`$, then (3) holds for any $`m1`$.
Recently Jenkins presented an elementary argument based on the identity (1) and standard formulas for the action of half-integral weight Hecke operators. Jenkinsโ result recovers the congruences obtained by Edixhoven in the case $`m=1`$. More precisely, he proves the following.
###### Theorem (Jenkins).
If $`\chi _d(p)=\chi _D(p)0`$, then
(4)
$$A(D,p^{2n}d)=p^nA(p^{2n}D,d),$$
and, therefore, (3) holds for $`m=1`$.
In this paper we prove the following congruences.
###### Theorem 1.
Let $`d`$ and $`D`$ be fundamental discriminants.
a. If $`\chi _d(p)=\chi _D(p)`$, then (3) holds for any $`m1`$.
b. If $`\chi _d(p)=\chi _D(p)0`$, then for any $`m1`$
$$A_m(D,p^{2n+2}d)A_m(D,p^{2n}d)0modp^n.$$
Our argument is also elementary and uses nothing but the identity (1) and some facts about the action of half-integral weight Hecke operators. The assumption that both discriminants are fundamental is not essential. It, however, allows to simplify and generalize the argument. For instance, Jenkinsโ identity (4) under this assumption follows at once from the definitions (see (8) below). Note that our Theorem 1a essentially coincides with the result of Edixhoven. This provides a reason to speculate that the arithmetic of half-integral weight Hecke operators is somehow connected with the $`p`$-adic geometry. Such a connection, if it really exists, looks as an enticing subject to investigate.
We now turn to the latter part of Onoโs question. He noticed \[7, Example 7.15\] that, for $`p11`$, the maximum congruence modulus in (2) exceeds $`p`$ and called for explanations. Recently Boylan found a pretty exact answer in the case $`p=2`$. Combining the result of Jenkins, a recent result of the author and a theorem of Serre we prove the following qualitative result, which uniformly explains the phenomenon without providing information about the specific congruence moduli.
###### Theorem 2.
Let $`p11`$. If $`\chi _d(p)=\chi _D(p)0`$, then the $`p`$-adic limit
$$\underset{n\mathrm{}}{lim}p^nA(D,p^{2n}d)=0.$$
2. Proofs
Theorem 1 follows at once from the following proposition.
###### Proposition 1.
Let $`p`$ be a prime, and let $`m1`$ and $`n0`$ be integers. Let $`d`$ and $`D`$ be fundamental discriminants (i.e. $`d`$ is the discriminant of $`(\sqrt{d})`$, and $`D`$ is either $`1`$ or the discriminant of $`(\sqrt{D})`$).
a. If $`\chi _d(p)=\chi _D(p)`$ then
(5)
$$A_m(D,p^{2n}d)=p^nA_m(p^{2n}D,d).$$
b. If $`\chi _d(p)=\chi _D(p)0`$ then
$$A_m(D,p^{2n+2}d)A_m(D,p^{2n}d)=p^{n+1}A_m(p^{2n+2}D,d)+p^nA_m(p^{2n}D,d).$$
###### Proof of Proposition 1.
Let $`F=b(n)q^n`$ be a (holomorphic or nearly holomorphic) modular form of weight $`k+1/2`$ for an integer $`k0`$, which belongs to the Kohnen plus-space (i.e. $`b(n)=0`$ if $`(1)^kn2,3mod4`$). The Hecke operator $`T^+(m)`$ acts on the Kohnen plus-space and sends $`F`$ to $`F|_{k+1/2}T^+(m)=b_m^+(n)q^n`$. If $`(1)^kn`$ is a fundamental discriminant, then
(6)
$$b_m^+(n)=\underset{l|m}{}\left(\frac{(1)^kn}{l}\right)l^{k1}b\left(\frac{m^2}{l^2}n\right).$$
This formula follows from \[4, Th. 4.5\], where it is proved in the equivalent language of Jacobi forms. Although formally the quoted theorem applies only to holomorphic half-integral weight modular forms, as it is mentioned in , nothing changes if we allow the pole at infinity. Also note that the technique of Jacobi forms from covers only the case of odd $`k`$. However, if $`k=0`$ (the only even $`k`$ which we need here) an equivalent formula may be found in Zagierโs paper \[9, proof of Th. 7\]. If $`k=0`$, non-trivial denominators apparently appear in the right-hand side of (6). In order to get rid of these denominators and to keep our notations compatible with those of we renormalize the Hecke operators by
$$T(m)=\{\begin{array}{cc}mT^+(m)\hfill & \text{if }k=0\hfill \\ T^+(m)\hfill & \text{otherwise}.\hfill \end{array}$$
It follows from (6) and the definition of the quantities $`A_{p^n}(D,d)`$ and $`B_{p^n}(D,d)`$ that under the assumptions of Proposition 1
$`A_{p^n}(D,d)={\displaystyle \underset{i=0}{\overset{n}{}}}\chi _D(p^{ni})p^iA(p^{2i}D,d)`$
$`B_{p^n}(D,d)={\displaystyle \underset{i=0}{\overset{n}{}}}\chi _d(p^{ni})B(D,p^{2i}d).`$
These equations combined with (1) imply that for any $`n0`$
(8)
$$\underset{i=0}{\overset{n}{}}\chi _D(p^{ni})p^iA(p^{2i}D,d)=\underset{i=0}{\overset{n}{}}\chi _d(p^{ni})A(D,p^{2i}d),$$
and an induction argument in $`n`$ finishes the proof of Proposition 1 in the case when $`m=1`$.
We now generalize the above argument to the case of arbitrary integer $`m1`$. Recall the usual relation between Hecke operators acting on the Kohnen plus-space of modular forms of weight $`k+1/2`$ (see i.e. \[4, Cor. 1 to Th. 4.5\]):
(9)
$$T^+(u)T^+(u^{})=\underset{c|(u,u^{})}{}c^{2k1}T^+(uu^{}/c^2).$$
In particular, for $`k=0,1`$ and integers $`n,s0`$
$$\underset{i=0}{\overset{\mathrm{min}(n,s)}{}}p^iT(p^{n+s2i})=T(p^n)T(p^s),$$
and the equations (LABEL:X) generalize to
$`{\displaystyle \underset{i=0}{\overset{\mathrm{min}(n,s)}{}}}p^iA_{p^{n+s2i}}(D,d)={\displaystyle \underset{i=0}{\overset{n}{}}}\chi _D(p^{ni})p^iA_{p^s}(p^{2i}D,d)`$
$`{\displaystyle \underset{i=0}{\overset{\mathrm{min}(n,s)}{}}}p^iB_{p^{n+s2i}}(D,d)={\displaystyle \underset{i=0}{\overset{n}{}}}\chi _d(p^{ni})B_{p^s}(D,p^{2i}d).`$
As previously, (1) implies that the left-hand sides of (LABEL:XX) are equal by absolute value and have opposite signs, and we obtain the following generalization of (8) for $`n,s0`$
(11)
$$\underset{i=0}{\overset{n}{}}\chi _D(p^{ni})p^iA_{p^s}(p^{2i}D,d)=\underset{i=0}{\overset{n}{}}\chi _d(p^{ni})A_{p^s}(D,p^{2i}d).$$
An induction argument in $`n`$ finishes the proof of Proposition 1 in the case when $`m=p^s`$ with $`s>0`$.
Assume now that $`m=p^sm_0`$ with $`pm_0`$ and $`s0`$. It follows from (9) that
$$T(p^sm_0)=T(p^s)T(m_0)$$
for any $`k0`$. It follows that we can multiply the indices in (11) and (LABEL:XX) by $`m_0`$, as we have had begun the whole argument with the consideration of $`f_d|_{\frac{1}{2}}T(m_0)`$ and $`g_D|_{\frac{3}{2}}T(m_0)`$ instead of $`f_d`$ and $`g_D`$ correspondingly. This implies the following generalization of (8),(11) for $`m_01`$, $`pm_0`$, $`s0`$ and any $`n0`$:
$$\underset{i=0}{\overset{n}{}}\chi _D(p^{ni})p^iA_{m_0p^s}(p^{2i}D,d)=\underset{i=0}{\overset{n}{}}\chi _d(p^{ni})A_{m_0p^s}(D,p^{2i}d),$$
and an induction argument in $`n`$ completes the proof of Proposition 1.
###### Proof of Theorem 2.
It follows from \[5, Theorem 3b\] that, since $`pdD`$, the formal power series in $`q`$
$$F=\underset{{\scriptscriptstyle \genfrac{}{}{0.0pt}{}{n>0}{p|n}}}{}\underset{{\scriptscriptstyle \genfrac{}{}{0.0pt}{}{l|n}{(p,l)=1}}}{}\left(\frac{D}{l}\right)l^1A(\frac{n^2}{l^2}D,d)q^n$$
is a $`p`$-adic cusp form of weight $`0`$. A result of Serre \[8, Th. 7; Rem., p.216\] implies that, for $`p11`$, the $`p`$-adic limit
$$\underset{n\mathrm{}}{lim}F|U^n=0,$$
where $`U`$ denotes Atkinโs $`U`$-operator $`\left(a(n)q^n\right)|U=a(pn)q^n`$. Thus the coefficient of $`q^{p^n}`$ in $`F`$ approaches $`0`$ $`p`$-adically as $`n\mathrm{}`$. That is
$$\underset{n\mathrm{}}{lim}A(p^{2n}D,d)=0.$$
The latter equality combined with Jenkinsโ identity (4) completes the proof of Theorem 2. โ
|
warning/0506/math0506570.html
|
ar5iv
|
text
|
# Generalized diagonal crossed products and smash products for quasi-Hopf algebras. Applications
## 1. Introduction
Quasi-bialgebras and quasi-Hopf algebras were introduced by Drinfeld in , in connection with the Knizhnik-Zamolodchikov equations, but also as very natural (especially from the tensor-categorical point of view) generalizations of bialgebras and Hopf algebras. Let $`k`$ be a field, $`H`$ an associative algebra and $`\mathrm{\Delta }:HHH`$ and $`\epsilon :Hk`$ two algebra morphisms. Roughly speaking, $`H`$ is a quasi-bialgebra if the category $`{}_{H}{}^{}`$ of left $`H`$-modules, equipped with the tensor product of vector spaces endowed with the diagonal $`H`$-module structure given via $`\mathrm{\Delta }`$, and with unit object $`k`$ viewed as a left $`H`$-module via $`\epsilon `$, is a monoidal category (if we impose the associativity constraints to be the trivial ones, we obtain the usual concept of bialgebra). The comultiplication $`\mathrm{\Delta }`$ is not coassociative but is quasi-coassociative in the sense that $`\mathrm{\Delta }`$ is coassociative up to conjugation by an invertible element $`\mathrm{\Phi }HHH`$. Note that the definition of a quasi-bialgebra or quasi-Hopf algebra is not self-dual.
Actions and coactions on algebras are an important part of the theory of Hopf algebras, and they have been extended to quasi-Hopf algebras: module algebras have been studied in , while (bi) comodule algebras were introduced in .
Over a finite dimensional Hopf algebra $`H`$, speaking about module algebras or comodule algebras is the same thing, since a left (right) $`H`$-module algebra is the same as a right (left) $`H^{}`$-comodule algebra. This does no longer hold over quasi-Hopf algebras, where a comodule algebra is an associative algebra but a module algebra is associative only in a tensor category, so in general being nonassociative as an algebra (for instance, the nonassociative algebra of octonions is such a module algebra over a certain quasi-Hopf algebra, cf. ). This fact leads to the following situation: a concept, construction, result etc. from the theory of Hopf algebras might admit more different generalizations when passing to quasi-Hopf algebras.
Such a situation occurs in this paper. To explain it, we recall some facts from . If $`H`$ is a finite dimensional quasi-Hopf algebra and $`๐ธ`$ is an $`H`$-bicomodule algebra, Hausser and Nill introduced the so-called diagonal crossed products $`H^{}๐ธ`$, $`H^{}\text{ }\text{ }๐ธ`$, $`๐ธH^{}`$ and $`๐ธ\text{ }\text{ }H^{}`$, which are all (isomorphic) associative algebras and have the property that for $`๐ธ=H`$ they are realizations of the quantum double of $`H`$, which has been introduced before by Majid in in the form of an implicit Tannaka-Krein reconstruction procedure. Also, if $`๐`$ and $`๐
`$ are a right and respectively a left $`H`$-comodule algebra, Hausser and Nill introduced an associative algebra structure on $`๐H^{}๐
`$, denoted by $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$ and called the two-sided crossed product, which has the property that with respect to the natural bicomodule algebra structure on $`๐๐
`$ one has an algebra isomorphism $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
(๐๐
)H^{}`$. Their motivation for introducing these constructions was the need to extend to the quasi-Hopf setting some models of Hopf spin chains and lattice current algebras from algebraic quantum field theory (see the introduction of for details). For this purpose, one of the key results in was that the two-sided crossed products can be iterated, providing thus a local net of associative algebras, with quantum double cosymmetry.
Now, if $`H`$ is a finite dimensional Hopf algebra, the construction $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$ may be described equivalently with module algebras instead of comodule algebras, and it becomes a two-sided smash product $`A\mathrm{\#}H\mathrm{\#}B`$ (where $`A`$ and $`B`$ are a left, respectively a right $`H`$-module algebra), with multiplication given by
$$(a\mathrm{\#}h\mathrm{\#}b)(a{}_{}{}^{}\mathrm{\#}h{}_{}{}^{}\mathrm{\#}b{}_{}{}^{})=a(h_1a{}_{}{}^{})\mathrm{\#}h_2h_1^{}\mathrm{\#}(bh_2^{})b{}_{}{}^{},$$
for all $`a,a{}_{}{}^{}A`$, $`h,h{}_{}{}^{}H`$ and $`b,b{}_{}{}^{}B`$.
It is this construction that we first wanted to generalize to quasi-Hopf algebras (where it will be different from the two-sided crossed product of Hausser and Nill). The need for such a construction arose as follows. It was proved in that, for a finite dimensional Hopf algebra $`H`$, the category $`{}_{H^{}}{}^{H^{}}_{H^{}}^{H^{}}`$ of $`H^{}`$-Hopf bimodules is isomorphic to the category of left modules over a two-sided smash product $`H^{}\mathrm{\#}(HH^{op})\mathrm{\#}H^{op}`$. We wanted a similar result for the category $`{}_{H^{}}{}^{H^{}}_{H^{}}^{H^{}}`$ for $`H`$ a finite dimensional quasi-Hopf algebra, but observed that we could not use the two-sided crossed product of Hausser and Nill, we needed a generalization of the two-sided crossed product from Hopf algebras in the other direction (the one based on module algebras and not on comodule algebras). After constructing this two-sided smash product $`A\mathrm{\#}H\mathrm{\#}B`$, we wanted to express it as some sort of diagonal crossed product $`(AB)H`$, and we were led naturally to consider a generalized diagonal crossed product $`๐๐ธ`$, where $`๐`$ is an $`H`$-bimodule algebra and $`๐ธ`$ is an $`H`$-bicomodule algebra.
We describe now more formally the structure of this paper ($`H`$ will be a fixed quasi-Hopf algebra or sometimes only a quasi-bialgebra). In Section 3 we introduce the left and right generalized diagonal crossed products $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$ (which will turn out to be isomorphic), where $`๐`$ is an $`H`$-bimodule algebra and $`๐ธ`$ is an associative algebra endowed with a two-sided coaction of $`H`$ on it, and we prove their associativity. If $`๐ธ`$ is an $`H`$-bicomodule algebra, one can construct out of it two two-sided coactions $`\delta _l`$ and $`\delta _r`$, hence we have four generalized diagonal crossed products $`๐๐ธ`$, $`๐\text{ }\text{ }๐ธ`$, $`๐ธ๐`$ and $`๐ธ\text{ }\text{ }๐`$.
In Section 4 we construct, starting with a bicomodule algebra $`๐ธ`$, two left $`HH^{op}`$-comodule algebra structures on $`๐ธ`$, denoted by $`๐ธ_1`$ and $`๐ธ_2`$. Regarding $`๐`$ as a left $`HH^{op}`$-module algebra, we identify $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ with the generalized smash products (in the sense of ) $`๐<๐ธ_1`$ and $`๐<๐ธ_2`$. We prove that $`๐ธ_1`$ and $`๐ธ_2`$ are twist equivalent as left $`HH^{op}`$-comodule algebras, and we obtain that $`๐๐ธ๐\text{ }\text{ }๐ธ`$ as algebras.
In Section 5 we consider a left $`H`$-module algebra $`A`$ and a right $`H`$-module algebra $`B`$. We first describe a slight generalization of the two-sided crossed product $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$, replacing $`H^{}`$ by $`๐`$, and call this algebra the generalized two-sided crossed product. Then we construct the two-sided generalized smash product $`A<๐ธ\text{ }>\text{ }B`$, which for $`๐ธ=H`$ is exactly the two-sided smash product $`A\mathrm{\#}H\mathrm{\#}B`$ that we needed.
In Section 6 we prove the algebra isomorphisms $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐
๐(๐๐
)`$ and $`A<๐ธ\text{ }>\text{ }B(AB)๐ธ`$, obtaining in particular that the generalized diagonal crossed product $`(AB)(๐๐
)`$ is isomorphic with both $`A<(๐๐
)\text{ }>\text{ }B`$ and $`๐\text{ }>\text{ }(AB)\text{ }<\text{ }๐
`$.
In Section 7 we study the invariance under twisting of our constructions.
Starting with Section 8 we move to applications. We prove first that both two-sided products (the generalized two-sided crossed product and the two-sided generalized smash product) may be written as some iterated products. Together with the fact that a generalized smash product $`A<๐ธ`$ becomes a right $`H`$-comodule algebra (and similarly for $`๐ธ\text{ }>\text{ }B`$), this allows us to obtain a very easy, conceptual and constructive proof of the theorem of Hausser and Nill concerning iterated two-sided crossed products. As a by-product of our approach, we obtain also that the iterated products arising in this theorem are actually isomorphic to a two-sided generalized smash product.
In Section 9 we prove what was our original motivation for this paper, namely that the category $`{}_{H^{}}{}^{H^{}}_{H^{}}^{H^{}}`$ of $`H^{}`$-Hopf bimodules over a finite dimensional quasi-Hopf algebra $`H`$ is isomorphic to $`{}_{H^{}\mathrm{\#}(HH^{op})\mathrm{\#}\overline{H^{}}}{}^{}`$. Along the way we obtain some other results of independent interest, such as the description of left modules over a two-sided smash product.
In Section 10 we prove that, if $`(H,๐ธ,C)`$ is a so-called Yetter-Drinfeld datum (here, $`C`$ is an $`H`$-bimodule coalgebra) with $`C`$ finite dimensional, then the category $`{}_{๐ธ}{}^{}๐ดD(H)^C`$ of (generalized) Yetter-Drinfeld modules is isomorphic to the category of left modules over the generalized diagonal crossed product $`C^{}๐ธ`$.
Some remarks on techniques are in order. What is characteristic in the approach of Hausser and Nill to their constructions is the systematic use of the so-called โgenerating matrixโ formalism of the St. Petersburg school (the use of $`\delta `$-implementers, $`\lambda \rho `$-intertwiners etc). The replacement of $`H^{}`$ by an arbitrary $`H`$-bimodule algebra in our definition of the generalized diagonal crossed products makes the use of this formalism impossible, so most of our proofs are different in spirit from the ones of Hausser and Nill, and often easier (just compare our proof of the theorem concerning iterated two-sided crossed products with the original one in ), providing thus also an alternative approach to the constructions of Hausser and Nill. Another alternative approach has been provided by Schauenburg in (using categorical techniques).
## 2. Preliminaries
In this section we recall some definitions and results and fix notation used throughout the paper.
### 2.1. Quasi-bialgebras and quasi-Hopf algebras
We work over a field $`k`$. All algebras, linear spaces etc. will be over $`k`$; unadorned $``$ means $`_k`$. Following Drinfeld , a quasi-bialgebra is a fourtuple $`(H,\mathrm{\Delta },\epsilon ,\mathrm{\Phi })`$, where $`H`$ is an associative algebra with unit, $`\mathrm{\Phi }`$ is an invertible element in $`HHH`$, and $`\mathrm{\Delta }:HHH`$ and $`\epsilon :Hk`$ are algebra homomorphisms satisfying the identities
(2.1) $`(id\mathrm{\Delta })(\mathrm{\Delta }(h))=\mathrm{\Phi }(\mathrm{\Delta }id)(\mathrm{\Delta }(h))\mathrm{\Phi }^1,`$
(2.2) $`(id\epsilon )(\mathrm{\Delta }(h))=h,(\epsilon id)(\mathrm{\Delta }(h))=h,`$
for all $`hH`$, and $`\mathrm{\Phi }`$ has to be a normalized $`3`$-cocycle, in the sense that
(2.3) $`(1\mathrm{\Phi })(id\mathrm{\Delta }id)(\mathrm{\Phi })(\mathrm{\Phi }1)=(idid\mathrm{\Delta })(\mathrm{\Phi })(\mathrm{\Delta }idid)(\mathrm{\Phi }),`$
(2.4) $`(id\epsilon id)(\mathrm{\Phi })=11.`$
The identities (2.2), (2.3) and (2.4) also imply that
(2.5)
$$(\epsilon idid)(\mathrm{\Phi })=(idid\epsilon )(\mathrm{\Phi })=11.$$
The map $`\mathrm{\Delta }`$ is called the coproduct or the comultiplication, $`\epsilon `$ the counit and $`\mathrm{\Phi }`$ the reassociator. As for bialgebras (see ) we denote $`\mathrm{\Delta }(h)=h_1h_2`$, but since $`\mathrm{\Delta }`$ is only quasi-coassociative we adopt the further convention (summation understood):
$$(\mathrm{\Delta }id)(\mathrm{\Delta }(h))=h_{(1,1)}h_{(1,2)}h_2,(id\mathrm{\Delta })(\mathrm{\Delta }(h))=h_1h_{(2,1)}h_{(2,2)},$$
for all $`hH`$. We will denote the tensor components of $`\mathrm{\Phi }`$ by capital letters, and those of $`\mathrm{\Phi }^1`$ by small letters, namely
$`\mathrm{\Phi }=X^1X^2X^3=T^1T^2T^3=Y^1Y^2Y^3=\mathrm{}`$
$`\mathrm{\Phi }^1=x^1x^2x^3=t^1t^2t^3=y^1y^2y^3=\mathrm{}`$
The quasi-bialgebra $`H`$ is called a quasi-Hopf algebra if there exists an anti-automorphism $`S`$ of the algebra $`H`$ and elements $`\alpha ,\beta H`$ such that, for all $`hH`$, we have:
(2.6) $`S(h_1)\alpha h_2=\epsilon (h)\alpha \text{ and }h_1\beta S(h_2)=\epsilon (h)\beta ,`$
(2.7) $`X^1\beta S(X^2)\alpha X^3=1\text{ and }S(x^1)\alpha x^2\beta S(x^3)=1.`$
For a quasi-Hopf algebra the antipode is determined uniquely up to a transformation $`\alpha U\alpha `$, $`\beta \beta U^1`$, $`S(h)US(h)U^1`$, where $`UH`$ is invertible. The axioms for a quasi-Hopf algebra imply that $`\epsilon (\alpha )\epsilon (\beta )=1`$, so, by rescaling $`\alpha `$ and $`\beta `$, we may assume without loss of generality that $`\epsilon (\alpha )=\epsilon (\beta )=1`$ and $`\epsilon S=\epsilon `$.
Together with a quasi-bialgebra or a quasi-Hopf algebra $`H=(H,\mathrm{\Delta },\epsilon ,\mathrm{\Phi },S,\alpha ,\beta )`$ we also have $`H^{op}`$, $`H^{cop}`$ and $`H^{op,cop}`$ as quasi-bialgebras (respectively quasi-Hopf algebras), where โopโ means opposite multiplication and โcopโ means opposite comultiplication. The structures are obtained by putting $`\mathrm{\Phi }_{op}=\mathrm{\Phi }^1`$, $`\mathrm{\Phi }_{cop}=(\mathrm{\Phi }^1)^{321}`$, $`\mathrm{\Phi }_{op,cop}=\mathrm{\Phi }^{321}`$, $`S_{op}=S_{cop}=(S_{op,cop})^1=S^1`$, $`\alpha _{op}=S^1(\beta )`$, $`\beta _{op}=S^1(\alpha )`$, $`\alpha _{cop}=S^1(\alpha )`$, $`\beta _{cop}=S^1(\beta )`$, $`\alpha _{op,cop}=\beta `$ and $`\beta _{op,cop}=\alpha `$.
Next we recall that the definition of a quasi-bialgebra or quasi-Hopf algebra is โtwist covariantโ in the following sense. An invertible element $`FHH`$ is called a gauge transformation or twist if $`(\epsilon id)(F)=(id\epsilon )(F)=1`$. If $`H`$ is a quasi-bialgebra or a quasi-Hopf algebra and $`F=F^1F^2HH`$ is a gauge transformation with inverse $`F^1=G^1G^2`$, then we can define a new quasi-bialgebra (respectively quasi-Hopf algebra) $`H_F`$ by keeping the multiplication, unit, counit (and antipode in the case of a quasi-Hopf algebra) of $`H`$ and replacing the comultiplication, reassociator and the elements $`\alpha `$ and $`\beta `$ by
(2.8) $`\mathrm{\Delta }_F(h)=F\mathrm{\Delta }(h)F^1,`$
(2.9) $`\mathrm{\Phi }_F=(1F)(id\mathrm{\Delta })(F)\mathrm{\Phi }(\mathrm{\Delta }id)(F^1)(F^11),`$
(2.10) $`\alpha _F=S(G^1)\alpha G^2,\text{ }\beta _F=F^1\beta S(F^2).`$
It is known that the antipode of a Hopf algebra is an anti-coalgebra morphism. For a quasi-Hopf algebra, we have the following: there exists a gauge transformation $`fHH`$ such that
(2.11)
$$f\mathrm{\Delta }(S(h))f^1=(SS)(\mathrm{\Delta }^{cop}(h))\text{,\hspace{0.25em}\hspace{0.25em}\hspace{0.25em}for all }hH\text{.}$$
The element $`f`$ can be computed explicitly. First set
$`A^1A^2A^3A^4=(\mathrm{\Phi }1)(\mathrm{\Delta }idid)(\mathrm{\Phi }^1),`$
$`B^1B^2B^3B^4=(\mathrm{\Delta }idid)(\mathrm{\Phi })(\mathrm{\Phi }^11),`$
and then define $`\gamma ,\delta HH`$ by
(2.12)
$$\gamma =S(A^2)\alpha A^3S(A^1)\alpha A^4\mathrm{and}\delta =B^1\beta S(B^4)B^2\beta S(B^3).$$
Then $`f`$ and $`f^1`$ are given by the formulae
(2.13) $`f`$ $`=`$ $`(SS)(\mathrm{\Delta }^{cop}(x^1))\gamma \mathrm{\Delta }(x^2\beta S(x^3)),`$
(2.14) $`f^1`$ $`=`$ $`\mathrm{\Delta }(S(x^1)\alpha x^2)\delta (SS)(\mathrm{\Delta }^{cop}(x^3)).`$
Moreover, $`f`$ satisfies the following relations:
(2.15)
$$f\mathrm{\Delta }(\alpha )=\gamma ,\mathrm{\Delta }(\beta )f^1=\delta .$$
Furthermore the corresponding twisted reassociator (see (2.9)) is given by
(2.16)
$$\mathrm{\Phi }_f=(SSS)(X^3X^2X^1).$$
### 2.2. Smash products
Suppose that $`(H,\mathrm{\Delta },\epsilon ,\mathrm{\Phi })`$ is a quasi-bialgebra. If $`U,V,W`$ are left (right) $`H`$-modules, define $`a_{U,V,W},๐_{U,V,W}:(UV)WU(VW)`$,
$`a_{U,V,W}((uv)w)=\mathrm{\Phi }(u(vw)),`$
$`๐_{U,V,W}((uv)w)=(u(vw))\mathrm{\Phi }^1.`$
The category $`{}_{H}{}^{}`$ ($`_H`$) of left (right) $`H`$-modules becomes a monoidal category (see for the terminology) with tensor product $``$ given via $`\mathrm{\Delta }`$, associativity constraints $`a_{U,V,W}`$ ($`๐_{U,V,W}`$), unit $`k`$ as a trivial $`H`$-module and the usual left and right unit constraints.
Now, let $`H`$ be a quasi-bialgebra. We say that a $`k`$-vector space $`A`$ is a left $`H`$-module algebra if it is an algebra in the monoidal category $`{}_{H}{}^{}`$, that is $`A`$ has a multiplication and a usual unit $`1_A`$ satisfying the following conditions:
(2.17) $`(aa{}_{}{}^{})a{}_{}{}^{\prime \prime }=(X^1a)[(X^2a{}_{}{}^{})(X^3a{}_{}{}^{\prime \prime })],`$
(2.18) $`h(aa{}_{}{}^{})=(h_1a)(h_2a{}_{}{}^{}),`$
(2.19) $`h1_A=\epsilon (h)1_A,`$
for all $`a,a{}_{}{}^{},a^{\prime \prime }A`$ and $`hH`$, where $`haha`$ is the left $`H`$-module structure of $`A`$. Following we define the smash product $`A\mathrm{\#}H`$ as follows: as vector space $`A\mathrm{\#}H`$ is $`AH`$ (elements $`ah`$ will be written $`a\mathrm{\#}h`$) with multiplication given by
(2.20)
$$(a\mathrm{\#}h)(a{}_{}{}^{}\mathrm{\#}h{}_{}{}^{})=(x^1a)(x^2h_1a{}_{}{}^{})\mathrm{\#}x^3h_2h{}_{}{}^{},$$
for all $`a,a{}_{}{}^{}A`$, $`h,h{}_{}{}^{}H`$. This $`A\mathrm{\#}H`$ is an associative algebra with unit $`1_A\mathrm{\#}1_H`$ and it is defined by a universal property (as Heyneman and Sweedler did for Hopf algebras), see . It is easy to see that $`H`$ is a subalgebra of $`A\mathrm{\#}H`$ via $`h1\mathrm{\#}h`$, $`A`$ is a $`k`$-subspace of $`A\mathrm{\#}H`$ via $`aa\mathrm{\#}1`$ and the following relations hold:
(2.21)
$$(a\mathrm{\#}h)(1\mathrm{\#}h{}_{}{}^{})=a\mathrm{\#}hh{}_{}{}^{},(1\mathrm{\#}h)(a\mathrm{\#}h{}_{}{}^{})=h_1a\mathrm{\#}h_2h{}_{}{}^{},$$
for all $`aA`$, $`h,h{}_{}{}^{}H`$.
For further use we need the notion of right $`H`$-module algebra. Let $`H`$ be a quasi-bialgebra. We say that a $`k`$-linear space $`B`$ is a right $`H`$-module algebra if $`B`$ is an algebra in the monoidal category $`_H`$, i.e. $`B`$ has a multiplication and a usual unit $`1_B`$ satisfying the following conditions:
(2.22) $`(bb{}_{}{}^{})b{}_{}{}^{\prime \prime }=(bx^1)[(b{}_{}{}^{}x^2)(b{}_{}{}^{\prime \prime }x^3)],`$
(2.23) $`(bb{}_{}{}^{})h=(bh_1)(b{}_{}{}^{}h_2),`$
(2.24) $`1_Bh=\epsilon (h)1_B,`$
for all $`b,b^{},b^{\prime \prime }B`$ and $`hH`$, where $`bhbh`$ is the right $`H`$-module structure of $`B`$. Also, we can define a (right-handed) smash product $`H\mathrm{\#}B`$ as follows: as vector space $`H\mathrm{\#}B`$ is $`HB`$ (elements $`hb`$ will be written $`h\mathrm{\#}b`$) with multiplication:
(2.25)
$$(h\mathrm{\#}b)(h{}_{}{}^{}\mathrm{\#}b{}_{}{}^{})=hh_1^{}x^1\mathrm{\#}(bh_2^{}x^2)(b{}_{}{}^{}x^3),$$
for all $`b,b{}_{}{}^{}B`$, $`h,h{}_{}{}^{}H`$. This $`H\mathrm{\#}B`$ is an associative algebra with unit $`1_H\mathrm{\#}1_B`$. In fact, one can see that $`B^{op}`$ becomes a left $`H^{op,cop}`$-module algebra and under the trivial permutation of tensor factors we have $`(B^{op}\mathrm{\#}H^{op,cop})^{op}=H\mathrm{\#}B`$.
### 2.3. Comodule algebras and generalized smash products
Recall from the notion of comodule algebra over a quasi-bialgebra.
###### Definition 2.1.
Let $`H`$ be a quasi-bialgebra. A unital associative algebra $`๐`$ is called a right $`H`$-comodule algebra if there exist an algebra morphism $`\rho :๐๐H`$ and an invertible element $`\mathrm{\Phi }_\rho ๐HH`$ such that:
(2.26) $`\mathrm{\Phi }_\rho (\rho id)(\rho (๐))=(id\mathrm{\Delta })(\rho (๐))\mathrm{\Phi }_\rho ,\text{ }\text{ }๐๐\text{,}`$
$`(1_๐\mathrm{\Phi })(id\mathrm{\Delta }id)(\mathrm{\Phi }_\rho )(\mathrm{\Phi }_\rho 1_H)`$
(2.27) $`=(idid\mathrm{\Delta })(\mathrm{\Phi }_\rho )(\rho idid)(\mathrm{\Phi }_\rho ),`$
(2.28) $`(id\epsilon )\rho =id,`$
(2.29) $`(id\epsilon id)(\mathrm{\Phi }_\rho )=(idid\epsilon )(\mathrm{\Phi }_\rho )=1_๐1_H.`$
Similarly, a unital associative algebra $`๐
`$ is called a left $`H`$-comodule algebra if there exist an algebra morphism $`\lambda :๐
H๐
`$ and an invertible element $`\mathrm{\Phi }_\lambda HH๐
`$ such that the following relations hold:
(2.30) $`(id\lambda )(\lambda (๐))\mathrm{\Phi }_\lambda =\mathrm{\Phi }_\lambda (\mathrm{\Delta }id)(\lambda (๐)),\text{ }\text{ }๐๐
\text{,}`$
$`(1_H\mathrm{\Phi }_\lambda )(id\mathrm{\Delta }id)(\mathrm{\Phi }_\lambda )(\mathrm{\Phi }1_๐
)`$
(2.31) $`=(idid\lambda )(\mathrm{\Phi }_\lambda )(\mathrm{\Delta }idid)(\mathrm{\Phi }_\lambda ),`$
(2.32) $`(\epsilon id)\lambda =id,`$
(2.33) $`(id\epsilon id)(\mathrm{\Phi }_\lambda )=(\epsilon idid)(\mathrm{\Phi }_\lambda )=1_H1_๐
.`$
When $`H`$ is a quasi-bialgebra, particular examples of left and right $`H`$-comodule algebras are given by $`๐=๐
=H`$ and $`\rho =\lambda =\mathrm{\Delta }`$, $`\mathrm{\Phi }_\rho =\mathrm{\Phi }_\lambda =\mathrm{\Phi }`$.
For a right $`H`$-comodule algebra $`(๐,\rho ,\mathrm{\Phi }_\rho )`$ we will denote
$$\rho (๐)=๐_0๐_1,\text{ }(\rho id)(\rho (๐))=๐_{0,0}๐_{0,1}๐_1\text{ etc.}$$
for any $`๐๐`$. Similarly, for a left $`H`$-comodule algebra $`(๐
,\lambda ,\mathrm{\Phi }_\lambda )`$, if $`๐๐
`$ then we will denote
$$\lambda (๐)=๐_{[1]}๐_{[0]},\text{ }(id\lambda )(\lambda (๐))=๐_{[1]}๐_{[0,1]}๐_{[0,0]}\text{ etc.}$$
In analogy with the notation for the reassociator $`\mathrm{\Phi }`$ of $`H`$, we will write
$`\mathrm{\Phi }_\rho =\stackrel{~}{X}_\rho ^1\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3=\stackrel{~}{Y}_\rho ^1\stackrel{~}{Y}_\rho ^2\stackrel{~}{Y}_\rho ^3=\mathrm{}`$
$`\mathrm{\Phi }_\rho ^1=\stackrel{~}{x}_\rho ^1\stackrel{~}{x}_\rho ^2\stackrel{~}{x}_\rho ^3=\stackrel{~}{y}_\rho ^1\stackrel{~}{y}_\rho ^2\stackrel{~}{y}_\rho ^3=\mathrm{}`$
and similarly for the element $`\mathrm{\Phi }_\lambda `$ of a left $`H`$-comodule algebra $`๐
`$. When there is no danger of confusion we will omit the subscripts $`\rho `$ or $`\lambda `$ for the tensor components of the elements $`\mathrm{\Phi }_\rho `$, $`\mathrm{\Phi }_\lambda `$ or for the tensor components of the elements $`\mathrm{\Phi }_\rho ^1`$, $`\mathrm{\Phi }_\lambda ^1`$.
If $`๐`$ is a right $`H`$-comodule algebra then we define the elements $`\stackrel{~}{p}_\rho ,\stackrel{~}{q}_\rho ๐H`$ as follows:
(2.34)
$$\stackrel{~}{p}_\rho =\stackrel{~}{p}_\rho ^1\stackrel{~}{p}_\rho ^2=\stackrel{~}{x}_\rho ^1\stackrel{~}{x}_\rho ^2\beta S(\stackrel{~}{x}_\rho ^3),\text{ }\stackrel{~}{q}_\rho =\stackrel{~}{q}_\rho ^1\stackrel{~}{q}_\rho ^2=\stackrel{~}{X}_\rho ^1S^1(\alpha \stackrel{~}{X}_\rho ^3)\stackrel{~}{X}_\rho ^2.$$
By \[7, Lemma 9.1\], we have the following relations, for all $`๐๐`$:
(2.35) $`\rho (๐_{<0>})\stackrel{~}{p}_\rho [1_๐S(๐_{<1>})]=\stackrel{~}{p}_\rho [๐1_H],`$
(2.36) $`[1_๐S^1(๐_{<1>})]\stackrel{~}{q}_\rho \rho (๐_{<0>})=[๐1_H]\stackrel{~}{q}_\rho ,`$
(2.37) $`\rho (\stackrel{~}{q}_\rho ^1)\stackrel{~}{p}_\rho [1_๐S(\stackrel{~}{q}_\rho ^2)]=1_๐1_H,`$
(2.38) $`[1_๐S^1(\stackrel{~}{p}_\rho ^2)]\stackrel{~}{q}_\rho \rho (\stackrel{~}{p}_\rho ^1)=1_๐1_H,`$
$`\mathrm{\Phi }_\rho (\rho id_H)(\stackrel{~}{p}_\rho )(\stackrel{~}{p}_\rho id_H)`$
(2.39) $`=(id_๐\mathrm{\Delta })(\rho (\stackrel{~}{x}_\rho ^1)\stackrel{~}{p}_\rho )(1_๐g^1S(\stackrel{~}{x}_\rho ^3)g^2S(\stackrel{~}{x}_\rho ^2)),`$
$`(\stackrel{~}{q}_\rho 1_H)(\rho id_H)(\stackrel{~}{q}_\rho )\mathrm{\Phi }_\rho ^1`$
(2.40) $`=[1_๐S^1(f^2\stackrel{~}{X}_\rho ^3)S^1(f^1\stackrel{~}{X}_\rho ^2)](id_๐\mathrm{\Delta })(\stackrel{~}{q}_\rho \rho (\stackrel{~}{X}_\rho ^1)),`$
where $`f=f^1f^2`$ is the element defined in (2.13) and $`f^1=g^1g^2`$.
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra and $`๐
`$ a left $`H`$-comodule algebra. Denote by $`A<๐
`$ the $`k`$-vector space $`A๐
`$ with multiplication:
(2.41)
$$(a<๐)(a^{}<๐^{})=(\stackrel{~}{x}_\lambda ^1a)(\stackrel{~}{x}_\lambda ^2๐_{[1]}a^{})<\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{},$$
for all $`a,a^{}A`$ and $`๐,๐^{}๐
`$. By , $`A<๐
`$ is an associative algebra with unit $`1_A<1_๐
`$. If we take $`๐
=H`$ then $`A<H`$ is just the smash product $`A\mathrm{\#}H`$. For this reason the algebra $`A<๐
`$ is called the generalized smash product of $`A`$ and $`๐
`$.
Similarly, if $`B`$ is a right $`H`$-module algebra and $`๐`$ is a right $`H`$-comodule algebra, then we denote by $`๐\text{ }>\text{ }B`$ the $`k`$-vector space $`๐B`$ with the newly defined multiplication
(2.42)
$$(๐\text{ }>\text{ }b)(๐{}_{}{}^{}\text{ }>\text{ }b{}_{}{}^{})=๐๐_0^{}\stackrel{~}{x}_\rho ^1\text{ }>\text{ }(b๐_1^{}\stackrel{~}{x}_\rho ^2)(b{}_{}{}^{}\stackrel{~}{x}_\rho ^3),$$
for all $`๐,๐{}_{}{}^{}๐`$ and $`b,b{}_{}{}^{}B`$. It is easy to see that $`๐\text{ }>\text{ }B`$ is an associative algebra with unit $`1_๐\text{ }>\text{ }1_B`$. Of course, if $`๐=H`$ then $`H\text{ }>\text{ }B=H\mathrm{\#}B`$ as algebras.
### 2.4. Bimodule algebras and bicomodule algebras
The following definition was introduced in under the name โquasi-commuting pair of $`H`$-coactionsโ.
###### Definition 2.2.
Let $`H`$ be a quasi-bialgebra. By an $`H`$-bicomodule algebra $`๐ธ`$ we mean a quintuple $`(\lambda ,\rho ,\mathrm{\Phi }_\lambda ,\mathrm{\Phi }_\rho ,\mathrm{\Phi }_{\lambda ,\rho })`$, where $`\lambda `$ and $`\rho `$ are left and right $`H`$-coactions on $`๐ธ`$, respectively, and where $`\mathrm{\Phi }_\lambda HH๐ธ`$, $`\mathrm{\Phi }_\rho ๐ธHH`$ and $`\mathrm{\Phi }_{\lambda ,\rho }H๐ธH`$ are invertible elements, such that:
* $`(๐ธ,\lambda ,\mathrm{\Phi }_\lambda )`$ is a left $`H`$-comodule algebra;
* $`(๐ธ,\rho ,\mathrm{\Phi }_\rho )`$ is a right $`H`$-comodule algebra;
* the following compatibility relations hold:
(2.43) $`\mathrm{\Phi }_{\lambda ,\rho }(\lambda id)(\rho (u))=(id\rho )(\lambda (u))\mathrm{\Phi }_{\lambda ,\rho },\text{ }\text{ }u๐ธ\text{,}`$
$`(1_H\mathrm{\Phi }_{\lambda ,\rho })(id\lambda id)(\mathrm{\Phi }_{\lambda ,\rho })(\mathrm{\Phi }_\lambda 1_H)`$
(2.44) $`=(idid\rho )(\mathrm{\Phi }_\lambda )(\mathrm{\Delta }idid)(\mathrm{\Phi }_{\lambda ,\rho }),`$
$`(1_H\mathrm{\Phi }_\rho )(id\rho id)(\mathrm{\Phi }_{\lambda ,\rho })(\mathrm{\Phi }_{\lambda ,\rho }1_H)`$
(2.45) $`=(idid\mathrm{\Delta })(\mathrm{\Phi }_{\lambda ,\rho })(\lambda idid)(\mathrm{\Phi }_\rho ).`$
As pointed out in , if $`๐ธ`$ is a bicomodule algebra then, in addition, we have that
(2.46)
$$(id_Hid_๐ธ\epsilon )(\mathrm{\Phi }_{\lambda ,\rho })=1_H1_๐ธ,\text{ }(\epsilon id_๐ธid_H)(\mathrm{\Phi }_{\lambda ,\rho })=1_๐ธ1_H.$$
As a first example of a bicomodule algebra is $`๐ธ=H`$, $`\lambda =\rho =\mathrm{\Delta }`$ and $`\mathrm{\Phi }_\lambda =\mathrm{\Phi }_\rho =\mathrm{\Phi }_{\lambda ,\rho }=\mathrm{\Phi }`$. For the left and right comodule algebra structures of $`๐ธ`$ we will use notation as above. For simplicity we denote
$`\mathrm{\Phi }_{\lambda ,\rho }=\mathrm{\Theta }^1\mathrm{\Theta }^2\mathrm{\Theta }^3=๐ฏ^1๐ฏ^2๐ฏ^3=\overline{\mathrm{\Theta }}^1\overline{\mathrm{\Theta }}^2\overline{\mathrm{\Theta }}^3,`$
$`\mathrm{\Phi }_{\lambda ,\rho }^1=\theta ^1\theta ^2\theta ^3=\stackrel{~}{\theta }^1\stackrel{~}{\theta }^2\stackrel{~}{\theta }^3=\overline{\theta }^1\overline{\theta }^2\overline{\theta }^3.`$
As we mentioned before, if $`H`$ is a quasi-bialgebra then so is $`H^{op}`$, where โopโ means the opposite multiplication. The reassociator of $`H^{op}`$ is $`\mathrm{\Phi }_{op}=\mathrm{\Phi }^1`$. Hence $`HH^{op}`$ is a quasi-bialgebra with reassociator
(2.47)
$$\mathrm{\Phi }_{HH^{op}}=(X^1x^1)(X^2x^2)(X^3x^3).$$
If we identify left $`HH^{op}`$-modules with $`H`$-bimodules, then the category of $`H`$-bimodules, $`{}_{H}{}^{}_{H}^{}`$, is monoidal, the associativity constraints being given by $`๐_{U,V,W}^{}:(UV)WU(VW)`$,
(2.48)
$$๐_{}^{}{}_{U,V,W}{}^{}((uv)w)=\mathrm{\Phi }(u(vw))\mathrm{\Phi }^1,$$
for any $`U,V,W{}_{H}{}^{}_{H}^{}`$ and $`uU`$, $`vV`$ and $`wW`$. Therefore, we can define algebras in the category of $`H`$-bimodules. Such an algebra will be called an $`H`$-bimodule algebra. More exactly, a $`k`$-vector space $`๐`$ is an $`H`$-bimodule algebra if $`๐`$ is an $`H`$-bimodule (denote the actions by $`h\phi `$ and $`\phi h`$, for $`hH`$ and $`\phi ๐`$) which has a multiplication and a usual unit $`1_๐`$ such that for all $`\phi ,\phi ^{},\phi ^{\prime \prime }๐`$ and $`hH`$ the following relations hold:
(2.49) $`(\phi \phi ^{})\phi ^{\prime \prime }=(X^1\phi x^1)[(X^2\phi ^{}x^2)(X^3\phi ^{\prime \prime }x^3)],`$
(2.50) $`h(\phi \phi ^{})=(h_1\phi )(h_2\phi ^{}),\text{ }(\phi \phi ^{})h=(\phi h_1)(\phi ^{}h_2),`$
(2.51) $`h1_๐=\epsilon (h)1_๐,\text{ }1_๐h=\epsilon (h)1_๐.`$
Let $`H`$ be a quasi-bialgebra. Then $`H^{}`$, the linear dual of $`H`$, is an $`H`$-bimodule via the $`H`$-actions
(2.52)
$$h\phi ,h{}_{}{}^{}=\phi (h{}_{}{}^{}h),\text{ }\phi h,h{}_{}{}^{}=\phi (hh{}_{}{}^{}),$$
for all $`\phi H^{}`$ and $`h,h{}_{}{}^{}H`$. The convolution $`\phi \psi ,h=\phi (h_1)\psi (h_2)`$, $`\phi ,\psi H^{}`$, $`hH`$, is a multiplication on $`H^{}`$; it is not in general associative, but with this multiplication $`H^{}`$ becomes an $`H`$-bimodule algebra.
## 3. Generalized diagonal crossed products
In order to define the generalized diagonal crossed products we need the notion of two-sided coaction.
Let $`H`$ be a quasi-bialgebra and $`๐ธ`$ a unital associative algebra. Recall from that a two-sided coaction of $`H`$ on $`๐ธ`$ is a pair $`(\delta ,\mathrm{\Psi })`$ where $`\delta :๐ธH๐ธH`$ is an algebra map and $`\mathrm{\Psi }H^2๐ธH^2`$ is an invertible element such that the following relations hold:
(3.1) $`(id_H\delta id_H)(\delta (u))\mathrm{\Psi }=\mathrm{\Psi }(\mathrm{\Delta }id_๐ธ\mathrm{\Delta })(\delta (u)),u๐ธ,`$
$`(1_H\mathrm{\Psi }1_H)(id_H\mathrm{\Delta }id_๐ธ\mathrm{\Delta }id_H)(\mathrm{\Psi })(\mathrm{\Phi }id_๐ธ\mathrm{\Phi }^1)`$
(3.2) $`=(id_Hid_H\delta id_Hid_H)(\mathrm{\Psi })(\mathrm{\Delta }id_Hid_๐ธid_H\mathrm{\Delta })(\mathrm{\Psi }),`$
(3.3) $`(\epsilon id_๐ธ\epsilon )\delta =id_๐ธ,`$
$`(id_H\epsilon id_๐ธ\epsilon id_H)(\mathrm{\Psi })`$
(3.4) $`=(\epsilon id_Hid_๐ธid_H\epsilon )(\mathrm{\Psi })=1_H1_๐ธ1_H.`$
If $`H`$ is a quasi-bialgebra then to any $`H`$-bicomodule algebra $`(๐ธ,\lambda ,\rho ,\mathrm{\Phi }_\lambda ,\mathrm{\Phi }_\rho ,\mathrm{\Phi }_{\lambda ,\rho })`$ one can associate (see ) two two-sided $`H`$-coactions, denoted by $`(\delta _l,\mathrm{\Psi }_l)`$ and $`(\delta _r,\mathrm{\Psi }_r)`$. More precisely
(3.5)
$$\{\begin{array}{cc}\delta _l=(\lambda id_H)\rho ,\hfill & \\ \mathrm{\Psi }_l:=(id_H\lambda id_H^2)\left((\mathrm{\Phi }_{\lambda ,\rho }1_H)(\lambda id_H^2)(\mathrm{\Phi }_\rho ^1)\right)[\mathrm{\Phi }_\lambda 1_H^2],\hfill & \end{array}$$
and
(3.6)
$$\{\begin{array}{cc}\delta _r=(id_H\rho )\lambda ,\hfill & \\ \mathrm{\Psi }_r=(id_H^2\rho id_H)\left((1_H\mathrm{\Phi }_{\lambda ,\rho }^1)(id_H^2\rho )(\mathrm{\Phi }_\lambda )\right)[1_H^2\mathrm{\Phi }_\rho ^1].\hfill & \end{array}$$
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`(\delta ,\mathrm{\Psi })`$ a two-sided coaction of $`H`$ on a unital associative algebra $`๐ธ`$. Denote $`\delta (u):=u_{(1)}u_{(0)}u_{(1)}`$, for all $`u๐ธ`$, $`\mathrm{\Psi }=\mathrm{\Psi }^1\mathrm{}\mathrm{\Psi }^5`$, $`\mathrm{\Psi }^1=\overline{\mathrm{\Psi }}^1\mathrm{}\overline{\mathrm{\Psi }}^5`$, and then define
(3.7) $`\mathrm{\Omega }_\delta =\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5=\overline{\mathrm{\Psi }}^1\overline{\mathrm{\Psi }}^2\overline{\mathrm{\Psi }}^3S^1(f^1\overline{\mathrm{\Psi }}^4)S^1(f^2\overline{\mathrm{\Psi }}^5),`$
(3.8) $`\mathrm{\Omega }_\delta ^{}=\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5=S^1(\mathrm{\Psi }^1g^1)S^1(\mathrm{\Psi }^2g^2)\mathrm{\Psi }^3\mathrm{\Psi }^4\mathrm{\Psi }^5.`$
Here $`f=f^1f^2`$ is the twist defined in (2.13) and $`f^1=g^1g^2`$ is its inverse.
We denote by $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$ the $`k`$-vector spaces $`๐๐ธ`$ and respectively $`๐ธ๐`$, furnished with the multiplications given respectively by:
$`(\phi _\delta u)(\phi ^{}_\delta u{}_{}{}^{})`$
(3.9) $`=(\mathrm{\Omega }_\delta ^1\phi \mathrm{\Omega }_\delta ^5)(\mathrm{\Omega }_\delta ^2u_{(1)}\phi ^{}S^1(u_{(1)})\mathrm{\Omega }_\delta ^4)_\delta \mathrm{\Omega }_\delta ^3u_{(0)}u^{},`$
$`(u_\delta \phi )(u^{}_\delta \phi ^{})`$
(3.10) $`=uu_{(0)}^{}\mathrm{\Omega }_\delta ^3_\delta (\mathrm{\Omega }_\delta ^2S^1(u_{(1)}^{})\phi u_{(1)}^{}\mathrm{\Omega }_\delta ^4)(\mathrm{\Omega }_\delta ^1\phi ^{}\mathrm{\Omega }_\delta ^5),`$
for all $`u,u^{}๐ธ`$ and $`\phi ,\phi ^{}๐`$, where we write $`\phi _\delta u`$ and $`u_\delta \phi `$ in place of $`\phi u`$ and respectively $`u\phi `$ to distinguish the new algebraic structures, and where $`\mathrm{\Omega }_\delta =\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5`$ and $`\mathrm{\Omega }_\delta ^{}=\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5`$ are the elements defined by (3.7) and (3.8), respectively. We call $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$ the left, and respectively right, generalized diagonal crossed product between $`๐`$ and $`๐ธ`$.
The following (technical) lemma, expressing some relations fulfilled by the elements $`\mathrm{\Omega }_\delta `$ and $`\mathrm{\Omega }_\delta ^{}`$, will be essential in the sequel. It will help us to prove that the generalized diagonal crossed products defined above are associative algebras, and moreover it will allow us to regard an $`H`$-bicomodule algebra $`๐ธ`$, in two ways, as a left $`HH^{op}`$-comodule algebra. We would like to stress that for these two aims, the explicit formulae for $`\mathrm{\Omega }_\delta `$ and $`\mathrm{\Omega }_\delta ^{}`$ are not so important, any other elements satisfying the relations in the lemma (plus some other minor conditions) are equally good, so it would be a natural question to ask whether there exist other such elements.
###### Lemma 3.1.
Let $`H`$ be a quasi-Hopf algbera, $`๐ธ`$ a unital associative algebra and $`(\delta ,\mathrm{\Psi })`$ a two-sided coaction of $`H`$ on $`๐ธ`$.
* Let $`\mathrm{\Omega }_\delta =\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5=\overline{\mathrm{\Omega }}_\delta ^1\mathrm{}\overline{\mathrm{\Omega }}_\delta ^5`$ be the element defined by (3.7). Then for all $`u๐ธ`$ the following relations hold:
$`\mathrm{\Omega }_\delta ^1u_{(1)}\mathrm{\Omega }_\delta ^2u_{(0,1)}\mathrm{\Omega }_\delta ^3u_{(0,0)}S^1(u_{(0,1)})\mathrm{\Omega }_\delta ^4S^1(u_{(1)})\mathrm{\Omega }_\delta ^5`$
(3.11) $`=u_{(1)_1}\mathrm{\Omega }_\delta ^1u_{(1)_2}\mathrm{\Omega }_\delta ^2u_{(0)}\mathrm{\Omega }_\delta ^3\mathrm{\Omega }_\delta ^4S^1(u_{(1)})_2\mathrm{\Omega }_\delta ^5S^1(u_{(1)})_1,`$
$`X^1(\overline{\mathrm{\Omega }}_\delta ^1)_1\mathrm{\Omega }_\delta ^1X^2(\overline{\mathrm{\Omega }}_\delta ^1)_2\mathrm{\Omega }_\delta ^2X^3\overline{\mathrm{\Omega }}_\delta ^2(\mathrm{\Omega }_\delta ^3)_{(1)}\overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_{(0)}^3S^1((\mathrm{\Omega }_\delta ^3)_{(1)})\overline{\mathrm{\Omega }}_\delta ^4x^3`$
$`\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^5)_2x^2\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^5)_1x^1=\overline{\mathrm{\Omega }}_\delta ^1(\overline{\mathrm{\Omega }}_\delta ^2)_1\mathrm{\Omega }_\delta ^1(\overline{\mathrm{\Omega }}_\delta ^2)_2\mathrm{\Omega }_\delta ^2`$
(3.12) $`\overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_\delta ^3\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^4)_2\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^4)_1\overline{\mathrm{\Omega }}_\delta ^5.`$
* Let $`\mathrm{\Omega }_\delta ^{}=\mathrm{\Omega }_\delta ^1\mathrm{}\mathrm{\Omega }_\delta ^5=\overline{\mathrm{\Omega }}_\delta ^1\mathrm{}\overline{\mathrm{\Omega }}_\delta ^5`$ be the element defined by (3.8). Then for all $`u๐ธ`$ the following relations hold:
$`\mathrm{\Omega }_\delta ^1S^1(u_{(1)})\mathrm{\Omega }_\delta ^2S^1(u_{(0,1)})u_{(0,0)}\mathrm{\Omega }_\delta ^3u_{(0,1)}\mathrm{\Omega }_\delta ^4u_{(1)}\mathrm{\Omega }_\delta ^5`$
(3.13) $`=S^1(u_{(1)})_2\mathrm{\Omega }_\delta ^1S^1(u_{(1)})_1\mathrm{\Omega }_\delta ^2\mathrm{\Omega }_\delta ^3u_{(0)}\mathrm{\Omega }_\delta ^4u_{(1)_1}\mathrm{\Omega }_\delta ^5u_{(1)_2},`$
$`X^3\overline{\mathrm{\Omega }}_\delta ^1X^2(\overline{\mathrm{\Omega }}_\delta ^2)_2\mathrm{\Omega }_\delta ^1X^1(\overline{\mathrm{\Omega }}_\delta ^2)_1\mathrm{\Omega }_\delta ^2\mathrm{\Omega }_\delta ^3\overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^4)_1x^1`$
$`\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^4)_2x^2\overline{\mathrm{\Omega }}_\delta ^5x^3=(\overline{\mathrm{\Omega }}_\delta ^1)_1\mathrm{\Omega }_\delta ^1(\overline{\mathrm{\Omega }}_\delta ^1)_2\mathrm{\Omega }_\delta ^2\overline{\mathrm{\Omega }}_\delta ^2S^1((\mathrm{\Omega }_\delta ^3)_{(1)})`$
(3.14) $`(\mathrm{\Omega }_\delta ^3)_{(0)}\overline{\mathrm{\Omega }}_\delta ^3(\mathrm{\Omega }_\delta ^3)_{(1)}\overline{\mathrm{\Omega }}_\delta ^4\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^5)_1\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^5)_2.`$
###### Proof.
We will prove only (a), (b) being similar. The relation (3.11) follows easily by applying (3.7), (3.1) and (2.11), the details are left to the reader. We prove now (3.12). We will not perform all the computations, but we will point out the relations that are used at every step. So, we compute:
$`X^1(\overline{\mathrm{\Omega }}_\delta ^1)_1\mathrm{\Omega }_\delta ^1X^2(\overline{\mathrm{\Omega }}_\delta ^1)_2\mathrm{\Omega }_\delta ^2X^3\overline{\mathrm{\Omega }}_\delta ^2(\mathrm{\Omega }_\delta ^3)_{(1)}\overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_{(0)}^3`$
$`S^1((\mathrm{\Omega }_\delta ^3)_{(1)})\overline{\mathrm{\Omega }}_\delta ^4x^3\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^5)_2x^2\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^5)_1x^1`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.7},\text{2.11})}{=}}`$ $`X^1\overline{\mathrm{\Psi }}_1^1\overline{๐ฟ}^1X^2\overline{\mathrm{\Psi }}_2^1\overline{๐ฟ}^2X^3\overline{\mathrm{\Psi }}^2\overline{๐ฟ}_{(1)}^3\overline{\mathrm{\Psi }}^3\overline{๐ฟ}_{(0)}^3S^1(f^1\overline{\mathrm{\Psi }}^4\overline{๐ฟ}_{(1)}^3)x^3`$
$`S^1(F^1f_1^2\overline{\mathrm{\Psi }}_1^5\overline{๐ฟ}^4)x^2S^1(F^2f_2^2\overline{\mathrm{\Psi }}_2^5\overline{๐ฟ}^5)x^1`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.2})}{=}}`$ $`\overline{๐ฟ}^1\overline{๐ฟ}_1^2\overline{\mathrm{\Psi }}^1\overline{๐ฟ}_2^2\overline{\mathrm{\Psi }}^2\overline{๐ฟ}^3\overline{\mathrm{\Psi }}^3S^1(S(x^3)f^1X^1\overline{๐ฟ}_1^4\overline{\mathrm{\Psi }}^4)`$
$`S^1(S(x^2)F^1f_1^2X^2\overline{๐ฟ}_2^4\overline{\mathrm{\Psi }}^5)S^1(S(x^1)F^2f_2^2X^3\overline{๐ฟ}^5)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.9},\text{2.16},\text{2.11})}{=}}`$ $`\overline{๐ฟ}^1\overline{๐ฟ}_1^2\overline{\mathrm{\Psi }}^1\overline{๐ฟ}_2^2\overline{\mathrm{\Psi }}^2\overline{๐ฟ}^3\overline{\mathrm{\Psi }}^3S^1(f^1\overline{\mathrm{\Psi }}^4)S^1(F^1\overline{๐ฟ}^4)_2`$
$`S^1(f^2\overline{\mathrm{\Psi }}^5)S^1(F^1\overline{๐ฟ}^4)_1S^1(F^2\overline{๐ฟ}^5)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.7})}{=}}`$ $`\overline{\mathrm{\Omega }}_\delta ^1(\overline{\mathrm{\Omega }}_\delta ^2)_1\mathrm{\Omega }_\delta ^1(\overline{\mathrm{\Omega }}_\delta ^2)_2\mathrm{\Omega }_\delta ^2\overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_\delta ^3\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^4)_2\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^4)_1\overline{\mathrm{\Omega }}_\delta ^5,`$
as claimed. We denoted by $`\overline{๐ฟ}^1\mathrm{}\overline{๐ฟ}^5`$ another copy of $`\mathrm{\Psi }^1`$ and by $`F^1F^2`$ another copy of the Drinfeld twist $`f`$ defined in (2.13). โ
Suppose now that $`๐ธ`$ is an $`H`$-bicomodule algebra and let $`(\delta ,\mathrm{\Psi })=(\delta _{l/r},\mathrm{\Psi }_{\lambda /r})`$ be the two-sided coactions defined by (3.5) and (3.6), respectively. For simplicity we denote $`\mathrm{\Omega }=\mathrm{\Omega }_{\delta _l}`$, $`\omega =\mathrm{\Omega }_{\delta _r}`$, $`\mathrm{\Omega }^{}=\mathrm{\Omega }_{\delta _l}^{}`$ and $`\omega ^{}=\mathrm{\Omega }_{\delta _r}^{}`$. Concretely, the elements $`\mathrm{\Omega },\omega H^2๐ธH^2`$ come out as
$`\mathrm{\Omega }=(\stackrel{~}{X}_\rho ^1)_{[1]_1}\stackrel{~}{x}_\lambda ^1\theta ^1(\stackrel{~}{X}_\rho ^1)_{[1]_2}\stackrel{~}{x}_\lambda ^2\theta _{[1]}^2`$
(3.15) $`(\stackrel{~}{X}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2S^1(f^1\stackrel{~}{X}_\rho ^2\theta ^3)S^1(f^2\stackrel{~}{X}_\rho ^3),`$
$`\omega =\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\lambda ^2\mathrm{\Theta }^1(\stackrel{~}{x}_\lambda ^3)_0\stackrel{~}{X}_\rho ^1\mathrm{\Theta }_0^2`$
(3.16) $`S^1(f^1(\stackrel{~}{x}_\lambda ^3)_{1_1}\stackrel{~}{X}_\rho ^2\mathrm{\Theta }_1^2)S^1(f^2(\stackrel{~}{x}_\lambda ^3)_{1_2}\stackrel{~}{X}_\rho ^3\mathrm{\Theta }^3),`$
where $`\mathrm{\Phi }_\rho =\stackrel{~}{X}_\rho ^1\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3`$, $`\mathrm{\Phi }_\lambda ^1=\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\lambda ^2\stackrel{~}{x}_\lambda ^3`$, $`\mathrm{\Phi }_{\lambda ,\rho }=\mathrm{\Theta }^1\mathrm{\Theta }^2\mathrm{\Theta }^3`$, $`\mathrm{\Phi }_{\lambda ,\rho }^1=\theta ^1\theta ^2\theta ^3`$ and $`f=f^1f^2`$ is the twist defined in (2.13).
For further use we record the fact that the formulae in Lemma 3.1 (a) specialize to $`(\delta _{l/r},\mathrm{\Psi }_{l/r})`$ as follows (for all $`u๐ธ`$):
$`\mathrm{\Omega }^1u_{0_{[1]}}\mathrm{\Omega }^2u_{0_{[0]_{0_{[1]}}}}\mathrm{\Omega }^3u_{0_{[0]_{0_{[0]}}}}S^1(u_{0_{[0]_1}})\mathrm{\Omega }^4S^1(u_1)\mathrm{\Omega }^5`$
(3.17) $`=u_{0_{[1]_1}}\mathrm{\Omega }^1u_{0_{[1]_2}}\mathrm{\Omega }^2u_{0_{[0]}}\mathrm{\Omega }^3\mathrm{\Omega }^4S^1(u_1)_2\mathrm{\Omega }^5S^1(u_1)_1,`$
$`X^1\overline{\mathrm{\Omega }}_1^1\mathrm{\Omega }^1X^2\overline{\mathrm{\Omega }}_2^1\mathrm{\Omega }^2X^3\overline{\mathrm{\Omega }}^2\mathrm{\Omega }_{0_{[1]}}^3\overline{\mathrm{\Omega }}^3\mathrm{\Omega }_{0_{[0]}}^3S^1(\mathrm{\Omega }_1^3)\overline{\mathrm{\Omega }}^4x^3`$
(3.18) $`\mathrm{\Omega }^4\overline{\mathrm{\Omega }}_2^5x^2\mathrm{\Omega }^5\overline{\mathrm{\Omega }}_1^5x^1=\overline{\mathrm{\Omega }}^1\overline{\mathrm{\Omega }}_1^2\mathrm{\Omega }^1\overline{\mathrm{\Omega }}_2^2\mathrm{\Omega }^2\overline{\mathrm{\Omega }}^3\mathrm{\Omega }^3\mathrm{\Omega }^4\overline{\mathrm{\Omega }}_2^4\mathrm{\Omega }^5\overline{\mathrm{\Omega }}_1^4\overline{\mathrm{\Omega }}^5,`$
and respectively
$`\omega ^1u_{[1]}\omega ^2u_{[0]_{0_{[1]}}}\omega ^3u_{[0]_{0_{[0]_0}}}S^1(u_{[0]_{0_{[0]_1}}})\omega ^4S^1(u_{[0]_1})\omega ^5`$
(3.19) $`=u_{[1]_1}\omega ^1u_{[1]_2}\omega ^2u_{[0]_0}\omega ^3\omega ^4S^1(u_{[0]_1})_2\omega ^5S^1(u_{[0]_1})_1,`$
$`\overline{\omega }_1^1\omega ^1\overline{\omega }_2^1\omega ^2\overline{\omega }^2\omega _{[1]}^3\overline{\omega }^3\omega _{[0]_0}^3S^1(\omega _{[0]_1}^3)\overline{\omega }^4\omega ^4\overline{\omega }_2^5\omega ^5\overline{\omega }_1^5`$
(3.20) $`=x^1\overline{\omega }^1x^2\overline{\omega }_1^2\omega ^1x^3\overline{\omega }_2^2\omega ^2\overline{\omega }^3\omega ^3\omega ^4\overline{\omega }_2^4X^3\omega ^5\overline{\omega }_1^4X^2\overline{\omega }^5X^1,`$
where we denoted by $`\mathrm{\Omega }=\mathrm{\Omega }^1\mathrm{}\mathrm{\Omega }^5=\overline{\mathrm{\Omega }}^1\mathrm{}\overline{\mathrm{\Omega }}^5`$ the element defined in (3.15) and by $`\omega =\omega ^1\mathrm{}\omega ^5=\overline{\omega }^1\mathrm{}\overline{\omega }^5`$ the element defined in (3.16).
If $`(๐ธ,\lambda ,\rho ,\mathrm{\Phi }_\lambda ,\mathrm{\Phi }_\rho ,\mathrm{\Phi }_{\lambda ,\rho })`$ is an $`H`$-bicomodule algebra then it is not hard to see that $`๐ธ^{op,cop}:=(๐ธ^{\mathrm{op}},\tau _{๐ธ,H}\rho ,\tau _{H,๐ธ}\lambda ,\mathrm{\Phi }_\rho ^{321},\mathrm{\Phi }_\lambda ^{321},\mathrm{\Phi }_{\lambda ,\rho }^{321})`$ is an $`H^{\mathrm{op},\mathrm{cop}}`$-bicomodule algebra (by $`\tau _{X,Y}:XYYX`$ we denoted the switch map $`xyyx`$). Moreover, in $`H^{op,cop}`$ we have that the Drinfeld twist (defined for an arbitrary quasi-Hopf algebra in (2.13)) is given by $`f_{op,cop}=f_{21}^1=g^2g^1`$, where $`f`$ is the Drinfeld twist of $`H`$. Now, if we denote by $`\mathrm{\Omega }_{op,cop}`$ and $`\omega _{op,cop}`$ the elements $`\mathrm{\Omega }_{\delta _{l/r}}`$ corresponding to the $`H^{op,cop}`$-bicomodule algebra $`๐ธ^{op,cop}`$, then one can easily check that
$$\mathrm{\Omega }^{}=(\omega _{op,cop})^{54321}\mathrm{and}\omega ^{}=(\mathrm{\Omega }_{op,cop})^{54321},$$
so we restrict to the study of the elements $`\mathrm{\Omega }`$, $`\omega `$ and their associated constructions.
Finally, for this particular situation we denote $`๐_{\delta _l}๐ธ=๐๐ธ`$, $`๐_{\delta _r}๐ธ=๐\text{ }\text{ }๐ธ`$, $`๐ธ_{\delta _l}๐=๐ธ๐`$ and $`๐ธ_{\delta _r}๐=๐ธ\text{ }\text{ }๐`$, where $`๐`$ is an arbitrary $`H`$-bimodule algebra. So the first two constructions are left generalized diagonal crossed products and the last two are right generalized diagonal crossed products. For example, the multiplications in $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ are given by
$`(\phi u)(\phi ^{}u{}_{}{}^{})`$
(3.21) $`=(\mathrm{\Omega }^1\phi \mathrm{\Omega }^5)(\mathrm{\Omega }^2u_{0_{[1]}}\phi ^{}S^1(u_1)\mathrm{\Omega }^4)\mathrm{\Omega }^3u_{0_{[0]}}u{}_{}{}^{},`$
$`(\phi \text{ }\text{ }u)(\phi ^{}\text{ }\text{ }u{}_{}{}^{})`$
(3.22) $`=(\omega ^1\phi \omega ^5)(\omega ^2u_{[1]}\phi ^{}S^1(u_{[0]_1})\omega ^4)\text{ }\text{ }\omega ^3u_{[0]_0}u^{},`$
for all $`\phi ,\phi ^{}๐`$ and $`u,u^{}๐ธ`$, where we write $`\phi u`$ and $`\phi \text{ }\text{ }u`$ in place of $`\phi u`$ to distinguish the new algebraic structures.
We are now ready to show that the generalized diagonal crossed products are unital associative algebras.
###### Proposition 3.2.
Let $`H`$ be a quasi-Hopf algebra, $`๐ธ`$ a unital associative algebra and $`(\delta ,\mathrm{\Psi })`$ a two-sided coaction of $`H`$ on $`๐ธ`$. Consider $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$, the $`k`$-vector spaces $`๐๐ธ`$ and respectively $`๐ธ๐`$, endowed with the multiplications defined in (3.9) and (3.10), respectively. Then these products define on $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$ two structures of associative algebra with unit $`1_๐_\delta 1_๐ธ`$ (respectively $`1_๐ธ_\delta 1_๐`$), containing $`๐ธ1_๐_\delta ๐ธ`$ (respectively $`๐ธ๐ธ_\delta 1_๐`$) as unital subalgebra.
Consequently, if $`๐ธ`$ is an $`H`$-bicomodule algebra and $`๐`$ is an $`H`$-bimodule algebra then $`๐๐ธ`$, $`๐\text{ }\text{ }๐ธ`$, $`๐ธ๐`$ and $`๐ธ\text{ }\text{ }๐`$ are associative algebras containing $`๐ธ`$ as unital subalgebra.
###### Proof.
We will give the proof only for $`๐_\delta ๐ธ`$, the one for $`๐ธ_\delta ๐`$ being similar (it will use the relations satisfied by $`\mathrm{\Omega }_\delta ^{}`$, instead of the ones satisfied by $`\mathrm{\Omega }_\delta `$). For $`\phi ,\phi ^{},\phi ^{\prime \prime }๐`$ and $`u,u^{},u^{\prime \prime }๐ธ`$ we compute:
$`(\phi _\delta u)[(\phi ^{}_\delta u^{})(\phi ^{\prime \prime }_\delta u^{\prime \prime })]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.9})}{=}}`$ $`(\phi _\delta u)[(\mathrm{\Omega }_\delta ^1\phi ^{}\mathrm{\Omega }_\delta ^5)(\mathrm{\Omega }_\delta ^2u_{(1)}^{}\phi ^{\prime \prime }S^1(u_{(1)}^{})\mathrm{\Omega }_\delta ^4)_\delta \mathrm{\Omega }_\delta ^3u_{(0)}^{}u^{\prime \prime }]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.9},\text{2.50})}{=}}`$ $`(\overline{\mathrm{\Omega }}_\delta ^1\phi \overline{\mathrm{\Omega }}_\delta ^5)[((\overline{\mathrm{\Omega }}_\delta ^2)_1u_{(1)_1}\mathrm{\Omega }_\delta ^1\phi ^{}\mathrm{\Omega }_\delta ^5S^1(u_{(1)})_1(\overline{\mathrm{\Omega }}_\delta ^4)_1)((\overline{\mathrm{\Omega }}_\delta ^2)_2u_{(1)_2}`$
$`\times \mathrm{\Omega }_\delta ^2u_{(1)}^{}\phi ^{\prime \prime }S^1(u_{(1)}^{})\mathrm{\Omega }_\delta ^4S^1(u_{(1)})_2(\overline{\mathrm{\Omega }}_\delta ^4)_2)]_\delta \overline{\mathrm{\Omega }}^3_\delta u_{(0)}\mathrm{\Omega }^3_\delta u^{}_{(0)}u^{\prime \prime }`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.11})}{=}}`$ $`(\overline{\mathrm{\Omega }}_\delta ^1\phi \overline{\mathrm{\Omega }}_\delta ^5)[(\overline{\mathrm{\Omega }}_\delta ^2)_1\mathrm{\Omega }_\delta ^1u_{(1)}\phi ^{}S^1(u_{(1)})\mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^4)_1)`$
$`((\overline{\mathrm{\Omega }}_\delta ^2)_2\mathrm{\Omega }_\delta ^2u_{(0,1)}u_{(1)}^{}\phi ^{\prime \prime }S^1(u_{(0,1)}u{}_{}{}^{}{}_{(1)}{}^{})\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^4)_2)]`$
$`_\delta \overline{\mathrm{\Omega }}_\delta ^3\mathrm{\Omega }_\delta ^3u_{(0,0)}u_{(0)}^{}u^{\prime \prime }`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.12},\text{2.49})}{=}}`$ $`[((\overline{\mathrm{\Omega }}_\delta ^1)_1\mathrm{\Omega }_\delta ^1\phi \mathrm{\Omega }_\delta ^5(\overline{\mathrm{\Omega }}_\delta ^5)_1)((\overline{\mathrm{\Omega }}_\delta ^1)_2\mathrm{\Omega }_\delta ^2u_{(1)}\phi ^{}S^1(u_{(1)})\mathrm{\Omega }_\delta ^4(\overline{\mathrm{\Omega }}_\delta ^5)_2)]`$
$`(\overline{\mathrm{\Omega }}_\delta ^2(\mathrm{\Omega }_\delta ^3)_{(1)}u_{(0,1)}u_{(1)}^{}\phi ^{\prime \prime }S^1((\mathrm{\Omega }_\delta ^3)_{(1)}u_{(0,1)}u_{(1)}^{})\overline{\mathrm{\Omega }}_\delta ^4)`$
$`_\delta \overline{\mathrm{\Omega }}_\delta ^3(\mathrm{\Omega }_\delta ^3)_{(0)}u_{(0,0)}u_{(0)}^{}u^{\prime \prime }`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.50},\text{3.9})}{=}}`$ $`[(\mathrm{\Omega }_\delta ^1\phi \mathrm{\Omega }_\delta ^5)(\mathrm{\Omega }_\delta ^2u_{(1)}\phi ^{}S^1(u_{(1)})\mathrm{\Omega }_\delta ^4)_\delta \mathrm{\Omega }_\delta ^3u_{(0)}u^{}](\phi ^{\prime \prime }_\delta u^{\prime \prime })`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.9})}{=}}`$ $`[(\phi _\delta u)(\phi ^{}_\delta u^{})](\phi ^{\prime \prime }_\delta u^{\prime \prime }).`$
The fact that $`1_๐_\delta 1_๐ธ`$ is the unit follows easily from the (co) unit axioms. โ
###### Remark 3.3.
In the algebras $`๐_\delta ๐ธ`$ and $`๐ธ_\delta ๐`$ we have $`(\phi _\delta 1_๐ธ)(1_๐_\delta u)=\phi _\delta u`$ and $`(u_\delta 1_๐)(1_๐ธ_\delta \phi )=u_\delta u`$, for all $`\phi ๐`$ and $`u๐ธ`$.
###### Examples 3.4.
1) As we mentioned before, if $`H`$ is a quasi-Hopf algebra then $`H^{}`$ is an $`H`$-bimodule algebra, hence it makes sense to consider the algebras $`H^{}_\delta ๐ธ`$ and $`๐ธ_\delta H^{}`$, which are exactly the left and right diagonal crossed products constructed in . For this reason we called the algebras in Proposition 3.2 the generalized diagonal crossed products.
2) Let $`A`$ be a left $`H`$-module algebra. Then $`A`$ becomes an $`H`$-bimodule algebra, where the right $`H`$-action is given via $`\epsilon `$. In this particular case $`AH`$ and $`A\text{ }\text{ }H`$ coincide both to the smash product algebra $`A\mathrm{\#}H`$. Moreover, if we replace the quasi-Hopf algebra $`H`$ by an arbitrary $`H`$-bicomodule algebra $`๐ธ`$, then $`A๐ธ`$ and $`A\text{ }\text{ }๐ธ`$ coincide with the generalized smash product algebra $`A<๐ธ`$. Therefore, the generalized diagonal crossed products may be viewed as a generalization of the (generalized) smash product.
3) As we have already mentioned, $`H`$ itself is an $`H`$-bicomodule algebra. So, in this case, the multiplications of the generalized diagonal crossed products $`๐H`$ and $`๐\text{ }\text{ }H`$ specialize to
(3.23) $`(\phi h)(\phi ^{}h^{})=(\mathrm{\Omega }^1\phi \mathrm{\Omega }^5)(\mathrm{\Omega }^2h_{(1,1)}\phi ^{}S^1(h_2)\mathrm{\Omega }^4)\mathrm{\Omega }^3h_{(1,2)}h^{},`$
(3.24) $`(\phi \text{ }\text{ }h)(\phi ^{}\text{ }\text{ }h^{})=(\omega ^1\phi \omega ^5)(\omega ^2h_1\phi ^{}S^1(h_{(2,2)})\omega ^4)\text{ }\text{ }\omega ^3h_{(2,1)}h^{},`$
for all $`\phi ,\phi {}_{}{}^{}๐`$ and $`h,h^{}H`$, where $`\mathrm{\Omega }=\mathrm{\Omega }^1\mathrm{}\mathrm{\Omega }^5,\omega =\omega ^1\mathrm{}\omega ^5H^5`$ are now given by:
(3.25) $`\mathrm{\Omega }=X_{(1,1)}^1x^1y^1X_{(1,2)}^1x^2y_1^2X_2^1x^3y_2^2S^1(f^1X^2y^3)S^1(f^2X^3),`$
(3.26) $`\omega =x^1x^2Y^1x_1^3X^1Y_1^2S^1(f^1x_{(2,1)}^3X^2Y_2^2)S^1(f^2x_{(2,2)}^3X^3Y^3),`$
and where $`f=f^1f^2`$ is the twist defined in (2.13).
4) Let $`H`$ be an ordinary Hopf algebra with bijective antipode, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra in the usual (Hopf) sense. In this case the multiplications of $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ coincide, and are given by
(3.27)
$$(\phi u)(\phi {}_{}{}^{}u{}_{}{}^{})=\phi (u_{\{1\}}\phi {}_{}{}^{}S^1(u_{\{1\}}))u_{\{0\}}u{}_{}{}^{},$$
for all $`\phi ,\phi {}_{}{}^{}๐`$ and $`u,u{}_{}{}^{}๐ธ`$, where
$$u_{\{1\}}u_{\{0\}}u_{\{1\}}:=u_{0_{[1]}}u_{0_{[0]}}u_1=u_{[1]}u_{[0]_0}u_{[0]_1}.$$
This construction appears in , in a slightly different form (namely, with $`S`$ instead of $`S^1`$), under the name โgeneralized twisted smash productโ (a particular case, when $`๐ธ=H`$, was introduced in ).
Let $`H`$ be a quasi-Hopf algebra. For an $`H`$-bicomodule algebra $`๐ธ`$ and an $`H`$-bimodule algebra $`๐`$ the multiplications of the right generalized diagonal crossed products $`๐ธ๐`$ and $`๐ธ\text{ }\text{ }๐`$ are the following. If $`\mathrm{\Omega }^{}=\mathrm{\Omega }^1\mathrm{}\mathrm{\Omega }^5`$ and $`\omega ^{}=\omega ^1\mathrm{}\omega ^5`$ we then have
$`(u\phi )(u^{}\phi ^{})`$
(3.28) $`=uu_{0_{[0]}}^{}\mathrm{\Omega }^3(\mathrm{\Omega }^2S^1(u_{0_{[1]}}^{})\phi u_1^{}\mathrm{\Omega }^4)(\mathrm{\Omega }^1\phi ^{}\mathrm{\Omega }^5),`$
$`(u\text{ }\text{ }\phi )(u^{}\text{ }\text{ }\phi ^{})`$
(3.29) $`=uu_{[0]_0}^{}\omega ^3\text{ }\text{ }(\omega ^2S^1(u_{[1]}^{})\phi u_{[0]_1}^{}\omega ^4)(\omega ^1\phi ^{}\omega ^5),`$
for all $`u,u^{}๐ธ`$ and $`\phi ,\phi ^{}๐`$. We know from Proposition 3.2 that $`๐ธ๐`$ and $`๐ธ\text{ }\text{ }๐`$ are associative algebras with unit $`1_๐ธ1_๐`$ and $`1_๐ธ\text{ }\text{ }1_๐`$, respectively, containing $`๐ธ`$ as unital subalgebra. In fact, under the trivial permutation of tensor factors we have that
(3.30)
$$๐ธ๐(๐^{op}\text{ }\text{ }๐ธ^{op,cop})^{op},๐ธ\text{ }\text{ }๐(๐^{op}๐ธ^{op,cop})^{op},$$
where the left generalized diagonal crossed products are made over $`H^{op,cop}`$. Note that $`๐^{op}`$ becomes an $`H^{op,cop}`$-bimodule algebra via the actions $`h_{op}\phi _{op}h^{}=h^{}\phi h`$, for all $`h,h^{}H`$ and $`\phi ๐`$. In the sequel we will restrict to the study of the left generalized diagonal crossed products.
###### Remark 3.5.
Let $`H`$ be a quasi-Hopf algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. In , Hausser and Nill proved that the two left (right) diagonal crossed products $`H^{}๐ธ`$ ($`๐ธH^{}`$) and $`H^{}\text{ }\text{ }๐ธ`$ ($`๐ธ\text{ }\text{ }H^{}`$) are isomorphic as algebras, and then that these four diagonal crossed products are isomorphic as algebras. We will prove in the next section that such result is also true for generalized diagonal crossed products, but as a consequence of the fact that the (generalized) diagonal crossed products can be written as some generalized smash products, and of an explicit algebra isomorphism between $`๐๐ธ`$ and $`๐ธ๐`$.
###### Remark 3.6.
There exists a very general scheme, due to Schauenburg , for constructing associative algebras starting with a monoidal category acting on a category of modules, and it is likely that the generalized diagonal crossed products fit into this scheme. However, we have chosen to prove the associativity of $`๐_\delta ๐ธ`$ by direct computation, first because Schauenburgโs machinery is itself quite complicated, and second because the difficulty of our proof lies actually only in Lemma 3.1, which is needed anyway in the next section.
If $`H`$ is a finite dimensional quasi-Hopf algebra and $`๐ธ`$ is an $`H`$-bicomodule algebra, Hausser and Nill constructed a map $`\mathrm{\Gamma }`$ from $`H^{}`$ to the diagonal crossed product $`๐ธH^{}`$, having the property that $`๐ธH^{}`$ is generated as algebra by $`๐ธ`$ and $`\mathrm{\Gamma }(H^{})`$. Such a map may also be constructed for the generalized diagonal crossed products. We need first the following result.
###### Lemma 3.7.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then, for all $`\phi ๐`$, we have
$$\phi 1_๐ธ=(1_๐\stackrel{~}{q}_\rho ^1)((\stackrel{~}{p}_\rho ^1)_{[1]}\phi \stackrel{~}{q}_\rho ^2S^1(\stackrel{~}{p}_\rho ^2)(\stackrel{~}{p}_\rho ^1)_{[0]}),$$
where $`\stackrel{~}{p}_\rho `$ and $`\stackrel{~}{q}_\rho `$ are given by (2.34).
###### Proof.
We compute:
$`(1_๐\stackrel{~}{q}_\rho ^1)((\stackrel{~}{p}_\rho ^1)_{[1]}\phi \stackrel{~}{q}_\rho ^2S^1(\stackrel{~}{p}_\rho ^2)(\stackrel{~}{p}_\rho ^1)_{[0]})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`(\stackrel{~}{q}_\rho ^1)_{0_{[1]}}(\stackrel{~}{p}_\rho ^1)_{[1]}\phi \stackrel{~}{q}_\rho ^2S^1(\stackrel{~}{p}_\rho ^2)S^1((\stackrel{~}{q}_\rho ^1)_1)(\stackrel{~}{q}_\rho ^1)_{0_{[0]}}(\stackrel{~}{p}_\rho ^1)_{[0]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.37})}{=}}`$ $`\phi 1_๐ธ,`$
which finishes the proof. โ
###### Proposition 3.8.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Define the map $`\mathrm{\Gamma }:๐๐๐ธ`$,
(3.31)
$$\mathrm{\Gamma }(\phi )=(\stackrel{~}{p}_\rho ^1)_{[1]}\phi S^1(\stackrel{~}{p}_\rho ^2)(\stackrel{~}{p}_\rho ^1)_{[0]},$$
for all $`\phi ๐`$. Then $`๐๐ธ`$ is generated as algebra by $`๐ธ`$ and $`\mathrm{\Gamma }(๐)`$.
###### Proof.
By the previous lemma it follows that
$$\phi 1_๐ธ=(1_๐\stackrel{~}{q}_\rho ^1)\mathrm{\Gamma }(\phi \stackrel{~}{q}_\rho ^2),$$
for all $`\phi ๐`$, so for $`\phi ๐`$ and $`u๐ธ`$ we can write
$$\phi u=(1_๐\stackrel{~}{q}_\rho ^1)\mathrm{\Gamma }(\phi \stackrel{~}{q}_\rho ^2)(1_๐u),$$
finishing the proof. โ
We will see other properties of the map $`\mathrm{\Gamma }`$ in subsequent sections.
We prove now a sort of associativity property of generalized diagonal crossed products with respect to tensoring by an arbitrary associative algebra.
###### Proposition 3.9.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra, $`๐ธ`$ an $`H`$-bicomodule algebra and $`C`$ an associative algebra. On $`๐ธC`$ we have a (canonical) $`H`$-bicomodule algebra structure, yielding algebra isomorphisms
(3.32) $`๐(๐ธC)(๐๐ธ)C,`$
(3.33) $`๐\text{ }\text{ }(๐ธC)(๐\text{ }\text{ }๐ธ)C,`$
defined by the trivial identifications.
###### Proof.
The $`H`$-bicomodule algebra structure on $`๐ธC`$ is given in such a way that everything that happens on $`C`$ is trivial, for instance the right $`H`$-comodule algebra structure is:
$`\rho _{๐ธC}:๐ธC(๐ธC)H,`$
$`\rho _{๐ธC}(uc)=(u_0c)u_1,u๐ธ,cC,`$
$`(\mathrm{\Phi }_\rho )_{๐ธC}(๐ธC)HH,`$
$`(\mathrm{\Phi }_\rho )_{๐ธC}=(\stackrel{~}{X}_\rho ^11_C)\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3,`$
and one can easily check that indeed $`๐ธC`$ becomes an $`H`$-bicomodule algebra. Also, it is easy to see that the elements $`\mathrm{\Omega }`$ and $`\omega `$ for $`๐ธC`$ are given by
$`\mathrm{\Omega }_{๐ธC}=\mathrm{\Omega }^1\mathrm{\Omega }^2(\mathrm{\Omega }^31_C)\mathrm{\Omega }^4\mathrm{\Omega }^5,`$
$`\omega _{๐ธC}=\omega ^1\omega ^2(\omega ^31_C)\omega ^4\omega ^5,`$
where $`\mathrm{\Omega }=\mathrm{\Omega }^1\mathrm{}\mathrm{\Omega }^5`$ and $`\omega =\omega ^1\mathrm{}\omega ^5`$ are the ones for $`๐ธ`$. Using this one obtains that the multiplications in $`๐(๐ธC)`$ and respectively $`๐\text{ }\text{ }(๐ธC)`$ coincide with those in $`(๐๐ธ)C`$ respectively $`(๐\text{ }\text{ }๐ธ)C`$ via the trivial identifications. โ
## 4. Generalized diagonal crossed products as generalized smash products
Let $`H`$ be a quasi-Hopf algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. We define two left $`HH^{op}`$-coactions on $`๐ธ`$, as follows:
$`\lambda _1,\lambda _2:๐ธ(HH^{op})๐ธ,`$
$`\lambda _1(u)=(u_{0_{[1]}}S^1(u_1))u_{0_{[0]}}:=u_{(1)}u_{(0)},`$
$`\lambda _2(u)=(u_{[1]}S^1(u_{[0]_1}))u_{[0]_0}:=u^{(1)}u^{(0)},`$
for all $`u๐ธ`$ (of course, in the Hopf case these two coactions coincide).
If we look at the element $`\mathrm{\Omega }H^2๐ธH^2`$ given by (3.15) and consider the element $`(\mathrm{\Omega }^1\mathrm{\Omega }^5)(\mathrm{\Omega }^2\mathrm{\Omega }^4)\mathrm{\Omega }^3`$, then one can check that this element is invertible in $`(HH^{op})(HH^{op})๐ธ`$, its inverse being given by
$$(\mathrm{\Theta }^1\stackrel{~}{X}_\lambda ^1(\stackrel{~}{x}_\rho ^1)_{[1]_1}S^1(\stackrel{~}{x}_\rho ^3g^2))(\mathrm{\Theta }_{[1]}^2\stackrel{~}{X}_\lambda ^2(\stackrel{~}{x}_\rho ^1)_{[1]_2}S^1(\mathrm{\Theta }^3\stackrel{~}{x}_\rho ^2g^1))\mathrm{\Theta }_{[0]}^2\stackrel{~}{X}_\lambda ^3(\stackrel{~}{x}_\rho ^1)_{[0]},$$
where $`f^1=g^1g^2`$ is the element given by (2.14). We will denote by $`\mathrm{\Phi }_{\lambda _1}(HH^{op})(HH^{op})๐ธ`$ this inverse.
Similarly, if we look at the element $`\omega `$ given by (3.16) and consider the element $`(\omega ^1\omega ^5)(\omega ^2\omega ^4)\omega ^3`$, then one can check that this element is invertible in $`(HH^{op})(HH^{op})๐ธ`$, with inverse defined by
$$(\stackrel{~}{Y}_\lambda ^1S^1(\theta ^3\stackrel{~}{y}_\rho ^3(\stackrel{~}{Y}_\lambda ^3)_{1_2}g^2))(\theta ^1\stackrel{~}{Y}_\lambda ^2S^1(\theta _1^2\stackrel{~}{y}_\rho ^2(\stackrel{~}{Y}_\lambda ^3)_{1_1}g^1))\theta _0^2\stackrel{~}{y}_\rho ^1(\stackrel{~}{Y}_\lambda ^3)_0.$$
We will denote by $`\mathrm{\Phi }_{\lambda _2}(HH^{op})(HH^{op})๐ธ`$ this inverse.
The next proposition generalizes the corresponding result obtained for Hopf algebras in .
###### Proposition 4.1.
With notation as above, $`(๐ธ,\lambda _1,\mathrm{\Phi }_{\lambda _1})`$ and respectively $`(๐ธ,\lambda _2,\mathrm{\Phi }_{\lambda _2})`$ are left $`HH^{op}`$-comodule algebras, denoted by $`๐ธ_1`$ respectively $`๐ธ_2`$.
###### Proof.
It is easy to see that $`\lambda _1`$ and $`\lambda _2`$ are algebra maps, and also that the conditions (2.32) and (2.33) in the definition of a left comodule algebra are satisfied. Then the conditions (2.30) and (2.31) for $`(๐ธ,\lambda _1,\mathrm{\Phi }_{\lambda _1})`$ (respectively for $`(๐ธ,\lambda _2,\mathrm{\Phi }_{\lambda _2})`$) to be a left $`HH^{op}`$-comodule algebra are equivalent to the relations (3.17) and (3.18) fulfilled by $`\mathrm{\Omega }`$ (respectively to the relations (3.19) and (3.20) fulfilled by $`\omega `$). โ
We are now able to express the (generalized) diagonal crossed products over $`H`$ as some generalized smash products over $`HH^{op}`$.
###### Proposition 4.2.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. View $`๐`$ as a left $`HH^{op}`$-module algebra with action $`(hh^{})\phi =h\phi h^{}`$ for all $`h,h^{}H`$ and $`\phi ๐`$, and consider the two left $`HH^{op}`$-comodule algebras $`๐ธ_1`$ and $`๐ธ_2`$ obtained from $`๐ธ`$ as above. Then we have algebra isomorphisms
$`๐๐ธ๐<๐ธ_1,๐\text{ }\text{ }๐ธ๐<๐ธ_2,`$
defined by the trivial identifications.
###### Proof.
We only prove the first isomorphism, the second being similar. The multiplication in $`๐<๐ธ_1`$ looks as follows (for all $`\phi ,\phi ^{}๐`$ and $`u,u^{}๐ธ`$):
$`(\phi <u)(\phi ^{}<u^{})`$
$`=`$ $`((\stackrel{~}{x}_\lambda ^1)_{๐ธ_1}\phi )((\stackrel{~}{x}_\lambda ^2)_{๐ธ_1}u_{(1)}\phi ^{})<(\stackrel{~}{x}_\lambda ^3)_{๐ธ_1}u_{(0)}u^{}`$
$`=`$ $`((\mathrm{\Omega }^1\mathrm{\Omega }^5)\phi )((\mathrm{\Omega }^2\mathrm{\Omega }^4)(u_{0_{[1]}}S^1(u_1))\phi ^{})<\mathrm{\Omega }^3u_{0_{[0]}}u^{}`$
$`=`$ $`(\mathrm{\Omega }^1\phi \mathrm{\Omega }^5)(\mathrm{\Omega }^2u_{0_{[1]}}\phi ^{}S^1(u_1)\mathrm{\Omega }^4)<\mathrm{\Omega }^3u_{0_{[0]}}u^{},`$
and via the trivial identification this is exactly the multiplication of $`๐๐ธ`$. โ
Recall that, for a finite dimensional quasi-Hopf algebra $`H`$, the quantum double $`D(H)`$ was first introduced by Majid in by an implicit Tannaka-Krein reconstruction procedure, and more explicit descriptions were obtained afterwards by Hausser and Nill in , . Actually, Hausser and Nill provided four explicit realizations of $`D(H)`$, two built on $`H^{}H`$ and two on $`HH^{}`$; all are, as algebras, diagonal crossed products, namely the two realizations built on $`H^{}H`$ coincide with $`H^{}H`$ and $`H^{}\text{ }\text{ }H`$ and the two built on $`HH^{}`$ coincide with $`HH^{}`$ and $`H\text{ }\text{ }H^{}`$.
On the other hand, it was proved in that the Drinfeld double of a finite dimensional Hopf algebra may be written as a generalized smash product. As a corollary to the previous proposition, we obtain a generalization of this result for quasi-Hopf algebras.
###### Corollary 4.3.
If $`H`$ is a finite dimensional quasi-Hopf algebra, then the quantum double $`D(H)`$ may be written as a generalized smash product.
###### Proof.
In the previous proposition take $`๐=H^{}`$, $`๐ธ=H`$ and use the fact that $`H^{}H`$ and $`H^{}\text{ }\text{ }H`$ are realizations for $`D(H)`$. โ
Let us also record the fact that the two left $`HH^{op}`$-comodule algebra structures on $`H`$ are defined as follows:
$`\lambda _1,\lambda _2:H(HH^{op})H,`$
$`\lambda _1(h)=(h_{(1,1)}S^1(h_2))h_{(1,2)},`$
$`\lambda _2(h)=(h_1S^1(h_{(2,2)}))h_{(2,1)},`$
for all $`hH`$, and
$`\mathrm{\Phi }_{\lambda _1},\mathrm{\Phi }_{\lambda _2}(HH^{op})(HH^{op})H,`$
$`\mathrm{\Phi }_{\lambda _1}=(Y^1X^1x_{(1,1)}^1S^1(x^3g^2))(Y_1^2X^2x_{(1,2)}^1S^1(Y^3x^2g^1))Y_2^2X^3x_2^1,`$
$`\mathrm{\Phi }_{\lambda _2}=(Y^1S^1(x^3y^3Y_{(2,2)}^3g^2))(x^1Y^2S^1(x_2^2y^2Y_{(2,1)}^3g^1))x_1^2y^1Y_1^3,`$
where $`f^1=g^1g^2`$ is the element given by (2.14).
Let again $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. We intend to prove that the two generalized left diagonal crossed products $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ are isomorphic as algebras, using their description as generalized smash products. First we need a result on generalized smash products. Namely, let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra, $`๐
`$ a left $`H`$-comodule algebra and $`UH๐
`$ an invertible element such that $`(\epsilon id_๐
)(U)=1_๐
`$. If we define a map
$`\lambda ^{}:๐
H๐
,\lambda ^{}(๐)=U\lambda (๐)U^1,`$
then, by , this is a new left $`H`$-comodule algebra structure on $`๐
`$, with
$`\mathrm{\Phi }_\lambda ^{}=(1_HU)(id_H\lambda )(U)\mathrm{\Phi }_\lambda (\mathrm{\Delta }id_๐
)(U^1),`$
which will be denoted by $`๐
^{}`$ (and we will say that $`๐
`$ and $`๐
^{}`$ are โtwist equivalentโ). We then may consider the generalized smash products $`A<๐
`$ and $`A<๐
^{}`$.
###### Proposition 4.4.
The map
$`f:A<๐
A<๐
^{},`$
$`f(a<๐)=U(a<๐)=U^1a<U^2๐`$
is an algebra isomorphism, and moreover $`f(1_A<๐)=1_A<๐`$, for all $`๐๐
`$ (that is, $`A<๐
`$ and $`A<๐
^{}`$ are equivalent extensions of $`๐
`$).
###### Proof.
Follows by a direct computation. โ
In view of this proposition, it will be sufficient to prove that if $`๐ธ`$ is an $`H`$-bicomodule algebra, then the two left $`HH^{op}`$-comodule algebras $`๐ธ_1`$ and $`๐ธ_2`$ constructed before are twist equivalent. To prove this, we need first a technical lemma (a part of it will be used also in a subsequent section).
###### Lemma 4.5.
Let $`H`$ be a quasi-Hopf algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Consider the elements $`\mathrm{\Omega }`$ and $`\omega `$ given by (3.15) and (3.16). Then the following hold:
$`\mathrm{\Theta }_1^1\mathrm{\Omega }^1\mathrm{\Theta }_2^1\mathrm{\Omega }^2\mathrm{\Theta }^2\mathrm{\Omega }^3\mathrm{\Omega }^5S^1(\mathrm{\Theta }^3)_1\mathrm{\Omega }^4S^1(\mathrm{\Theta }^3)_2`$
$`=\mathrm{\Theta }_1^1\stackrel{~}{x}_\lambda ^1\mathrm{\Theta }_2^1\stackrel{~}{x}_\lambda ^2\overline{\mathrm{\Theta }}^1\stackrel{~}{X}_\rho ^1\mathrm{\Theta }_0^2(\stackrel{~}{x}_\lambda ^3)_0\overline{\mathrm{\Theta }}^2S^1(f^2\stackrel{~}{X}_\rho ^3\mathrm{\Theta }^3)`$
(4.1) $`S^1(f^1\stackrel{~}{X}_\rho ^2\mathrm{\Theta }_1^2(\stackrel{~}{x}_\lambda ^3)_1\overline{\mathrm{\Theta }}^3),`$
$`\mathrm{\Theta }_1^1\mathrm{\Omega }^1\theta ^1S^1(\theta ^3)\mathrm{\Omega }^5S^1(\mathrm{\Theta }^3)_1\mathrm{\Theta }_2^1\mathrm{\Omega }^2\theta _{0_{[1]}}^2S^1(\theta _1^2)\mathrm{\Omega }^4S^1(\mathrm{\Theta }^3)_2`$
(4.2) $`\mathrm{\Theta }^2\mathrm{\Omega }^3\theta _{0_{[0]}}^2=\omega ^1\omega ^5\omega ^2\mathrm{\Theta }^1S^1(\mathrm{\Theta }^3)\omega ^4\omega ^3\mathrm{\Theta }^2.`$
###### Proof.
The relation (4.1) follows by applying (2.11), (2.45) and (2.44), we leave the details to the reader. We prove now (4.2). We compute:
$`\mathrm{\Theta }_1^1\mathrm{\Omega }^1\theta ^1S^1(\theta ^3)\mathrm{\Omega }^5S^1(\mathrm{\Theta }^3)_1`$
$`\mathrm{\Theta }_2^1\mathrm{\Omega }^2\theta _{0_{[1]}}^2S^1(\theta _1^2)\mathrm{\Omega }^4S^1(\mathrm{\Theta }^3)_2\mathrm{\Theta }^2\mathrm{\Omega }^3\theta _{0_{[0]}}^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{4.1})}{=}}`$ $`\mathrm{\Theta }_1^1\stackrel{~}{x}_\lambda ^1\theta ^1S^1(f^2\stackrel{~}{X}_\rho ^3\mathrm{\Theta }^3\theta ^3)\mathrm{\Theta }_2^1\stackrel{~}{x}_\lambda ^2\overline{\mathrm{\Theta }}^1\theta _{0_{[1]}}^2`$
$`S^1(f^1\stackrel{~}{X}_\rho ^2\mathrm{\Theta }_1^2(\stackrel{~}{x}_\lambda ^3)_1\overline{\mathrm{\Theta }}^3\theta _1^2)\stackrel{~}{X}_\rho ^1\mathrm{\Theta }_0^2(\stackrel{~}{x}_\lambda ^3)_0\overline{\mathrm{\Theta }}^2\theta _{0_{[0]}}^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.44})}{=}}`$ $`\stackrel{~}{x}_\lambda ^1\stackrel{~}{\mathrm{\Theta }}^1\theta ^1S^1(f^2\stackrel{~}{X}_\rho ^3(\stackrel{~}{x}_\lambda ^3)_1\mathrm{\Theta }^3\stackrel{~}{\mathrm{\Theta }}^3\theta ^3)\stackrel{~}{x}_\lambda ^2\mathrm{\Theta }^1\stackrel{~}{\mathrm{\Theta }}_{[1]}^2\overline{\mathrm{\Theta }}^1\theta _{0_{[1]}}^2`$
$`S^1(f^1\stackrel{~}{X}_\rho ^2(\stackrel{~}{x}_\lambda ^3)_{0_1}\mathrm{\Theta }_1^2\stackrel{~}{\mathrm{\Theta }}_{[0]_1}^2\overline{\mathrm{\Theta }}^3\theta _1^2)\stackrel{~}{X}_\rho ^1(\stackrel{~}{x}_\lambda ^3)_{0_0}\mathrm{\Theta }_0^2\stackrel{~}{\mathrm{\Theta }}_{[0]_0}^2\overline{\mathrm{\Theta }}^2\theta _{0_{[0]}}^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.43})}{=}}`$ $`\stackrel{~}{x}_\lambda ^1S^1(f^2\stackrel{~}{X}_\rho ^3(\stackrel{~}{x}_\lambda ^3)_1\mathrm{\Theta }^3)\stackrel{~}{x}_\lambda ^2\mathrm{\Theta }^1\overline{\mathrm{\Theta }}^1`$
$`S^1(f^1\stackrel{~}{X}_\rho ^2(\stackrel{~}{x}_\lambda ^3)_{0_1}\mathrm{\Theta }_1^2\overline{\mathrm{\Theta }}^3)\stackrel{~}{X}_\rho ^1(\stackrel{~}{x}_\lambda ^3)_{0_0}\mathrm{\Theta }_0^2\overline{\mathrm{\Theta }}^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.26})}{=}}`$ $`\stackrel{~}{x}_\lambda ^1S^1(f^2(\stackrel{~}{x}_\lambda ^3)_{1_2}\stackrel{~}{X}_\rho ^3\mathrm{\Theta }^3)\stackrel{~}{x}_\lambda ^2\mathrm{\Theta }^1\overline{\mathrm{\Theta }}^1`$
$`S^1(f^1(\stackrel{~}{x}_\lambda ^3)_{1_1}\stackrel{~}{X}_\rho ^2\mathrm{\Theta }_1^2\overline{\mathrm{\Theta }}^3)(\stackrel{~}{x}_\lambda ^3)_0\stackrel{~}{X}_\rho ^1\mathrm{\Theta }_0^2\overline{\mathrm{\Theta }}^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.16})}{=}}`$ $`\omega ^1\omega ^5\omega ^2\overline{\mathrm{\Theta }}^1S^1(\overline{\mathrm{\Theta }}^3)\omega ^4\omega ^3\overline{\mathrm{\Theta }}^2,`$
as required. โ
###### Proposition 4.6.
Let $`H`$ be a quasi-Hopf algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then the left $`HH^{op}`$-comodule algebras $`๐ธ_1`$ and $`๐ธ_2`$ are twist equivalent. More exactly, for the element $`U(HH^{op})๐ธ`$ given by
$$U=(\mathrm{\Theta }^1S^1(\mathrm{\Theta }^3))\mathrm{\Theta }^2,$$
we have that
$`\lambda _2(u)=U\lambda _1(u)U^1,u๐ธ,`$
$`\mathrm{\Phi }_{\lambda _2}=(1U)(id\lambda _1)(U)\mathrm{\Phi }_{\lambda _1}(\mathrm{\Delta }id)(U^1).`$
###### Proof.
The first relation follows immediately from (2.43), and the second is equivalent to the relation (4.2) proved in the previous lemma. โ
As a consequence of these results and (3.30), we obtain:
###### Corollary 4.7.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then the two generalized left (right) diagonal crossed products $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ ($`๐ธ๐`$ and $`๐ธ\text{ }\text{ }๐`$, respectively) are isomorphic as algebras, and moreover they are equivalent extensions of $`๐ธ`$.
###### Remark 4.8.
Let $`H`$ be a quasi-Hopf algebra, $`๐`$ an $`H`$-bimodule algebra and $`๐ธ`$ an $`H`$-bicomodule algebra with $`\mathrm{\Phi }_{\lambda ,\rho }=1_H1_๐ธ1_H`$. Then, by (2.43) it follows that
$$(\lambda id)\rho =(id\rho )\lambda ,$$
and by (4.2) it follows that $`\mathrm{\Omega }=\omega `$. So, in this case we have that $`๐๐ธ`$ and $`๐\text{ }\text{ }๐ธ`$ are not only isomorphic, but they actually coincide, and that $`๐ธ_1`$ and $`๐ธ_2`$ also coincide. An example of such an $`๐ธ`$ is the tensor product $`๐๐
`$, where $`๐`$ is a right comodule algebra and $`๐
`$ is a left comodule algebra, see . We will encounter another example in a subsequent section.
We end this section by showing that the left generalized diagonal crossed products are isomorphic, as algebras, to the right generalized diagonal crossed products.
Let $`H`$ be a quasi-Hopf algebra, $`๐ธ`$ a unital associative algebra and $`(\delta ,\mathrm{\Psi })`$ a two-sided coaction of $`H`$ on $`๐ธ`$. We associate to $`(\delta ,\mathrm{\Psi })`$ the elements $`p_\delta ,q_\delta H๐ธH`$ as follows:
(4.3) $`p_\delta =p_\delta ^1p_\delta ^2p_\delta ^3=\mathrm{\Psi }^2S^1(\mathrm{\Psi }^1\beta )\mathrm{\Psi }^3\mathrm{\Psi }^4\beta S(\mathrm{\Psi }^5),`$
(4.4) $`q_\delta =q_\delta ^1q_\delta ^2q_\delta ^3=S(\overline{\mathrm{\Psi }}^1)\alpha \overline{\mathrm{\Psi }}^2\overline{\mathrm{\Psi }}^3S^1(\alpha \overline{\mathrm{\Psi }}^5)\overline{\mathrm{\Psi }}^4.`$
By we have the following relations, for all $`u๐ธ`$:
(4.5) $`p_\delta (1_Hu1_H)=\delta (u_{(0)})p_\delta [S^1(u_{(1)})1_๐ธS(u_{(1)})],`$
(4.6) $`(1_Hu1_H)q_\delta =[S(u_{(1)})1_๐ธS^1(u_{(1)})]q_\delta \delta (u_{(0)}),`$
$`[S(\overline{\mathrm{\Psi }}^2)f^1S(\overline{\mathrm{\Psi }}^1)f^21_๐ธS^1(F^2\overline{\mathrm{\Psi }}^5)S^1(F^1\overline{\mathrm{\Psi }}^4)]`$
(4.7) $`\times (\mathrm{\Delta }id_๐ธ\mathrm{\Delta })(q_\delta \delta (\overline{\mathrm{\Psi }}^3))=[1_Hq_\delta 1_H](id_H\delta id_H)(q_\delta )\mathrm{\Psi },`$
(4.8) $`\delta (q_\delta ^2)p_\delta [S^1(q_\delta ^1)1_๐ธS(q_\delta ^3)]=1_H1_๐ธ1_H,`$
(4.9) $`[S(p_\delta ^1)1_๐ธS^1(p_\delta ^3)]q_\delta \delta (p_\delta ^2)=1_H1_๐ธ1_H,`$
where $`f=f^1f^2=F^1F^2`$ is the Drinfeld twist defined in (2.13). Moreover, the definitions of $`q_\delta `$ and of a two-sided coaction imply
$`q_\delta ^1\mathrm{\Psi }^1(q_\delta ^2)_{(1)}\mathrm{\Psi }^2(q_\delta ^2)_{(0)}\mathrm{\Psi }^3(q_\delta ^2)_{(1)}\mathrm{\Psi }^4q_\delta ^3\mathrm{\Psi }^5`$
(4.10) $`=S(\overline{\mathrm{\Psi }}^1)q_L^1\overline{\mathrm{\Psi }}_1^2q_L^2\overline{\mathrm{\Psi }}_2^2\overline{\mathrm{\Psi }}^3q_R^1\overline{\mathrm{\Psi }}_1^4S^1(\overline{\mathrm{\Psi }}^5)q_R^2\overline{\mathrm{\Psi }}_2^4,`$
where $`q_L=q_L^1q_L^2:=S(x^1)\alpha x^2x^3`$ and $`q_R=q_R^1q_R^2:=X^1S^1(\alpha X^3)X^2`$. Finally, we need the formulae
(4.11) $`(S(h_1)1_H)q_L\mathrm{\Delta }(h_2)`$ $`=`$ $`(1h)q_L,`$
(4.12) $`(1_HS^1(h_2))q_R\mathrm{\Delta }(h_1)`$ $`=`$ $`(h1_H)q_R,`$
for all $`hH`$, which have been established in .
###### Proposition 4.9.
Let $`H`$ be a quasi-Hopf algebra, $`(\delta ,\mathrm{\Psi })`$ a two-sided coaction of $`H`$ on an associative unital algebra $`๐ธ`$, and $`๐`$ an $`H`$-bimodule algebra. Then the map $`\vartheta :๐_\delta ๐ธ๐ธ_\delta ๐`$ defined for all $`\phi ๐`$ and $`u๐ธ`$ by
$$\vartheta (\phi _\delta u)=q_\delta ^2u_{(0)}S^1(q_\delta ^1u_{(1)})\phi q_\delta ^3u_{(1)}$$
is an algebra isomorphism. In particular, if $`๐ธ`$ is an $`H`$-bicomodule algebra then we get that all four generalized diagonal crossed products $`๐๐ธ`$, $`๐ธ๐`$, $`๐\text{ }\text{ }๐ธ`$ and $`๐ธ\text{ }\text{ }๐`$ are isomorphic as unital algebras.
###### Proof.
We show that $`\vartheta `$ is multiplicative. For any $`\phi ,\phi ^{}๐`$ and $`u,u^{}๐ธ`$ we have:
$`\vartheta ((\phi _\delta u)(\phi ^{}_\delta u^{}))`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.9},\text{3.7})}{=}}`$ $`\vartheta ((\overline{\mathrm{\Psi }}^1\phi S^1(f^2\overline{\mathrm{\Psi }}^5))(\overline{\mathrm{\Psi }}^2u_{(1)}\phi ^{}S^1(f^1\overline{\mathrm{\Psi }}^4u_{(1)}))_\delta \overline{\mathrm{\Psi }}^3u_{(0)}u^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.50},\text{2.11})}{=}}`$ $`q_\delta ^2\overline{\mathrm{\Psi }}_{(0)}^3u_{(0,0)}u_{(0)}^{}_\delta (S^1(F^2(q_\delta ^1)_2\overline{\mathrm{\Psi }}_{(1)_2}^3u_{(0,1)_2}u_{(1)_2}^{}g^2)\overline{\mathrm{\Psi }}^1`$
$`\phi S^1(f^2\overline{\mathrm{\Psi }}^5)(q_\delta ^3)_1\overline{\mathrm{\Psi }}_{(1)_1}^3u_{(0,1)_1}u_{(1)_1}^{})(S^1(F^1(q_\delta ^1)_1\overline{\mathrm{\Psi }}_{(1)_1}^3u_{(0,1)_1}u_{(1)_1}^{}g^1)`$
$`\times \overline{\mathrm{\Psi }}^2u_{(1)}\phi ^{}S^1(f^1\overline{\mathrm{\Psi }}^4u_{(1)})(q_\delta ^3)_2\overline{\mathrm{\Psi }}_{(1)_2}^3u_{(0,1)_2}u_{(1)_2}^{}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{4.7})}{=}}`$ $`q_\delta ^2(Q_\delta ^2)_{(0)}\mathrm{\Psi }^3u_{(0,0)}u_{(0)}^{}_\delta (S^1(q_\delta ^1(Q_\delta ^2)_{(1)}\mathrm{\Psi }^2u_{(0,1)_2}u_{(1)_2}^{}g^2)\phi `$
$`q_\delta ^3(Q_\delta ^2)_{(1)}\mathrm{\Psi }^4u_{(0,1)_1}u_{(1)_1}^{})(S^1(Q_\delta ^1\mathrm{\Psi }^1u_{(0,1)_1}u_{(1)_1}^{}g^1)u_{(1)}\phi ^{}`$
$`S^1(u_{(1)})Q_\delta ^3\mathrm{\Psi }^5u_{(0,1)_2}u_{(1)_2}^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{4.10},\text{3.2})}{=}}`$ $`q_\delta ^2u_{(0)}\overline{\mathrm{\Psi }}^3u_{(0)}^{}_\delta (S^1(q_\delta ^1q_L^2u_{(1)_{(2,2)}}\overline{\mathrm{\Psi }}_2^2u_{(1)_2}^{}g^2)`$
$`\phi q_\delta ^3q_R^1u_{(1)_{(1,1)}}\overline{\mathrm{\Psi }}_1^4u_{(1)_1}^{})(S^1(q_L^1u_{(1)_{(2,1)}}\overline{\mathrm{\Psi }}_1^2u_{(1)_1}^{}g^1)u_{(1)_1}\overline{\mathrm{\Psi }}^1\phi ^{}`$
$`S^1(u_{(1)_2}\overline{\mathrm{\Psi }}^5)q_R^2u_{(1)_{(1,2)}}\overline{\mathrm{\Psi }}_2^4u_{(1)_2}^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{4.11},\text{4.12},\text{4.10})}{=}}`$ $`q_\delta ^2u_{(0)}(Q_\delta ^2)_{(0)}\mathrm{\Psi }^3u_{(0)}^{}_\delta (S^1(q_\delta ^1u_{(1)}(Q_\delta ^2)_{(1)}\mathrm{\Psi }^2u_{(1)_2}^{}g^2)\phi `$
$`q_\delta ^3u_{(1)}(Q_\delta ^2)_{(1)}\mathrm{\Psi }^4u_{(1)_1}^{})(S^1(Q_\delta ^1\mathrm{\Psi }^1u_{(1)_1}^{}g^1)\phi ^{}Q_\delta ^3\mathrm{\Psi }^5u_{(1)_2}^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.1},\text{3.8})}{=}}`$ $`q_\delta ^2u_{(0)}(Q_\delta ^2)_{(0)}u_{(0,0)}^{}\mathrm{\Omega }^3_\delta (\mathrm{\Omega }^2S^1(q_\delta ^1u_{(1)}(Q_\delta ^2)_{(1)}u_{(0,1)}^{})\phi `$
$`q_\delta ^3u_{(1)}(Q_\delta ^2)_{(1)}u_{(0,1)}^{}\mathrm{\Omega }^4)(\mathrm{\Omega }^1S^1(Q_\delta ^1u_{(1)}^{})\phi ^{}Q_\delta ^3u_{(1)}^{}\mathrm{\Omega }^5)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.10})}{=}}`$ $`(q_\delta ^2u_{(0)}_\delta S^1(q_\delta ^1u_{(1)})\phi q_\delta ^3u_{(1)})`$
$`\times (Q_\delta ^2u_{(0)}^{}_\delta S^1(Q_\delta ^1u_{(1)}^{})\phi ^{}Q_\delta ^3u_{(1)}^{})=\vartheta (\phi _\delta u)\vartheta (\phi ^{}_\delta u^{}),`$
as needed. (We denoted by $`Q_\delta ^1Q_\delta ^2Q_\delta ^3`$ another copy of $`q_\delta `$ and by $`F^1F^2`$ another copy of $`f`$).
It is easy to see that the unit and counit properties imply $`\vartheta (1_๐_\delta 1_๐ธ)=1_๐ธ_\delta 1_๐`$, so it remains to show that $`\vartheta `$ is bijective. To this end, define $`\vartheta ^1:๐ธ_\delta ๐๐_\delta ๐ธ`$ given for all $`u๐ธ`$ and $`\phi ๐`$ by
$$\vartheta ^1(u_\delta \phi )=u_{(1)}p_\delta ^1\phi S^1(u_{(1)}p_\delta ^3)_\delta u_{(0)}p_\delta ^2,$$
where $`p_\delta =p_\delta ^1p_\delta ^2p_\delta ^3`$ is the element defined in (4.3).
We claim that $`\vartheta `$ and $`\vartheta ^1`$ are inverses. Indeed, $`\vartheta \vartheta ^1=id_{๐ธ_\delta ๐}`$ because of (4.6) and (4.9), and $`\vartheta \vartheta ^1=id_{๐_\delta ๐ธ}`$ because of (4.5) and (4.8) (we leave the verification of the details to the reader). โ
## 5. Generalized two-sided crossed product and two-sided generalized smash product
Let $`H`$ be a finite dimensional quasi-bialgebra and $`(๐,\rho ,\mathrm{\Phi }_\rho )`$,$`(๐
,\lambda ,\mathrm{\Phi }_\lambda )`$ a right and a left $`H`$-comodule algebra, respectively. As in the case of a bialgebra, the right $`H`$-coaction $`(\rho ,\mathrm{\Phi }_\rho )`$ on $`๐`$ induces a left $`H^{}`$-action $`:H^{}๐๐`$ defined by
(5.1)
$$\phi ๐=\phi (๐_1)๐_0,$$
for all $`\phi H^{}`$ and $`๐๐`$, where $`\rho (๐)=๐_0๐_1`$ for all $`๐๐`$. Similarly, the left $`H`$-coaction $`(\lambda ,\mathrm{\Phi }_\lambda )`$ on $`๐
`$ provides a right $`H^{}`$-action $`:๐
H^{}๐
`$ defined by
(5.2)
$$๐\phi =\phi (๐_{[1]})๐_{[0]},$$
for all $`\phi H^{}`$ and $`๐๐
`$, where we now denote $`\lambda (๐)=๐_{[1]}๐_{[0]}`$ for all $`๐๐
`$. Following \[7, Proposition 11.4 (ii)\] we define an algebra structure on the $`k`$-vector space $`๐H^{}๐
`$. This algebra is denoted by $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$ and its multiplication is defined by
(5.3) $`(๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)(๐^{}\text{ }>\text{ }\phi ^{}\text{ }<\text{ }๐^{})`$
$`=`$ $`๐(\phi _1๐^{})\stackrel{~}{x}_\rho ^1\text{ }>\text{ }(\stackrel{~}{x}_\lambda ^1\phi _2\stackrel{~}{x}_\rho ^2)(\stackrel{~}{x}_\lambda ^2\phi _1^{}\stackrel{~}{x}_\rho ^3)\text{ }<\text{ }\stackrel{~}{x}_\lambda ^3(๐\phi _2^{})๐^{},`$
for all $`๐,๐^{}๐`$, $`๐,๐^{}๐
`$ and $`\phi ,\phi ^{}H^{}`$, where we write $`๐\text{ }>\text{ }\phi \text{ }<\text{ }๐`$ for $`๐\phi ๐`$ when viewed as an element of $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$. The unit of the algebra $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$ is $`1_๐\text{ }>\text{ }\epsilon \text{ }<\text{ }1_๐
`$. Hausser and Nill called this algebra the two-sided crossed product. They proved that $`๐๐
`$ is an $`H`$-bicomodule algebra (here $`\mathrm{\Phi }_{\lambda ,\rho }`$ is trivial) and the diagonal crossed product $`(๐๐
)H^{}`$ is isomorphic, as an algebra, to the two-sided crossed product $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$.
This construction admits a slight generalization, as follows. Let $`H`$ be a quasi-bialgebra, $`๐`$ a right $`H`$-comodule algebra, $`๐
`$ a left $`H`$-comodule algebra and $`๐`$ an $`H`$-bimodule algebra. On $`๐๐๐
`$ define a multiplication by
(5.4) $`(๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)(๐^{}\text{ }>\text{ }\phi ^{}\text{ }<\text{ }๐^{})`$
$`=`$ $`๐๐_0^{}\stackrel{~}{x}_\rho ^1\text{ }>\text{ }(\stackrel{~}{x}_\lambda ^1\phi ๐_1^{}\stackrel{~}{x}_\rho ^2)(\stackrel{~}{x}_\lambda ^2๐_{[1]}\phi ^{}\stackrel{~}{x}_\rho ^3)\text{ }<\text{ }\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{},`$
for all $`๐,๐^{}๐`$, $`๐,๐^{}๐
`$ and $`\phi ,\phi ^{}๐`$, where we write $`๐\text{ }>\text{ }\phi \text{ }<\text{ }๐`$ for $`๐\phi ๐`$. Then one can prove by a direct computation that this multiplication yields an associative algebra with unit $`1_๐\text{ }>\text{ }1_๐\text{ }<\text{ }1_๐
`$, denoted by $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐
`$ and called the generalized two-sided crossed product. It is obvious that for $`H`$ finite dimensional and $`๐=H^{}`$ we recover the two-sided crossed product $`๐\text{ }>\text{ }H^{}\text{ }<\text{ }๐
`$ of Hausser and Nill.
We construct now a different kind of two-sided product, using โdualโ objects, that is by replacing comodule algebras by module algebras and the bimodule algebra by a bicomodule algebra.
###### Proposition 5.1.
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra, $`B`$ a right $`H`$-module algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. If we define on $`A๐ธB`$ a multiplication, by
$`(a<u\text{ }>\text{ }b)(a{}_{}{}^{}<u{}_{}{}^{}\text{ }>\text{ }b{}_{}{}^{})`$
(5.5) $`=(\stackrel{~}{x}_\lambda ^1a)(\stackrel{~}{x}_\lambda ^2u_{[1]}\theta ^1a{}_{}{}^{})<\stackrel{~}{\mathrm{x}}_\lambda ^3u_{[0]}\theta ^2u{}_{}{}^{}{}_{0}{}^{}\stackrel{~}{x}_{\rho }^{1}\text{ }>\text{ }(b\theta ^3u{}_{}{}^{}{}_{1}{}^{}\stackrel{~}{x}_{\rho }^{2})(b{}_{}{}^{}\stackrel{~}{x}_\rho ^3),`$
for all $`a,a{}_{}{}^{}A`$, $`u,u{}_{}{}^{}๐ธ`$ and $`b,b{}_{}{}^{}B`$ (where we write $`a<u\text{ }>\text{ }b`$ for $`aub`$), and we denote this structure on $`A๐ธB`$ by $`A<๐ธ\text{ }>\text{ }B`$, then $`A<๐ธ\text{ }>\text{ }B`$ is an associative algebra with unit $`1_A<1_๐ธ\text{ }>\text{ }1_B`$.
###### Proof.
For all $`a,a^{},a^{\prime \prime }A`$, $`u,u^{},u^{\prime \prime }๐ธ`$ and $`b,b^{},b^{\prime \prime }B`$ we compute:
$`[(a<u\text{ }>\text{ }b)(a^{}<u^{}\text{ }>\text{ }b^{})](a^{\prime \prime }<u^{\prime \prime }\text{ }>\text{ }b^{\prime \prime })`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{5.5})}{=}}`$ $`\{\stackrel{~}{y}_\lambda ^1[(\stackrel{~}{x}_\lambda ^1a)(\stackrel{~}{x}_\lambda ^2u_{[1]}\theta ^1a^{})]\}[\stackrel{~}{y}_\lambda ^2(\stackrel{~}{x}_\lambda ^3)_{[1]}u_{[0,1]}\theta _{[1]}^2`$
$`\times u_{0_{[1]}}^{}(\stackrel{~}{x}_\rho ^1)_{[1]}\overline{\theta }^1a^{\prime \prime }]<\stackrel{~}{y}^3_\lambda (\stackrel{~}{x}_\lambda ^3)_{[0]}u_{[0,0]}\theta ^2_{[0]}u^{}_{0_{[0]}}(\stackrel{~}{x}_\rho ^1)_{[0]}\overline{\theta }^2`$
$`\times u_0^{\prime \prime }\stackrel{~}{y}_\rho ^1\text{ }>\text{ }\{[(b\theta ^3u_1^{}\stackrel{~}{x}_\rho ^2)(b^{}\stackrel{~}{x}_\rho ^3)]\overline{\theta }^3u_1^{\prime \prime }\stackrel{~}{y}_\rho ^2\}(b^{\prime \prime }\stackrel{~}{y}_\rho ^3)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.18},\text{2.23},\text{2.17},\text{2.22})}{=}}`$ $`[(X^1(\stackrel{~}{y}_\lambda ^1)_1\stackrel{~}{x}_\lambda ^1a]\{[X^2(\stackrel{~}{y}_\lambda ^1)_2\stackrel{~}{x}_\lambda ^2u_{[1]}\theta ^1a^{}][X^3\stackrel{~}{y}_\lambda ^2(\stackrel{~}{x}_\lambda ^3)_{[1]}u_{[0,1]}`$
$`\times \theta _{[1]}^2u_{0_{[1]}}^{}(\stackrel{~}{x}_\rho ^1)_{[1]}\overline{\theta }^1a^{\prime \prime }]\}<\stackrel{~}{y}^3_\lambda (\stackrel{~}{x}_\lambda ^3)_{[0]}u_{[0,0]}\theta ^2_{[0]}u^{}_{0_{[0]}}(\stackrel{~}{x}_\rho ^1)_{[0]}\overline{\theta }^2u^{\prime \prime }_0\stackrel{~}{y}^1_\rho `$
$`\text{ }>\text{ }[b\theta ^3u_1^{}\stackrel{~}{x}_\rho ^2\overline{\theta }_1^3u_{1_1}^{\prime \prime }(\stackrel{~}{y}_\rho ^2)_1x^1]\{[(b^{}\stackrel{~}{x}_\rho ^3\overline{\theta }_2^3u_{1_2}^{\prime \prime }(\stackrel{~}{y}_\rho ^2)_2x^2](b^{\prime \prime }\stackrel{~}{y}_\rho ^3x^3)\}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.31})}{=}}`$ $`(\stackrel{~}{y}_\lambda ^1a)\{[(\stackrel{~}{y}_\lambda ^2)_1\stackrel{~}{x}_\lambda ^1u_{[1]}\theta ^1a^{}][(\stackrel{~}{y}_\lambda ^2)_2\stackrel{~}{x}_\lambda ^2u_{[0,1]}\theta _{[1]}^2u_{0_{[1]}}^{}`$
$`\times (\stackrel{~}{x}_\rho ^1)_{[1]}\overline{\theta }^1a^{\prime \prime }]\}<\stackrel{~}{y}^3_\lambda \stackrel{~}{x}^3_\lambda u_{[0,0]}\theta ^2_{[0]}u^{}_{0_{[0]}}(\stackrel{~}{x}_\rho ^1)_{[0]}\overline{\theta }^2u^{\prime \prime }_0\stackrel{~}{y}^1_\rho `$
$`\text{ }>\text{ }[b\theta ^3u_1^{}\stackrel{~}{x}_\rho ^2\overline{\theta }_1^3u_{1_1}^{\prime \prime }(\stackrel{~}{y}_\rho ^2)_1x^1]\{[b^{}\stackrel{~}{x}_\rho ^3\overline{\theta }_2^3u_{1_2}^{\prime \prime }(\stackrel{~}{y}_\rho ^2)_2x^2](b^{\prime \prime }\stackrel{~}{y}_\rho ^3x^3)\}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.30},\text{2.44},\text{2.18})}{=}}`$ $`(\stackrel{~}{y}_\lambda ^1a)\{\stackrel{~}{y}_\lambda ^2u_{[1]}\theta ^1[(\stackrel{~}{x}_\lambda ^1a^{})(\stackrel{~}{x}_\lambda ^2\mathrm{\Theta }^1u_{0_{[1]}}^{}(\stackrel{~}{x}_\rho ^1)_{[1]}\overline{\theta }^1a^{\prime \prime })]\}`$
$`<\stackrel{~}{y}_\lambda ^3u_{[0]}\theta ^2(\stackrel{~}{x}_\lambda ^3)_0\mathrm{\Theta }^2u_{0_{[0]}}^{}(\stackrel{~}{x}_\rho ^1)_{[0]}\overline{\theta }^2u_0^{\prime \prime }\stackrel{~}{y}_\rho ^1\text{ }>\text{ }[b\theta ^3(\stackrel{~}{x}_\lambda ^3)_1`$
$`\times \mathrm{\Theta }^3u_1^{}\stackrel{~}{x}_\rho ^2\overline{\theta }_1^3u_{1_1}^{\prime \prime }(\stackrel{~}{y}_\rho ^2))_1x^1]\{[b{}_{}{}^{}\stackrel{~}{x}_\rho ^3\overline{\theta }_2^3u_{1_2}^{\prime \prime }(\stackrel{~}{y}_\rho ^2)_2x^2](b^{\prime \prime }\stackrel{~}{y}_\rho ^3x^3)\}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.43},\text{2.45},\text{2.26})}{=}}`$ $`(\stackrel{~}{y}_\lambda ^1a)\{\stackrel{~}{y}_\lambda ^2u_{[1]}\theta ^1[((\stackrel{~}{x}_\lambda ^1a^{})(\stackrel{~}{x}_\lambda ^2u_{[1]}^{}\overline{\theta }^1a^{\prime \prime })]\}<\stackrel{~}{y}_\lambda ^3u_{[0]}\theta ^2`$
$`\times (\stackrel{~}{x}_\lambda ^3)_0u_{[0]_0}^{}\overline{\theta }_0^2u_{0,0}^{\prime \prime }\stackrel{~}{x}_\rho ^1\stackrel{~}{y}_\rho ^1\text{ }>\text{ }[b\theta ^3(\stackrel{~}{x}_\lambda ^3)_1u_{[0]_1}^{}\overline{\theta }_1^2`$
$`\times u_{0,1}^{\prime \prime }\stackrel{~}{x}_\rho ^2(\stackrel{~}{y}_\rho ^2)_1x^1]\{[b^{}\overline{\theta }^3u_1^{\prime \prime }\stackrel{~}{x}_\rho ^3(\stackrel{~}{y}_\rho ^2)_2x^2](b^{\prime \prime }\stackrel{~}{y}_\rho ^3x^3)\}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.27},\text{2.23})}{=}}`$ $`(\stackrel{~}{y}_\lambda ^1a)\{\stackrel{~}{y}_\lambda ^2u_{[1]}\theta ^1[(\stackrel{~}{x}_\lambda ^1a^{})(\stackrel{~}{x}_\lambda ^2u_{[1]}^{}\overline{\theta }^1a^{\prime \prime })]\}<\stackrel{~}{y}_\lambda ^3u_{[0]}\theta ^2`$
$`\times (\stackrel{~}{x}_\lambda ^3u_{[0]}^{}\overline{\theta }^2u_0^{\prime \prime }\stackrel{~}{y}_\rho ^1)_0\stackrel{~}{x}_\rho ^1\text{ }>\text{ }[b\theta ^3(\stackrel{~}{x}_\lambda ^3u_{[0]}^{}\overline{\theta }^2u_0^{\prime \prime }\stackrel{~}{y}_\rho ^1)_1\stackrel{~}{x}_\rho ^2]`$
$`\times \{[(b^{}\overline{\theta }^3u_1^{\prime \prime }\stackrel{~}{y}_\rho ^2)(b^{\prime \prime }\stackrel{~}{y}_\rho ^3)]\stackrel{~}{x}_\rho ^3\}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{5.5})}{=}}`$ $`(a<u\text{ }>\text{ }b)[(\stackrel{~}{x}_\lambda ^1a^{})(\stackrel{~}{x}_\lambda ^2u_{[1]}\overline{\theta }^1a^{\prime \prime })<\stackrel{~}{x}_\lambda ^3u_{[0]}^{}\overline{\theta }^2u_0^{\prime \prime }\stackrel{~}{y}_\rho ^1`$
$`\text{ }>\text{ }(b^{}\overline{\theta }^3u_1^{\prime \prime }\stackrel{~}{y}_\rho ^2)(b^{\prime \prime }\stackrel{~}{y}_\rho ^3)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{5.5})}{=}}`$ $`(a<u\text{ }>\text{ }b)[(a^{}<u^{}\text{ }>\text{ }b^{})(a^{\prime \prime }<u^{\prime \prime }\text{ }>\text{ }b^{\prime \prime })].`$
Finally, by (2.28), (2.29), (2.32), (2.33) and (2.46) it follows that $`1_A<1_๐ธ\text{ }>\text{ }1_B`$ is the unit of $`A<๐ธ\text{ }>\text{ }B`$. โ
###### Remarks 5.2.
(i) The generalized two-sided crossed product $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐
`$ cannot be particularized for $`๐=k`$ or $`๐
=k`$ because, in general, $`k`$ is not a right or left $`H`$-comodule algebra (in fact, we can do that if and only if we work with a quasi-Hopf algebra which is a twisted Hopf algebra, i.e. it is of the form $`H_F`$ where $`H`$ is an ordinary Hopf algebra and $`FHH`$ is a twist on $`H`$). For the algebra $`A<๐ธ\text{ }>\text{ }B`$, we can take $`A=k`$ or $`B=k`$. In these cases we obtain the right or left generalized smash products $`๐ธ\text{ }>\text{ }B`$ and $`A<๐ธ`$, respectively. For this reason we call the algebra $`A<๐ธ\text{ }>\text{ }B`$ the two-sided generalized smash product. Note that, in the Hopf case, the multiplication of $`A<๐ธ\text{ }>\text{ }B`$ is given by
$$(a<u\text{ }>\text{ }b)(a^{}<u^{}\text{ }>\text{ }b^{})=a(u_{[1]}a^{})<u_{[0]}u_0^{}\text{ }>\text{ }(bu_1^{})b^{},$$
for all $`a,a{}_{}{}^{}A`$, $`u,u{}_{}{}^{}๐ธ`$ and $`b,b{}_{}{}^{}B`$.
(ii) Let $`๐ธ=H`$. In this particular case we will denote the algebra $`A<H\text{ }>\text{ }B`$ by $`A\mathrm{\#}H\mathrm{\#}B`$ (the elements will be written $`a\mathrm{\#}h\mathrm{\#}b`$, $`aA`$, $`hH`$, $`bB`$) and will call it the two-sided smash product. Our terminology is based on the fact that when we take $`A=k`$ or $`B=k`$ the resulting algebra is the right or left smash product algebra. Note that the multiplication of $`A\mathrm{\#}H\mathrm{\#}B`$ is defined by
$$(a\mathrm{\#}h\mathrm{\#}b)(a{}_{}{}^{}\mathrm{\#}h{}_{}{}^{}\mathrm{\#}b{}_{}{}^{})=(x^1a)(x^2h_1y^1a^{})\mathrm{\#}x^3h_2y^2h_1^{}z^1\mathrm{\#}(by^3h_2^{}z^2)(b^{}z^3),$$
for all $`a,a{}_{}{}^{}A`$, $`h,h{}_{}{}^{}H`$ and $`b,b{}_{}{}^{}B`$. It follows that the canonical maps $`i:A\mathrm{\#}HA\mathrm{\#}H\mathrm{\#}B`$ and $`j:H\mathrm{\#}BA\mathrm{\#}H\mathrm{\#}B`$, $`i(a\mathrm{\#}h)=a\mathrm{\#}h\mathrm{\#}1_B`$ and $`j(h\mathrm{\#}b)=1_A\mathrm{\#}h\mathrm{\#}b`$, are algebra morphisms.
In the Hopf case the multiplication of the two-sided smash product is defined by
$$(a\mathrm{\#}h\mathrm{\#}b)(a^{}\mathrm{\#}h^{}\mathrm{\#}b^{})=a(h_1a^{})\mathrm{\#}h_2h_1^{}\mathrm{\#}(bh_2^{})b^{}.$$
## 6. Two-sided products vs generalized diagonal crossed products
As mentioned before, Hausser and Nill proved that a two-sided crossed product over a quasi-Hopf algebra is isomorphic to a right diagonal crossed product. We prove now that a generalized two-sided crossed product is isomorphic to a left generalized diagonal crossed product. Namely, let $`H`$ be a quasi-bialgebra, $`๐`$ an $`H`$-bimodule algebra, $`(๐,\rho ,\mathrm{\Phi }_\rho )`$ a right $`H`$-comodule algebra and $`(๐
,\lambda ,\mathrm{\Phi }_\lambda )`$ a left $`H`$-comodule algebra. Then, by , $`๐๐
`$ becomes an $`H`$-bicomodule algebra, with the following structure: $`\rho (๐๐)=(๐_0๐)๐_1`$, $`\lambda (๐๐)=๐_{[1]}(๐๐_{[0]})`$, $`\mathrm{\Phi }_\rho =(\stackrel{~}{X}_\rho ^11_๐
)\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3`$, $`\mathrm{\Phi }_\lambda =\stackrel{~}{X}_\lambda ^1\stackrel{~}{X}_\lambda ^2(1_๐\stackrel{~}{X}_\lambda ^3)`$, $`\mathrm{\Phi }_{\lambda ,\rho }=1_H(1_๐1_๐
)1_H`$, for all $`๐๐`$ and $`๐๐
`$.
###### Proposition 6.1.
If $`H`$ is a quasi-Hopf algebra and $`๐`$, $`๐`$, $`๐
`$ are as above, then the generalized two-sided crossed product $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐
`$ is isomorphic as an algebra to the generalized left diagonal crossed product $`๐(๐๐
)`$.
###### Proof.
Define the map $`\nu :๐\text{ }>\text{ }๐\text{ }<\text{ }๐
๐(๐๐
)`$,
$`\nu (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)=\phi S^1(๐_1\stackrel{~}{p}_\rho ^2)(๐_0\stackrel{~}{p}_\rho ^1๐),`$
for all $`๐๐`$, $`๐๐
`$ and $`\phi ๐`$, where $`\stackrel{~}{p}=\stackrel{~}{p}_\rho ^1\stackrel{~}{p}_\rho ^2`$ is the element defined in (2.34). We prove that $`\nu `$ is an algebra map. For $`๐,๐^{}๐`$, $`๐,๐^{}๐
`$, $`\phi ,\phi ^{}๐`$, we compute:
$`\nu ((๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)(๐^{}\text{ }>\text{ }\phi ^{}\text{ }<\text{ }๐^{}))`$
$`=`$ $`\nu (๐๐_0^{}\stackrel{~}{x}_\rho ^1\text{ }>\text{ }(\stackrel{~}{x}_\lambda ^1\phi ๐_1^{}\stackrel{~}{x}_\rho ^2)(\stackrel{~}{x}_\lambda ^2b_{[1]}\phi ^{}\stackrel{~}{x}_\rho ^3)\text{ }<\text{ }\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{})`$
$`=`$ $`[(\stackrel{~}{x}_\lambda ^1\phi ๐_1^{}\stackrel{~}{x}_\rho ^2)(\stackrel{~}{x}_\lambda ^2b_{[1]}\phi ^{}\stackrel{~}{x}_\rho ^3)]S^1(๐_1๐_{0,1}^{}(\stackrel{~}{x}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)`$
$`(๐_0๐_{0,0}^{}(\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.50},\text{2.11})}{=}}`$ $`[\stackrel{~}{x}_\lambda ^1\phi ๐_1^{}\stackrel{~}{x}_\rho ^2S^1(g^2)S^1((\stackrel{~}{x}_\rho ^1)_{1_2}(\stackrel{~}{p}_\rho ^2)_2)S^1(๐_{1_2}๐_{0,1_2}^{})S^1(f^2)]`$
$`[\stackrel{~}{x}_\lambda ^2b_{[1]}\phi ^{}\stackrel{~}{x}_\rho ^3S^1(g^1)S^1((\stackrel{~}{x}_\rho ^1)_{1_1}(\stackrel{~}{p}_\rho ^2)_1)S^1(๐_{1_1}๐_{0,1_1}^{})S^1(f^1)]`$
$`(๐_0๐_{0,0}^{}(\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.39},\text{2.26})}{=}}`$ $`[\stackrel{~}{x}_\lambda ^1\phi ๐_1^{}S^1(\stackrel{~}{p}_\rho ^2)S^1(๐_{0,1}^{})S^1(f^2a_{1_2}\stackrel{~}{X}_\rho ^3)]`$
$`[\stackrel{~}{x}_\lambda ^2๐_{[1]}\phi ^{}S^1(f^1a_{1_1}\stackrel{~}{X}_\rho ^2๐_{0,0,1}^{}(\stackrel{~}{p}_\rho ^1)_1\stackrel{~}{P}_\rho ^2)]`$
$`(๐_0\stackrel{~}{X}_\rho ^1๐_{0,0,0}^{}(\stackrel{~}{p}_\rho ^1)_0\stackrel{~}{P}_\rho ^1\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.26},\text{2.35})}{=}}`$ $`[\stackrel{~}{x}_\lambda ^1\phi S^1(f^2\stackrel{~}{X}_\rho ^3๐_1\stackrel{~}{p}_\rho ^2)]`$
$`[\stackrel{~}{x}_\lambda ^2๐_{[1]}\phi ^{}S^1(f^1\stackrel{~}{X}_\rho ^2๐_{0,1}(\stackrel{~}{p}_\rho ^1)_1๐_1^{}\stackrel{~}{P}_\rho ^2)]`$
$`(\stackrel{~}{X}_\rho ^1๐_{0,0}(\stackrel{~}{p}_\rho ^1)_0๐_0^{}\stackrel{~}{P}_\rho ^1\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{}),`$
where $`\stackrel{~}{P}_\rho ^1\stackrel{~}{P}_\rho ^2`$ is another copy of $`\stackrel{~}{p}_\rho `$.
On the other hand, we have seen in Remark 4.8 that for the bicomodule algebra $`๐๐
`$ we have $`\mathrm{\Omega }=\omega `$ and the multiplications in $`๐(๐๐
)`$ and $`๐\text{ }\text{ }(๐๐
)`$ coincide. One can check that $`\omega H^2(๐๐
)H^2`$ for $`๐๐
`$ is obtained by
(6.1) $`\omega =\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\lambda ^2(\stackrel{~}{X}_\rho ^1\stackrel{~}{x}_\lambda ^3)S^1(f^1\stackrel{~}{X}_\rho ^2)S^1(f^2\stackrel{~}{X}_\rho ^3),`$
and then we compute (using the formula for $``$):
$`\nu (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)\nu (๐^{}\text{ }>\text{ }\phi ^{}\text{ }<\text{ }๐^{})`$
$`=`$ $`[\phi S^1(๐_1\stackrel{~}{p}_\rho ^2)(๐_0\stackrel{~}{p}_\rho ^1๐)][\phi ^{}S^1(๐_1^{}\stackrel{~}{P}_\rho ^2)(๐_0^{}\stackrel{~}{P}_\rho ^1๐^{})]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.22},\text{6.1})}{=}}`$ $`[\stackrel{~}{x}_\lambda ^1\phi S^1(f^2\stackrel{~}{X}_\rho ^3๐_1\stackrel{~}{p}_\rho ^2)][\stackrel{~}{x}_\lambda ^2๐_{[1]}\phi ^{}S^1(f^1\stackrel{~}{X}_\rho ^2๐_{0,1}(\stackrel{~}{p}_\rho ^1)_1๐_1^{}\stackrel{~}{P}_\rho ^2)]`$
$`(\stackrel{~}{X}_\rho ^1๐_{0,0}(\stackrel{~}{p}_\rho ^1)_0๐_0^{}\stackrel{~}{P}_\rho ^1\stackrel{~}{x}_\lambda ^3๐_{[0]}๐^{}),`$
hence $`\nu `$ is multiplicative. It obviously satisfies $`\nu (1_๐\text{ }>\text{ }1_๐\text{ }<\text{ }1_๐
)=1_๐(1_๐1_๐
)`$, hence it is an algebra map.
We prove now that $`\nu `$ is bijective. Define $`\nu ^1:๐(๐๐
)๐\text{ }>\text{ }๐\text{ }<\text{ }๐
`$,
$$\nu ^1(\phi (๐๐))=\stackrel{~}{q}_\rho ^1๐_0\text{ }>\text{ }\phi \stackrel{~}{q}_\rho ^2๐_1\text{ }<\text{ }๐,$$
for all $`๐๐`$, $`๐๐
`$, $`\phi ๐`$, where $`\stackrel{~}{q}_\rho =\stackrel{~}{q}_\rho ^1\stackrel{~}{q}_\rho ^2`$ is the element defined in (2.34). We claim that $`\nu `$ and $`\nu ^1`$ are inverses. Indeed,
$`\nu \nu ^1(\phi (๐๐))`$
$`=`$ $`\phi \stackrel{~}{q}_\rho ^2๐_{<1>}S^1(\stackrel{~}{p}_\rho ^2)S^1(๐_{0,1})S^1((\stackrel{~}{q}_\rho ^1)_1)((\stackrel{~}{q}_\rho ^1)_0๐_{0,0}\stackrel{~}{p}_\rho ^1๐)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.35})}{=}}`$ $`\phi \stackrel{~}{q}_\rho ^2S^1(\stackrel{~}{p}_\rho ^2)S^1((\stackrel{~}{q}_\rho ^1)_1)((\stackrel{~}{q}_\rho ^1)_0\stackrel{~}{p}_\rho ^1๐๐)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.37})}{=}}`$ $`\phi (๐๐),`$
and
$`\nu ^1\nu (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)`$
$`=`$ $`\stackrel{~}{q}_\rho ^1๐_{0,0}(\stackrel{~}{p}_\rho ^1)_0\text{ }>\text{ }\phi S^1(\stackrel{~}{p}_\rho ^2)S^1(๐_1)\stackrel{~}{q}_\rho ^2๐_{0,1}(\stackrel{~}{p}_\rho ^1)_1\text{ }<\text{ }๐`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.36})}{=}}`$ $`๐\stackrel{~}{q}_\rho ^1(\stackrel{~}{p}_\rho ^1)_0\text{ }>\text{ }\phi S^1(\stackrel{~}{p}_\rho ^2)\stackrel{~}{q}_\rho ^2(\stackrel{~}{p}_\rho ^1)_1\text{ }<\text{ }๐`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.38})}{=}}`$ $`๐\text{ }>\text{ }\phi \text{ }<\text{ }๐,`$
and this finishes the proof. โ
###### Remark 6.2.
Let $`H`$, $`๐`$, $`๐`$, $`๐
`$ be as above and consider the map $`\mathrm{\Gamma }`$ as in Proposition 3.8, with $`๐ธ`$ taken to be $`๐๐
`$. Then, due to the particular structure of $`๐ธ`$, the map $`\mathrm{\Gamma }:๐๐(๐๐
)`$ is given by
$$\mathrm{\Gamma }(\phi )=\phi S^1(\stackrel{~}{p}_\rho ^2)(\stackrel{~}{p}_\rho ^11_๐
),$$
for all $`\phi ๐`$, where $`\stackrel{~}{p}_\rho `$ is the one corresponding to $`๐`$. Then, using this formula and (3.21), one verifies that the isomorphism $`\nu `$ from Proposition 6.1 reduces to
$$\nu (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)=๐\mathrm{\Gamma }(\phi )๐,๐๐,๐๐
,\phi ๐,$$
where we suppressed the embeddings of $`๐`$ and $`๐
`$ into $`๐(๐๐
)`$ (this generalizes , Proposition 11.4).
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra and $`B`$ a right $`H`$-module algebra. Then $`AB`$ becomes an $`H`$-bimodule algebra via the $`H`$-actions
(6.2)
$$h(ab)h{}_{}{}^{}=habh{}_{}{}^{},\text{ }\text{ }aA\text{}h,h{}_{}{}^{}H\text{}bB\text{.}$$
###### Proposition 6.3.
Let $`H`$ be a quasi-Hopf algebra, $`A`$ a left $`H`$-module algebra, $`B`$ a right $`H`$-module algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then the two-sided generalized smash product $`A<๐ธ\text{ }>\text{ }B`$ is isomorphic as an algebra to the generalized diagonal crossed product $`(AB)๐ธ`$.
###### Proof.
Define $`\mu :(AB)๐ธA<๐ธ\text{ }>\text{ }B`$,
(6.3)
$$\text{ }\mu ((ab)u)=\mathrm{\Theta }^1a<\stackrel{~}{q}_\rho ^1\mathrm{\Theta }_0^2u_0\text{ }>\text{ }bS^1(\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2\mathrm{\Theta }_1^2u_1,$$
for all $`aA`$, $`bB`$ and $`u๐ธ`$, where $`\stackrel{~}{q}_\rho =\stackrel{~}{q}_\rho ^1\stackrel{~}{q}_\rho ^2`$ is the element defined in (2.34). We will prove that $`\mu `$ is an algebra isomorphism. First, observe that the multiplication of $`(AB)๐ธ`$ is defined by
$`((ab)u)((a{}_{}{}^{}b{}_{}{}^{})u{}_{}{}^{})`$
(6.4) $`=[(\mathrm{\Omega }^1a)(\mathrm{\Omega }^2u_{0_{[1]}}a{}_{}{}^{})(b\mathrm{\Omega }^5)(b{}_{}{}^{}S^1(u_1)\mathrm{\Omega }^4)]\mathrm{\Omega }^3u_{0_{[0]}}u{}_{}{}^{},`$
for all $`a,a{}_{}{}^{}A`$, $`b,b{}_{}{}^{}B`$ and $`u,u{}_{}{}^{}๐ธ`$. By using (4.1), (2.40), (2.44) and several times (2.36) and (2.26), we obtain that
$`\mathrm{\Theta }_1^1\mathrm{\Omega }^1\mathrm{\Theta }_2^1\mathrm{\Omega }^2\stackrel{~}{q}_\rho ^1(\mathrm{\Theta }^2\mathrm{\Omega }^3)_0\mathrm{\Omega }^5S^1(\mathrm{\Theta }^3)_1(\stackrel{~}{q}_\rho ^2)_1(\mathrm{\Theta }^2\mathrm{\Omega }^3)_{1_1}\mathrm{\Omega }^4S^1(\mathrm{\Theta }^3)_2`$
$`\times (\stackrel{~}{q}_\rho ^2)_2(\mathrm{\Theta }^2\mathrm{\Omega }^3)_{1_2}=\stackrel{~}{x}_\lambda ^1\mathrm{\Theta }^1\stackrel{~}{x}_\lambda ^2๐ฏ^1\mathrm{\Theta }_{[1]}^2\overline{\mathrm{\Theta }}^1\stackrel{~}{x}_\lambda ^3\stackrel{~}{q}_\rho ^1(๐ฏ^2\mathrm{\Theta }_{[0]}^2\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_0\stackrel{~}{x}_\rho ^1`$
(6.5) $`S^1(๐ฏ^3\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2(๐ฏ^2\mathrm{\Theta }_{[0]}^2\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_1\stackrel{~}{x}_\rho ^2S^1(\overline{\mathrm{\Theta }}^3)\stackrel{~}{Q}_\rho ^2\overline{\mathrm{\Theta }}_1^2\stackrel{~}{x}_\rho ^3,`$
where we denote by $`\stackrel{~}{Q}_\rho ^1\stackrel{~}{Q}_\rho ^2`$ another copy of $`\stackrel{~}{q}_\rho `$. On the other hand, by (2.26), (2.43) and (2.36) it follows that
$`\overline{\mathrm{\Theta }}^1u_{0_{[1]}}(\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_0\stackrel{~}{x}_\rho ^1u_{0_{[0]_0}}u{}_{}{}^{}{}_{0}{}^{}(\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_1\stackrel{~}{x}_\rho ^2u_{0_{[0]_{1_1}}}u_{1_1}^{}{}_{}{}^{}`$
$`S^1(\overline{\mathrm{\Theta }}^3u_1)\stackrel{~}{Q}_\rho ^2\overline{\mathrm{\Theta }}_1^2\stackrel{~}{x}_\rho ^3u_{0_{[0]_{1_2}}}u{}_{}{}^{}{}_{1_2}{}^{}=u_{[1]}\overline{\mathrm{\Theta }}^1(u_{[0]}\stackrel{~}{Q}_\rho ^1)_0(\overline{\mathrm{\Theta }}^2u{}_{}{}^{})_{0,0}\stackrel{~}{x}_\rho ^1`$
(6.6) $`(u_{[0]}\stackrel{~}{Q}_\rho ^1)_1(\overline{\mathrm{\Theta }}^2u{}_{}{}^{})_{0,1}\stackrel{~}{x}_\rho ^2S^1(\overline{\mathrm{\Theta }}^3)\stackrel{~}{Q}_\rho ^2(\overline{\mathrm{\Theta }}^2u{}_{}{}^{})_1\stackrel{~}{x}_\rho ^3,`$
for all $`u,u^{^{}}๐ธ`$. Finally, using (2.34), (2.45), (2.6) and (2.46), one checks that
(6.7)
$$๐ฏ^1\stackrel{~}{q}_\rho ^1๐ฏ_0^2S^1(๐ฏ^3)\stackrel{~}{q}_\rho ^2๐ฏ_1^2=(\stackrel{~}{q}_\rho ^1)_{[1]}\theta ^1(\stackrel{~}{q}_\rho ^1)_{[0]}\theta ^2\stackrel{~}{q}_\rho ^2\theta ^3.$$
Now, for all $`a,a^{}A`$, $`u,u^{}๐ธ`$ and $`b,b^{}B`$ we compute:
$`\mu (((ab)u)((a^{}b^{})u^{}))`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.4},\text{6.3})}{=}}`$ $`(\mathrm{\Theta }_1^1\mathrm{\Omega }^1a)(\mathrm{\Theta }_2^1\mathrm{\Omega }^2u_{0_{[1]}}a^{})<\stackrel{~}{q}_\rho ^1(\mathrm{\Theta }^2\mathrm{\Omega }^3)_0u_{0_{[0]_0}}u_0^{}`$
$`\text{ }>\text{ }(b\mathrm{\Omega }^5S^1(\mathrm{\Theta }^3)_1(\stackrel{~}{q}_\rho ^2)_1(\mathrm{\Theta }^2\mathrm{\Omega }^3)_{1_1}u_{0_{[0]_{1_1}}}u_{1_1}^{})`$
$`(b{}_{}{}^{}S^1(u_1)\mathrm{\Omega }^4S^1(\mathrm{\Theta }^3)_2(\stackrel{~}{q}_\rho ^2)_2(\mathrm{\Theta }^2\mathrm{\Omega }^3)_{1_2}u_{0_{[0]_{1_2}}}u_{1_2}^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.5})}{=}}`$ $`(\stackrel{~}{x}_\lambda ^1\mathrm{\Theta }^1a)(\stackrel{~}{x}_\lambda ^2๐ฏ^1\mathrm{\Theta }_{[1]}^2\overline{\mathrm{\Theta }}^1u_{0_{[1]}}a{}_{}{}^{})<\stackrel{~}{x}_\lambda ^3\stackrel{~}{q}_\rho ^1๐ฏ_0^2\mathrm{\Theta }_{[0]_0}^2(\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_0`$
$`\stackrel{~}{x}_\rho ^1u_{0_{[0]_0}}u_0^{}\text{ }>\text{ }(bS^1(๐ฏ^3\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2๐ฏ_1^2\mathrm{\Theta }_{[0]_1}^2(\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2)_1\stackrel{~}{x}_\rho ^2`$
$`u_{0_{[0]_{1_1}}}u_{1_1}^{})(b^{}S^1(\overline{\mathrm{\Theta }}^3u_1)\stackrel{~}{Q}_\rho ^2\overline{\mathrm{\Theta }}_1^2\stackrel{~}{x}_\rho ^3u_{0_{[0]_{1_2}}}u_{1_2}^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.6})}{=}}`$ $`(\stackrel{~}{x}_\lambda ^1\mathrm{\Theta }^1a)(\stackrel{~}{x}_\lambda ^2๐ฏ^1\mathrm{\Theta }_{[1]}^2u_{[1]}\overline{\mathrm{\Theta }}^1a^{})<\stackrel{~}{x}_\lambda ^3\stackrel{~}{q}_\rho ^1๐ฏ_0^2\mathrm{\Theta }_{[0]_0}^2(u_{[0]}\stackrel{~}{Q}_\rho ^1)_0`$
$`(\overline{\mathrm{\Theta }}^2u^{})_{0,0}\stackrel{~}{x}_\rho ^1\text{ }>\text{ }(bS^1(๐ฏ^3\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2๐ฏ_1^2\mathrm{\Theta }_{[0]_1}^2(u_{[0]}\stackrel{~}{Q}_\rho ^1)_1(\overline{\mathrm{\Theta }}^2u^{})_{0,1}\stackrel{~}{x}_\rho ^2)`$
$`(b^{}S^1(\overline{\mathrm{\Theta }}^3)\stackrel{~}{Q}_\rho ^2(\overline{\mathrm{\Theta }}^2u^{})_1\stackrel{~}{x}_\rho ^3)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.7},\text{5.5},\text{2.43})}{=}}`$ $`(\mathrm{\Theta }^1a<\stackrel{~}{q}_\rho ^1\mathrm{\Theta }_0^2u_0\text{ }>\text{ }bS^1(\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2\mathrm{\Theta }_1^2u_1)`$
$`(\overline{\mathrm{\Theta }}^1a^{}<\stackrel{~}{Q}_\rho ^1\overline{\mathrm{\Theta }}_0^2u_0^{}\text{ }>\text{ }b^{}S^1(\overline{\mathrm{\Theta }}^3)\stackrel{~}{Q}_\rho ^2\overline{\mathrm{\Theta }}_1^2u_1^{})`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.3})}{=}}`$ $`\mu ((ab)u)\mu ((a^{}b^{})u^{}),`$
as claimed. The (co) unit axioms imply $`\mu ((1_A1_B)1_๐ธ)=1_A<1_๐ธ\text{ }>\text{ }1_B`$, so it remains to show that $`\mu `$ is bijective. To this end, define $`\mu ^1:A<๐ธ\text{ }>\text{ }B(AB)๐ธ`$,
(6.8)
$$\mu ^1(a<u\text{ }>\text{ }b)=(\theta ^1abS^1(\theta ^3u_1\stackrel{~}{p}_\rho ^2))\theta ^2u_0\stackrel{~}{p}_\rho ^1,$$
for all $`aA`$, $`u๐ธ`$ and $`bB`$, where $`\stackrel{~}{p}_\rho =\stackrel{~}{p}_\rho ^1\stackrel{~}{p}_\rho ^2`$ is the element defined in (2.34). We show that $`\mu `$ and $`\mu ^1`$ are inverses. Indeed,
$`\mu \mu ^1(a<u\text{ }>\text{ }b)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.8},\text{6.3})}{=}}`$ $`a<\stackrel{~}{q}_\rho ^1u_{0,0}(\stackrel{~}{p}_\rho ^1)_0\text{ }>\text{ }bS^1(u_1\stackrel{~}{p}_\rho ^2)\stackrel{~}{q}_\rho ^2u_{0,1}(\stackrel{~}{p}_\rho ^1)_1`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.36},\text{2.38})}{=}}`$ $`a<u\text{ }>\text{ }b,`$
for all $`aA`$, $`u๐ธ`$, $`bB`$, and similarly
$`\mu ^1\mu ((ab)u)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{6.3},\text{6.8})}{=}}`$ $`[\theta ^1\mathrm{\Theta }^1abS^1(\mathrm{\Theta }^3)\stackrel{~}{q}_\rho ^2(\mathrm{\Theta }^2u)_1S^1(\theta ^3(\stackrel{~}{q}_\rho ^1)_1(\mathrm{\Theta }^2u)_{0,1})\stackrel{~}{p}_\rho ^2)]`$
$`\theta ^2(\stackrel{~}{q}_\rho ^1)_0(\mathrm{\Theta }^2u)_{0,0}\stackrel{~}{p}_\rho ^1`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.35},\text{2.37})}{=}}`$ $`(ab)u,`$
and this finishes our proof. โ
As a consequence of the two propositions, we obtain the following result:
###### Corollary 6.4.
Let $`H`$ be a quasi-Hopf algebra, $`A`$ a left $`H`$-module algebra, $`B`$ a right $`H`$-module algebra, $`๐`$ a right $`H`$-comodule algebra and $`๐
`$ a left $`H`$-comodule algebra. Then we have algebra isomorphisms
$`A<(๐๐
)\text{ }>\text{ }B(AB)(๐๐
)๐\text{ }>\text{ }(AB)\text{ }<\text{ }๐
.`$
## 7. Invariance under twisting
In this section we prove that the generalized diagonal crossed products and the two-sided smash products are, in certain senses, invariant under twisting (such a result has also been proved by Hausser and Nill in for their diagonal crossed products, with a different method, and by the authors in for smash products).
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra, $`๐`$ an $`H`$-bimodule algebra and $`FHH`$ a gauge transformation. If we introduce on $`A`$ another multiplication, by $`aa^{}=(G^1a)(G^2a^{})`$ for all $`a,a^{}A`$, where $`F^1=G^1G^2`$, and denote by $`A_{F^1}`$ this structure, then, as in , one can prove that $`A_{F^1}`$ becomes a left $`H_F`$-module algebra, with the same unit and $`H`$-action as for $`A`$. If we introduce on $`๐`$ another multiplication, by $`\phi \phi ^{}=(G^1\phi F^1)(G^2\phi ^{}F^2)`$ for all $`\phi ,\phi ^{}๐`$, and denote this by $`{}_{F}{}^{}๐_{F^1}^{}`$, then $`{}_{F}{}^{}๐_{F^1}^{}`$ is an $`H_F`$-bimodule algebra (for instance, if $`๐=H^{}`$, then $`{}_{F}{}^{}๐_{F^1}^{}`$ is just $`(H_F)^{}`$). Moreover, if we regard $`๐`$ as a left $`HH^{op}`$-module algebra and $`{}_{F}{}^{}๐_{F^1}^{}`$ as a left $`H_FH_F^{op}`$-module algebra, then $`{}_{F}{}^{}๐_{F^1}^{}`$ coincides with $`๐_{T^1}`$, where $`T`$ is the gauge transformation on $`HH^{op}`$ given by $`T=(F^1G^1)(F^2G^2)`$, and using the identification $`H_F(H_F)^{op}(HH^{op})_T`$.
Suppose that we have also a left $`H`$-comodule algebra $`๐
`$; then, by , on the algebra structure of $`๐
`$ one can introduce a left $`H_F`$-comodule algebra structure (denoted in what follows by $`๐
^{F^1}`$) by putting $`\lambda ^{F^1}=\lambda `$ and $`\mathrm{\Phi }_\lambda ^{F^1}=\mathrm{\Phi }_\lambda (F^11_๐
)`$.
###### Proposition 7.1.
With notation as above, we have an algebra isomorphism
$`A<๐
A_{F^1}<๐
^{F^1},`$
obtained from the trivial identification.
###### Proof.
Check directly that the multiplication in $`A_{F^1}<๐
^{F^1}`$ coincides, via the trivial identification, with the one in $`A<๐
`$. โ
Similarly, if $`๐`$ is a right $`H`$-comodule algebra, by one can introduce on the algebra structure of $`๐`$ a right $`H_F`$-comodule algebra structure (denoted by $`{}_{}{}^{F}๐`$) by putting $`{}_{}{}^{F}\rho =\rho `$ and $`{}_{}{}^{F}\mathrm{\Phi }_{\rho }^{}=(1_๐F)\mathrm{\Phi }_\rho `$.
Also, one can check that if $`๐ธ`$ is an $`H`$-bicomodule algebra, the left and right $`H_F`$-comodule algebras $`๐ธ^{F^1}`$ and $`{}_{}{}^{F}๐ธ`$ actually define the structure of an $`H_F`$-bicomodule algebra on $`๐ธ`$, denoted by $`{}_{}{}^{F}๐ธ_{}^{F^1}`$, which has the same $`\mathrm{\Phi }_{\lambda ,\rho }`$ as $`๐ธ`$.
Suppose now that $`H`$ is a quasi-Hopf algebra. Transforming this $`H_F`$-bicomodule algebra $`{}_{}{}^{F}๐ธ_{}^{F^1}`$, as in a previous section, into the two left $`H_FH_F^{op}`$-comodule algebras $`({}_{}{}^{F}๐ธ_{}^{F^1})_1`$ and $`({}_{}{}^{F}๐ธ_{}^{F^1})_2`$, by using the identification $`H_FH_F^{op}(HH^{op})_T`$ as before and the fact, observed in , that the Drinfeld twist $`f_F`$ on $`H_F`$ depends on the one on $`H`$ by the formula $`f_F=(SS)(F_{21}^1)fF^1`$, we may obtain algebra isomorphisms
$$({}_{}{}^{F}๐ธ_{}^{F^1})_1(๐ธ_1)^{T^1},({}_{}{}^{F}๐ธ_{}^{F^1})_2(๐ธ_2)^{T^1},$$
defined by the trivial identifications.
As a consequence, using the expressions of the generalized left diagonal crossed products as generalized smash products, we obtain the following result:
###### Proposition 7.2.
With notation as before, the algebra isomorphisms
$$๐๐ธ_F๐_{F^1}{}_{}{}^{F}๐ธ_{}^{F^1},๐\text{ }\text{ }๐ธ{}_{F}{}^{}๐_{F^1}^{}\text{ }\text{ }{}_{}{}^{F}๐ธ_{}^{F^1},$$
are defined by the trivial identifications.
Suppose again that $`H`$ is a quasi-bialgebra, $`A`$ is a left $`H`$-module algebra and $`FHH`$ is a gauge transformation. Suppose now that we also have a right $`H`$-module algebra $`B`$. If we introduce on $`B`$ another multiplication, by $`bb^{}=(bF^1)(b^{}F^2)`$ for all $`b,b^{}B`$, denoting this structure by $`{}_{F}{}^{}B`$, then $`{}_{F}{}^{}B`$ becomes a right $`H_F`$-module algebra with the same unit and right $`H`$-action as for $`B`$. So, we have the following type of invariance under twisting for two-sided smash products:
###### Proposition 7.3.
With notation as before, we have an algebra isomorphism
$`\phi :A\mathrm{\#}H\mathrm{\#}BA_{F^1}\mathrm{\#}H_F\mathrm{\#}_FB,`$
$`\phi (a\mathrm{\#}h\mathrm{\#}b)=F^1a\mathrm{\#}F^2hG^1\mathrm{\#}bG^2,aA,hH,bB.`$
In particular, by taking $`B=k`$ or respectively $`A=k`$, we have algebra isomorphisms
$$A\mathrm{\#}HA_{F^1}\mathrm{\#}H_F,H\mathrm{\#}BH_F\mathrm{\#}_FB.$$
###### Proof.
Follows by a direct computation, similar to the one in . โ
## 8. Iterated products
It was proved in that, if $`H`$ is a quasi-bialgebra and $`A`$ is a left $`H`$-module algebra, then $`A\mathrm{\#}H`$ becomes a right $`H`$-comodule algebra, with structure:
$`\rho :A\mathrm{\#}H(A\mathrm{\#}H)H,\rho (a\mathrm{\#}h)=(x^1a\mathrm{\#}x^2h_1)x^3h_2,\text{ }\text{ }aA\text{}hH\text{,}`$
$`\mathrm{\Phi }_\rho =(1_A\mathrm{\#}X^1)X^2X^3(A\mathrm{\#}H)HH.`$
Similarly, one can prove that if $`B`$ is a right $`H`$-module algebra, then $`H\mathrm{\#}B`$ becomes a left $`H`$-comodule algebra, with structure:
$`\lambda :H\mathrm{\#}BH(H\mathrm{\#}B),\lambda (h\mathrm{\#}b)=h_1x^1(h_2x^2\mathrm{\#}bx^3),\text{ }\text{ }hH\text{}bB\text{,}`$
$`\mathrm{\Phi }_\lambda =X^1X^2(X^3\mathrm{\#}1_B)HH(H\mathrm{\#}B).`$
In the sequel we need some more general results, that we are stating now (the proof is similar to the one in ).
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then $`A<๐ธ`$ becomes a right $`H`$-comodule algebra, with structure defined for all $`aA`$ and $`u๐ธ`$ by:
$`\rho :A<๐ธ(A<๐ธ)H,\text{ }\rho (a<u)=(\theta ^1a<\theta ^2u_0)\theta ^3u_1,`$
$`\mathrm{\Phi }_\rho =(1_A<\stackrel{~}{X}_\rho ^1)\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3(A<๐ธ)HH.`$
Similarly, let $`H`$ be a quasi-bialgebra, $`B`$ a right $`H`$-module algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Then $`๐ธ\text{ }>\text{ }B`$ becomes a left $`H`$-comodule algebra, with structure defined for all $`u๐ธ`$ and $`bB`$ by:
$`\lambda :๐ธ\text{ }>\text{ }BH(๐ธ\text{ }>\text{ }B),\lambda (u\text{ }>\text{ }b)=u_{[1]}\theta ^1(u_{[0]}\theta ^2\text{ }>\text{ }b\theta ^3),`$
$`\mathrm{\Phi }_\lambda =\stackrel{~}{X}_\lambda ^1\stackrel{~}{X}_\lambda ^2(\stackrel{~}{X}_\lambda ^3\text{ }>\text{ }1_B)HH(๐ธ\text{ }>\text{ }B).`$
We are now ready to prove that the two-sided generalized smash product can be written (in two ways) as an iterated generalized smash product.
###### Proposition 8.1.
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra, $`B`$ a right $`H`$-module algebra and $`๐ธ`$ an $`H`$-bicomodule algebra. Consider the right and left $`H`$-comodule algebras $`A<๐ธ`$ and $`๐ธ\text{ }>\text{ }B`$ as above. Then we have algebra isomorphisms
$$A<๐ธ\text{ }>\text{ }B(A<๐ธ)\text{ }>\text{ }B,A<๐ธ\text{ }>\text{ }BA<(๐ธ\text{ }>\text{ }B),$$
given by the trivial identifications. In particular, we have
$$A\mathrm{\#}H\mathrm{\#}B(A\mathrm{\#}H)\text{ }>\text{ }B,A\mathrm{\#}H\mathrm{\#}BA<(H\mathrm{\#}B).$$
###### Proof.
We will prove the first isomorphism, the second is similar. We compute the multiplication in $`(A<๐ธ)\text{ }>\text{ }B`$. For $`a,a{}_{}{}^{}A`$, $`b,b{}_{}{}^{}B`$ and $`u,u{}_{}{}^{}๐ธ`$ we have:
$`((a<u)\text{ }>\text{ }b)((a{}_{}{}^{}<u{}_{}{}^{})\text{ }>\text{ }b{}_{}{}^{})`$
$`=`$ $`(a<u)(a{}_{}{}^{}<u{}_{}{}^{})_0(1_A<\stackrel{~}{x}_\rho ^1)\text{ }>\text{ }(b(a{}_{}{}^{}<u{}_{}{}^{})_1\stackrel{~}{x}_\rho ^2)(b{}_{}{}^{}\stackrel{~}{x}_\rho ^3)`$
$`=`$ $`(a<u)(\theta ^1a{}_{}{}^{}<\theta ^2u{}_{}{}^{}{}_{0}{}^{}\stackrel{~}{x}_{\rho }^{1})\text{ }>\text{ }(b\theta ^3u{}_{}{}^{}{}_{1}{}^{}\stackrel{~}{x}_{\rho }^{2})(b{}_{}{}^{}\stackrel{~}{x}_\rho ^3)`$
$`=`$ $`((\stackrel{~}{x}_\lambda ^1a)(\stackrel{~}{x}_\lambda ^2u_{[1]}\theta ^1a{}_{}{}^{})<\stackrel{~}{x}_\lambda ^3u_{[0]}\theta ^2u{}_{}{}^{}{}_{0}{}^{}\stackrel{~}{x}_{\rho }^{1})\text{ }>\text{ }(b\theta ^3u{}_{}{}^{}{}_{1}{}^{}\stackrel{~}{x}_{\rho }^{2})(b{}_{}{}^{}\stackrel{~}{x}_\rho ^3).`$
Via the trivial identification, this is exactly the multiplication of $`A<๐ธ\text{ }>\text{ }B`$. โ
Recall from the definition and properties of the so-called quasi-smash product, but in a more general form. Let $`H`$ be a quasi-bialgebra, $`๐`$ a right $`H`$-comodule algebra and $`๐`$ an $`H`$-bimodule algebra. Define a multiplication on $`๐๐`$ by
(8.1)
$$(๐\overline{\mathrm{\#}}\phi )(๐{}_{}{}^{}\overline{\mathrm{\#}}\phi {}_{}{}^{})=๐๐{}_{}{}^{}{}_{0}{}^{}\stackrel{~}{x}_{\rho }^{1}\overline{\mathrm{\#}}(\phi ๐{}_{}{}^{}{}_{1}{}^{}\stackrel{~}{x}_{\rho }^{2})(\phi {}_{}{}^{}\stackrel{~}{x}_\rho ^3),๐,๐^{^{}}๐,\phi ,\phi {}_{}{}^{}๐,$$
where we write $`๐\text{ }\overline{\mathrm{\#}}\text{ }\phi `$ for $`๐\phi `$, and denote this structure by $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐`$. Then $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐`$ becomes a left $`H`$-module algebra with unit $`1_๐\text{ }\overline{\mathrm{\#}}\text{ }1_๐`$ and with left $`H`$-action
$$h(๐\text{ }\overline{\mathrm{\#}}\text{ }\phi )=a\text{ }\overline{\mathrm{\#}}\text{ }h\phi ,๐๐,hH,\phi ๐.$$
Note that for $`๐=H^{}`$ we obtain the quasi-smash product $`๐\text{ }\overline{\mathrm{\#}}\text{ }H^{}`$ from . Also, by taking $`B`$ a right $`H`$-module algebra and $`๐=B`$ as an $`H`$-bimodule algebra with trivial left $`H`$-action, $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐`$ is exactly the generalized smash product $`๐\text{ }>\text{ }B`$.
We need the left-handed version of the above construction too. Namely, if $`H`$ is a quasi-bialgebra, $`๐
`$ a left $`H`$-comodule algebra and $`๐`$ an $`H`$-bimodule algebra, define a multiplication on $`๐๐
`$ by
(8.2)
$$(\phi \text{ }\overline{\mathrm{\#}}\text{ }๐)(\phi {}_{}{}^{}\text{ }\overline{\mathrm{\#}}\text{ }๐{}_{}{}^{})=(\stackrel{~}{x}_\lambda ^1\phi )(\stackrel{~}{x}_\lambda ^2๐_{[1]}\phi {}_{}{}^{})\text{ }\overline{\mathrm{\#}}\text{ }\stackrel{~}{x}_\lambda ^3๐_{[0]}๐{}_{}{}^{},\phi ,\phi {}_{}{}^{}๐,๐,๐{}_{}{}^{}๐
,$$
where we write $`\phi \text{ }\overline{\mathrm{\#}}\text{ }๐`$ for $`\phi ๐`$, and denote this structure by $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐
`$. Then $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐
`$ becomes a right $`H`$-module algebra with unit $`1_๐\text{ }\overline{\mathrm{\#}}\text{ }1_๐
`$ and with right $`H`$-action
$$(\phi \text{ }\overline{\mathrm{\#}}\text{ }๐)h=\phi h\text{ }\overline{\mathrm{\#}}\text{ }๐,\phi ๐,hH,๐๐
.$$
By taking $`A`$ a left $`H`$-module algebra and $`๐=A`$ as an $`H`$-bimodule algebra with trivial right $`H`$-action, $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐
`$ is exactly the generalized smash product $`A<๐
`$.
By , a two-sided crossed product may be written as a generalized smash product. We have a similar result for generalized two-sided crossed products, which allows us to write them (in two ways) as generalized smash products.
###### Proposition 8.2.
Let $`H`$ be a quasi-bialgebra, $`๐`$ a right $`H`$-comodule algebra, $`๐
`$ a left $`H`$-comodule algebra and $`๐`$ an $`H`$-bimodule algebra. Consider the left and right $`H`$-module algebras $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐`$ and $`๐\text{ }\overline{\mathrm{\#}}\text{ }๐
`$ as above. Then we have algebra isomorphisms
$$๐\text{ }>\text{ }๐\text{ }<\text{ }๐
(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐
,๐\text{ }>\text{ }๐\text{ }<\text{ }๐
๐\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }๐
),$$
obtained from the trivial identifications.
###### Proof.
Follows by direct computations. โ
We now apply the above results. In , Hausser and Nill generalized to the setting of quasi-Hopf algebras some models of Hopf spin chains and lattice current algebras. The key result for this was the next Theorem, concerning iterated two-sided crossed products (with $`H`$ finite dimensional and $`๐=H^{}`$). The original proof of this theorem is quite difficult to read, being written in the formalism of universal intertwiners. Using our results, we are now able to obtain for free a conceptual proof of the Theorem, together with the explicit form of the structures that appear at (i) and (ii).
###### Theorem 8.3.
(Hausser and Nill). Let $`H`$ be a quasi-bialgebra, $`๐`$ an $`H`$-bimodule algebra, $`๐`$ a right $`H`$-comodule algebra, $`๐น`$ an $`H`$-bicomodule algebra and $``$ a left $`H`$-comodule algebra. Then:
* $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐น`$ admits a right $`H`$-comodule algebra structure;
* $`๐น\text{ }>\text{ }๐\text{ }<\text{ }`$ admits a left $`H`$-comodule algebra structure;
* there is an algebra isomorphism (given by the trivial identification)
$$(๐\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }๐\text{ }<\text{ }๐\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }๐\text{ }<\text{ }).$$
###### Proof.
Writting $`๐\text{ }>\text{ }๐\text{ }<\text{ }๐น`$ as $`(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น`$, we obtain that this is a right $`H`$-comodule algebra (being a generalized smash product between a left $`H`$-module algebra and an $`H`$-bicomodule algebra), and we can write explicitly its structure:
$`\rho :๐\text{ }>\text{ }๐\text{ }<\text{ }๐น(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น((๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น)H(๐\text{ }>\text{ }๐\text{ }<\text{ }๐น)H,`$
$`\rho (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐)=(๐\text{ }>\text{ }\theta ^1\phi \text{ }<\text{ }\theta ^2๐_0)\theta ^3๐_1,๐๐,\phi ๐,๐๐น,`$
$`\mathrm{\Phi }_\rho =(1_๐\text{ }>\text{ }1_๐\text{ }<\text{ }\stackrel{~}{X}_\rho ^1)\stackrel{~}{X}_\rho ^2\stackrel{~}{X}_\rho ^3(๐\text{ }>\text{ }๐\text{ }<\text{ }๐น)HH.`$
Similarly, writing $`๐น\text{ }>\text{ }๐\text{ }<\text{ }`$ as $`๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ })`$, we obtain that this is a left $`H`$-comodule algebra, with structure:
$`\lambda :๐น\text{ }>\text{ }๐\text{ }<\text{ }๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ })H(๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }))H(๐น\text{ }>\text{ }๐\text{ }<\text{ }),`$
$`\lambda (๐\text{ }>\text{ }\phi \text{ }<\text{ }๐ )=๐_{[1]}\theta ^1(๐_{[0]}\theta ^2\text{ }>\text{ }\phi \theta ^3\text{ }<\text{ }๐ ),๐๐น,\phi ๐,๐ ,`$
$`\mathrm{\Phi }_\lambda =\stackrel{~}{X}_\lambda ^1\stackrel{~}{X}_\lambda ^2(\stackrel{~}{X}_\lambda ^3\text{ }>\text{ }1_๐\text{ }<\text{ }1_{})HH(๐น\text{ }>\text{ }๐\text{ }<\text{ }).`$
To prove (iii), we will use the identifications appearing in our results:
$`(๐\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }๐\text{ }<\text{ }((๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น)\text{ }>\text{ }๐\text{ }<\text{ }`$
$`((๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น)\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ })(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }),`$
and
$`๐\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }๐\text{ }<\text{ })๐\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }))`$
$`(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<(๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }))(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }).`$
So, we have proved that the two iterated generalized two-sided crossed products that appear in (iii) are both isomorphic as algebras (via the trivial identifications) to the two-sided generalized smash product $`(๐\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ })`$. โ
Using the same results, we obtain another relation between the generalized two-sided crossed product and the two-sided generalized smash product. More exactly, let $`H`$ be a quasi-bialgebra, $`๐`$ an $`H`$-bimodule algebra, $`A`$ a left $`H`$-module algebra, $`B`$ a right $`H`$-module algebra and $`๐ธ`$ and $`๐น`$ two $`H`$-bicomodule algebras. As we have seen before, $`A<๐ธ`$ (respectively $`๐น\text{ }>\text{ }B`$) becomes a right (respectively left) $`H`$-comodule algebra, so we may consider the generalized two-sided crossed product $`(A<๐ธ)\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }B)`$. On the other hand, by the above Theorem of Hausser and Nill, $`๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น`$ becomes a right $`H`$-comodule algebra and a left $`H`$-comodule algebra, but actually, using the explicit formulae for its structures that we gave, one can prove that it is even an $`H`$-bicomodule algebra, with $`\mathrm{\Phi }_{\lambda ,\rho }=1_H(1_๐ธ\text{ }>\text{ }1_๐\text{ }<\text{ }1_๐น)1_H`$, so we may consider the two-sided generalized smash product $`A<(๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }B`$.
###### Proposition 8.4.
We have an algebra isomorphism
$$(A<๐ธ)\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }B)A<(๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }B$$
obtained from the trivial identification. In particular, we have
$$(A\mathrm{\#}H)\text{ }>\text{ }H^{}\text{ }<\text{ }(H\mathrm{\#}B)A<(H\text{ }>\text{ }H^{}\text{ }<\text{ }H)\text{ }>\text{ }B.$$
###### Proof.
This may be proved by computing explicitly the multiplication rules in the two algebras and noting that they coincide. Alternatively, we provide a conceptual proof, by a sequence of identifications using the above results. We compute:
$`A<(๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }BA<((๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }B)`$
$`A<(((๐ธ\text{ }\overline{\mathrm{\#}}\text{ }๐)<๐น)\text{ }>\text{ }B)A<((๐ธ\text{ }\overline{\mathrm{\#}}\text{ }๐)<(๐น\text{ }>\text{ }B))`$
$`A<(๐ธ\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }B))A<(๐ธ\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }(๐น\text{ }>\text{ }B)))`$
$`(A<๐ธ)\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }(๐น\text{ }>\text{ }B))(A<๐ธ)\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }B),`$
where the fourth and the fifth identities hold because the left $`H`$-comodule algebra structures on $`(๐ธ\text{ }>\text{ }๐\text{ }<\text{ }๐น)\text{ }>\text{ }B`$, $`๐ธ\text{ }>\text{ }๐\text{ }<\text{ }(๐น\text{ }>\text{ }B)`$ and $`๐ธ\text{ }>\text{ }(๐\text{ }\overline{\mathrm{\#}}\text{ }(๐น\text{ }>\text{ }B))`$ coincide (via the trivial identifications). โ
## 9. $`H^{}`$-Hopf bimodules
Let $`H`$ be a finite dimensional quasi-bialgebra and $`A`$ a left $`H`$-module algebra. Recall from the category $`_A^H^{}`$, whose objects are vector spaces $`M`$, such that $`M`$ is a right $`H^{}`$-comodule (i.e. $`M`$ is a left $`H`$-module, with action denoted by $`hmhm`$) and $`A`$ acts on $`M`$ to the right (denote by $`mama`$ this action) such that $`m1_A=m`$ for all $`mM`$ and the following relations hold:
(9.1) $`(ma)a^{}=(X^1m)[(X^2a)(X^3a^{})],`$
(9.2) $`h(ma)=(h_1m)(h_2a),`$
for all $`a,a^{}A`$, $`mM`$, $`hH`$. Similarly, the category $`{}_{A}{}^{}_{}^{H^{}}`$ consists of vector spaces $`M`$, such that $`M`$ is a right $`H^{}`$-comodule and $`A`$ acts on $`M`$ to the left (denote by $`amam`$ this action) such that $`1_Am=m`$ for all $`mM`$ and the following relations hold:
(9.3) $`a(a^{}m)=[(x^1a)(x^2a^{})](x^3m),`$
(9.4) $`h(am)=(h_1a)(h_2m),`$
for all $`a,a^{}A`$, $`mM`$, $`hH`$. From the description of left modules over $`A\mathrm{\#}H`$ in , it is clear that $`{}_{A}{}^{}_{}^{H^{}}`$$`{}_{A\mathrm{\#}H}{}^{}`$. If $`H`$ is a quasi-Hopf algebra, by we have an isomorphism of categories $`_A^H^{}_{A\mathrm{\#}H}`$. In what follows we need a description of $`_A^H^{}`$ as a category of left modules over a right smash product.
###### Proposition 9.1.
Let $`H`$ be a quasi-Hopf algebra and $`A`$ a left $`H`$-module algebra. Define on $`A`$ a new multiplication, by putting
(9.5) $`aa^{}=(g^1a^{})(g^2a),a,a^{}A,`$
where $`f^1=g^1g^2`$ is given by (2.14), and denote this new structure by $`\overline{A}`$. Then $`\overline{A}`$ becomes a right $`H`$-module algebra, with the same unit as $`A`$ and right $`H`$-action given by $`ah=S(h)a`$, for all $`aA`$, $`hH`$.
###### Proof.
A straightforward computation, using (2.11) and (2.16). โ
###### Definition 9.2.
Let $`H`$ be a quasi-bialgebra and $`B`$ a right $`H`$-module algebra. We say that $`M`$, a $`k`$-linear space, is a left $`H,B`$-module if
* $`M`$ is a left $`H`$-module with action denoted by $`hmhm`$;
* $`B`$ acts weakly on $`M`$ from the left, i.e. there exists a $`k`$-linear map $`BMM`$, denoted by $`bmbm`$, such that $`1_Bm=m`$ for all $`mM`$;
* the following compatibility conditions hold:
(9.6) $`b(b^{}m)=x^1([(bx^2)(b^{}x^3)]m),`$
(9.7) $`b(hm)=h_1[(bh_2)m],`$
for all $`b,b^{}B`$, $`hH`$, $`mM`$. The category of all left $`H,B`$-modules, morphisms being the $`H`$-linear maps that preserve the $`B`$-action, will be denoted by $`{}_{H,B}{}^{}`$.
###### Proposition 9.3.
If $`H`$, $`B`$ are as above, then the categories $`{}_{H,B}{}^{}`$ and $`{}_{H\mathrm{\#}B}{}^{}`$ are isomorphic. The isomorphism is given as follows. If $`M`$$`{}_{H\mathrm{\#}B}{}^{}`$, define $`hm=(h\mathrm{\#}1)m`$ and $`bm=(1\mathrm{\#}b)m`$. Conversely, if $`M`$$`{}_{H,B}{}^{}`$, define $`(h\mathrm{\#}b)m=h(bm)`$.
###### Proof.
Straightforward computation. โ
###### Proposition 9.4.
If $`H`$ is a finite dimensional quasi-Hopf algebra and $`A`$ is a left $`H`$-module algebra, then $`_A^H^{}`$ is isomorphic to $`{}_{H\mathrm{\#}\overline{A}}{}^{}`$, where $`\overline{A}`$ is the right $`H`$-module algebra constructed in Proposition 9.1. The correspondence is given as follows (we fix $`\{e_i\}`$ a basis in $`H`$ with $`\{e^i\}`$ a dual basis in $`H^{}`$):
$``$ If $`M{}_{H\mathrm{\#}\overline{A}}{}^{}`$, then $`M`$ becomes an object in $`_A^H^{}`$ with the following structures (we denote by $`hmhm`$ the left $`H`$-module structure of $`M`$ and by $`amam`$ the weak left $`\overline{A}`$-action on $`M`$ arising from Proposition 9.3):
$`MMH^{},m{\displaystyle \underset{i=1}{\overset{n}{}}}e_ime^i,mM,`$
$`MAM,mama=q^1((S(q^2)a)m),`$
where $`q_R=q^1q^2=X^1S^1(\alpha X^3)X^2HH`$ (it is the element $`\stackrel{~}{q}_\rho `$ given by (2.34) corresponding to $`๐=H`$).
$``$ Conversely, if $`M_A^H^{}`$, denoting the $`H^{}`$-comodule structure of $`M`$ by $`MMH^{}`$, $`mm_{(0)}m_{(1)}`$, and the weak right $`A`$-action on $`M`$ by $`mama`$, then $`M`$ becomes an object in $`{}_{H\mathrm{\#}\overline{A}}{}^{}`$ with the following structures (again via Proposition 9.3): $`M`$ is a left $`H`$-module with action $`hm=m_{(1)}(h)m_{(0)}`$, and the weak left $`\overline{A}`$-action on $`M`$ is given by
$`am=(p^1m)(p^2a),a\overline{A},mM,`$
where $`p_R=p^1p^2=x^1x^2\beta S(x^3)HH`$ (it is the element $`\stackrel{~}{p}_\rho `$ given by (2.34) corresponding to $`๐=H`$).
###### Proof.
Assume first that $`M{}_{H\mathrm{\#}\overline{A}}{}^{}`$; then we have, by Propositions 9.3 and 9.1:
(9.8) $`a(a^{}m)=x^1([(g^1S(x^3)a^{})(g^2S(x^2)a)]m),`$
(9.9) $`a(hm)=h_1[(S(h_2)a)m],`$
for all $`a,a^{}A`$, $`hH`$, $`mM`$. We have to prove that $`M_A^H^{}`$. To prove (9.1), we compute (denoting by $`Q^1Q^2`$ another copy of $`q_R`$):
$`(ma)a^{}`$ $`=`$ $`Q^1[(S(Q^2)a^{})(q^1[(S(q^2)a)m])]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.9})}{=}}`$ $`Q^1q_1^1[(S(q_2^1)S(Q^2)a^{})((S(q^2)a)m)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.8})}{=}}`$ $`Q^1q_1^1x^1[((g^1S(x^3)S(q^2)a)(g^2S(x^2)S(Q^2q_2^1)a^{}))m]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.40})}{=}}`$ $`q^1X_1^1[((g^1S(X_{(2,2)}^1)S(q_2^2)f^1X^2a)`$
$`(g^2S(X_{(2,1)}^1)S(q_1^2)f^2X^3a^{}))m]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.11})}{=}}`$ $`q^1X_1^1[((S(q^2X_2^1)_1X^2a)(S(q^2X_2^1)_2X^3a^{}))m]`$
$`=`$ $`q^1X_1^1[(S(q^2X_2^1)((X^2a)(X^3a^{})))m]`$
$`=`$ $`q^1[X_1^1[(S(X_2^1)(S(q^2)((X^2a)(X^3a^{}))))m]]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.9})}{=}}`$ $`q^1[(S(q^2)((X^2a)(X^3a^{})))(X^1m)]`$
$`=`$ $`(X^1m)((X^2a)(X^3a^{})),q.e.d.`$
To prove (9.2), we compute:
$`(h_1m)(h_2a)`$ $`=`$ $`q^1((S(q^2)h_2a)(h_1m))`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.9})}{=}}`$ $`q^1h_{(1,1)}((S(h_{(1,2)})S(q^2)h_2a)m)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.36})}{=}}`$ $`hq^1((S(q^2)a)m)`$
$`=`$ $`h(ma),q.e.d.`$
Obviously $`m1_A=m`$, for all $`mM`$, hence indeed $`M_A^H^{}`$.
Conversely, assume that $`M_A^H^{}`$, that is
(9.10) $`(ma)a^{}=(X^1m)[(X^2a)(X^3a^{})],`$
(9.11) $`h(ma)=(h_1m)(h_2a),`$
for all $`mM`$, $`a,a^{}A`$, $`hH`$, and we have to prove that
(9.12) $`a(a^{}m)=x^1([(g^1S(x^3)a^{})(g^2S(x^2)a)]m),`$
(9.13) $`a(hm)=h_1[(S(h_2)a)m],`$
for all $`a,a^{}A`$, $`hH`$, $`mM`$.
To prove (9.12), we compute (denoting by $`P^1P^2`$ another copy of $`p_R`$):
$`a(a^{}m)`$ $`=`$ $`(p^1[(P^1m)(P^2a^{})])(p^2a)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.11})}{=}}`$ $`[(p_1^1P^1m)(p_2^1P^2a^{})](p^2a)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.10})}{=}}`$ $`(X^1p_1^1P^1m)[(X^2p_2^1P^2a^{})(X^3p^2a)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.39})}{=}}`$ $`(x_1^1p^1m)[(x_{(2,1)}^1p_1^2g^1S(x^3)a^{})(x_{(2,2)}^1p_2^2g^2S(x^2)a)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.11})}{=}}`$ $`x^1[(p^1m)[(p_1^2g^1S(x^3)a^{})(p_2^2g^2S(x^2)a)]]`$
$`=`$ $`x^1[((g^1S(x^3)a^{})(g^2S(x^2)a))m],q.e.d.`$
To prove (9.13), we compute:
$`h_1[(S(h_2)a)m]`$ $`=`$ $`h_1[(p^1m)(p^2S(h_2)a)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.11})}{=}}`$ $`(h_{(1,1)}p^1m)(h_{(1,2)}p^2S(h_2)a)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.35})}{=}}`$ $`(p^1hm)(p^2a)`$
$`=`$ $`a(hm),q.e.d.`$
Obviously $`1_Am=m`$, for all $`mM`$, hence indeed $`M{}_{H\mathrm{\#}\overline{A}}{}^{}`$.
In order to prove that $`_A^H^{}`$$`{}_{H\mathrm{\#}\overline{A}}{}^{}`$, the only things left to prove are the following:
* If $`M`$$`{}_{H\mathrm{\#}\overline{A}}{}^{}`$, then $`am=am`$, for all $`aA`$, $`mM`$;
* If $`M_A^H^{}`$, then $`ma=ma`$, for all $`aA`$, $`mM`$.
To prove (1), we compute:
$`am`$ $`=`$ $`(p^1m)(p^2a)`$
$`=`$ $`q^1[(S(q^2)p^2a)(p^1m)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.9})}{=}}`$ $`q^1p_1^1[(S(p_2^1)S(q^2)p^2a)m]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.38})}{=}}`$ $`am,q.e.d.`$
To prove (2), we compute:
$`ma`$ $`=`$ $`q^1[(S(q^2)a)m]`$
$`=`$ $`q^1[(p^1m)(p^2S(q^2)a)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.11})}{=}}`$ $`(q_1^1p^1m)(q_2^1p^2S(q^2)a)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.37})}{=}}`$ $`ma,`$
and the proof is finished. โ
We will need the description of left modules over a two-sided smash product.
###### Definition 9.5.
Let $`H`$ be a quasi-bialgebra, $`A`$ a left $`H`$-module algebra and $`B`$ a right $`H`$-module algebra. Define the category $`{}_{A,H,B}{}^{}`$ as follows: an object in this category is a left $`H`$-module $`M`$, with action denoted by $`hmhm`$, and we have left weak actions of $`A`$ and $`B`$ on $`M`$, denoted by $`amam`$ and $`bmbm`$, such that:
* $`M`$$`{}_{A\mathrm{\#}H}{}^{}`$, that is the relations (9.3) and (9.4) hold;
* $`M`$$`{}_{H\mathrm{\#}B}{}^{}`$, that is the relations (9.6) and (9.7) hold;
* the following compatibility condition holds:
(9.14) $`b(am)=(y^1a)[y^2((by^3)m)],`$
for all $`aA`$, $`bB`$, $`mM`$. The morphisms in this category are the $`H`$-linear maps compatible with the two weak actions.
###### Proposition 9.6.
If $`H`$, $`A`$, $`B`$ are as above, then $`{}_{A\mathrm{\#}H\mathrm{\#}B}{}^{}{}_{A,H,B}{}^{}`$, the isomorphism being given as follows:
* If $`M{}_{A\mathrm{\#}H\mathrm{\#}B}{}^{}`$, define $`am=(a\mathrm{\#}1\mathrm{\#}1)m`$, $`hm=(1\mathrm{\#}h\mathrm{\#}1)m`$, $`bm=(1\mathrm{\#}1\mathrm{\#}b)m`$.
* Conversely, if $`M`$$`{}_{A,H,B}{}^{}`$, define $`(a\mathrm{\#}h\mathrm{\#}b)m=a(h(bm))`$.
###### Proof.
Straightforward computation, using the formula for the multiplication in $`A\mathrm{\#}H\mathrm{\#}B`$. Let us point out how the condition (9.14) occurs:
$`b(am)`$ $`=`$ $`(1\mathrm{\#}1\mathrm{\#}b)((a\mathrm{\#}1\mathrm{\#}1)m)`$
$`=`$ $`[(1\mathrm{\#}1\mathrm{\#}b)(a\mathrm{\#}1\mathrm{\#}1)]m`$
$`=`$ $`(y^1a\mathrm{\#}y^2\mathrm{\#}by^3)m`$
$`=`$ $`(y^1a)(y^2((by^3)m)),`$
which is exactly (9.14). โ
Let $`H`$ be a finite dimensional quasi-bialgebra and $`A`$, $`D`$ two left $`H`$-module algebras. It is obvious that $`{}_{A}{}^{}_{}^{H^{}}`$ coincides with the category of left $`A`$-modules within the monoidal category $`{}_{H}{}^{}`$, and similarly $`_D^H^{}`$ coincides with the category of right $`D`$-modules within $`{}_{H}{}^{}`$. Hence, we can introduce the following new category:
###### Definition 9.7.
If $`H`$, $`A`$, $`D`$ are as above, define $`{}_{A}{}^{}_{D}^{H^{}}`$ as the category of $`AD`$-bimodules within the monoidal category $`{}_{H}{}^{}`$, that is $`M`$$`{}_{A}{}^{}_{D}^{H^{}}`$ if and only if $`M`$$`{}_{A}{}^{}_{}^{H^{}}`$, $`M`$$`_D^H^{}`$ and the following relation holds:
(9.15) $`(am)d=(X^1a)[(X^2m)(X^3d)],`$
for all $`aA`$, $`mM`$, $`dD`$, where $`amam`$ and $`mdmd`$ are the weak actions.
###### Proposition 9.8.
Let $`H`$ be a finite dimensional quasi-Hopf algebra and $`A`$, $`D`$ two left $`H`$-module algebras. Then we have an isomorphism of categories $`{}_{A}{}^{}_{D}^{H^{}}{}_{A\mathrm{\#}H\mathrm{\#}\overline{D}}{}^{}`$, where $`\overline{D}`$ is the right $`H`$-module algebra as in Proposition 9.1.
###### Proof.
Since $`{}_{A}{}^{}_{}^{H^{}}{}_{A\mathrm{\#}H}{}^{}`$ and $`_D^H^{}{}_{H\mathrm{\#}\overline{D}}{}^{}`$, the only thing left to prove is that the compatibility (9.14) in $`{}_{A,H,\overline{D}}{}^{}`$ is equivalent to the compatibility (9.15) in $`{}_{A}{}^{}_{D}^{H^{}}`$. Let us first note the following easy consequences of (2.3), (2.6):
(9.16) $`X^1p_1^1X^2p_2^1X^3p^2=y^1y_1^2p^1y_2^2p^2S(y^3),`$
(9.17) $`q_1^1y^1q_2^1y^2S(q^2y^3)=X^1q^1X_1^2S(q^2X_2^2)X^3,`$
where $`p_R=p^1p^2`$ and $`q_R=q^1q^2`$ are the elements given by (2.34) for $`๐=H`$.
Let now $`M{}_{A}{}^{}_{D}^{H^{}}`$, with right $`D`$-action on $`M`$ denoted by $`mdmd`$. Then, by Proposition 9.4, the weak left $`\overline{D}`$-action on $`M`$ is given by $`dm=(p^1m)(p^2d)`$. We check (9.14); we compute:
$`d(am)`$ $`=`$ $`(p^1(am))(p^2d)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.4})}{=}}`$ $`[(p_1^1a)(p_2^1m)](p^2d)`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.15})}{=}}`$ $`(X^1p_1^1a)[(X^2p_2^1m)(X^3p^2d)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.16})}{=}}`$ $`(y^1a)[(y_1^2p^1m)(y_2^2p^2S(y^3)d)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.2})}{=}}`$ $`(y^1a)[y^2((p^1m)(p^2S(y^3)d))]`$
$`=`$ $`(y^1a)[y^2((S(y^3)d)m)]`$
$`=`$ $`(y^1a)[y^2((dy^3)m)],q.e.d.`$
Conversely, assume that $`M{}_{A\mathrm{\#}H\mathrm{\#}\overline{D}}{}^{}`$, and denote the actions of $`A`$, $`H`$, $`\overline{D}`$ on $`M`$ by $`am`$, $`hm`$, $`dm`$ respectively. Then, by Proposition 9.4, the right $`D`$-action on $`M`$ is given by $`md=q^1((S(q^2)d)m)`$. To check (9.15), we compute:
$`(am)d`$ $`=`$ $`q^1[(S(q^2)d)(am)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.14})}{=}}`$ $`q^1[(y^1a)(y^2((S(q^2)dy^3)m))]`$
$`=`$ $`q^1[(y^1a)(y^2((S(q^2y^3)d)m))]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.4})}{=}}`$ $`(q_1^1y^1a)[q_2^1y^2((S(q^2y^3)d)m)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.17})}{=}}`$ $`(X^1a)[q^1X_1^2((S(q^2X_2^2)X^3d)m)]`$
$`=`$ $`(X^1a)[q^1X_1^2((S(q^2)X^3dX_2^2)m)]`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{9.7})}{=}}`$ $`(X^1a)[q^1((S(q^2)X^3d)(X^2m))]`$
$`=`$ $`(X^1a)[(X^2m)(X^3d)],q.e.d.`$
and the proof is finished. โ
Let $`H`$ be a finite dimensional quasi-bialgebra and $`๐`$, $`๐`$ two $`H`$-bimodule algebras. Define the category $`{}_{๐}{}^{H^{}}_{๐}^{H^{}}`$ as the category of $`๐๐`$-bimodules within the monoidal category $`{}_{H}{}^{}_{H}^{}`$. By regarding $`๐`$ and $`๐`$ as left module algebras over $`HH^{op}`$, it is easy to see that $`{}_{๐}{}^{H^{}}_{๐}^{H^{}}{}_{๐}{}^{}_{๐}^{(HH^{op})^{}}`$. Hence, as a consequence of Proposition 9.8, we finally obtain:
###### Theorem 9.9.
If $`H`$ is a finite dimensional quasi-Hopf algebra and $`๐`$, $`๐`$ are two $`H`$-bimodule algebras, then we have an isomorphism of categories $`{}_{๐}{}^{H^{}}_{๐}^{H^{}}{}_{๐\mathrm{\#}(HH^{op})\mathrm{\#}\overline{๐}}{}^{}`$. In particular, we have $`{}_{H^{}}{}^{H^{}}_{H^{}}^{H^{}}{}_{H^{}\mathrm{\#}(HH^{op})\mathrm{\#}\overline{H^{}}}{}^{}`$.
## 10. Yetter-Drinfeld modules as modules over a generalized diagonal crossed product
If $`H`$ is a quasi-bialgebra, then the category of $`(H,H)`$-bimodules, $`{}_{H}{}^{}_{H}^{}`$, is monoidal. The associativity constraints are given by (2.48). A coalgebra in the category of $`(H,H)`$-bimodules will be called an $`H`$-bimodule coalgebra. More precisely, an $`H`$-bimodule coalgebra $`C`$ is an $`(H,H)`$-bimodule (denote the actions by $`hc`$ and $`ch`$) with a comultiplication $`\underset{ยฏ}{\mathrm{\Delta }}:CCC`$ and a counit $`\underset{ยฏ}{\epsilon }:Ck`$ satisfying the following relations, for all $`cC`$ and $`hH`$:
(10.1) $`\mathrm{\Phi }(\underset{ยฏ}{\mathrm{\Delta }}id)(\underset{ยฏ}{\mathrm{\Delta }}(c))\mathrm{\Phi }^1=(id\underset{ยฏ}{\mathrm{\Delta }})(\underset{ยฏ}{\mathrm{\Delta }}(c)),`$
(10.2) $`\underset{ยฏ}{\mathrm{\Delta }}(hc)=h_1c_{\underset{ยฏ}{1}}h_2c_{\underset{ยฏ}{2}},\underset{ยฏ}{\mathrm{\Delta }}(ch)=c_{\underset{ยฏ}{1}}h_1c_{\underset{ยฏ}{2}}h_2,`$
(10.3) $`\underset{ยฏ}{\epsilon }(hc)=\epsilon (h)\underset{ยฏ}{\epsilon }(c),\underset{ยฏ}{\epsilon }(ch)=\underset{ยฏ}{\epsilon }(c)\epsilon (h),`$
where we used the Sweedler-type notation $`\underset{ยฏ}{\mathrm{\Delta }}(c)=c_{\underset{ยฏ}{1}}c_{\underset{ยฏ}{2}}`$. An example of an $`H`$-bimodule coalgebra is $`H`$ itself.
Our next definition extends the definition of Yetter-Drinfeld modules from .
###### Definition 10.1.
Let $`H`$ be a quasi-bialgebra, $`C`$ an $`H`$-bimodule coalgebra and $`๐ธ`$ an $`H`$-bicomodule algebra. A left-right Yetter-Drinfeld module is a $`k`$-vector space $`M`$ with the following additional structure:
* $`M`$ is a left $`๐ธ`$-module; we write $``$ for the left $`๐ธ`$-action;
* we have a $`k`$-linear map $`\rho _M:MMC`$, $`\rho _M(m)=m_{(0)}m_{(1)}`$, called the right $`C`$-coaction on $`M`$, such that for all $`mM`$, $`\underset{ยฏ}{\epsilon }(m_{(1)})m_{(0)}=m`$ and
$`(\theta ^2m_{(0)})_{(0)}(\theta ^2m_{(0)})_{(1)}\theta ^1\theta ^3m_{(1)}`$
(10.4) $`=\stackrel{~}{x}_\rho ^1(\stackrel{~}{x}_\lambda ^3m)_{(0)}\stackrel{~}{x}_\rho ^2(\stackrel{~}{x}_\lambda ^3m)_{(1)_{\underset{ยฏ}{1}}}\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\rho ^3(\stackrel{~}{x}_\lambda ^3m)_{(1)_{\underset{ยฏ}{2}}}\stackrel{~}{x}_\lambda ^2,`$
* the following compatibility relation holds:
(10.5)
$$u_0m_{(0)}u_1m_{(1)}=(u_{[0]}m)_{(0)}(u_{[0]}m)_{(1)}u_{[1]},$$
for all $`u๐ธ`$, $`mM`$. $`{}_{๐ธ}{}^{}๐ดD(H)^C`$ will be the category of left-right Yetter-Drinfeld modules and maps preserving the actions by $`๐ธ`$ and the coactions by $`C`$.
Let $`H`$ be a quasi-bialgebra, $`๐ธ`$ an $`H`$-bicomodule algebra and $`C`$ an $`H`$-bimodule coalgebra. Let us call the threetuple $`(H,๐ธ,C)`$ a Yetter-Drinfeld datum. We note that, for an arbitrary $`H`$-bimodule coalgebra $`C`$, the linear dual space of $`C`$, $`C^{}`$, is an $`H`$-bimodule algebra. The multiplication of $`C^{}`$ is the convolution, that is $`(c^{}d^{})(c)=c^{}(c_{\underset{ยฏ}{1}})d^{}(c_{\underset{ยฏ}{2}})`$, the unit is $`\underset{ยฏ}{\epsilon }`$ and the left and right $`H`$-module structures are given by $`(hc^{}h{}_{}{}^{})(c)=c^{}(h{}_{}{}^{}ch)`$, for all $`h,h^{}H`$, $`c^{},d^{}C^{}`$, $`cC`$.
In the rest of this section we establish that if $`H`$ is a quasi-Hopf algebra and $`C`$ is finite dimensional then the category $`{}_{๐ธ}{}^{}๐ดD(H)^C`$ is isomorphic to the category of left $`C^{}๐ธ`$-modules, $`{}_{C^{}๐ธ}{}^{}`$. First some lemmas.
###### Lemma 10.2.
Let $`H`$ be a quasi-Hopf algebra and $`(H,๐ธ,C)`$ a Yetter-Drinfeld datum. We have a functor $`F:{}_{๐ธ}{}^{}๐ดD(H)^C{}_{C^{}๐ธ}{}^{}`$, given by F(M)=M as $`k`$-module, with the $`C^{}๐ธ`$-module structure defined by
(10.6)
$$(c^{}u)m:=c^{},\stackrel{~}{q}_\rho ^2(um)_{(1)}\stackrel{~}{q}_\rho ^1(um)_{(0)},$$
for all $`c^{}C^{}`$, $`u๐ธ`$ and $`mM`$, where $`\stackrel{~}{q}_\rho =\stackrel{~}{q}_\rho ^1\stackrel{~}{q}_\rho ^2`$ is the element defined in (2.34). $`F`$ transforms a morphism to itself.
###### Proof.
Let $`\stackrel{~}{Q}_\rho ^1\stackrel{~}{Q}_\rho ^2`$ be another copy of $`\stackrel{~}{q}_\rho `$. For all $`c^{},d^{}C^{}`$, $`u,u^{}๐ธ`$ and $`mM`$ we compute:
$`[(c^{}u)(d^{}u^{})]m`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`[(\mathrm{\Omega }^1c^{}\mathrm{\Omega }^5)(\mathrm{\Omega }^2u_{0_{[1]}}d^{}S^1(u_1)\mathrm{\Omega }^4)\mathrm{\Omega }^3u_{0_{[0]}}u^{}]m`$
$`=`$ $`d^{},S^1(u_1)\mathrm{\Omega }^4(\stackrel{~}{q}_\rho ^2)_2(\mathrm{\Omega }^3u_{0_{[0]}}u{}_{}{}^{}m)_{(1)_{\underset{ยฏ}{2}}}\mathrm{\Omega }^2u_{0_{[1]}}`$
$`c^{},\mathrm{\Omega }^5(\stackrel{~}{q}_\rho ^2)_1(\mathrm{\Omega }^3u_{0_{[0]}}u{}_{}{}^{}m)_{(1)_{\underset{ยฏ}{1}}}\mathrm{\Omega }^1\stackrel{~}{q}_\rho ^1(\mathrm{\Omega }^3u_{0_{[0]}}u^{}m)_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.15})}{=}}`$ $`d^{},S^1(f^1\stackrel{~}{X}_\rho ^2\theta ^3u_1)(\stackrel{~}{q}_\rho ^2)_2((\stackrel{~}{X}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(1)_{\underset{ยฏ}{2}}}(\stackrel{~}{X}_\rho ^1)_{[1]_{\underset{ยฏ}{2}}}`$
$`\times \stackrel{~}{x}_\lambda ^2\theta _{[1]}^2u_{0_{[1]}}c^{},S^1(f^2\stackrel{~}{X}_\rho ^3)(\stackrel{~}{q}_\rho )_1((\stackrel{~}{X}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(1)_{\underset{ยฏ}{1}}}`$
$`\stackrel{~}{x}_\lambda ^1\theta ^1\stackrel{~}{q}^1_\rho ((\stackrel{~}{X}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.5},\text{2.40})}{=}}`$ $`d^{},S^1(\theta ^3u_1)\stackrel{~}{Q}_\rho ^2\stackrel{~}{x}_\rho ^3(\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(1)_{\underset{ยฏ}{2}}}\stackrel{~}{x}_\lambda ^2\theta _{[1]}^2u_{0_{[1]}}`$
$`c^{},\stackrel{~}{q}_\rho ^2(\stackrel{~}{Q}_\rho ^1)_1\stackrel{~}{x}_\rho ^2(\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(1)_{\underset{ยฏ}{1}}}\stackrel{~}{x}_\lambda ^1\theta ^1`$
$`\stackrel{~}{q}_\rho ^1(\stackrel{~}{Q}_\rho ^1)_0\stackrel{~}{x}_\rho ^1(\stackrel{~}{x}_\lambda ^3\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.4})}{=}}`$ $`d^{},S^1(\theta ^3u_1)\stackrel{~}{Q}_\rho ^2\overline{\theta }^3(\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(1)}\theta _{[1]}^2u_{0_{[1]}}`$
$`c^{},\stackrel{~}{q}_\rho ^2(\stackrel{~}{Q}_\rho ^1)_1[\overline{\theta }^2(\theta _{[0]}^2u_{0_{[0]}}u^{}m)_{(0)}]_{(1)}\overline{\theta }^1\theta ^1`$
$`\stackrel{~}{q}_\rho ^1(\stackrel{~}{Q}_\rho ^1)_0[\overline{\theta }^2(\theta _{[0]}^2u_{0_{[0]}}u{}_{}{}^{}m)_{(0)}]_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.5},\text{2.45})}{=}}`$ $`d^{},S^1(\alpha \stackrel{~}{X}_\rho ^3\theta ^3u_1)\stackrel{~}{X}_\rho ^2\overline{\theta }^3\theta _1^2u_{0,1}(u{}_{}{}^{}m)_{(1)}`$
$`c^{},\stackrel{~}{q}_\rho ^2[(\stackrel{~}{X}_\rho ^1)_{[0]}\overline{\theta }^2\theta _0^2u_{0,0}(u{}_{}{}^{}m)_{(0)}]_{(1)}(\stackrel{~}{X}_\rho ^1)_{[1]}\overline{\theta }^1\theta ^1`$
$`\stackrel{~}{q}_\rho ^1[(\stackrel{~}{X}_\rho ^1)_{[0]}\overline{\theta }^2\theta _0^2u_{0,0}(u{}_{}{}^{}m)_{(0)}]_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.45},\text{2.26})}{=}}`$ $`d^{},S^1(\alpha \theta _2^3u_{1_2}\stackrel{~}{X}_\rho ^3)\theta _1^3u_{1_1}\stackrel{~}{X}_\rho ^2(u{}_{}{}^{}m)_{(1)}`$
$`c^{},\stackrel{~}{q}_\rho ^2[\theta ^2u_0\stackrel{~}{X}_\rho ^1(u{}_{}{}^{}m)_{(0)}]_{(1)}\theta ^1\stackrel{~}{q}_\rho ^1[\theta ^2u_0\stackrel{~}{X}_\rho ^1(u{}_{}{}^{}m)_{(0)}]_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.6},\text{2.34})}{=}}`$ $`c^{},\stackrel{~}{q}_\rho ^2[u\stackrel{~}{Q}_\rho ^1(u{}_{}{}^{}m)_{(0)}]_{(1)}d^{},\stackrel{~}{Q}_\rho ^2(u{}_{}{}^{}m)_{(1)}`$
$`\stackrel{~}{q}_\rho ^1[u\stackrel{~}{Q}_\rho ^1(u{}_{}{}^{}m)_{(0)}]_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.6})}{=}}`$ $`d^{},\stackrel{~}{Q}_\rho ^2(u{}_{}{}^{}m)_{(1)}(c^{}u)[\stackrel{~}{Q}_\rho ^1(u{}_{}{}^{}m)_{(0)}]=(c^{}u)[(d^{}u{}_{}{}^{})m],`$
as needed. It is not hard to see that $`(\underset{ยฏ}{\epsilon }1_๐ธ)m=m`$ for all $`mM`$, so $`M`$ is a left $`C^{}๐ธ`$-module. The fact that a morphism in $`{}_{๐ธ}{}^{}๐ดD(H)^C`$ becomes a morphism in $`{}_{C^{}๐ธ}{}^{}`$ can be proved more easily, we leave the details to the reader. โ
###### Lemma 10.3.
Let $`H`$ be a quasi-Hopf algebra and $`(H,๐ธ,C)`$ a Yetter-Drinfeld datum and assume $`C`$ is finite dimensional. We have a functor $`G:{}_{C^{}๐ธ}{}^{}{}_{๐ธ}{}^{}๐ดD(H)^C`$, given by $`G(M)=M`$ as $`k`$-module, with structure maps defined by
(10.7) $`um=(\underset{ยฏ}{\epsilon }u)m,`$
(10.8) $`\rho _M:MMC,\text{ }\rho _M(m)={\displaystyle \underset{i=1}{\overset{n}{}}}(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})mS^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]},`$
for $`mM`$ and $`u๐ธ`$. Here $`\stackrel{~}{p}_\rho =\stackrel{~}{p}_\rho ^1\stackrel{~}{p}_\rho ^2`$ is the element defined in (2.34), $`\{c_i\}_{i=\overline{1,n}}`$ is a basis of $`C`$ and $`\{c^i\}_{i=\overline{1,n}}`$ is the corresponding dual basis of $`C^{}`$. $`G`$ transforms a morphism to itself.
###### Proof.
The most difficult part of the proof is to show that $`G(M)`$ satisfies the relations (10.4) and (10.5). It is then straightforward to show that a map in $`{}_{C^{}๐ธ}{}^{}`$ is also a map in $`{}_{๐ธ}{}^{}๐ดD(H)^C`$, and that $`G`$ is a functor.
It is not hard to see that (2.45), (2.6) and (2.46) imply
(10.9)
$$\overline{\theta }^1\theta ^1\overline{\theta }^2\theta _0^2\stackrel{~}{p}_\rho ^1\overline{\theta }^3\theta _1^2\stackrel{~}{p}_\rho ^2S(\theta ^3)=(\stackrel{~}{p}_\rho ^1)_{[1]}(\stackrel{~}{p}_\rho ^1)_{[0]}\stackrel{~}{p}_\rho ^2.$$
Write $`\stackrel{~}{p}_\rho =\stackrel{~}{p}_\rho ^1\stackrel{~}{p}_\rho ^2=\stackrel{~}{P}_\rho ^1\stackrel{~}{P}_\rho ^2`$. For all $`mM`$ we compute:
$`(\theta ^2m_{(0)})_{(0)}(\theta ^2m_{(0)})_{(1)}\theta ^1\theta ^3m_{(1)}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}((\underset{ยฏ}{\epsilon }\theta ^2)(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})m)_{(0)}((\underset{ยฏ}{\epsilon }\theta ^2)(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})m)_{(1)}\theta ^1`$
$`\theta ^3S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21},\text{10.8})}{=}}`$ $`{\displaystyle \underset{i,j=1}{\overset{n}{}}}(c^j(\stackrel{~}{P}_\rho ^1)_{[0]})(c^i(\theta _0^2\stackrel{~}{p}_\rho ^1)_{[0]})mS^1(\stackrel{~}{P}_\rho ^2)c_j(\stackrel{~}{P}_\rho ^1)_{[1]}\theta ^1`$
$`\theta ^3S^1(\theta _1^2\stackrel{~}{p}_\rho ^2)c_i(\theta _0^2\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21},\text{3.15})}{=}}`$ $`{\displaystyle \underset{i,j=1}{\overset{n}{}}}[c^jc^i(\stackrel{~}{X}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3(\overline{\theta }^2(\stackrel{~}{P}_\rho ^1)_{[0]_0}\theta _0^2\stackrel{~}{p}_\rho ^1)_{[0]}]mS^1(f^2\stackrel{~}{X}_\rho ^3\stackrel{~}{P}_\rho ^2)`$
$`c_j(\stackrel{~}{X}_\rho ^1)_{[1]_1}\stackrel{~}{x}_\lambda ^1\overline{\theta }^1(\stackrel{~}{P}_\rho ^1)_{[1]}\theta ^1\theta ^3S^1(f^1\stackrel{~}{X}_\rho ^2\overline{\theta }^3(\stackrel{~}{P}_\rho ^1)_{[0]_1}\theta _1^2\stackrel{~}{p}_\rho ^2)c_i`$
$`(\stackrel{~}{X}_\rho ^1)_{[1]_2}\stackrel{~}{x}_\lambda ^2(\overline{\theta }^2(\stackrel{~}{P}_\rho ^1)_{[0]_0}\theta _0^2\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.43},\text{10.9},\text{2.30})}{=}}`$ $`{\displaystyle \underset{i,j=1}{\overset{n}{}}}[c^jc^i(\stackrel{~}{X}_\rho ^1(\stackrel{~}{P}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3]mS^1(f^2\stackrel{~}{X}_\rho ^3\stackrel{~}{P}_\rho ^2)c_j`$
$`(\stackrel{~}{X}_\rho ^1(\stackrel{~}{P}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]_1}\stackrel{~}{x}_\lambda ^1S^1(f^1\stackrel{~}{X}_\rho ^2(\stackrel{~}{P}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{X}_\rho ^1(\stackrel{~}{P}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]_2}\stackrel{~}{x}_\lambda ^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.39})}{=}}`$ $`{\displaystyle \underset{i,j=1}{\overset{n}{}}}[c^jc^i((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3]m\stackrel{~}{x}_\rho ^2S^1(f^2((\stackrel{~}{x}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)_2g^2)c_j`$
$`((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]_1}\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\rho ^3S^1(f^1((\stackrel{~}{x}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)_1g^1)c_i((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]_2}\stackrel{~}{x}_\lambda ^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.11},\text{10.2})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}[c^i((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3]m\stackrel{~}{x}_\rho ^2(S^1((\stackrel{~}{x}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)c_i`$
$`((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{1}}\stackrel{~}{x}^1_\lambda \stackrel{~}{x}^3_\rho (S^1((\stackrel{~}{x}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)c_i((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{2}}\stackrel{~}{x}^2_\lambda `$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}[(\stackrel{~}{x}_\rho ^1)_{0_{[1]}}c^iS^1((\stackrel{~}{x}_\rho ^1)_1)((\stackrel{~}{x}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[0]}\stackrel{~}{x}_\lambda ^3]m`$
$`\stackrel{~}{x}_\rho ^2(S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{1}}\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\rho ^3(S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{2}}\stackrel{~}{x}_\lambda ^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}[(\underset{ยฏ}{\epsilon }\stackrel{~}{x}_\rho ^1)(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})(\underset{ยฏ}{\epsilon }\stackrel{~}{x}_\lambda ^3)]m`$
$`\stackrel{~}{x}_\rho ^2(S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{1}}\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\rho ^3(S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]})_{\underset{ยฏ}{2}}\stackrel{~}{x}_\lambda ^2`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.7},\text{10.8})}{=}}`$ $`\theta _\rho ^1(\stackrel{~}{x}_\lambda ^3m)_{(0)}\stackrel{~}{x}_\rho ^2(\stackrel{~}{x}_\lambda ^3m)_{(1)_{\underset{ยฏ}{1}}}\stackrel{~}{x}_\lambda ^1\stackrel{~}{x}_\rho ^3(\stackrel{~}{x}_\lambda ^3m)_{(1)_{\underset{ยฏ}{2}}}\stackrel{~}{x}_\lambda ^2.`$
Similarly, we compute:
$`u_0m_{(0)}u_1m_{(1)}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(\underset{ยฏ}{\epsilon }u_0)(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})mu_1S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(u_{0,0_{[1]}}c^iS^1(u_{0,1})u_{0,0_{[0]}}(\stackrel{~}{p}_\rho ^1)_{[0]})m`$
$`u_1S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(c^i(u_{0,0}\stackrel{~}{p}_\rho ^1)_{[0]})mu_1S^1(u_{0,1}\stackrel{~}{p}_\rho ^2)c_i(u_{0,0}\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.35})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(c^i(\stackrel{~}{p}_\rho ^1u)_{[0]})mS^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1u)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})(\underset{ยฏ}{\epsilon }u_{[0]})mS^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}u_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.8})}{=}}`$ $`(u_{[0]}m)_{(0)}(u_{[0]}m)_{(1)}u_{[1]},`$
for all $`u๐ธ`$ and $`mM`$, and this finishes the proof. โ
The next result generalizes \[8, Proposition 3.12\], which is recovered by taking $`C=๐ธ=H`$.
###### Theorem 10.4.
Let $`H`$ be a quasi-Hopf algebra and $`(H,๐ธ,C)`$ a Yetter-Drinfeld datum, assuming $`C`$ to be finite dimensional. Then the categories $`{}_{๐ธ}{}^{}๐ดD(H)^C`$ and $`{}_{C^{}๐ธ}{}^{}`$ are isomorphic.
###### Proof.
We have to verify that the functors $`F`$ and $`G`$ defined in Lemmas 10.2 and 10.3 are inverse to each other. Let $`M{}_{๐ธ}{}^{}๐ดD(H)^C`$. The structures on $`G(F(M))`$ (using first Lemma 10.2 and then Lemma 10.3) are denoted by $`^{}`$ and $`\rho _{M}^{}{}_{}{}^{}`$. For any $`u๐ธ`$ and $`mM`$ we have that
$$u{}_{}{}^{}m=(\underset{ยฏ}{\epsilon }u)m=\underset{ยฏ}{\epsilon },\stackrel{~}{q}_\rho ^2(um)_{(1)}\stackrel{~}{q}_\rho ^1(um)_{(0)}=um$$
because $`\underset{ยฏ}{\epsilon }(hc)=\epsilon (h)\underset{ยฏ}{\epsilon }(c)`$ and $`\underset{ยฏ}{\epsilon }(m_{(1)})m_{(0)}=m`$ for all $`hH`$, $`cC`$, $`mM`$. We now compute for $`mM`$ that
$`\rho {}_{}{}^{}{}_{M}{}^{}(m)`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})mS^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.6})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}c^i,\stackrel{~}{q}_\rho ^2((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(1)}\stackrel{~}{q}_\rho ^1((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(0)}S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.5})}{=}}`$ $`\stackrel{~}{q}_\rho ^1(\stackrel{~}{p}_\rho ^1)_0m_{(0)}S^1(\stackrel{~}{p}_\rho ^2)\stackrel{~}{q}_\rho ^2(\stackrel{~}{p}_\rho ^1)_1m_{(1)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.38})}{=}}`$ $`m_{(0)}m_{(1)}=\rho _M(m).`$
Conversely, take $`M{}_{C^{}๐ธ}{}^{}`$. We want to show that $`F(G(M))=M`$. If we denote the left $`C^{}๐ธ`$-action on $`F(G(M))`$ by $``$, then, using Lemmas 10.2 and 10.3 we find, for all $`c^{}C^{}`$, $`u๐ธ`$ and $`mM`$:
$`(c^{}u)m`$
$`=`$ $`c^{},\stackrel{~}{q}_\rho ^2(um)_{(1)}\stackrel{~}{q}_\rho ^1(um)_{(0)}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}c^{},\stackrel{~}{q}_\rho ^2S^1(\stackrel{~}{p}_\rho ^2)c_i(\stackrel{~}{p}_\rho ^1)_{[1]}(\underset{ยฏ}{\epsilon }\stackrel{~}{q}_\rho ^1)(c^i(\stackrel{~}{p}_\rho ^1)_{[0]})(\underset{ยฏ}{\epsilon }u)m`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{3.21})}{=}}`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}c^{},\stackrel{~}{q}_\rho ^2S^1((\stackrel{~}{q}_\rho ^1)_1\stackrel{~}{p}_\rho ^2)c_i((\stackrel{~}{q}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[1]}`$
$`(c^i((\stackrel{~}{q}_\rho ^1)_0\stackrel{~}{p}_\rho ^1)_{[0]})(\underset{ยฏ}{\epsilon }u)m`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.37},\text{3.21})}{=}}`$ $`(c^{}1_๐ธ)(\underset{ยฏ}{\epsilon }u)m=(c^{}u)m,`$
and this finishes our proof. โ
There is a relation between the functor $`F`$ from Lemma 10.2 and the map $`\mathrm{\Gamma }`$ as in Proposition 3.8.
###### Proposition 10.5.
Let $`H`$ be a quasi-Hopf algebra, $`(H,๐ธ,C)`$ a Yetter-Drinfeld datum and $`M`$ an object in $`{}_{๐ธ}{}^{}๐ดD(H)^C`$; consider the map $`\mathrm{\Gamma }:C^{}C^{}๐ธ`$ as in Proposition 3.8. Then the left $`C^{}๐ธ`$-module structure on $`M`$ given in Lemma 10.2 and the map $`\mathrm{\Gamma }`$ are related by the formula:
$`\mathrm{\Gamma }(c^{})m=c^{},m_{(1)}m_{(0)},`$
for all $`c^{}C^{}`$ and $`mM`$.
###### Proof.
We compute:
$`\mathrm{\Gamma }(c^{})m`$ $`=`$ $`((\stackrel{~}{p}_\rho ^1)_{[1]}c^{}S^1(\stackrel{~}{p}_\rho ^2)(\stackrel{~}{p}_\rho ^1)_{[0]})m`$
$`=`$ $`(\stackrel{~}{p}_\rho ^1)_{[1]}c^{}S^1(\stackrel{~}{p}_\rho ^2),\stackrel{~}{q}_\rho ^2((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(1)}\stackrel{~}{q}_\rho ^1((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(0)}`$
$`=`$ $`c^{},S^1(\stackrel{~}{p}_\rho ^2)\stackrel{~}{q}_\rho ^2((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(1)}(\stackrel{~}{p}_\rho ^1)_{[1]}\stackrel{~}{q}_\rho ^1((\stackrel{~}{p}_\rho ^1)_{[0]}m)_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{10.5})}{=}}`$ $`c^{},S^1(\stackrel{~}{p}_\rho ^2)\stackrel{~}{q}_\rho ^2(\stackrel{~}{p}_\rho ^1)_1m_{(1)}\stackrel{~}{q}_\rho ^1(\stackrel{~}{p}_\rho ^1)_0m_{(0)}`$
$`{\displaystyle \genfrac{}{}{0pt}{}{(\text{2.38})}{=}}`$ $`c^{},m_{(1)}m_{(0)},`$
finishing the proof. โ
|
warning/0506/gr-qc0506019.html
|
ar5iv
|
text
|
# The world problem: on the computability of the topology of 4-manifolds
## 1 Introduction
A theorem proved by Markov on the non-classifiability of the 4-manifolds implies that, given some comprehensive specification for the topology of a manifold (such as its triangulation, a la Regge calculus, or instructions for constructing it via cutting and gluing simpler spaces) there exists no general algorithm to decide whether the manifold is homeomorphic to some other manifold . The impossibility of classifying the 4-manifolds is a well-known topological result, the proof of which, however, may not be well known in the physics community. It is potentially a result of profound physical implications, as the universe certainly appears to be a manifold of at least four dimensions. The burgeoning quest for the topology of the universe is still in its infancy; Markovโs theorem may ultimately bear upon what can be deduced about it. Already Markovโs theorem impacts certain approaches to quantum gravity. On the basis of this theorem, and consideration of hypothetical quantum superpositions of manifolds, Penrose has heuristically argued that the universe is fundamentally non-computable . As another example, in analogy with Feynmanโs sum over histories approach to quantum mechanics, the Euclidean path integral approach to quantum gravity requires a sum over all possible topologies, with appropriate weighting, in order to calculate expectation values. However, Markovโs theorem implies inherent difficulties in computing such a summation, as it would be impossible to decide whether a particular topology had been counted more than once .
Owing to its theorized physical significance, the computability and tractability of this sum over topologies has received some attention in the literature. Although direct summation of the series is non-computable, it is unknown whether it might nonetheless be deducible by indirect means, perhaps as the computable limit of some sequence; failing in that, it has been implied that the sum can nevertheless be approximated to any desired order of accuracy . However, without a systematic way to proceed, there is no guarantee that such an approximation could be carried through in finite time. To obviate such difficulty, it has been proposed to relax the condition of homeomorphy, when classifying the manifolds, and instead classify them according to a weaker condition, in terms of their triangulation . But such a classification scheme would keep infinite redundancy of physically distinct manifolds in the series and it is not clear how to interpret the resulting sum. More recently, partly sidestepping the issue of computability, deductions have been made about the density of topologies per โnormalized volumeโ โ a geometric quantity โ in the context of a saddle-point approximation to the Euclidean path integral . The above work was motivated by the tantalizing possibility that this sum over topologies might determine the value of the cosmological constant .
Manifold non-classifiability represents a fascinating juxtaposition of theoretical computer science with physics. The intent here is to outline a proof that will establish a correspondence between Turing machines and 4-manifolds such that deciding whether a manifold is homeomorphic to a certain other manifold is tantamount to deciding whether the corresponding Turing machine halts; to the authorโs knowledge this illuminating point has not been explicitly made elsewhere. It is further hoped that the proof sketched here will provide insight into the physical relevance of Markovโs theorem.
This paper is organized as follows. In Section 2, Turing machines, and the unsolvability of the halting problem, are reviewed. In Section 3 it is shown that if the group triviality problem could be solved then the halting problem could be solved. In Section 4 it is shown that if the 4-manifold homeomorphy problem could be solved then the group triviality problem could be solved. These results are discussed in Section 5.
## 2 Turing machines
A Turing machine is a formal idealization of a computer . In its simplest formulation, a Turing machine consists of a linear tape divided into squares onto which symbols have been printed, and a movable head that scans each square one at a time. The sequence of symbols initially printed on the tape can be considered the input of the Turing machine. The head can overwrite the current scanned square, move one square to the right, or move one square to the left, depending on its internal state and its programmed instructions. Let the $`h+1`$ possible states of the machine be denoted by $`q_0,\mathrm{}q_h`$ and the $`k+1`$ possible symbols printed on the tape be denoted by $`s_0,\mathrm{}s_k`$. The instructions followed by the machine can be conceived as a list of if-then statements of the form: โif the current state is $`q_i`$ and the current scanned symbol is $`s_j`$ then \[either move a square or print a symbol\] and change to state $`q_k`$ โ. After updating its state and its current scanned symbol, the machine repeats the process, reviewing the list of if-then statements. This goes on forever or until the machine arrives at a $`(q_i,s_j)`$ pair for which it has no instructions, at which point it halts. Note that, although more properly referred to as a program, by convention the term โTuring machineโ is taken to be synonymous with its hardwired instructions.
Consider, as Turing did, machines designed to output a sequence of symbols, potentially never ending, as the digits of a real number. Its output can be printed on every other square of the tape, while the rest of the squares are reserved for โscratch paperโ. Rather than print the entire sequence continuously, these machines will print only $`j`$ digits, given the integer $`j`$ as an input (i.e., initially printed on some of the tape squares). All such machines, which input an integer and output a digit, can themselves be ordered and numbered by integers. Turing provided a specific way to encode the instructions which uniquely characterize each Turing machine into the digits of a (very large) integer; these integers can then be ordered and renumbered by consecutive integers โ call them $`\tau _i`$.
A Turing machine, which can examine another Turing machine by reviewing the latterโs specifications on tape, cannot in general decide whether an arbitrary Turing machine will complete its computation and halt on a given input, or go into an infinite loop without ever printing any output. This assertion can be proven by contradiction. Assume the existence of a machine algorithm that decides, in a finite number of steps, whether a given machine will halt on a given input. A machine $`\delta `$ can then be constructed which, given an input integer $`n`$, operates as follows. $`\delta `$ initializes a counter $`j`$ to 1, checks to see whether $`\tau _1`$ halts on input 1, and if so increments $`j`$ by 1. $`\delta `$ then checks to see whether each subsequent machine $`\tau _i`$ halts on input $`j`$, in order, incrementing $`j`$ for each halting machine. When $`\tau _i`$ is determined not to halt, $`j`$ remains at the same value and the next machine $`\tau _{i+1}`$ is checked. Finally, $`\delta `$ checks to see whether $`\tau _n`$ halts on input $`j`$, where $`j`$ now equals one plus the number of halting machines up through $`\tau _{n1}`$. If $`\delta `$ decides that $`\tau _n`$ halts, then $`\delta `$ prints the $`j`$th digit computed by $`\tau _n`$ and then halts itself. Otherwise, $`\delta `$ just halts. Note that, in the former case, as part of $`\delta `$โs assigned task, $`\delta `$ must effectively emulate machine $`\tau _n`$. (Turing proved it is possible to design a machine such as $`\delta `$ to emulate any other arbitrary machine $`\tau _i`$ on command.) By assumption, $`\delta `$ can perform all of the above operations in a finite number of steps.
Since $`\delta `$ is essentially a machine that outputs a digit on being input an integer, $`\delta `$ itself ranks among the $`\tau _i`$ machines described previously. Now give $`\delta `$ input $`k`$, such that the $`k`$th halting machine is $`\delta `$. $`\delta `$ will proceed by computing the first digit output from the first halting machine, the first two digits output from the second halting machine, and so forth, up to the first $`k1`$ digits output from the $`(k1)`$th halting machine. In so doing, $`\delta `$ will have computed the first $`k1`$ digits of its own output sequence. Now $`\delta `$ must compute the first $`k`$ digits of the $`k`$th halting machine, itself. According to the algorithm by which $`\delta `$ is defined, $`\delta `$ must recompute the first $`k1`$ digits of its output sequence. Then to compute the $`k`$th digit, $`\delta `$ must recompute the first $`k1`$ digits of its output sequence. And so forth, ad infinitum. We have arrived at a contradiction: the assumption that $`\delta `$ will halt on all input implies that $`\delta `$ will not halt on at least one input.
Alternatively, the unsolvability of the halting problem can be understood using Cantorโs diagonal argument. If one attempts to enumerate all of the sequences computed by halting machines, i.e. put them on a one-to-one correspondence with the integers, one can always use a machine such as $`\delta `$ to construct a sequence not on the list - i.e., $`1\delta (j)`$, if the output digits are binary digits. This would imply that the computable sequences are uncountably infinite and, as there is at least one Turing machine for each such sequence, that Turing machines are also uncountable. However, since Turing machines are finitely specified, they must be countable: a contradiction, proving again that the halting problem is unsolvable.
## 3 Semigroups and groups
A few definitions are in order. A semigroup is a set of elements for which a binary operator has been defined so as to satisfy closure and associativity; equivalently, it is a group in which elements are not required to have inverses. A finitely generated semigroup or group, generally infinite, albeit discrete, and non-Abelian, has a finite alphabet of generators. Its elements can be represented as โwordsโ, i.e. strings โspelled outโ by products of generators. A finitely presented semigroup or group is specified by a finite number of generators and a finite number of relations, where relations are equations between words. The word problem for semigroups or groups is the problem of finding a general algorithm which, by successive application of the relations, can decide whether two arbitrary words are equal (in a finite number of steps).
The following proof of the unsolvability of the semigroup word problem proceeds very much like that of Post but has been modified to connect it more directly with the halting problem. Consider a semigroup $`\mathrm{\Gamma }_\tau `$ with generators $`q_0,q_1,\mathrm{}q_h`$, $`s_0,s_1,\mathrm{}s_k`$, and $`l`$. Each $`q_i`$ will represent a state of a Turing machine, each $`s_j`$ will represent a symbol on the tape, $`s_0`$ will represent a blank, and $`l`$ will represent the left and right bounds of the string of symbols input to the machine.
All of the operations of a Turing machine $`\tau `$ can then be represented by relations in $`\mathrm{\Gamma }_\tau `$. The action of printing over symbol $`s_b`$ with symbol $`s_d`$ can be represented by the following relation,
$$q_as_b=q_cs_d$$
(1)
where $`a`$ and $`c`$ have some specific values between 1 and $`h`$, and likewise $`b`$ and $`d`$ between 0 and $`k`$. In accordance with Turingโs convention, all machine actions will be accompanied by a simultaneous change of state. Similarly, the action of moving to the left one space can be represented by the following $`h+2`$ relations.
$`s_iq_as_b=q_cs_is_b,i=0,1,\mathrm{},h`$ (2)
$`lq_as_b=lq_cs_0s_b`$ (3)
And the action of moving to the right one space can be represented by the following $`h+2`$ relations.
$`q_as_bs_i=s_bq_cs_i,i=0,1,\mathrm{},h`$ (4)
$`q_as_bl=s_bq_cs_0l`$ (5)
This completes the semigroup โemulationโ of a Turing machine.
For the purpose of investigating the halting problem, Iโm going to introduce two new generators with the unconventional notation $``$ and $``$ , for reasons that will soon become clear. For every $`q_as_b`$ pair that does not appear in the left hand side of equations (1-5), add the relation:
$$q_as_b=s_b$$
(6)
Now add the following $`2h+3`$ relations:
$`s_i=,i=0,1,\mathrm{},h`$ (7)
$`l=l`$ (8)
$`s_i=,i=0,1,\mathrm{},h`$ (9)
In effect, $``$ devours all symbols to its left. If it comes to the end-marker $`l`$, it mutates into $``$ . $``$ devours all symbols to its right.
The outcome is that if any word $`\omega _\iota `$ corresponds to an input $`\iota `$ on which the associated Turing machine halts, then it can be shown to be equivalent, by repeated application of the above relations (7-9), to the word $`ll`$. If a word does not correspond to an input on which the associated Turing machine halts, then it is not equivalent to the word $`ll`$. By convention, $`q_0`$ is reserved for the halting state, so the relation $`ll=q_0`$ might be added - then $`\omega _\iota =q_0`$ in $`\mathrm{\Gamma }_\tau `$ if and only if $`\tau `$ halts on input $`\iota `$. An algorithm that could solve the word problem for semigroups, therefore, could solve the halting problem for Turing machines.
The above result for semigroups has direct implications for groups. For each finitely presented semigroup $`\mathrm{\Gamma }_\tau `$ described above, there is a prescription for constructing a finitely presented group $`G_\tau ^{}`$ such that for every generator and relation in $`\mathrm{\Gamma }_\tau `$ there is a corresponding generator and relation in $`G_\tau ^{}`$, and the following theorem holds: There exist words $`u_\iota `$ and $`v_\iota `$ in the finitely presented group $`G_\tau ^{}`$ that are equal if and only if $`\omega _\iota =q_0`$ in the finitely presented semigroup $`\mathrm{\Gamma }_\tau `$ . Equivalently, $`w_\iota u_\iota v_\iota ^1=1`$ in $`G_\tau ^{}`$ if and only if $`\omega _\iota =q_0`$ in $`\mathrm{\Gamma }_\tau `$. Further, for each finitely presented group $`G_\tau ^{}`$ and each word $`w_\iota `$ in $`G_\tau ^{}`$ there is a prescription for constructing a finitely presented group $`G_\tau (w_\iota )`$ such that for every generator and relation in $`G_\tau ^{}`$ there is a corresponding generator and relation in $`G_\tau (w_\iota )`$ and the following theorem holds: $`G_\tau (w_\iota )`$ is trivial, i.e. contains only the identity element, if and only if $`w_\iota =1`$ in $`G_\tau ^{}`$ . It follows that the triviality of finitely presented groups is algorithmically undecidable.
## 4 Manifolds
Each element of the fundamental group of a manifold represents an equivalence class of closed paths in the manifold that can be continuously deformed into one another, i.e., a homotopy class of closed paths. As an example, a trivial element in a fundamental group represents a class of paths that can be contracted to a point, and a trivial fundamental group implies a simply connected manifold. As another example, the infinite cyclic group, which can be finitely presented by one generator and no relations, is the fundamental group of a hypersphere with one arcwise connected handle: each element of the group, equal to the generator raised to some power $`p`$, corresponds with the homotopy class of paths that wind about the handle $`p`$ times (and negative powers will be said to correspond to counterwindings, described below). It will be shown that for any given finitely presented group, a manifold can always be constructed for which the given group is fundamental. The prescription can be summarized as attaching to a hypersphere a handle for each generator of the group, followed by further surgery to accommodate each relation.
The following construction is homeomorphic to that of Markov, but the method of construction has been streamlined for pedagogical purposes. Consider an arbitrary finitely presented group of the form
$$G=\{g_1,\mathrm{},g_m|r_1,\mathrm{},r_n\}$$
(10)
where each $`r_i`$ is a word representing a relation of the form $`r_i=1`$ and is called a relator. Beginning with the 4-sphere, $`S^4`$, for each generator $`g_i`$ attach a handle of the form $`H_i=S^3\times [1,+1]`$. Each such attachment is performed by removing from $`S^4`$ two non-intersecting, open 4-balls and identifying the resulting 3-spherical boundaries with the ends of $`H_i`$. Calling the former $`S^4`$ region $`A`$, the attachments are subject to the conditions that no two handles intersect, and the intersection of each handle with $`A`$ is a union of two 3-spheres: $`H_iH_j=0`$, $`ij`$, $`AH_i=S^3\times \{1,+1\}`$. In this manner a manifold can be handily constructed for each free fundamental group of the form $`\{g_1,\mathrm{},g_n|\}`$. To understand this, note that the construction thus far is homeomorphic to the connected sum of $`m`$ copies of $`S^3\times S^1`$, then use the fact that the fundamental group of the cross product of manifolds is the free product of the fundamental groups of the manifolds, while the fundamental group of the connected sum of manifolds is the direct product of the fundamental groups of the manifolds.
An arbitrary word can be represented by a closed path in the above construction as follows. Consider a path that begins at some point inside $`A`$. Reading the word from left to right, represent each generator $`g_i`$ of positive power $`p`$ by a path that enters its associated handle $`H_i`$ at $`S^2\times \{1\}`$ , then exits $`H_i`$ at $`S^2\times \{+1\}`$ , then circles back around and repeats $`p1`$ times. Represent negative powers $`p`$ the same way but switch $`S^2\times \{1\}`$ and $`S^2\times \{+1\}`$ (hence negative powers โunwindโ positive powers). After exiting the handle for the $`p`$th time, continue the path to the handle associated with the next generator in the word, and repeat the winding process, continuing in this way until the last generator in the word has been represented. Finally, join the end of the path with its starting point to close the loop.
A relator of the finitely presented fundamental group, being a word equated with the identity, corresponds to paths that can be continuously deformed to a point. Obviously such deformation of a path through a handle is obstructed; some topological surgery will be necessary to bypass the obstruction. For each relator $`r_j`$ , gouge out a region from the above constructed manifold (call the manifold $`M`$) along the vicinity of a path representative of $`r_j`$ such that the gouged-out region is homeomorphic to $`U^3\times S^2`$ , where $`U^3`$ is the open 3-ball. Simultaneously, in a copy of $`S^4`$ , gouge out a similar $`U^3\times S^1`$ region; call this manifold $`O_j`$. Finally, identify the $`S^2\times S^1`$ boundary of the gouged-out region in $`M`$ (call this boundary $`T_j`$ ) with the $`S^2\times S^1`$ boundary of the gouged-out region in $`O_j`$. Note that $`O_j`$ is simply connected. (To see this, consider that the only conceivably non-trivial closed path in $`O_j`$ is one that interlocks with the loop formed by the gouged-out region. But the former can be continuously deformed to the boundary of the latter, whereupon it can be made to encircle a cross-section homeomorphic to $`S^2`$, and thereon contracted to a point.) Any path in the homotopy class of paths associated with the relator $`r_j`$ can now be continuously deformed to the surface of $`T_j`$ , then contracted to a point in $`O_j`$ . Repeat this surgery for each relator, in this way gluing to $`M`$, $`n`$ copies of $`O_j`$. This completes the construction. It can be verified, by considering the fundamental groups of the subspaces that cover $`M`$ , that the fundamental group of $`M`$ is the given group $`G`$ as advertised.
If two manifolds are homeomorphic, their fundamental groups are isomorphic. But the converse is not necessarily true, thus the non-classifiability of the manifolds does not immediately follow from the non-classifiability of their fundamental groups. Fortunately for the purposes of this proof, the manifolds constructed above have the following critical property. First consider another manifold formed by gouging out from $`S^4`$, $`m`$ non-intersecting regions homeomorphic to $`U^3\times S^1`$, and gluing the remaining boundaries to those of an identical copy; call the resulting manifold $`N_m`$. Given one of the previously constructed manifolds $`M`$ such that its fundamental group $`G`$ has $`m`$ generators, if $`G`$ is trivial then, it turns out, $`M`$ must be homeomorphic to $`N_m`$ .
To come full circle, let the fundamental group of the manifold $`M`$ represent a Turing machine: let $`M=M_\tau (w_\iota )`$ such that its fundamental group is $`G_\tau (w_\iota )`$ , as described in Section 2. Call $`M_\tau (w_\iota )`$ a Turing manifold. Call $`N_{m(\tau ,\iota )}`$, where $`m(\tau ,\iota )`$ is the number of generators required to represent the Turing machine $`\tau `$ with input $`\iota `$ by $`G_\tau (w_\iota )`$, a halting manifold. It follows that the Turing manifold $`M_\tau (w_\iota )`$ is homeomorphic to the halting manifold $`N_{m(\tau ,\iota )}`$ if and only if Turing machine $`\tau `$ halts on input $`\iota `$ .
## 5 Discussion
A sketch of a proof has been given for the non-classifiability of the 4-manifolds, by way of a topological construction whereby a 4-manifold represents a Turing machine. More precisely, a Turing machine has been encoded into a finitely presented semigroup, which has been encoded into a finitely presented group, which along with a particular Turing input has been encoded into another finitely presented group, which has been encoded into a 4-manifold. The chain of encodings is such that solving the homeomorphy problem for 4-manifolds would solve the halting problem for Turing machines, which is unsolvable. Expressed more intuitively, the essence of the problem is that the topology of a 4-manifold is potentially so rich that its complexity can rival that of any computer program intended to analyze it. Inputting the specifications of a 4-manifold to such a computer program can, in a sense, be equivocated with inputting a computer program to a computer program โ an enterprise subject to logical paradoxes and limitations of the kind brought to light by Turing.
Regarding the physical applicability of Markovโs theorem, while the constructions considered above are compact 4-manifolds, spacetime is often considered to be non-compact, and is sometimes speculated to have hidden extra dimensions. Markovโs proof applies equally well to higher dimensional manifolds - consider $`M\times S^{d4}`$, where $`d>4`$ \- as well as non-compact manifolds - consider $`M\mathrm{\#}R^4`$. Granted Markovโs theorem only applies to manifolds that are permitted to be non-simply connected, but there is a strong possibility that the universe lives in this category. On the cosmic scale, the universe may be multiply-connected ; on the stellar scale, black hole interiors may be topologically nontrivial, though such nontriviality might be rendered undetectable by event horizons (on the other hand, traversable worm holes might exist ); on the subatomic scale, particles are sometimes speculated to be topological geons ; and on the Planck scale, spacetime foam is conjectured to perturb the local topology to no end .
It is conceivable that some physical criteria could be found which would restrict permissible 4-manifolds to classifiable manifolds. For example, if a strict interpretation of causality is imposed, in the form of the conditions of isochrony and the exclusion of closed timelike curves, then it can be shown that the allowed 4-manifolds are constrained to those of the form $`C\times [0,1]`$, $`C\times [0,\mathrm{})`$, and $`C\times (\mathrm{},\mathrm{})`$, where $`C`$ is a 3-manifold . These manifolds are classifiable if the 3-manifolds are classifiable; although whether the 3-manifolds are classifiable is still an open question. Note that the proof of Markovโs theorem, as sketched above, is not applicable to 3-manifolds; for example, the three-dimensional analog of $`O_j`$ is not simply connected, as required. In a sense, there is not enough โroomโ in a 3-manifold to topologically encode a Turing machine, and so there is hope that 3-manifolds might be classifiable. However, whether the universe obeys the previously mentioned interpretation of causality is unknown. These particular conditions may be too restrictive; they would preclude Wheelerโs spacetime foam, as well as other exotic but physically motivated topological proposals. In summary, on the basis of current physical knowledge, the non-classifiability of the 4-manifolds remains relevant.
I wish to thank Steve Carlip for inspiring discussion. This work was supported in part by a National Research Council Associateship Award at the Goddard Space Flight Center, funded by NASA Space Sciences grant ATP02-0043-0056, and in part by Department of Energy grant DE-FG02-91ER40674.
## References
|
warning/0506/astro-ph0506230.html
|
ar5iv
|
text
|
# Long-term optical/IR variability of the Be/X-ray binary LS V +44 17/RX J0440.9+4431
## 1 Introduction
LS V +44 17 is a relatively bright V=10.8 B0 star that is associated with the X-ray source RX J0440.9+4431. RX J0440.9+4431/LS V +44 17 belongs to the subgroup of high-mass X-ray binaries known as Be/X-ray binaries. These systems consist of a neutron star orbiting a O9e-B2e star. The letter $`e`$ stands for emission, as instead of the normal photospheric absorption lines the optical spectra of Be stars display emission lines. Although helium and iron are occasionally seen in emission, the hydrogen lines, especially those of the Balmer series, constitute the signature for which Be stars are renowned (Porter & Rivinius 2003). Strong infrared emission is another defining characteristic of Be stars. The origin of these two observational properties (emission lines and infrared excess) resides in a gaseous, equatorially concentrated circumstellar disc around the OB star. This disc acts as a reservoir of material for accretion on to the compact object. Although the ultimate cause of the Be phenomenon is still not known, it is believed to be related to the rapid rotation of these stars. Recent studies (Townsend et al. 2004) indicate that Be stars may rotate much closer to break-up velocity that previously thought.
The optical/IR information on LS V +44 17 is very scarce. It mainly comes from surveys of the Galactic Plane or catalogues prepared for specific space missions. However, no variability studies of this system have ever been reported. Interest in LS V +44 17 grew when the first evidence that it might be an X-ray binary came to light (Motch et al. 1997). RX J0440.9+4431 is an X-ray pulsar with a spin period of 202 seconds and belongs to the poorly studied group of persistent Be/X-ray binaries (Reig & Roche 1999).
In this work we present the results of our monitoring programme on high-mass X-ray binaries for RX J0440.9+4431/LS V +44 17. We have performed a detailed analysis of its optical and infrared variability covering a period of almost 10 years. We have also made a comparative study of the long-term variability time scales of various Be/X-ray binaries.
## 2 Observations
### 2.1 Optical observations
Optical spectroscopic observations were obtained from 7 telescopes at 4 different observatories: from the Roque de los Muchachos observatory (ORM) in La Palma (Spain), observations were made with the 1.0m Jacobus Kapteyn Telescope, the 2.5m Isaac Newton Telescope, the 4.2m William Herschel Telescope (service time) and the 2.5m Nordic Optical Telescope; from the Skinakas observatory (SKI) in Crete (Greece) the data come from the 1.3m telescope; and from the Haut Provence observatory (OHP) in France the 1.52m and the 1.93m telescopes were employed. Finally one spectrum was taken from the 4m telescope of the Kitt Peak National Observatory (KPNO) in the USA. Table 1 gives the log of the spectroscopic observations. This table contains instrumental information together with the results of the spectral analysis: the equivalent width of the H$`\alpha `$ and H$`\beta `$ lines and an indication of the profile shape of the lines. Negative values indicate that the line is in emission. The reduction of the spectra was made using the STARLINK Figaro package (Shortridge et al. 2001), while their analysis was performed using the STARLINK Dipso package (Howarth et al. 1998).
Optical photometric observations were made using two photometric systems. Strรถmgren photometry ($`uvby`$) was obtained from the Skinakas observatory (SKI) on August 16, 1999. LS V +44 17 was also observed through the Johnson $`B`$, $`V`$, $`R`$ and $`I`$ filters on three occasions from the Skinakas observatory (see Table 2). The telescope was equipped with a 1024$`\times `$1024 SITe CCD chip, containing 24$`\mu `$m pixels. Reduction of the data was carried out using the IRAF tools for aperture photometry.
### 2.2 Infrared observations
Infrared photometry in the JHK bands was obtained as part of a monitoring programme of Be/X-ray binaries at the 1.5 m. Carlos Sรกnchez Telescope (TCS), located at the Teide Observatory in Tenerife, Spain. The instruments used were the Continuously Variable Filter Photometer (CVF) up to January 2001, and the CAIN-II camera, equiped with a $`256\times 256`$ HgCdTe (NICMOS 3) detector ever since. The last data in December 2004 were obtained with the recently comissioned FIN photometer.
Instrumental CVF and FIN magnitudes were transformed to the standard system defined by Kidger & Martรญn-Luis (2003). Instrumental CAIN magnitudes were obtained from the images by means of the IRAF tools for aperture photometry, and transformed to the standard system defined by Hunt et al. (1998). The accuracy of the standard JHK values in all three bands is 0.01, 0.03 and 0.02 mag. for CVF, CAIN-II and FIN data respectively. The obtained values are given in Table 2.
## 3 Results
### 3.1 Previous optical work
The first astronomical observations date back to the circa 1930. LS V +44 17 is mentioned in the Bergedorfer Spektral-Durchmusterung catalogue (BSD, Schwassmann & van Rhijn 1935), where a spectral type B0 is suggested. The first accurate photometric observations on the Johnson photometric system were performed by Bigay (1963) who gave $`V=10.78`$, $`(BV)=0.61`$ and $`(UB)=0.36`$ for observations performed in 1953, although the star is referenced in Seyfert & Popper (1941) with a photographic magnitude of 11.3. LS V +44 17 is also found on one of the photographic plates of the Sandage two-colour survey of the Galactic plane (Lanning & Meakes 2001) with $`B=10.4`$ and $`(UB)=0.4`$, although these values are affected by large errors ($`\mathrm{ยฟ}\mathrm{}\pm 0.5`$ mag) owing to the uncertainties inherent to obtaining accurate visual estimates from photographic plates.
Further photometric data are provided by catalogues of optical surveys. However, varying results are obtained for different versions of the catalogue due to slightly different reduction methods. Another disadvantage of the catalogued values is the difficulty in defining the exact date of the observations. The Hipparcos and Tycho Catalogues (Hog et al. 2000) give $`V=10.71`$ and $`(BV)=0.44`$ for observations that took place some time from November 1989 to March 1993, whereas the USNO-A2.0 Catalogue (Monet et al. 1998) gives a blue magnitude of 12.2 and a red magnitude of 10.4 for the epoch 1953.025.
### 3.2 Spectral classification
The presence of H , He i and Si iv lines clearly demonstrates that RX J0440.9+4431 contains an early-type star (Fig. 1). He II lines, although present, are weak (only He ii 4686ร
is clearly detected, He ii 4200ร
is weak and He ii 4541ร
is buried in the noise), which implies that the spectral type must be earlier than B1 but later than O9.5. A spectral type in this range is also borne out by the weak Si iii 4552ร
. He ii 4686ร
is normally last seen at B0.5, whereas the ratio Si iii 4552ร
/Si iv 4089ร
increases smoothly as we progress toward later spectral types. A visual comparison of the relative strength of metallic lines of LS V +44 17 with those of MK standards (Walborn & Fitzpatrick 1990) resulted in B0.2 as the closest spectral type. In particular, the fact that the ratio of the C iii 4650ร
blend with respect to He ii 4686ร
is larger than one agrees more with a B0.2 than with a B0 type. As for the luminosity class, a main-sequence star is suggested by the fact that the ratios of He i 4026ร
over Si iv 4089ร
and He i 4121ร
over Si iv 4116ร
are larger than one. Therefore, we conclude that the optical counterpart to the Be/X-ray binary RX J0440.9+4431 is a B0.2V star.
### 3.3 Reddening and distance
Besides the H$`\alpha `$ and He I $`\lambda 6678`$ lines, the red-end spectrum (Fig.2) contains several strong diffuse interstellar bands (DIB), which can be used to estimate the amount of interstellar absorption toward the source (Herbig & Leka 1991; Galazutdinov et al. 2000). We used six different interstellar lines (6010 ร
, 6195 ร
, 6202 ร
, 6269 ร
, 6376/79 ร
and 6613 ร
) to derive a colour excess $`E(BV)`$ according to the linear relationships of Herbig (1975). Given the different amount of available data (the spectra covered different wavelength intervals) and different signal-to-noise values of the spectra, we obtained a mean $`E(BV)`$ by averaging all measurements. The resulting reddening was $`E(BV)=0.62\pm 0.03`$. For the sake of comparison, we also obtained the colour excess using only the average of the five measurements of the highest resolution spectra (those taken from the WHT on February 2, 2004). The resulting $`E(BV)`$ was 0.64$`\pm `$0.02. The errors are the weighted standard deviation of the results of the various lines used.
Estimating the reddening from photometric data in Be stars might be misleading as the circumstellar continuum emission affects the photometric colours and indices (Fabregat & Torrejรณn 1998). This effect is expected to be more distinct for longer wavelengths. Indeed, a B0.2V star has an intrinsic colour $`(BV)_0=0.25`$ (Wegner 1994). Taking the measured photometric magnitudes $`(BV)=0.63`$ (2004 observations) we derive an excess $`E(BV)0.9`$, somewhat larger than the value derived above. However, LS V +44 17 went through a low-activity optical state, presumably during the first half of 2001 as indicated by the low H$`\alpha `$ equivalent widths (see Fig. 5). Interpreting this optical minimum as a weakening of the disc we would expect that the IR magnitudes at that time should be very close to those of the underlying B star. Assuming the interstellar extinction law $`E(JK)=0.54E(BV)`$ and the intrinsic colour $`(JK)_0=0.16`$ for a main-sequence B0.2 star (Koornneef 1983), the observed $`(JK)=0.21`$ gives $`E(BV)=0.68\pm 0.03`$, in good agreement with the spectroscopically derived value.
Finally, taking the standard law $`A_V=3.1E(BV)`$ and assuming an average absolute magnitude for a B0.2V star of $`M_V=3.8`$ (Humphreys & McElroy 1984; Martins et al. 2005) the distance to RX J0440.9+4431 is estimated to be $``$ 3.3$`\pm `$0.5 kpc. This error includes those of $`m_V`$ (0.02) and $`A_V`$ (0.3), but assumes no error in the absolute magnitude $`M_V`$.
### 3.4 Rotational velocity
The rotational velocity was estimated by measuring the full width at half maximum of He I lines (see e.g. Steele et al. 1999). After correcting for instrumental resolution we obtained $`v\mathrm{sin}i=235\pm 15`$ km s<sup>-1</sup>, which compares favourably to the value of 246$`\pm `$16 km s<sup>-1</sup> given for weak-emission early-type shell stars (Mennickent et al. 1994). As a comparison, other rotational velocities in Be/X-ray binaries are: $`v\mathrm{sin}i=200\pm 30`$ km s<sup>-1</sup> in LS I +61 235/RX J0146.9+6121 (Reig et al. 1997a), $`v\mathrm{sin}i=290\pm 50`$ km s<sup>-1</sup> in V635 Cas/4U 0115+63 (Negueruela & Okazaki 2001), $`v\mathrm{sin}i=240\pm 20`$ km s<sup>-1</sup> in LS 992/RX J0812.4โ3114 (Reig et al. 2001), $`v\mathrm{sin}i=240\pm 20`$ km s<sup>-1</sup> in SAX J2103.5+4545 (Reig et al. 2004).
## 4 Discussion
We have monitored the Be/X-ray binary LS V +44 17 for the last 10 years. Our observations coincided with the latest stages of a declining disc phase. The slow and gradual decline of the EW(H$`\alpha `$) and IR colours seems to indicate that the mechanism that feeds the disc had already stopped when we started the monitoring of the source. The source entered a long period (1998-2003) of low optical/IR activity, where the line emission just filled in the underlying absorption expected from the photosphere of the B-type star and the IR magnitudes showed their lowest values. The equivalent width of the H$`\alpha `$ line (EW(H$`\alpha `$)) always remained negative, indicating that the complete loss of the disc did not occur. However, given the large observational gaps of our data we cannot rule out the possibility that such an event could have happened. The loss of the circumstellar disc could have occurred in early 2001 (the EW(H$`\alpha `$) was only โ0.2 ร
in August 2001). As mentioned before, in January 2001 the measured intrinsic IR colour $`(JK)0.16`$ agrees with a B0-B0.5 star (Koornneef 1983). In other words, in 2001 the underlying B-type star would have been exposed. Figure 5 shows the evolution of the H$`\alpha `$ equivalent width and the infrared magnitudes. The EW(H$`\alpha `$) and the IR magnitudes follow the same trend, namely, a slow decrease, reaching a minimum around MJD 52000 and a gradual increase. This long-term variability suggests the dissipation and subsequent formation of the circumstellar disc and sets a common origin (i.e., the circumstellar disc) for the H$`\alpha `$ emission and infrared excess.
### 4.1 The H$`\alpha `$ line
V/R variability is defined as the intensity variations of the two peaks (known as violet and red peak) in the split profile of a spectral line. In many Be stars, if monitored over a long enough periods of time, these variations are quasiperiodic (Okazaki 1997). The V/R ratio is defined as $`V/R=(I(V)I_c)/(I(R)I_c)`$, where $`I(V)`$, $`I(R)`$ and $`I_c`$ are the intensities of the violet peak, red peak and continuum, respectively.
The evolution of the H$`\alpha `$ profile throughout the period covered by our observations is presented in Fig.3. The vertical scale was left the same in all plots in order to show the variability in the strength of the line. Double-peak H$`\alpha `$-line profiles, both symmetric and asymmetric, are always present in LS V +44 17. Symmetric profiles are believed to be generated in quasi-Keplerian discs (see e.g Hummel 1994). Asymmetric profiles are associated with radial motion and/or distorted density distributions (Hanuschik et al. 1995; Hummel & Hanuschik 1997). The model that most successfully accounts for the long-term variability of these asymmetric profiles is the one-armed oscillation model (Okazaki 2000, and references therein).
In LS V +44 17, the ratio of the intensities of the violet over the red peak ($`V/R`$) hardly varied over almost 10 years. Excluding the first two spectra, the lines have $`|\mathrm{log}(V/R)|<0.05`$ (column 8 of Table 1). A symmetric profile does not necessarily mean the absence of the density wave as symmetric split profiles (the V=R phase) can occur during a fraction of the V/R cycle, more precisely, when the star lies between the observer and the high-density perturbation (Telting et al. 1994). These V=R phases represent a fraction of the entire V/R cycle. In LS V +44 17 an asymmetric profile was last seen in 1996. Since then only symmetric profiles are present. Thus it is very unlikely that the spectral state of the last 8 years correspond to a V/R phase. We conclude that the density wave faded away before the dissipation of the disc, perhaps because the disc became too tenuous to support a density wave.
Some spectra show the depression between the double peak profile extending below the stellar continuum, reminiscent of the shell profile. These types of lines are explained by partial absorption of the central star by the circumstellar disc as a consequence of a high inclination angle (see e.g. Hummel & Vrancken 2000). However, the profiles seen in LS V +44 17 cannot be considered as proper shell lines because they only occurred during the optical minimum. If ascribed to absorption by the disc itself then we should expect the central depression to become more apparent as the extent of the disc increases, i.e, as the EW(H$`\alpha `$) increases. No such trend is seen (Fig. 4). In addition, the width of the central reversal is considerably broader ($`FWHM200400`$ km s<sup>-1</sup>) than the typical value of shell profiles ($`FWHM\mathrm{ยก}\mathrm{}50`$ km s<sup>-1</sup>). Finally, none of the spectra of LS V +44 17 fulfil the criterion given by Hanuschik (1996) that in order for a profile to have shell characteristics the ratio $`I_\mathrm{p}/I_{\mathrm{cd}}`$ should be larger than 1.5. Therefore the apparent shell profiles in LS V +44 17 are likely to be due to the photospheric absorption line, which combines with a weak double-peaked emission.
Significant changes are apparent in the distance between the peaks and the strength of the line. The peak separation correlates with the intensity of the H$`\alpha `$ line (Fig. 4). As the EW(H$`\alpha `$) increases, the distance between peaks decreases. Interpreting the peak separation ($`\mathrm{\Delta }_{\mathrm{peak}}`$) as the outer radius ($`R_{\mathrm{out}}`$) of the emission line forming region (Huang 1972)
$$\frac{R_{\mathrm{out}}}{R_{}}=\sqrt{\frac{2v\mathrm{sin}i}{\mathrm{\Delta }_{\mathrm{peak}}}}$$
(1)
we conclude that lower velocities of the emitting components occur when the disc has developed, i.e, when the EW(H$`\alpha `$) is large. Despite a moderate increase of the EW(H$`\alpha `$) the absence of asymmetries indicates that fast radial displacements do not take place during the first instances of the formation of the disc.
The He i 6678ร
line also shows V/R variability. In general, it imitates the behaviour of the H$`\alpha `$ line. Since metallic lines are generated at smaller disc radii than the hydrogen lines (Hummel & Vrancken 1995; Jaschek & Jaschek 2004), the asymmetry of the He i line profiles indicates that the internal changes of the disc are global, affecting its entire structure.
### 4.2 Variability time scales
Although the spectroscopic data are distributed irregularly over the period of the reported observations, the smooth variations of the H$`\alpha `$ equivalent width and infrared brightness indicate that structural changes in the circumstellar disc of LS V +44 17 occur on time scales of years.
Table 3 shows the typical time scales associated with disc variability for a number of Be/X-ray binaries: $`T_{\mathrm{disc}}`$ is the typical duration of formation/dissipation of the circumstellar disc and $`T_{\mathrm{V}/\mathrm{R}}`$ represents the quasi period for $`V/R`$ variability. $`T_{\mathrm{disc}}`$ exhibits a good correlation with the orbital period. Systems with narrow orbits tend to show faster disc growth and dissipation cycles, while slower evolutionary time scales are associated with long orbital periods. This is in agreement with the disc truncation model (Okazaki & Negueruela 2001), which suggests a direct relationship between the size of the disc and the orbital period (see also Reig et al. 1997b, for observational evidence in this respect). Within the framework of the global one-armed oscillation model, the viscous excitation of a density wave is associated with longer time scales when the disc is larger (Okazaki 2000).
Although the orbital period of RX J0440.9+4431/LS V +44 17 is unknown, its classification as a persistent system (Reig & Roche 1999) with a the relatively long spin period implies that it must be long. A $`P_{\mathrm{orb}}=150200`$ d is estimated from the $`P_{\mathrm{spin}}P_{\mathrm{orb}}`$ diagram (Corbet 1986). For LS I +61 235 Corbetโs diagram implies $`P_{\mathrm{orb}}\mathrm{ยฟ}\mathrm{}200`$ d. The typical time scales associated with the evolution of the circumstellar disc in LS V +44 17 are rather long โ the EW(H$`\alpha `$) had not recovered from the initial pre-disc-loss phase values nine years later โ and agree with those of Be/X-ray binaries with wide orbits. In contrast, changes originated by the density wave are much faster. In LS V +44 17, the time elapsed between the slightly blue-dominated line of the first spectrum and the strongly red-dominated line of the second spectrum of Fig.3 is just 3 months.
None of the three Be/X-ray binaries that have gone through disc-loss phases and for which there is a good optical follow-up coverage, namely, X Per in 1990 (Clark et al. 2001), 4U 0115+63 in 1997 (Negueruela & Okazaki 2001) and 1A 0535+262 in 1998 (Haigh et al. 2004), exhibited asymmetric profiles during the initial stage of disc growth. After the disc loss phase the first asymmetric profile did not occur until the EW($`H\alpha `$) was $`67`$ ร
in 4U 0115+63 ($`P_{\mathrm{orb}}=24.3`$ d), $`710`$ ร
in 1A 0535+262 ($`P_{\mathrm{orb}}=111`$ d) and $`1012`$ ร
in X Per ($`P_{\mathrm{orb}}=250`$ d). In our monitoring of LS V +44 17/RX J0440.9+4431, asymmetric profiles are associated with the largest values of the EW(H$`\alpha `$). Below 8 ร
, only symmetric profiles are observed. The peak separation of the latest spectra imply a disc radius of $``$2 $`R_{}`$, assuming a Keplerian disc and Eq.(1).
In summary, the correlation of $`T_{\mathrm{disc}}`$ and the orbital period provides further observational evidence for the interaction of the neutron star with the circumstellar disc of its Be starโs companion, whilst the relationship betwen the EW(H$`\alpha `$) at which the first asymmetry appears with the orbital period implies that the density oscillations do not become observable until the disc has reached a critical size or density.
## 5 Conclusion
We have monitored the Be/X-ray binary LS V +44 17 for the last 10 years. The observations coincided with a period of low optical/IR activity, characterised by the likely loss of the Be starโs circumstellar disc and subsequent reformation. Since 2001 the envelope has been gradually growing as indicated by the increase of the equivalent width and the narrowing of the peak separation of the split H$`\alpha `$ line. The time scales for structural changes in the circumstellar disc of RX J0440.9+4431/LS V +44 17 compares favourably with those of Be/X-ray binaries with long orbital periods. While the formation/dissipation of the disc may last for several years, the line profile changes are much faster and, in general, depends on the duration of the active phase. The disappearance of the V/R varibility before the dissipation of the disc and the lack of asymmetric profiles of the latest observations even though the equivalent width of the H$`\alpha `$ line has increased up to $`6`$ ร
confirms the fact that the effects of the density perturbation do not manifest themselves until the disc is fully developed. By studying the characteristic variability time scales of a number of Be/X-ray binaries we have found further observational evidence of the influence of the neutron star on the envelope of the Be star.
###### Acknowledgements.
The authors thank M. Brotherton and P. Nandra for providing the Kitt Peak spectrum. The Kitt Peak National Observatory, a division of the National Optical Astronomy Observatories, is operated by the Association of Universities for Research in Astronomy, Inc. under cooperative agreement with the National Science Foundation. IN is a researcher of the programme Ramรณn y Cajal, funded by the Spanish Ministerio de Educaciรณn y Ciencia and the University of Alicante, with partial support from the Generalitat Valenciana and the European Regional Development Fund (ERDF/FEDER). This research is partially supported by the MEC through grant ESP-2002-04124-C03-03. Skinakas Observatory is a collaborative project of the University of Crete, the Foundation for Research and Technology-Hellas and the Max-Planck-Institut fรผr Extraterrestrische Physik. The WHT spectrum was obtained as part of the ING service programme. Based in part on observations made at Observatoire de Haute Provence (CNRS), France. The Carlos Sanchez Telescope is operated at the Teide Observatory by the Instituto de Astrofรญsica de Canarias.
|
warning/0506/cond-mat0506467.html
|
ar5iv
|
text
|
# Resonant x-ray scattering spectra from multipole orderings: Np M4,5 edges in NpO2
## I Introduction
Resonant x-ray scattering (RXS) technique has attracted much attention to study spin and orbital properties of 3d transition-metal compounds. RXS at the $`K`$-edge is described by a second-order optical process that an incident x-ray excites a $`1s`$ core electron to unoccupied $`4p`$ states and then the $`4p`$ electron is recombined with the core hole with emitting x-ray in the dipole process ($`E1`$). It became widely known after the observation of intensities on orbital-ordering superlattice spots at Mn K-edge in LaMnO<sub>3</sub>.Murakami et al. (1998) At the earlier stage, the spectra were interpreted as a direct observation of orbital ordering.Ishihara and Maekawa (1998) However, subsequent theoretical studies based on band structure calculations revealed that the spectra are a direct reflection of lattice distortion,Elfimov et al. (1999); Benfatto et al. (1999); Takahashi et al. (1999) since the $`4p`$ state in the intermediate state is influenced not by the orbital ordering of $`3d`$ electrons but by lattice distortion through the hybridization with the $`2p`$ state at neighboring oxygen sites.
Different from transition-metal compounds, M<sub>4,5</sub> edges are available for forbidden reflection Bragg spots in actinide compounds.Isaacs et al. (1990); Mannix et al. (1999); Longfield et al. (2002) The RXS spectra are more directly reflecting multipole orderings of $`5f`$ states, since the $`E`$1 process involves a transition from the $`3d`$-core to $`5f`$ states. Each actinide atom usually carries local multipole moments, which can order at low temperatures due to intersite interactions such as exchange interactions. For such localized electron systems, RXS amplitudes are given by summing up contributions at each site. The crystal electric field (CEF) and the intersite interaction can be safely neglected in the intermediate state, because they are much smaller than the intra-atomic Coulomb interaction. Therefore, it may be reasonable to assume that the intermediate state preserves the rotational invariance. Under the assumption, we derive an expression for the RXS amplitude in the $`E`$1 process to characterize the spectra. Although the expression is essentially the same as the formula by Hannon et al.,Hannon et al. (1989) the present form is useful to calculate energy profiles with taking full account of multiplet structures. Using this expression together with a microscopic model, we calculate the RXS spectra in the triple-k multipole ordering phase in NpO<sub>2</sub>.
NpO<sub>2</sub> undergoes a second-order phase transition below $`T_\text{0}=25.5`$ K. Osborne and E. F. Westrum (1953); Erdรถs et al. (1980) Since Np ions are Kramers ions in the $`(5f)^3`$ configuration, a magnetic ground state is naturally expected. However, neither Mรถssbauer spectroscopyDunlap et al. (1968); Friedt et al. (1985) nor neutron diffraction experimentsCox and Frazer (1967); Heaton et al. (1967) could detect any evidence of the sizable magnetic moment. Actually, the former experiment gave an estimate of the upper limit of the magnitude of the magnetic moment $`0.01\mu _\text{B}`$, which was too small to explain the effective paramagnetic moment $`2.95\mu _\text{B}`$.Ross and Lam (1967) Another complication is that a muon spin relaxation ($`\mu `$SR) experiment has suggested the low-temperature phase of breaking time-reversal symmetry.Kopmann et al. (1998)
A natural way to reconcile with the above observations is to introduce the higher-rank multipole ordering rather than the dipole moment. Actually, Santini and Amoretti proposed a octupole ordering of $`\mathrm{\Gamma }_2(xyz)`$ symmetry.Santini and Amoretti (2000, 2002) However, this phase can be ruled out because it gives rise to no RXS intensity. Recently, Paixรฃo et al. have reported that a longitudinal triple-k octupole ordering accounts well for their RXS experiment.Paixรฃo et al. (2002) The reason for anticipating triple-k orderings is that it excludes a crystal distortion or a shift of oxygen positions, which is consistent with the experiment. Experimental data obtained from the <sup>17</sup>O NMR spectrum, which indicate the existence of two inequivalent oxygen sites, support the occurrence of the triple-k octupole ordering phase.Tokunaga et al. (2005) Some theoretical works also have lent support to realization of this type of the phase.Kiss and Fazekas (2003); Kubo and Hotta (2005)
Assuming the $`\mathrm{\Gamma }_8`$-quartet ground state, we explicitly construct a triple-k octupole ordering state. This state is found to simultaneously carry a finite quadrupole moment, which generates the RXS intensity. Since the RXS amplitudes are characterized by three terms, the scalar, dipole, and quadrupole ones, it is not necessary to assume the existence of the hexadecapole moment instead of the quadrupole moment. We calculate the energy profiles with taking full account of multiplet structures in the intermediate state. We obtain spikes-like curves at Np $`M_4`$ edges for smaller values of the core-level width $`\mathrm{\Gamma }`$ as a reflection of multiplet structures. They are found to merge into a single peak with $`\mathrm{\Gamma }>1`$ eV. The energy profile with $`\mathrm{\Gamma }2`$ eV seems to agree with the experiment. The azimuthal-angle dependence of the RXS spectra is obtained in agreement with the previous analysis.Paixรฃo et al. (2002); Caciuffo et al. (2003) The present analysis provides a sound basis to the previous phenomenological analysis.
The present paper is organized as follows. In Sec. II, we present an expression for the RXS amplitude, which is useful to calculate the energy profiles. In Sec. III, we analyze the RXS spectra in the triple-k octupole ordering of NpO<sub>2</sub> on the basis of a localized electron model. Section IV is devoted to concluding remarks. In Appendix, we derive the general expression of RXS characterizing energy profiles.
## II Theoretical Framework of RXS
### II.1 Second-order optical process
In the resonant process, an incident photon with energy $`\mathrm{}\omega `$, wavevector k, and polarization vector $`\mathit{ฯต}`$ excites a core electron to an empty valence shell of the intermediate state, then the excited electron falls into the core state emitting a photon having the same energy, wavevector $`\text{k}^{}`$, and polarization vector $`\mathit{ฯต}^{}`$. For example, at $`M_{4,5}`$ edges in actinide compounds, a $`3d`$ core electron is promoted to partially filled $`5f`$ states at each site by the $`E`$1 transition. The definition of a geometrical arrangement adopted here is found in Fig. 1. The RXS amplitude is assumed as a sum of contributions from individual ions. Since the dipole matrix element involves well-localized wavefunction of core states, the assumption seems quite reasonable. Accordingly, the RXS intensity observed in the experiment may be expressed for the scattering vector G ($`=\text{k}^{}\text{k}`$) as
$$I(\mathit{ฯต}^{},\mathit{ฯต},\text{G},\omega )\left|\frac{1}{\sqrt{N}}\underset{j}{}\text{e}^{i\text{G}\text{r}_j}M_j(\mathit{ฯต}^{},\mathit{ฯต},\omega )\right|^2,$$
(1)
where $`M_j(\mathit{ฯต}^{},\mathit{ฯต},\omega )`$ represents the RXS amplitude at site $`j`$ with $`N`$ being the number of sites. For the $`E`$1 transition, it is expressed asBlume (1985); Blume and Gibbs (1988); Hannon et al. (1989); Hill and McMorrow (1996)
$$M_j(\mathit{ฯต}^{},\mathit{ฯต},\omega )=\underset{\alpha ^{},\alpha }{}ฯต_\alpha ^{}ฯต_\alpha ^{}\underset{\mathrm{\Lambda }}{}\frac{\psi _0|x_{\alpha ,j}|\mathrm{\Lambda }\mathrm{\Lambda }|x_{\alpha ^{},j}|\psi _0}{\mathrm{}\omega (E_\mathrm{\Lambda }E_0)+i\mathrm{\Gamma }},$$
(2)
where the dipole operators $`x_{\alpha ,j}`$โs are defined as $`x_{1,j}=x_j`$, $`x_{2,j}=y_j`$, and $`x_{3,j}=z_j`$ in the coordinate frame fixed to the crystal axes with the origin located at the center of site $`j`$. The $`|\psi _0`$ represents the ground state with energy $`E_0`$, while $`|\mathrm{\Lambda }`$ represents the intermediate state with energy $`E_\mathrm{\Lambda }`$. The $`\mathrm{\Gamma }`$ describes the life-time broadening width of the core hole.
### II.2 Energy profiles
In localized models, the ground state and the intermediate state at each site are well specified by the eigenfunctions of the angular momentum operator, $`|J,m`$. The CEF and the intersite interaction usually lift the degeneracy in the ground state. Thus the ground state at site $`j`$ may be expressed as $`|\psi _0_j=_mc_j(m)|J,m`$. On the other hand, in the intermediate state, we can neglect the CEF and the intersite interaction in a good approximation, since their energies are much smaller than the intra-atomic Coulomb interaction and the spin-orbit interaction (SOI) which give rise to the multiplet structure. Thus the intermediate state preserves the rotational symmetry. Under the assumption, as derived in Appendix, we obtain a general expression of the scattering amplitude at site $`j`$:
$`M_j(\mathit{ฯต}^{},\mathit{ฯต},\omega )`$ $`=`$ $`\alpha _0(\omega )\mathit{ฯต}^{}\mathit{ฯต}`$ (3)
$``$ $`i\alpha _1(\omega )(\mathit{ฯต}^{}\times \mathit{ฯต})\psi _0|\text{J}|\psi _0`$
$`+`$ $`\alpha _2(\omega ){\displaystyle \underset{\nu }{}}P_\nu (\mathit{ฯต}^{},\mathit{ฯต})\psi _0|z_\nu |\psi _0,`$
where
$`z_1`$ $``$ $`Q_{x^2y^2}={\displaystyle \frac{\sqrt{3}}{2}}(J_x^2J_y^2),`$ (4a)
$`z_2`$ $``$ $`Q_{3z^2r^2}={\displaystyle \frac{1}{2}}(3J_z^2J(J+1)),`$ (4b)
$`z_3`$ $``$ $`Q_{yz}={\displaystyle \frac{\sqrt{3}}{2}}(J_yJ_z+J_zJ_y),`$ (4c)
$`z_4`$ $``$ $`Q_{zx}={\displaystyle \frac{\sqrt{3}}{2}}(J_zJ_x+J_xJ_z),`$ (4d)
$`z_5`$ $``$ $`Q_{xy}={\displaystyle \frac{\sqrt{3}}{2}}(J_xJ_y+J_yJ_x),`$ (4e)
and
$`P_1(\mathit{ฯต}^{},\mathit{ฯต})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}}(ฯต_x^{}ฯต_xฯต_y^{}ฯต_y),`$ (5a)
$`P_2(\mathit{ฯต}^{},\mathit{ฯต})`$ $`=`$ $`{\displaystyle \frac{1}{2}}(2ฯต_z^{}ฯต_zฯต_x^{}ฯต_xฯต_y^{}ฯต_y),`$ (5b)
$`P_3(\mathit{ฯต}^{},\mathit{ฯต})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}}(ฯต_y^{}ฯต_z+ฯต_z^{}ฯต_y),`$ (5c)
$`P_4(\mathit{ฯต}^{},\mathit{ฯต})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}}(ฯต_z^{}ฯต_x+ฯต_x^{}ฯต_z),`$ (5d)
$`P_5(\mathit{ฯต}^{},\mathit{ฯต})`$ $`=`$ $`{\displaystyle \frac{\sqrt{3}}{2}}(ฯต_x^{}ฯต_y+ฯต_y^{}ฯต_x).`$ (5e)
Here we have suppressed the dependence on $`j`$ in the right hand side of Eq. (3). The energy profiles are given by only three functions, $`\alpha _0(\omega )`$, $`\alpha _1(\omega )`$, and $`\alpha _2(\omega )`$, whose expressions are explicitly given in Appendix.
Several facts are immediately deduced from Eq. (3). First, since the scalar, dipole, and quadrupole terms exhaust the amplitude, the octupole ordering alone does not give rise to the RXS amplitude. Second, the choice of the CEF parameters in the ground state does not affect the shape of energy profiles $`\alpha _0(\omega )`$, $`\alpha _1(\omega )`$ and $`\alpha _2(\omega )`$, although it affects the expectation values of dipole and/or quadrupole operators. Third, $`\alpha _0(\omega )`$ has no contribution to the forbidden Bragg spots in the antiferro-type structure. In order to calculate the energy profiles, however, we need to know explicitly wavefunctions of the intermediate state, which are discussed in the next section.
### II.3 Absorption coefficient
Within the $`E1`$ transition, the absorption coefficient is given by
$$A(\omega )\underset{j}{}\underset{\alpha }{}\underset{\mathrm{\Lambda }}{}|\mathrm{\Lambda }|x_{\alpha ,j}|\psi _0|^2\frac{\mathrm{\Gamma }/\pi }{(\mathrm{}\omega E_\mathrm{\Lambda }+E_0)^2+\mathrm{\Gamma }^2},$$
(6)
where $`|\mathrm{\Lambda }`$ with energy $`E_\mathrm{\Lambda }`$ represents the final state, which is equivalent to the intermediate state of RXS. A comparison of Eq. (6) with Eq. (2) leads to
$$A(\omega )\mathrm{Im}\alpha _0(\omega ),$$
(7)
where Im $`X`$ denotes the imaginary part of $`X`$.
## III RXS spectra from NpO<sub>2</sub>
### III.1 Quartet ground state
NpO<sub>2</sub> has the CaF<sub>2</sub> type structure ($`F_{m\overline{3}m}`$) with a lattice constant $`a=5.431`$ ร
at room temperature, as schematically shown in Fig. 2.Osborne and E. F. Westrum (1953) Np ions are tetravalent in NpO<sub>2</sub>, as confirmed by the isomer shift in Mรถssbauer spectraDunlap et al. (1968) and by the neutron diffraction experiment.Delapalme et al. (1980) In a localized description, each Np ion is in the $`(5f)^3`$-configuration. The Hamiltonian of Np ions consists of the intra-atomic Coulomb interaction between $`5f`$ electrons in addition to the SOI of $`5f`$ electrons. The Slater integrals for the Coulomb interaction and the SOI parameters are evaluated within the Hatree-Fock approximation (HFA),Cowan (1981) and are listed in Table 1. Because the isotropic parts of the Coulomb interaction $`F^0`$โs are known to be well screened in solids compared to those of the anisotropic parts, the former quantities are multiplied by a factor 0.25 while the latterโs are by 0.8. Within the HFA, the ground state has the ten-fold degeneracy corresponding to $`J=9/2`$ multiplet. The choice of the multiplying factors does not alter this conclusion. Nagao and Igarashi (2003) Note that these states of $`J=9/2`$ are slightly deviated from those of the perfect Russell-Saunders (RS) coupling scheme with $`L=6`$ and $`S=3/2`$ due to the presence of the strong SOI. For instance, $`\text{L}^2`$ and $`\text{S}^2`$ take values 39.752 and 3.237 respectively, compared to the RS values 42 and 3.75.
In crystal, the ten-fold degeneracy is lifted by the CEF. Under the cubic symmetry, the CEF Hamiltonian $`H_{\text{C}EF}`$ may be expressed as
$$H_{\text{C}EF}=B_4(O_4^0+5O_4^4)+B_6(O_6^021O_6^4),$$
(8)
where $`O_k^q`$โs represent Stevens operator equivalence. Thereby the degenerate levels are split into one doublet $`\mathrm{\Gamma }_6`$ and two quartets $`\mathrm{\Gamma }_8^{(1)}`$ and $`\mathrm{\Gamma }_8^{(2)}`$. The level scheme has been analyzed by the inelastic neutron scattering, which yields an estimate of CEF parameters as $`B_4=3.03\times 10^2`$ meV and $`B_6=2.36\times 10^4`$ meV. Amoretti et al. (1992) The lowest levels are given by the $`\mathrm{\Gamma }_8^{(2)}`$, which is separated about 55 meV from another quartet $`\mathrm{\Gamma }_8^{(1)}`$. Diagonalizing Eq. (8), we obtain the bases of the lowest quartet as
$`|+`$ $`=`$ $`c_1|+{\displaystyle \frac{9}{2}}+c_2|+{\displaystyle \frac{1}{2}}+c_3|{\displaystyle \frac{7}{2}},`$ (9)
$`|+`$ $`=`$ $`c_1|{\displaystyle \frac{9}{2}}+c_2|{\displaystyle \frac{1}{2}}+c_3|+{\displaystyle \frac{7}{2}},`$ (10)
$`|`$ $`=`$ $`c_4|+{\displaystyle \frac{5}{2}}+c_5|{\displaystyle \frac{3}{2}},`$ (11)
$`|`$ $`=`$ $`c_4|{\displaystyle \frac{5}{2}}+c_5|+{\displaystyle \frac{3}{2}},`$ (12)
with $`c_1=0.2757,c_2=0.4483,c_3=0.8503,c_4=0.9751`$ and $`c_5=0.2216`$. State $`|m`$ denotes the eigenstate with $`J_z=m`$. Symbols $`\tau `$ ($`=\pm `$) and $`\sigma `$ ($`=,`$) are introduced to represent the state $`|\tau ,\sigma `$, which distinguish non-Kramersโ and Kramersโ pairs, respectively.
### III.2 Triple-k structure
The four-fold degeneracy in the ground $`\mathrm{\Gamma }_8^{(2)}`$ quartet may be lifted by the intersite interaction, giving rise to induced multipole moments. Actually, several experiments tell us that the time-reversal symmetry is broken with nearly zero dipole moment in the ordered phase below $`T_0=25.5`$ K.Kopmann et al. (1998); Tokunaga et al. (2005) These observations lead Santini and Amoretti to propose the antiferro ordering of $`T_{xyz}`$-type ($`T_{xyz}\frac{\sqrt{15}}{6}\overline{J_xJ_yJ_z}`$). Santini and Amoretti (2000, 2002) Here the overline on operators means symmetrization, for example, $`\overline{J_xJ_y^2}=J_xJ_y^2+J_yJ_xJ_y+J_y^2J_x`$.Shiina et al. (1997). Unfortunately, this phase would not give rise to the RXS intensities observed in the experiments.
An important observation is that no external distortion from cubic structure exists in the ordered phase, that is, the unit cell remains cubic below $`T_0`$. This leads us to consider the triple-k ordering, since it allows the crystal to keep the cubic symmetry. As schematically shown in Fig. 3 (c), the triple-k structure is defined by all three members of the star of $`\text{k}=001`$ simultaneously present on each site of the lattice; there are four sublattices 1, 2, 3 and 4 at $`(0,0,0),(\frac{1}{2},\frac{1}{2},0),(0,\frac{1}{2},\frac{1}{2})`$ and $`(\frac{1}{2},0,\frac{1}{2})`$, respectively.
#### III.2.1 octupole ordering
We start by the octupole ordering of $`\mathrm{\Gamma }_{5u}`$-type proposed by Paixรฃo et al.Paixรฃo et al. (2002) The corresponding octupole operators are defined by
$`T_x^\beta `$ $`=`$ $`{\displaystyle \frac{\sqrt{15}}{6}}\overline{J_x(J_y^2J_z^2)},`$ (13a)
$`T_y^\beta `$ $`=`$ $`{\displaystyle \frac{\sqrt{15}}{6}}\overline{J_y(J_z^2J_x^2)},`$ (13b)
$`T_z^\beta `$ $`=`$ $`{\displaystyle \frac{\sqrt{15}}{6}}\overline{J_z(J_x^2J_y^2)}.`$ (13c)
Defining operators,
$$T_p=\{\begin{array}{ccc}\frac{1}{\sqrt{3}}\left(T_x^\beta +T_y^\beta +T_z^\beta \right)\hfill & \text{f}or& p=111,\hfill \\ \frac{1}{\sqrt{3}}\left(T_x^\beta T_y^\beta T_z^\beta \right)\hfill & \text{f}or& p=\overline{1}11,\hfill \\ \frac{1}{\sqrt{3}}\left(T_x^\beta +T_y^\beta T_z^\beta \right)\hfill & \text{f}or& p=1\overline{1}1,\hfill \\ \frac{1}{\sqrt{3}}\left(T_x^\beta T_y^\beta +T_z^\beta \right)\hfill & \text{f}or& p=11\overline{1},\hfill \end{array}$$
(14)
we assign them to each sublattice. Each $`T_p`$ operator has eigenvalues $`\pm t_1`$ ($`t_1=6.102`$) and doubly degenerated $`0`$. The eigenstates of eigenvalues $`\pm t_1`$ are connected to each other by the time-reversal operation, and so are two degenerate states of eigenvalue 0. For example, the eigenstate of eigenvalue $`t_1`$ for $`T_{111}`$ is explicitly written as
$`|t_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}\text{e}^{\text{i}\left(\theta _{111}\frac{\pi }{2}\right)}[|++\text{e}^{\text{i}\frac{\pi }{4}}|+]`$ (15)
$`{\displaystyle \frac{1}{2}}[|+\text{e}^{\text{i}\frac{3\pi }{4}}|],`$
with $`\theta _{111}`$ being an angle between vector $`(1,1,1)`$ and the $`z`$ axis, that is, $`\mathrm{cos}\theta _{111}=\sqrt{1/3}`$, $`\mathrm{sin}\theta _{111}=\sqrt{2/3}`$. This state is different from a ground state assumed by Lovesey et al, who considered the state deviating from $`\mathrm{\Gamma }_8`$ quartet.Lovesey et al. (2003) Using the eigenstates as bases, $`T_p`$ is represented as
$$T_p=\left(\begin{array}{cccc}t_1& 0& 0& 0\\ 0& +t_1& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right).$$
(16)
The ground state is given by assigning either of eigenstates of $`\pm t_1`$ to each sublattice; which eigenstate is relevant depends on the sign of acting mean field. As shown in Fig. 3, one longitudinal order and two transverse orders are possible in the triple-k ordering; for the longitudinal one, eigenstates of $`T_{111}`$, $`T_{\overline{1}11}`$, $`T_{1\overline{1}1}`$ and $`T_{11\overline{1}}`$ are assigned to sublattices 1, 3, 4 and 2, respectively; for two transverse orders, eigenstates of of $`T_{111}`$, $`T_{\overline{1}11}`$, $`T_{1\overline{1}1}`$, and $`T_{11\overline{1}}`$, are assigned to sublattices 1, 2, 3 and 4, and to 1, 4, 2 and 3, respectively.
Introducing the quadrupole operators,
$$Q_p=\{\begin{array}{ccc}\frac{1}{\sqrt{3}}\left(Q_{yz}+Q_{zx}+Q_{xy}\right)\hfill & \text{f}or& p=111,\hfill \\ \frac{1}{\sqrt{3}}\left(Q_{yz}Q_{zx}Q_{xy}\right)\hfill & \text{f}or& p=\overline{1}11,\hfill \\ \frac{1}{\sqrt{3}}\left(Q_{yz}+Q_{zx}Q_{xy}\right)\hfill & \text{f}or& p=1\overline{1}1,\hfill \\ \frac{1}{\sqrt{3}}\left(Q_{yz}Q_{zx}+Q_{xy}\right)\hfill & \text{f}or& p=11\overline{1},\hfill \end{array}$$
(17)
we can construct the quadrupole ordering state by assigning them to each sublattice in the same way as for octupole orderings. Since $`Q_p`$โs and $`T_p`$โs are simultaneously diagonalized because of commuting with each other, $`Q_p`$ could be represented as
$$Q_p=\left(\begin{array}{cccc}q_1& 0& 0& 0\\ 0& q_1& 0& 0\\ 0& 0& +q_1& 0\\ 0& 0& 0& +q_1\end{array}\right),$$
(18)
with $`q_1=8.273`$.
Let the octupole ordering be primarily realized. Then, each Np ion is in the eigenstate of the eigenvalue $`t_1`$ (or $`t_1`$). Since the state is also the eigenstate of the eigenvalue $`q_1`$, the quadrupole ordering is simultaneously induced. On the other hand, if the quadrupole order is primary, each Np ion is in the eigenstate of the eigenvalue $`q_1`$ or $`q_1`$. For the case of eigenvalue $`q_1`$, two eigensates are to be degenerate and give eigenvalues $`t_1`$ and $`t_1`$ to the octupole moment $`T_p`$, and thereby the net octupole moment becomes zero. For the case of $`q_1`$, two eigenstates are also to be degenerate and give the eigenvalue 0 to $`T_p`$. In either case, the quadrupole order carries no octupole order.
#### III.2.2 dipole ordering
Although the dipole ordering is ruled out in NpO<sub>2</sub>, it may be interesting to discuss here what happens in the dipole ordering. Introducing the dipole operators,
$$J_p=\{\begin{array}{ccc}\frac{1}{\sqrt{3}}\left(J_x+J_y+J_z\right)\hfill & \text{f}or& p=111,\hfill \\ \frac{1}{\sqrt{3}}\left(J_xJ_yJ_z\right)\hfill & \text{f}or& p=\overline{1}11,\hfill \\ \frac{1}{\sqrt{3}}\left(J_x+J_yJ_z\right)\hfill & \text{f}or& p=1\overline{1}1,\hfill \\ \frac{1}{\sqrt{3}}\left(J_xJ_y+J_z\right)\hfill & \text{f}or& p=11\overline{1},\hfill \end{array}$$
(19)
we can construct the dipole ordering state by assigning them to each sublattice in the same way as in the octupole ordering. Note that $`J_p`$ and $`Q_p`$ are simultaneously diagonalized, because both operators commute with each other. Within the bases of simultaneous eigenstates of $`J_p`$ and $`Q_p`$, the relevant operators are represented as
$`J_p`$ $`=`$ $`\left(\begin{array}{cccc}j_1& 0& 0& 0\\ 0& +j_1& 0& 0\\ 0& 0& j_2& 0\\ 0& 0& 0& +j_2\end{array}\right),`$ (24)
$`Q_p`$ $`=`$ $`\left(\begin{array}{cccc}q_1& 0& 0& 0\\ 0& q_1& 0& 0\\ 0& 0& +q_1& 0\\ 0& 0& 0& +q_1\end{array}\right),`$ (29)
$`T_p`$ $`=`$ $`\left(\begin{array}{cccc}0& t_1& 0& 0\\ t_1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),`$ (34)
where $`j_1=3.27`$, $`j_2=0.18`$ with parameters given in NpO<sub>2</sub>. The magnetic moment is evaluated on either of eigenstates of $`\pm j_1`$: $`L_p+2S_p=2.48`$ ($`L_p`$ and $`S_p`$ are defined as in the same way as $`J_p`$).
In the dipole ordering, the ground state is given by assigning one of the eigenstates of $`J_p`$โs to each sublattice. Since $`j_1`$ is much larger than $`j_2`$, the ground state is likely to be either of eigenstates of $`\pm j_1`$. It is obvious from Eqs. (29) and (34) that the dipole ordering induces the quadrupole moment but no octupole moment. Note that, if the quadrupole ordering is primary, no dipole moment is induced, because the doubly-degenerate eigenstates of $`Q_p`$ are the eigenstates of $`\pm j_1`$ of $`J_p`$.
### III.3 RXS spectra
Irrespective of whether the octupole or quadrupole ordering is realized, RXS amplitudes are generated at each site, according to Eq. (3). They are proportional to $`q_1\alpha _2(\omega )(P_3+P_4+P_5)`$ for the simultaneous eigenstate of $`T_{111}`$ and $`Q_{111}`$, to $`q_1\alpha _2(\omega )(P_3P_4P_5)`$ for the simultaneous eigenstate of $`T_{\overline{1}11}`$ and $`Q_{\overline{1}11}`$, to $`q_1\alpha _2(\omega )(P_3+P_4P_5)`$ for the simultaneous eigenstate of $`T_{1\overline{1}1}`$ and $`Q_{1\overline{1}1}`$, and to $`q_1\alpha _2(\omega )(P_3P_4+P_5)`$ for the simultaneous eigenstate of $`T_{11\overline{1}}`$ and $`Q_{11\overline{1}}`$. On the scattering vector $`\text{G}=(hh\mathrm{})`$ with $`h+\mathrm{}=odd`$, these amplitudes are summed up with a positive sign for sublattices 1 and 2 and with a negative sign for sublattices 3 and 4. Therefore, the total RXS amplitude becomes proportional to $`q_1\alpha _2(\omega )P_5`$ for the longitudinal order, while they are proportional to $`q_1\alpha _2(\omega )P_3`$ and $`q_1\alpha _2(\omega )P_4`$ for the two transverse orders. Note that a similar analysis is applied to the dipole ordering. In this case, both the dipole and quadrupole terms contribute to the amplitude. These results are summarized in Table 2. For the transverse case, our present treatment could be extended applying to the RXS spectra detected at Np $`M_4`$ edges in U<sub>0.75</sub>Np<sub>2</sub>O<sub>2</sub>.Wilkins et al. (2004) In this compound, the spectra may be interpreted as a consequence brought about by the transverse type of triple-k AFO ordering driven by the same ordering pattern at U sites.
Polarization dependences become particularly simple for $`\text{G}=(00\mathrm{})`$ ($`\mathrm{}=`$ odd) in the octupole and quadrupole orderings. They are explicitly written in the scattering geometry shown in Fig. 1 as $`P_3=0`$, $`P_4=0`$, $`P_5=(\sqrt{3}/2)\mathrm{sin}2\psi `$ in the $`\sigma \sigma ^{}`$ channel, while $`P_3=(\sqrt{3}/2)\mathrm{cos}\theta \mathrm{cos}\psi `$, $`P_4=(\sqrt{3}/2)\mathrm{cos}\theta \mathrm{sin}\psi `$, $`P_5=(\sqrt{3}/2)\mathrm{sin}\theta \mathrm{cos}2\psi `$ in the $`\sigma \pi ^{}`$ channel. Figure 4 shows the azimuthal-angle dependence of the spectra $`\text{G}=(003)`$ in comparison with the experiment. Paixรฃo et al. (2002); Caciuffo et al. (2003) The experimental data are well fitted by $`\mathrm{sin}^22\psi `$ in the $`\sigma `$-$`\sigma ^{}`$ channel, and $`\mathrm{sin}^2\theta \mathrm{cos}^22\psi `$ in the $`\sigma `$-$`\pi ^{}`$ channel. The two transverse orders cannot reproduce the experimental curves, as seen from panel (b). Paixรฃo et al. and Caciuffo et al. analyzed their experimental data and concluded that the longitudinal order gives rise to this dependence.Paixรฃo et al. (2002); Caciuffo et al. (2003) The present analysis confirms their result. Note that, based on a group theoretical point of view, Nikolaev and Michel have obtained the same result.Nikolaev and Michel (2003)
Now we discuss the energy profiles. In order to calculate them, we need the wavefunctions in the intermediate state. We first evaluate the Slater integrals for the Coulomb interaction and the SOI parameters within the HFA, which are shown in Table 3. These values are reduced by taking account of screening effects. The reduction factors are set the same as in the ground state. The Hamiltonian of the intermediate state, consisting of the full intra-atomic Coulomb interactions between $`5f`$-$`5f`$, $`5f`$-$`3d`$ and $`3d`$-$`3d`$ electrons as well as the SOI of $`5f`$ and $`3d`$ electrons, is represented by $`1001\times (2j_d+1)`$ microscopic states with the total angular momentum of the core hole $`j_d=3/2`$ and $`5/2`$ corresponding to the $`M_4`$ and the $`M_5`$ edges, respectively. Diagonalizing the Hamiltonian matrix, we obtain multiplet structures in the intermediate state. The $`\alpha _2(\omega )`$ is calculated by using Eq. (A.8).
The energy profile is proportional to $`|\alpha _2(\omega )|^2`$ in the octupole ordering phase. The calculated spectra around $`M_4`$ and $`M_5`$ edges are displayed with several choices of $`\mathrm{\Gamma }`$ values in Fig. 5. The origin of the energy is adjusted such that the peak of the RXS spectrum is located at the experimental peak position. Since there is no reliable estimation for the $`\mathrm{\Gamma }`$ value, we choose three typical values $`\mathrm{\Gamma }=0.01,0.5`$ and $`2.0`$ eV. The spike-like curves with $`\mathrm{\Gamma }=0.01`$ eV directly reflect the multiplet splittings of the intermediate states. For the $`M_4`$ edge, the choice $`\mathrm{\Gamma }=0.5`$ eV makes a multi-peak-structure line-shape. It merges into a single-peak structure around $`\mathrm{\Gamma }1.0`$ eV. The choice $`\mathrm{\Gamma }=2.0`$ eV corresponds to one of better fittings with the experimental line shape.Paixรฃo et al. (2002); Caciuffo et al. (2003) The core-level energy is adjusted such that the calculated peak at the $`M_4`$ edge with $`\mathrm{\Gamma }=2`$ eV coincides with the experimental one. Paixao et al. reported that the line shape is well fitted by a Lorentzian-squared rather than a Lorentzian one.Paixรฃo et al. (2002) As shown above, the line shape is basically determined by the multiplet structure, which is smeared by the life-time broadening. Whether it looks Lorentzian-squared or Lorentzian seems unimportant. As for the spectra at the $`M_5`$ edge, their shape depends rather sensitively on the value of $`\mathrm{\Gamma }`$ compared to that at the $`M_4`$ edge.
The energy profile in the dipole ordering is given by the sum of the dipole and quadrupole terms. However, $`|\alpha _1(\omega )|^2`$ is about two orders of magnitude larger than $`|\alpha _2(\omega )|^2`$. For instance, $`|\alpha _1(\omega )|^2192\times |\alpha _2(\omega )|^2`$ when $`\mathrm{\Gamma }=2.0`$ eV. Thus the dipole term usually dominates the quadrupole term. Although the dipole ordering is ruled out from experiments, we show $`|\alpha _1(\omega )|^2`$ in Fig. 6 as a reference. The peak at the $`M_4`$ edge with $`\mathrm{\Gamma }=2`$ eV is at 3847.5 eV, 0.7 eV higher than the peak position of $`|\alpha _2(\omega )|^2`$. Note that the spectral shape at the $`M_5`$ edge depends on $`\mathrm{\Gamma }`$ more sensitive than that at the $`M_4`$ edge.
### III.4 Absorption coefficient
The absorption coefficient is proportional to $`\mathrm{Im}\alpha _0(\omega )`$. We calculate $`\alpha _0(\omega )`$ from Eq. (A.4) in the same way as in the calculation of $`\alpha _1(\omega )`$ and $`\alpha _2(\omega )`$. The calculated results are shown in Fig. 7 at the $`M_4`$ and $`M_5`$ edges. The present calculation confirms the previous multiplet calculation by Lovesey et al., in which $`\text{Im}\alpha _0(\omega )`$ has been calculated at the $`M_\text{4}`$ edge for $`\mathrm{\Gamma }=0.7`$ eV.Lovesey et al. (2003) With increasing values of $`\mathrm{\Gamma }`$, the multiplet structure merges into a single peak. The peak position at the $`M_4`$ edge with $`\mathrm{\Gamma }=2`$ eV is about $`0.35`$ eV higher than that in $`|\alpha _2(\omega )|^2`$.
## IV Concluding remarks
In this paper, we have studied the RXS spectra at the Np M<sub>4,5</sub> edges in the triple-k multipole ordering phase of NpO<sub>2</sub>, on the basis of a localized electron model. We have derived an expression of scattering amplitudes in the $`E`$1 process, assuming that the rotational invariance is preserved in the intermediate states of the scattering process. This is a reasonable assumption when the multiplet energy is larger than those of the CEF and the intersite interaction. On the basis of this expression, we have analyzed the RXS spectra in NpO<sub>2</sub>. Assuming the $`\mathrm{\Gamma }_8`$-quartet ground state, we have constructed the triple-k ordering ground state. The energy profiles have been calculated by taking full account of the multiplet structure in the intermediate state, in agreement with the experiment.
RXS signals on multipole ordering superlattice spots have also been observed and analyzed at L<sub>2,3</sub> edges of rare-earth metals in their compounds such as CeB<sub>6</sub> and DyB<sub>2</sub>C<sub>2</sub>.Nakao et al. (2001); Yakhou et al. (2001); Tanaka et al. (1999); Hirota et al. (2000); Matsumura et al. (2005); Nagao and Igarashi (2001); Igarashi and Nagao (2002, 2003) The intermediate state is created by the transition from the $`2p`$-core to $`5d`$ states. Since the $`5d`$ states are considerably delocalized with forming energy bands, the assumption that the intermediate state preserves the rotational invariance becomes less accurate. An extension of the formula is left in future study.
###### Acknowledgements.
We thank M. Yokoyama and M. Takahashi for valuable discussions. This work was partially supported by a Grant-in-Aid for Scientific Research from the Ministry of Education, Science, Sports and Culture, Japan.
## Appendix A Derivation of Eq. (3)
We derive a general expression of RXS amplitude under the assumption that the intermediate state keeps the rotational symmetry at each site. The following derivation emphasizes the multiplet structure in the intermediate state. Thereby it is more general than the previous analyses, in which the fast collision approximation was adopted by replacing the multiplets with a single level.Hannon et al. (1989); Luo et al. (1993); Lovesey and Balcar (1996) A part of the results found in this Appendix were used in Ref. Nagao and Igarashi, 2005 when we analyzed the RXS spectra form URu<sub>2</sub>Si<sub>2</sub>.
Let the core hole be created at site $`j`$ in the intermediate state. We express the intermediate state as $`|\mathrm{\Lambda }=|J^{},M,i`$, where the magnitude $`J^{}`$ and the magnetic quantum number $`M`$ of total angular momentum (including a core-hole angular momentum) are good quantum numbers. To distinguish multiplets having the same $`J^{}`$ value but having the different energy, we introduce the index $`i`$. Defining $`M_{\alpha \alpha ^{}}`$ by $`M_j(\mathit{ฯต}^{},\mathit{ฯต},\omega )=_{\alpha \alpha ^{}}ฯต_\alpha ^{}ฯต_\alpha ^{}M_{\alpha \alpha ^{}}(j,\omega )`$, we rewrite Eq. (2) as
$`M_{\alpha \alpha ^{}}(j,\omega )`$ $`=`$ $`{\displaystyle \underset{J^{},M,i}{}}E_i(\omega ,J^{})\psi _0|x_{\alpha ,j}|J^{},M,i`$ (35)
$`\times J^{},M,i|x_{\alpha ^{},j}|\psi _0,`$
with
$$E_i(\omega ,J^{})=\frac{1}{\mathrm{}\omega (E_{J^{},i}E_0)+i\mathrm{\Gamma }}.$$
(36)
Assuming that the ground-state wavefunction is expressed as a linear combination of $`|J,m`$ at each site,
$$|\psi _0=\underset{m}{}c_j(m)|J,m,$$
(37)
and inserting this equation into Eq. (35), we obtain
$$M_{\alpha \alpha ^{}}(j,\omega )=\underset{m,m^{}}{}c_j^{}(m)c_j(m^{})M_{\alpha \alpha ^{}}^{m,m^{}}(\omega ),,$$
(38)
with
$`M_{\alpha \alpha ^{}}^{m,m^{}}(\omega )`$ $`=`$ $`{\displaystyle \underset{J^{}}{}}{\displaystyle \underset{i=1}{\overset{N_J^{}}{}}}E_i(\omega ,J^{}){\displaystyle \underset{M=J^{}}{\overset{J^{}}{}}}`$ (39)
$`\times J,m|x_\alpha |J^{},M,iJ^{},M,i|x_\alpha ^{}|J,m^{},`$
where the number of the multiplets having the value $`J`$ is denoted by $`N_J`$. We have suppressed the index $`j`$ specifying the core-hole site. The selection rule for the $`E`$1 process confines the range of the summation over $`J^{}`$ to $`J^{}=J,J\pm 1`$. The matrix element of the type $`J,m|x_\alpha |J^{},M`$ is analyzed by utilizing the Wigner-Eckart theorem for a vector operator with the use of the Wignerโs $`3j`$ symbol;Tinkham (1964)
$$J,m|s_\mu |J^{}M=(1)^{J^{}+m1}\left(\begin{array}{ccc}J^{}& 1& J\\ M& \mu & m\end{array}\right)(JV_1J^{})$$
(40)
with $`s_{\pm 1}=(1/\sqrt{2})(x\pm iy)`$, $`s_0=z`$. The symbol $`(JV_1J^{})`$ denotes the reduced matrix element of the set of irreducible tensor operator of the first rank. Because of the nature of the dipole operators, $`M^{m,m^{}}(\omega )\text{0}`$ only when $`|mm^{}|2`$. After lengthy calculation, we obtain
$`M_{\alpha ,\alpha ^{}}^{m,m}(\omega )`$ $`=`$ $`\left[{\displaystyle \frac{1}{3}}J(J+1)m^2\right]\alpha _2(\omega )M_{\alpha ,\alpha ^{}}^{3z^2r^2}`$ (41a)
$``$ $`\text{i}m\alpha _1(\omega )M_{\alpha ,\alpha ^{}}^z+\alpha _0(\omega )\delta _{\alpha ,\alpha ^{}},`$
$`M_{\alpha ,\alpha ^{}}^{m,m+1}(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{2}}f_m(2m+1)\alpha _2(\omega )(M^{zx}+\text{i}M^{yz})_{\alpha ,\alpha ^{}}`$ (41b)
$``$ $`\text{i}{\displaystyle \frac{1}{2}}f_m\alpha _1(\omega )(M^x+\text{i}M^y)_{\alpha ,\alpha ^{}},`$
$`M_{\alpha ,\alpha ^{}}^{m+1,m}(\omega )`$ $`=`$ $`{\displaystyle \frac{1}{2}}f_m(2m+1)\alpha _2(\omega )(M^{zx}\text{i}M^{yz})_{\alpha ,\alpha ^{}}`$ (41c)
$``$ $`\text{i}{\displaystyle \frac{1}{2}}f_m\alpha _1(\omega )(M^x\text{i}M^y)_{\alpha ,\alpha ^{}},`$
$`M_{\alpha ,\alpha ^{}}^{m,m+2}(\omega )`$ $`=`$ $`a_m^{\prime \prime }\alpha _2(\omega )(M^{x^2y^2}+\text{i}M^{xy})_{\alpha ,\alpha ^{}},`$ (41d)
$`M_{\alpha ,\alpha ^{}}^{m+2,m}(\omega )`$ $`=`$ $`a_m^{\prime \prime }\alpha _2(\omega )(M^{x^2y^2}\text{i}M^{xy})_{\alpha ,\alpha ^{}},`$ (41e)
where
$`f_m`$ $`=`$ $`\sqrt{(Jm)(J+m+1)},`$ (42)
$`a_m^{\prime \prime }`$ $`=`$ $`{\displaystyle \frac{1}{2}}f_mf_{m+1},`$ (43)
and the $`3\times 3`$ matrices, $`M^x`$, $`M^y`$, $`M^z`$, $`M^{xy}`$, $`M^{yz}`$, $`M^{zx}`$, $`M^{x^2y^2}`$, and $`M^{3z^2r^2}`$ are tabulated in Table 4.
The energy profiles are given by
$`\alpha _0(\omega )`$ $`=`$ $`{\displaystyle \frac{2}{3}}J(2J1)F_{J1}(\omega )+{\displaystyle \frac{2}{3}}J(J+1)F_J(\omega )`$ (44a)
$`+`$ $`{\displaystyle \frac{2}{3}}(2J^2+5J+3)F_{J+1}(\omega ),`$
$`\alpha _1(\omega )`$ $`=`$ $`(2J1)F_{J1}(\omega )F_J(\omega )`$ (44b)
$`+`$ $`(2J+3)F_{J+1}(\omega ),`$
$`\alpha _2(\omega )`$ $`=`$ $`{\displaystyle \frac{4}{3}}\left[F_{J1}(\omega )+F_J(\omega )F_{J+1}(\omega )\right],`$ (44c)
with
$$F_J^{}(\omega )=()^{JJ^{}}|(JV_1J^{})|^2\underset{i=1}{\overset{N_J^{}}{}}E_i(\omega ,J^{}).$$
(45)
Substituting Eqs. (41) into Eq. (38), we obtain the final expression Eq. (3).
|
warning/0506/hep-ph0506151.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
Multiple astrophysical experiments have measured the presence of Dark Matter (DM). The leading candidate for this DM is a particle. dmreview Existing studies have concentrated on a single particle providing the DM density of the universe but multiple particles are allowed. This knowledge should lead to a systematic search for invisible decays of particles and mesons known to exist. However, the only particles with reported invisible branching ratios or limits are the $`\pi ^0`$ and $`Z`$pdg
There are at least two pieces of evidence that the DM component of the universe may be lighter than the Minimal Supersymmetric Standard Model (MSSM) or minimal supergravity mediated supersymmetry breaking (mSUGRA) lightest neutralino that dominates DM studies. First, recent measurements of 511 keV gamma rays from the galactic center indicate a Gaussian profile of low-velocity positrons. Jean:2003ci Traditional MSSM or mSUGRA neutralinos are heavy, and their annihilation would produce too many high-energy $`\gamma `$-rays from neutral pions which decay to photons, as well as significant bremsstrahlung as their decay products slow until they are nearly at rest as is required explain the 511 keV line. This places such a scenario in strong conflict with EGRET upper limits on the higher-energy gamma flux. Strong:2004de ; Beacom:2004pe Thus, if a DM particle is responsible for the 511 keV line, it must be lighter than approximately 100 MeV. Second, recent analysis of DM flows and caustics indicate that the CDMS limit and DAMA evidence for DM can be compatible due to the lower detection threshold of DAMA. dmflows This effect is also enhanced if there is a flow of DM through our solar system.
Neutralinos in the general MSSM must have $`M_{\chi ^0}>6`$ GeV to obtain an appropriate relic density. Bottino:2002ry Constraints on even lighter DM comes from CUSB, which measured $`\mathrm{{\rm Y}}\gamma +\mathrm{invisible}`$ signals, giving the best sensitivity to DM lighter than approximately $`1.5\text{GeV}`$, and losing sensitivity as the DM mass increases due to the soft spectrum of Initial State Radiation (ISR) photons. ups1stogammainvis Modern b-factories can improve in this measurement by at least an order of magnitude.mySinglinoDM Searches for invisibly-decaying Higgs Bosons are sensitive if the Higgs is heavy, has significant coupling to the $`Z`$, and the Higgs decays dominantly to dark matter. invisiblehiggs Finally, the LEP single-photon counting measurements limit arbitrary Standard Model (SM)-DM interactions; however, the $`Z`$ invisible width dominates these measurements at LEP energies, and these experiments have no sensitivity if the SM-DM mediator does not couple to the electron. lepsinglephoton
These constraints can all be avoided in many models that are not the MSSM. Two attractive possibilities that can explain the above data are (1) a light neutralino with couplings to the SM mediated by a light scalar singlet. This may occur in the Next-to-Minimal Supersymmetric Model (NMSSM) and related models which solve the $`\mu `$ problem with a singlet; nmssm ; mySinglinoDM (2) light scalar DM coupled to the Standard Model through a new gauge boson $`U`$Boehm:2003hm ; morelightdm ; Fayet:1991ux In both cases there must be some small coupling to the Standard Model in order to avoid having the DM density over-close the universe. dmreview It is in general possible to couple the DM preferentially to some quarks and/or leptons but not others. This can result in invisible decays of some hadrons of a given spin $`J`$ and Charge$`\times `$Parity (CP) eigenvalue but not others, and little or no signal at direct detection experiments. Building such models is straightforward, and several already exist in the literature. nmssm ; Boehm:2003hm Our purpose in this letter is not to build such models, but point out several measurements that can be performed at colliders that are sensitive to light DM.
At $`e^+e^{}`$ detectors like BaBar, Belle, and CLEO, one can use ISR to explore energy regions below the nominal collider energy. isrtheory ; isrexpt With one ISR photon, these experiments deliver the same luminosity between 9.85 GeV and 10.58 GeV as they do from interactions without an ISR photon at their nominal center of mass energy, 10.58 GeV. In order to identify the presence of a bottomonium state without observing its decay, one must require an ISR photon of a specific energy and/or a radiative decay. A radiative decay is any transition from one quarkonium state to another. We present several techniques which can be used to suppress backgrounds when the ISR photon is lost because it is outside the detector acceptance. Radiative decays from the $`\mathrm{{\rm Y}}(4S)`$ will have similar statistics to ISR production of lower $`\mathrm{{\rm Y}}`$ resonances, but no radiative decays of the $`\mathrm{{\rm Y}}(4S)`$ have yet been discovered.
Another measurement that can be performed at B-factories is in $`bs`$ transitions such as $`B^+K^++invisible`$ and is sensitive to dark matter with masses up to 2.4 GeV, but is not sensitive to the $`J^{CP}`$ of the mediator.bsme This branching ratio may be 50 times larger than is expected from the Standard Model process with neutrinos.
The Standard Model expectation for $`\mathrm{{\rm Y}}`$ to invisible is $`\mathrm{\Gamma }(\mathrm{{\rm Y}}\nu \overline{\nu })=4.14\times 10^4\mathrm{\Gamma }(\mathrm{{\rm Y}}e^+e^{})1\times 10^5`$ with a theoretical uncertainty of only $`23\%`$, and is sensitive to the bottom squark mass and R-parity violation in SUSY theories. sminvisible
Expectations for branching ratios of hadrons into DM may be as large as a few percent.
## II Sources of Dark Matter
We assume that DM couples to the Standard Model through some mediating boson. On general model-independent grounds we expect this particle to be either a vector, scalar, or pseudo-scalar. If the mediator is a scalar or pseudo-scalar, only a $`SU(2)`$ doublet can couple to $`b\overline{b}`$ by gauge invariance at dimension 4. This scalar doublet will generally mix with the Standard Model Higgs, or the CP-odd $`A`$ of a Two Higgs Doublet Model. Therefore, we expect scalar and pseudo-scalar mediated DM to show up dominantly in interactions with heavy fermions such as $`b`$ quarks.
If the mediator is a vector gauge boson, giving $`b`$ and $`\overline{b}`$ equal and opposite charge under the new gauge group (which we assume to be a $`U(1)`$) is sufficient to introduce the proper couplings. Fayet:1991ux ; Boehm:2003hm One need not expect this gauge boson to couple to all fermions in the Standard Model equally. One might expect that the first two generations are not charged under this new group, in order to be consistent with precise measurements of the muon and electron anomalous magnetic moment, as well as the lack of unexplained vector resonances in hadronic data. This situation would result in extremely small DM-nucleon cross sections for direct detection experiments.
In order to get small DM masses in the MSSM and other models, one generally has to also bring down another particle mass for the purpose of getting a large enough annihilation or coannihilation cross section. However, in the case of DM masses less than $`M_\mathrm{{\rm Y}}/2=4.73`$ GeV that we consider, one generally cannot bring down a particle charged under the SM gauge groups without violating existing experimental constraints.<sup>1</sup><sup>1</sup>1The usual particles that are made light in supersymmetric models are the stau or a Higgs. The stau is often the next-to-lightest supersymmetric particle and undergoes t-channel coannihilation with the LSP. The Higgs mediates s-channel annihilation when there is significant higgsino or wino fraction in the LSP. Having a light bottom squark may be one exception to this assumption; Berger:2000mp however, a recent re-analysis of available data indicates that this solution is now disfavored. Janot:2004cy
The direct constraint on the annihilation mediator is the reason why existing constraints Bottino:2002ry require $`M_{\chi ^0}>6`$ GeV, despite the fact that the theory is consistent with a massless neutralino. Gogoladze:2002xp If the mediator has some mixing with a pure singlet state, these constraints can be largely avoided.
A general argument predicts that DM should be heavier than 2 GeV. Lee:1977ua This is based on DM annihilation couplings that are proportional to $`G_F`$. However, there is nothing that requires DM to have something to do with weak bosons or electroweak symmetry breaking. The argument that DM couplings must be proportional to $`G_F`$ is based solely on the coincidence that the DM annihilation cross section (c.f. Eq. 3) is similar in size to weak cross sections. This may simply be a numeric coincidence and annihilation cross sections need not be proportional to $`G_F`$. If we simply assume that DM exists and $`m_\chi <6`$ GeV is an allowed region, we are forced to recognize that it must have picobarn cross sections with some Standard Model particle. These expected cross sections are explored in the following section.
## III Dark Matter Coupling Expectations
The required rate of DM annihilation can be naively estimated. We will not accurately compute the relic density since we are not proposing a specific DM model, but one can get an order of magnitude estimate for s-wave annihilation using dmreview ; pdg
$$\mathrm{\Omega }_Xh^2\frac{0.1\mathrm{pb}c}{\sigma v}.$$
(1)
Where $`\mathrm{\Omega }_X=\rho _X/\rho _c`$ is the relic density for species $`X`$ relative to the critical density $`\rho _c`$, $`h`$ is the Hubble constant, and $`\sigma v`$ is the thermally averaged annihilation cross section of the DM into Standard Model particles. Using the central value of the WMAP wmap result for $`\mathrm{\Omega }_Xh^2=0.113`$, we can invert this equation and solve for the required annihilation cross section for light relics
$$\sigma v=0.88\mathrm{pb}.$$
(2)
The velocity $`v`$ appearing here is the Mรธller velocity, the relative velocity of annihilating particles at the temperature they froze-out. The approximate temperature at freeze-out is $`T=m_\chi /x_{FO}`$ where $`m_\chi `$ is the mass of the DM and $`x_{FO}`$ is an expansion parameter evaluated at the freeze-out temperature that is $`x_{FO}2025`$ depending on the model. Thus the average momentum for a fermion is $`k_BT`$ and therefore the average relative velocity is roughly $`1/x_{FO}`$. For $`x=20`$ at freeze-out we have:
$$\sigma (\chi \chi SM)18\mathrm{p}\mathrm{b}.$$
(3)
The invisible branching ratio of a hadron can then be estimated by assuming that the time-reversed reaction is the same, $`\sigma (f\overline{f}\chi \chi )\sigma (\chi \chi f\overline{f})`$. This assumption holds if $`m_\chi m_f`$ and $`M_\mathrm{{\rm Y}}4m_\chi ^2+6m_\chi T_{FO}`$. We assume that the DM mediator is not flavor changing and that annihilation occurs in the $`s`$ channel. <sup>2</sup><sup>2</sup>2A $`t`$-channel mediator is possible, but this requires that the mediator carry color and electromagnetic charge, and therefore is unlikely if we consider $`m_\chi <5`$ GeV. Therefore, the best-motivated hadrons to have an invisible width are same-flavor quark-antiquark bound states (quarkonia). The CERN Yellow Report provides a thorough review of quarkonium physics. Brambilla:2004wf
The invisible width of a hadron composed dominantly of $`q\overline{q}`$ is given approximately by:
$$\mathrm{\Gamma }(H\chi \chi )=f_H^2M_H\sigma (q\overline{q}\chi \chi )$$
(4)
where $`f_H`$ is the hadronic form factor for the state $`H`$, and $`M_H`$ is the hadronโs mass.
We can predict an approximate expectation for the branching ratios for narrow states. Some of the most promising are:
$$BR(\mathrm{{\rm Y}}(1S)\chi \chi )0.41\%BR(J/\mathrm{\Psi }\chi \chi )0.023\%BR(\eta \chi \chi )0.033\%$$
(5)
Branching ratios for scalars and pseudo-scalars tend to be smaller since those states are wider. This estimate does not take into account kinematic factors arising from the mediator mass and DM mass. These factors can both enhance or suppress these branching ratios.
If a particular hadron $`H`$ decays invisibly, then that hadron must mix into the mediator $`M`$ before decaying into Dark Matter. If $`M`$ does not violate the discrete symmetries $`C`$ and $`P`$, $`H`$ and $`M`$ mix only if they share the same spin, $`C`$, and $`P`$ eigenvalues. Therefore, the observation of an invisible decay not only constrains the mass of the dark matter and mediator, but may also uniquely identify the spin, $`C`$, and $`P`$ of the mediator. An invisible decay does not have sensitivity to the spin of the Dark Matter itself.
With running B-factories BaBar and Belle having roughly 400 $`\mathrm{fb}^1`$ recorded, these experiments may already have tens of thousands of DM production events, if the DM is kinematically accessible.
## IV Bottomonium production via ISR
Bottomonia can be identified by observing the particles emitted when it makes radiative transitions to lighter bottomonia. These transitions are show in Fig. 1. Since the states are fairly narrow, the energy of the photon radiated in a transition (or kinematics of a particle pair) gives a clean way to select specific quarkonia transitions. The CLEO experiment has provided measurements of most of the quarkonia transitions. quarkonium The number of ISR $`\mathrm{{\rm Y}}`$ production events collected by the BaBar and Belle experiments is now competitive with that collected by the traditional method of scanning the resonance. Furthermore, off-peak data can be used for the measurements we propose.
The ISR cross section for a particular final state $`f`$, with $`e^+e^{}`$ cross section $`\sigma _f(s)`$ is to first order isrtheory :
$$\frac{d\sigma (s,x)}{dx}=W(s,x)\sigma _f(s(1x))$$
(6)
where $`x=\frac{2E_\gamma }{\sqrt{s}}`$, $`E_\gamma `$ is the energy of the ISR photon in the CM frame, and $`\sqrt{s}`$ is the CM energy. The function
$$W(s,x)=\beta \left[(1+\delta )x^{(\beta 1)}1+\frac{x}{2}\right]$$
(7)
describes the energy spectrum of the ISR photons, where $`\beta =\frac{2\alpha }{\pi x}(2\mathrm{ln}\frac{\sqrt{s}}{m_e}1)`$ and $`\delta `$ take into account vertex and self-energy corrections; $`\alpha `$ is the electromagnetic coupling constant and $`m_e`$ is the mass of the electron. At the $`\mathrm{{\rm Y}}(4S)`$ energy, $`\beta =0.088`$ and $`\delta =0.067`$. By tagging the ISR photon, B-factories can explore all of the vector $`b\overline{b}`$ bound states.
This function (Eq. 7) is strongly peaked in the forward and backward directions, so ISR photons will be close to the beamline. Assuming a detector acceptance of $`0.9<\mathrm{cos}(\theta )<0.9`$, the fraction of the nominal luminosity at the $`\mathrm{{\rm Y}}(4S)`$ resonance delivered to the $`\mathrm{{\rm Y}}(1S)`$, $`\mathrm{{\rm Y}}(2S)`$ and $`\mathrm{{\rm Y}}(3S)`$ resonances is $`1.9\times 10^5`$, $`3.2\times 10^5`$ and $`5.0\times 10^5`$ respectively. This results in hundreds of thousands of events per resonance with current recorded luminosities.
If one does not require that the ISR photon is identified because it is not in the detector acceptance, the production of lower resonances is larger by a factor $`5`$ to $`7`$.isrexpt The fraction of the nominal luminosity at the $`\mathrm{{\rm Y}}(4S)`$ delivered to the $`\mathrm{{\rm Y}}(1S)`$, $`\mathrm{{\rm Y}}(2S)`$, and $`\mathrm{{\rm Y}}(3S)`$ resonances is $`8.5\times 10^5`$, $`1.5\times 10^4`$, and $`2.3\times 10^4`$.
One can then tag quarkonium states by looking for a particular radiative transition such as $`\mathrm{{\rm Y}}(2S)\chi _{b0}(1P)\gamma `$ or $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(1S)\pi ^+\pi ^{}`$. One can examine the decay modes of the tagged state in a decay-mode independent manner.
## V The kinematics of ISR production with radiative decays
The irreducible physics background for these invisible decays coming from a pair of neutrinos and pions is extremely small due to the fact that weak cross sections are suppressed by $`(M_\mathrm{{\rm Y}}/M_W)^20.01`$, the final state has high multiplicity, and our signal has resonant enhancement. For example $`\sigma (e^+e^{}\pi ^+\pi ^{}\nu \overline{\nu })10^6`$ pb before applying any cuts. Therefore, the dominant backgrounds will come from unrelated processes that do not actually have a neutrino or DM, or neutrinos from $`\tau `$ decays.
The knowledge that resonances are formed in our signal gives us the kinematic constraint that the square of the four-momenta forming a resonance must be equal to the resonance mass-squared. For the production of a single $`\mathrm{{\rm Y}}`$ resonance via ISR and not observing the ISR photon, there are 2 kinematic variables that are undetermined. These variables are associated with the four-vectors of the ISR photon and $`\mathrm{{\rm Y}}`$ which we presume decays invisibly. With $`n`$ intermediate resonances decaying radiatively to each other, in the event that the ISR photon is unobserved, we can predict all but $`2n`$ of these undetermined variables. The most important radiative decays of bottomonium are listed in Appendix A, sorted by cross section. Up to two intermediate resonances can be created with sizeable event rates.
### V.1 Resonance Constraints
We can either observe the ISR photon, or require that its angle with respect to the beamline is consistent with it being outside the detector acceptance using the kinematic constraints.
With one intermediate resonance and one radiative decay (e.g. $`\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)\pi ^+\pi ^{}`$), we define $`M_1`$ (e.g $`M_{\mathrm{{\rm Y}}(2S)}`$) to be the mass of the intermediate resonance and $`M_2`$ (e.g. $`M_{\mathrm{{\rm Y}}(1S)}`$) to be the mass of the final invisibly-decaying state. We can predict all but one kinematic variable using the measurement of the particles emitted in the radiative transition and the beam constraint. In the center-of-mass frame:
$$E_{\mathrm{ISR}}=\frac{sM_1^2}{2\sqrt{s}},\mathrm{cos}\theta =\frac{\sqrt{s}}{p_r}\frac{M_1^2M_2^2+M_r^2}{sM_1^2}\frac{E_r}{p_r}\frac{s+M_1^2}{sM_1^2}$$
(8)
where $`p_r^\mu =(E_r;\stackrel{}{p}_r)`$ is the sum of the four momenta of all the particles emitted in the radiative transition, $`M_r^2=p_{r\mu }p_r^\mu `$, $`p_r=|\stackrel{}{p}_r|`$, and $`s=M_{\mathrm{{\rm Y}}(4S)}^2`$ is the center-of-mass energy. Our predicted angle $`\theta `$ is the angle between the ISR photon and $`\stackrel{}{p}_r`$. $`\mathrm{cos}\theta `$ can be related to $`\mathrm{\Delta }M^2=M_1^2M_2^2`$. However, when expressed as an angle it is clear that it can still be used when the ISR photon is unobserved.
We can also invert Eq. 8 to come up with a cut on the energy of the radiated system
$$\frac{M_1^2M_2^2}{2\sqrt{s}}<E_r<\frac{\sqrt{s}}{2}\left(1\frac{M_2^2}{M_1^2}\right)$$
(9)
which is useful for single photon transitions.
Backgrounds without resonances can be distributed outside the physical region $`1.0\mathrm{cos}\theta 1.0`$ since for background without resonances, Eq. 8 does not describe any physical angle at all. Therefore, a cut requiring $`1.0\mathrm{cos}\theta 1.0`$ will suppress most backgrounds by a factor $`10^210^3`$ (c.f. Table. 1). If the ISR photon is unobserved we also know that it must be outside the detectorโs acceptance. Assuming the EM calorimeter extends between $`\theta _{\mathrm{min}}`$ and $`\theta _{\mathrm{max}}`$ as seen from the center-of-mass-frame, the angle between the ISR photon and the beamline ($`\theta _{\mathrm{ISR}}`$) must satisfy $`\theta _{\mathrm{ISR}}<\theta _{min}`$ or $`\theta _{\mathrm{ISR}}>\theta _{max}`$. Since the angle with respect to the beamline $`\theta _r`$ of $`\stackrel{}{p}_r`$ is measured, this amounts to the restriction:
$$|\theta \theta _r|<\theta _{min}\mathrm{or}|\theta +\theta _r|>\theta _{max}$$
(10)
Furthermore, the signal peaks in both these regions as $`(\theta \pm \theta _r)^2`$. This corresponds to the ISR photon being nearly parallel to one of the beamlines.
If the ISR photon is observed both $`E_{\mathrm{ISR}}`$ and $`\mathrm{cos}\theta `$ are measured and can be directly compared to (8). The final kinematic variable can be taken to be the angle between the plane defined by the beamline and $`\stackrel{}{p}_r`$, and the plane defined by beamline and $`\stackrel{}{p}_{\mathrm{ISR}}`$. Only if the ISR photon is observed can this angle be determined. This angle will only provide power in suppressing background if the background happens to peak in this variable. If the ISR photon is unobserved, this angle is not knowable in principle.
When two intermediate resonances are formed, and two radiative decays occur, all kinematic variables can be determined by measuring the energy and momentum of the particles in the radiative decay. The constraints above still apply for the first transition. The new constraint available allows us to predict the angle between the ISR photon and the second radiative decay:
$$\mathrm{cos}\theta ^{}=\frac{\sqrt{s}}{|\stackrel{}{r}_2|}\frac{M_2^2M_3^2+r_2^2+2r_1r_2}{sM_1^2}\frac{E_2}{|\stackrel{}{r}_2|}\frac{s+M_1^2}{sM_1^2}$$
(11)
We can further apply the same trick as with $`\mathrm{cos}\theta `$ to require the ISR photon to be in the beam line, replacing $`\theta `$ with $`\theta ^{}`$ in Eq. 10.
It should be noted that it is possible to emit two or more ISR photons. In this case, Eqs. 8-11 are accurate in the limit that the ISR photons are collinear.
### V.2 Missing Momentum and Standard Model Decay Backgrounds
Considering that there are invisible particles in the final states we are interested in, there is an irreducible background when the final state decays to visible Standard Model particles, but those particles lie outside the detector acceptance. A missing momentum cut can force the transverse momentum of the final state to be large enough that one can ensure that its decay products would be in the detector volume. However, this is almost completely useless for machines running at the $`\mathrm{{\rm Y}}(4S)`$ for the following reason: assume the final state undergoes a 2-body decay, those decay products lie exactly on the edge of a symmetric CLEO-like detector at $`\theta _{\mathrm{min}}`$ and $`\theta _{\mathrm{max}}=\pi \theta _{\mathrm{min}}`$, and the final state is at rest at the center. The cut required on the transverse component of the sum of radiative transition particles is:
$$(p_r)_T>\frac{1}{2}(E_{CM}E_{ISR}E_r)\mathrm{sin}\theta _{min}2\mathrm{G}\mathrm{e}\mathrm{V},$$
(12)
which is larger than the visible energy ($`E_{ISR}+E_r`$) in any radiative transition. Here $`E_{\mathrm{CM}}`$ is the center of mass energy, $`E_{\mathrm{ISR}}`$ is the energy of the ISR photon, and $`E_r`$ is the sum of energies of particles emitted in the transition(s). This background is further discussed in Sec. VI.3.
For a collider running at $`๐ช(30GeV)`$, requiring the ISR photon to be visible provides enough transverse momentum that the decay products of the final state must lie in the detector acceptance. However, the ISR production cross section of $`\mathrm{{\rm Y}}`$โs is reduced by a factor $`100`$. This requirement also eliminates most of the two-photon background, as discussed in Sec. VI.1.
## VI Invisible Upsilon Decays
To demonstrate that invisible widths can be measured using ISR and radiative decays, we concentrate on the modes $`\mathrm{{\rm Y}}(nS)\mathrm{{\rm Y}}(1S)\pi ^+\pi ^{}`$ (n=2,3) for colliders running at the $`\mathrm{{\rm Y}}(4S)`$ since these modes have the largest cross section (2.91pb, 0.784 pb). Many decays<sup>3</sup><sup>3</sup>3For instance $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\gamma \gamma \mathrm{{\rm Y}}(1S)\gamma \gamma \pi ^+\pi ^{}`$, $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\pi ^0\pi ^0\mathrm{{\rm Y}}(1S)\pi ^0\pi ^0\pi ^+\pi ^{}`$, $`\mathrm{{\rm Y}}(3S)h_b(1P)\pi ^+\pi ^{}\eta _b(1S)\pi ^+\pi ^{}\gamma `$, $`\mathrm{{\rm Y}}(3S)\chi _{b0}(2P)\gamma \eta _b(1S)\gamma \eta `$, $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\pi ^+\pi ^{}\chi _{b0}(1P)\pi ^+\pi ^{}\gamma `$. have the visible topology $`\pi ^+\pi ^{}+n\gamma `$, $`n0`$, which might be useful for triggering. These modes seem the most promising since the running B-factories BaBar and Belle do not have the sensitivity in their calorimeter that CLEO had, and they have excellent charged particle tracking resolution by design. These transitions were measured by CLEO.isrexpt A table of possible transitions and their tagging signatures is presented in Appendix A. The $`\mathrm{{\rm Y}}(3S)`$ mode has harder pions, and therefore may have a higher reconstruction or triggering efficiency than the $`\mathrm{{\rm Y}}(2S)`$ mode.
For background simulations we have used PYTHIA pythia and CompHEP comphep with $`\tau `$ decays simulated with TAUOLA tauola . We have smeared charged tracks according to the BaBar detector resolution babartdr
$$\frac{\sigma _{p_t}}{p_t}=0.21\%1.4\%p_t,$$
(13)
charged tracks must have $`p_t>100`$ MeV, and all objects must lie within the detector $`0.87<\mathrm{cos}\theta <0.96`$ from the center-of-mass frame (i.e. BaBar geometry). For photons we require $`E>20`$ MeV and smear their energy according to
$$\frac{\sigma _E}{E}=1.2\%\frac{1.0\%}{\left(E/\mathrm{GeV}\right)^{1/4}},\sigma _\theta =\sigma _\varphi =2\mathrm{m}\mathrm{r}\frac{3\mathrm{m}\mathrm{r}}{\sqrt{E/\mathrm{GeV}}}.$$
(14)
Finally, for charged tracks we do not differentiate $`\pi ^+`$, $`e^+`$, $`\mu ^+`$ or $`K^+`$ and assign each charged track to have a mass $`m_{\pi ^+}`$ after smearing its momentum, since tracking information is reliable but particle ID is not. These tracks generally are soft enough that they do not enter the calorimeter, or enter at a grazing angle.
The cuts proposed in Sec. V.1 are drastic and in general can result in a background suppression of $`10^5`$ or more, including effects of detector resolution. Due to the large size of these backgrounds, detailed detector resolution effects and multiple scattering will be very important. Therefore, each background should be measured directly in order to estimate the signal contamination. Therefore, we present estimates of each background and the level to which they can be suppressed including smearing. The exact numbers may change significantly when detector effects are taken into account.
Aside from the purely kinematic constraints presented in Sec. V.1, there is no further angular information from the matrix element that is useful to identify the signal. In the case of a final state $`\mathrm{{\rm Y}}`$ from a di-$`\pi `$ transition, there is no spin correlation between the outgoing pions and the final $`\mathrm{{\rm Y}}`$. The operator involving a coupling between the polarization of the $`\mathrm{{\rm Y}}`$ and the momenta of the pions is D-wave suppressed and measured to be very small. y3s1spipi Furthermore, the $`\pi \pi `$ invariant mass spectrum in the $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(1S)\pi \pi `$ transition is not well understood theoretically and may involve another intermediate state. Guo:2004dt
Including all backgrounds and a realistic smearing of energy and momenta, the measurements proposed can limit at $`2\sigma `$ sensitivity
$$BR(\mathrm{{\rm Y}}(1S)\mathrm{invisible})<0.113\%$$
(15)
using the $`\mathrm{{\rm Y}}(2S)`$ mode and
$$BR(\mathrm{{\rm Y}}(1S)\mathrm{invisible})<0.335\%$$
(16)
using the $`\mathrm{{\rm Y}}(3S)`$ mode. The combined $`2\sigma `$ sensitivity is then
$$BR(\mathrm{{\rm Y}}(1S)\mathrm{invisible})<0.107\%.$$
(17)
The dominant backgrounds are discussed in the following subsections. Their numeric importance and cuts needed to suppress them is summarized in Table 1.
### VI.1 Photon Fusion Background
The two photon fusion process occurs when both incoming beams emit a photon and those photons annihilate into electrons, muons, taus, or hadrons. This cross section is very large, in the hundreds of nanobarns. Furthermore, our signal spans the region $`0<Q^2<1`$ GeV<sup>2</sup> in which non-perturbative QCD effects dominate hadron production. Due to this, a reliable simulation of hadron production is not possible and in any case should not be relied upon due to non-perturbative effects. This background must be measured directly.
To demonstrate that this background can be overcome, we simulate 10 fb<sup>-1</sup> of the lepton<sup>4</sup><sup>4</sup>4Here lepton refers to an $`e`$ or $`\mu `$. production processes $`e^+e^{}e^+e^{}l^+l^{}`$, $`e^+e^{}e^+e^{}+\mathrm{hadrons}`$ and $`e^+e^{}e^+e^{}l^+l^{}\gamma `$. We simulate the first two backgrounds using PYTHIA, and the second using CompHEP <sup>5</sup><sup>5</sup>5The $`\gamma \gamma l^+l^{}\gamma `$ process is not included in PYTHIA. in the equivalent photon approximation. Budnev:1974de At these low energies, $`\pi /\mu `$ separation is generally unreliable since the muon is not energetic enough to reach the outer muon detector. The $`l^+l^{}`$ cross section is also about an order of magnitude larger than the $`\pi ^+\pi ^{}`$ cross section, making the $`l^+l^{}`$ the most important background in any case. The $`\mu ^+\mu ^{}`$ cross section is 38.5 nb for the BaBar detector geometry, assuming both muons are visible in the detector volume. This is sufficiently large that it overwhelms the physics signal that is normally triggered on at the B-factories (about 1 nb). Therefore, the rate must be reduced at the trigger level. Requiring an extra visible photon reduces this cross section to a triggerable level. In the case of the $`e^+e^{}e^+e^{}l^+l^{}`$ background, an extra photon comes from Initial/Final State Radiation.
Di-lepton events produced in photon fusion have the characteristic that the leptons are back-to-back in the plane perpendicular to the beamline. By contrast, our signal is a 3-body decay, so only a small fraction are back-to-back. Therefore, $`\stackrel{}{p}_r=0`$ and $`\mathrm{cos}\theta `$ (c.f. Eq. 8) will be very large. As can be seen in Table 1 a cut on $`\mathrm{cos}\theta `$ alone can remove this background. If detector resolution effects cause this background to bleed into the signal region, a cut $`\mathrm{\Delta }\varphi _{ll}<\pi `$, where $`\varphi `$ is the angle between the leptons in the plane perpendicular to the beamline can also remove this background. Therefore, this background should be carefully studied. This can be done by identifying singly-tagged photon fusion events, where one of the initial electrons is deflected into the detector. Experiments such as DELPHI have also employed far-forward particle detectors to identify the photon fusion signal when the electrons are deflected by a small angle (known as the Small Angle Tagger and Very Small Angle Tagger).
### VI.2 di-$`\tau `$ Background
The only irreducible physics background to these processes that have true neutrinos comes from $`\tau `$ decays. The dominant source is $`e^+e^{}\tau ^+\tau ^{}`$ via a virtual photon, where both $`\tau `$โs decay to pions. The total $`\tau ^+\tau ^{}`$ cross section is 993 pb at tree level. $`\tau `$โs can also be produced in photon fusion, with a cross-section of approximately 22 pb.
For example, we consider the di-$`\tau `$ background to the process $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(1S)\pi ^+\pi ^{}`$. We require exactly two charged tracks and one photon visible in the detector. Generally the photon comes from a $`\pi ^0`$ decay in which the other photon is outside the detectorโs acceptance, or final/initial state radiation.
The effect of the kinematic cuts proposed in Sec. V.1 are shown in Table 1. The di-$`\tau `$ background generally has very different kinematics than our signal, as well as extra $`\pi ^0`$โs. It can be reduced below 0.1 fb with these cuts.
### VI.3 Two-Body Decay Background
A background to all processes is true resonance production where the final state resonance decays via any 2-body decay and its decay products lie outside the detector acceptance. This background is irreducible, but is accurately measured in events with both the radiative decay of interest and the final hadron decaying to a 2-body state. This gives roughly 10 times the statistics on measuring this background, so it can be subtracted.
For 2-body decays this amounts to an irreducible background that has a branching ratio $`f_2\mathrm{\Omega }`$. Here $`f_2`$ is the fraction of 2-body final-state bottomonium decays plus multi-body decays which are arranged such that all decay products are outside the detector acceptance. $`\mathrm{\Omega }`$ is the fraction of the solid angle covered by the detector. This background can be directly measured by relying on the sample of non-invisibly decaying final state particles provided by the ISR + radiative transition technique. We take $`f_2=5\%`$ and $`\mathrm{\Omega }=91.5\%`$ for the BaBar detector geometry. The final decay is uncorrelated to its production mechanism and the radiative transition, and therefore is isotropic in detector angle $`\theta `$ and $`\varphi `$. Events with a radiative transition and visible bottomonium decay give a measurement of all decay channels of the final state particle, including effects of detector resolution for the radiative transition, ISR smear, and multi-body decays. It should be noted that one cannot simply take this background sample and pretend the beamline bisects the detector in a different direction. The asymmetric boost of modern B-factories changes the size and area of the would-be beamline and this must be taken into account.
### VI.4 Drell-Yan
Direct production of $`\nu \overline{\nu }`$ is small since the neutrinos must come from a $`Z`$ or $`W^\pm `$, which are heavy. For instance, BR$`(\mathrm{{\rm Y}}(1S)\nu \overline{\nu })1\times 10^5`$sminvisible Only the vector resonances have a sizeable branching fraction to neutrinos, since they can mix directly with the $`Z`$. Scalars and pseudo-scalars can only emit neutrinos in loop suppressed processes.
Modern B-factories do not have the sensitivity to test this branching ratio, and therefore it is not a background.
## VII Conclusions
Measurements of invisible branching ratios of mesons are extremely important, given the established evidence for Dark Matter (DM) and the knowledge that most DM scenarios require some Standard Model-DM interaction. Given tight constraints on flavor changing neutral currents, the most important mesons to examine are flavor neutral bound states of quarks. Running high-luminosity B-factories motivate looking for invisible decays of bottomonium first.
ISR and radiative decays provide a powerful method to measure the invisible branching fractions of the bottomonium resonances. If the DM is lighter than $`M_\mathrm{{\rm Y}}/2`$, annihilation of DM into standard model particles is expected to have a picobarn-scale cross section. While the sensitivity achievable is not capable of measuring the Standard Model $`\mathrm{{\rm Y}}\nu \overline{\nu }`$, decays to dark matter should be significantly stronger than decays to neutrinos, due to the $`(M_\mathrm{{\rm Y}}/M_Z)^2`$ suppression of the Standard Model process. These techniques can limit $`BR(\mathrm{{\rm Y}}(1S)\mathrm{invisible})<0.1\%`$, which is sensitive enough to discover dark matter if it couples in this manner.
DM with a mass $`M_\chi <5`$ GeV is generally allowed in models. Direct detection experiments are very insensitive in this mass region, and would also be insensitive if dark matter preferentially couples to heavy quarks. Therefore, alternative methods to discover DM are required if the DM is this light. We strongly encourage experimental teams at BaBar, Belle, and CLEO to pursue these techniques.
Acknowledgments
We thank Jack Gunion, Tao Han, Dan Hooper, and Steve Sekula for useful discussions; Dave Mattingly and Steve Sekula carefully reading this manuscript. This work was supported in part by DOE grant DEโFG03โ91ERโ40674, the Davis Institute for High Energy Physics, and the U.C. Davis Deanโs office.
## Appendix A Bottomonium Event Rates
In the following we give the expected cross section for bottomonium production assuming $`E_{\mathrm{CM}}=M_{\mathrm{{\rm Y}}(4S)}=10.58`$ GeV. It should be noted that both on-peak and off-peak data can be used for this analysis.
For each final state, a โtagging topologyโ is given, which is the set of particles visible in the detectorโs acceptance. In each case the particles in the tagging topology have a well-defined kinematics as outlined in Sec. V.1. In cases where $`\gamma _{\mathrm{ISR}}`$ is not listed in the tagging topology, the cross section includes both when the ISR photon is visible, and when it lies outside the detector acceptance. When $`\gamma _{\mathrm{ISR}}`$ is listed, the cross section corresponds to requiring the ISR photon to be visible for the BaBar detector geometry, $`0.87<\mathrm{cos}\theta <0.96`$ in the center-of-mass frame.
We catalog only existing, measured resonances and transitions of $`b\overline{b}`$ quarkonia here (with the exception of the undiscovered $`\eta _b`$). Other decay chains will be possible involving radiative decays of the $`\mathrm{{\rm Y}}(4S)`$ when those decays are discovered. This clean method of tagging the initial state will allow the discovery and cataloging of many more radiative decays, improving statistics from what is listed below.
### A-1 Vector Mediated Dark Matter
| Final state | Decay chain | Tagging topology | | $`\sigma _\mathrm{{\rm Y}}`$(pb) |
| --- | --- | --- | --- | --- |
| $`\mathrm{{\rm Y}}(1S)`$ | | $`\gamma _{\mathrm{ISR}}`$ | | 3.034 |
| | $`\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^+\pi ^{}`$ | | 2.91 |
| | $`\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^0\pi ^0`$ | | 1.4 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^+\pi ^{}`$ | | 0.784 |
| | $`\mathrm{{\rm Y}}(2S)\chi _{b1}(1P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\gamma `$ | 0.369 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^0\pi ^0`$ | | 0.36 |
| | $`\mathrm{{\rm Y}}(2S)\chi _{b2}(1P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\gamma `$ | 0.239 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b1}(2P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\gamma `$ | 0.168 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\gamma \gamma `$ | $`\pi ^+\pi ^{}`$ | 0.165 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b2}(2P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\gamma `$ | 0.142 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^+\pi ^{}`$ | $`\pi ^+\pi ^{}`$ | 0.0921 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\gamma \gamma `$ | $`\pi ^0\pi ^0`$ | 0.0788 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^0\pi ^0`$ | $`\pi ^+\pi ^{}`$ | 0.0658 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^+\pi ^{}`$ | $`\pi ^0\pi ^0`$ | 0.0441 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b1}(2P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\omega `$ | 0.0322 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\mathrm{{\rm Y}}(1S)`$ | $`\pi ^0\pi ^0`$ | $`\pi ^0\pi ^0`$ | 0.0315 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b2}(2P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\omega `$ | 0.0219 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b0}(2P)\mathrm{{\rm Y}}(1S)`$ | $`\gamma `$ | $`\gamma `$ | 0.0107 |
| Total | | | | 6.912 |
| $`\mathrm{{\rm Y}}(2S)`$ | | $`\gamma _{\mathrm{ISR}}`$ | | 2.465 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)`$ | $`\gamma \gamma `$ | | 0.875 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)`$ | $`\pi ^+\pi ^{}`$ | | 0.49 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b1}(2P)\mathrm{{\rm Y}}(2S)`$ | $`\gamma `$ | $`\gamma `$ | 0.415 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)`$ | $`\pi ^0\pi ^0`$ | | 0.35 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b2}(2P)\mathrm{{\rm Y}}(2S)`$ | $`\gamma `$ | $`\gamma `$ | 0.323 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b0}(2P)\mathrm{{\rm Y}}(2S)`$ | $`\gamma `$ | $`\gamma `$ | 0.0545 |
| Total | | | | 2.508 |
### A-2 Pseudoscalar Mediated Dark Matter
| Final state | Decay chain | Tagging topology | | $`\sigma _\mathrm{{\rm Y}}`$(pb) |
| --- | --- | --- | --- | --- |
| $`\eta _b(1S)`$ | $`\mathrm{{\rm Y}}(3S)h_b(1P)\eta _b(1S)`$ | $`\pi ^+\pi ^{}`$ | $`\gamma `$ | 0.00874 |
| | $`\mathrm{{\rm Y}}(3S)h_b(1P)\eta _b(1S)`$ | $`\pi ^0`$ | $`\gamma `$ | 0.00236 |
| | $`\mathrm{{\rm Y}}(3S)\chi _{b0}(2P)\eta _b(1S)`$ | $`\gamma `$ | $`\eta `$ | 0.00213 |
| Total | | | | 0.013 |
### A-3 Scalar Mediated Dark Matter
| Final state | Decay chain | Tagging topology | | $`\sigma _\mathrm{{\rm Y}}`$(pb) |
| --- | --- | --- | --- | --- |
| $`\chi _{b0}(1P)`$ | $`\mathrm{{\rm Y}}(2S)\chi _{b0}(1P)`$ | $`\gamma _{\mathrm{ISR}}`$ | $`\gamma `$ | 0.0937 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\chi _{b0}(1P)`$ | $`\gamma \gamma `$ | $`\gamma `$ | 0.0333 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\chi _{b0}(1P)`$ | $`\pi ^+\pi ^{}`$ | $`\gamma `$ | 0.0186 |
| | $`\mathrm{{\rm Y}}(3S)\mathrm{{\rm Y}}(2S)\chi _{b0}(1P)`$ | $`\pi ^0\pi ^0`$ | $`\gamma `$ | 0.0133 |
| Total | | | | 0.159 |
| $`\chi _{b0}(2P)`$ | $`\mathrm{{\rm Y}}(3S)\chi _{b0}(2P)`$ | $`\gamma _{\mathrm{ISR}}`$ | $`\gamma `$ | 0.188 |
| Total | | | | 0.188 |
|
warning/0506/hep-th0506148.html
|
ar5iv
|
text
|
# Classical Non-Local conserved charges in String Theory
## Abstract
We construct a conserved non local charge in $`AdS_5\times S_5`$ string theory.
String theory is the strongest candidate to describe all interactionsgsw . However, there is no crucial experiment giving full support of the theory which is still in need of a stronger basis for a full theory of nature. As happens in several field theories, non perturbative results may give important clues to the behaviour of the theory in particularly difficult situations. Several results have already been obtained from the idea of duality duality . Recently some new features connected with the high dimensionality of strings have led to further insight into the structure of the socalled brane cosmology wittenvariousdim ; horavawitten ; rs .
More recently, some authors are pursuing higher conservation laws roiban ; nappi ; breno , which proved of great help in theories of lower dimensionality grossneveu ; luescher ; abdalla ; aar . In case we can use higher conservation laws in a way similar to the one used in two dimensional space time, it is possible that further constraints in the dynamical behaviour of strings can be imposed and one can gather information based on more general grounds to be compared with observations. As an example, strong nonperturbative insight about gravity can be obtained from the holographic principle holo .
We begin our problem with a set of currents defined in $`AdS_5\times S_5`$ space described by the coset $`PSU(2,2|4)/SO(4,1)\times SO(5)`$ metsaev . The underlying string theoryhas been described accordingly nathan . The algebra $`psu(2,2|4)`$ has, under the discrete group $`Z_4`$, a discrete decomposition $`=_{i=0}^3_i`$, described by
$`t_\alpha _1,t_{\underset{ยฏ}{a}}_2,`$
$`t_{\widehat{\alpha }}_3,t_{\underset{ยฏ}{[ab]}}_0`$ (1)
where $`\underset{ยฏ}{a}`$ are indices parametrizing $`AdS_5\times S_5`$, $`\alpha `$ and $`\widehat{\alpha }`$ are the superspace connections. The non vanishing structure constants are well known breno .
In terms of the supercoset valued filed $`g(x,\theta ,\widehat{\theta })`$ we can define algebra valued currents $`๐=g^1dg`$, which in turn are can be decomposed according to (1). The resulting currents have been shown to obey the relations breno
$`_\mu J_1^\mu +[J_{0\mu },J_1^\mu ]`$ $`=`$ $`\epsilon ^{\mu \nu }[J_{2\mu },J_{3\nu }],`$
$`_\mu J_2^\mu +[J_{0\mu },J_2^\mu ]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\epsilon ^{\mu \nu }\left([J_{3\mu },J_{3\nu }][J_{1\mu },J_{1\nu }]\right),`$
$`_\mu J_3^\mu +[J_{0\mu },J_3^\mu ]`$ $`=`$ $`\epsilon ^{\mu \nu }[J_{2\mu },J_{1\nu }],`$ (2)
and
$`\epsilon ^{\mu \nu }\left(_\mu J_{1\nu }+[J_{0\mu },J_{1\nu }]\right)`$ $`=`$ $`\epsilon ^{\mu \nu }[J_{2\mu },J_{3\nu }],`$
$`\epsilon ^{\mu \nu }\left(_\mu J_{2\nu }+[J_{0\mu },J_{2\nu }]\right)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\epsilon ^{\mu \nu }\left([J_{3\mu },J_{3\nu }]+[J_{1\mu },J_{1\nu }]\right),`$
$`\epsilon ^{\mu \nu }\left(_\mu J_{3\nu }+[J_{0\mu },J_{3\nu }]\right)`$ $`=`$ $`\epsilon ^{\mu \nu }[J_{2\mu },J_{1\nu }],`$ (3)
in the underlying Minkowski space. Such relations keep some similarity with zero curvature conditions, well known in integrable models grossneveu and nonlinear sigma models luescher ; abdalla , implying, upon quantization, severe constraints upon the S-matrix elements aar .
Here, we try a constructive approach. First, we neglect terms related to gauge-field valued elements of the algebra, which simplifies the discussion. In fact, the relevant currents transform nontrivially under gauge transformations, which is mirrored in the fact that the derivatives in the conservation laws are covariant derivatives of the form $`+[๐_0,]`$. We also notice that some commutators are gauge valued, such as $`[๐_2,๐_2]`$, or $`[๐_1,๐_3]`$. In expression (2), there are several conservation laws inbuilt and we are going to construct one of them. We claim that a non local conserved charge should be described in terms of a gauge dressing of a combination of the following building blocks:
$`Q^{(1)}`$ $`=`$ $`2{\displaystyle J_3^0(t,x_1)๐x_1},`$
$`Q^{(2)}`$ $`=`$ $`{\displaystyle J_1^0(t,x_1)ฯต(x_1x_2)J_2^0(t,x_2)๐x_1๐x_2}{\displaystyle J_2^0(t,x_1)ฯต(x_1x_2)J_1^0(t,x_2)๐x_1๐x_2},`$
$`Q^{(3)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_1^0(t,x_1)ฯต(x_1x_2)J_1^0(t,x_2)ฯต(x_2x_3)J_1^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_3^0(t,x_1)ฯต(x_1x_2)J_3^0(t,x_2)ฯต(x_2x_3)J_1^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_3^0(t,x_1)ฯต(x_1x_2)J_1^0(t,x_2)ฯต(x_2x_3)J_3^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_1^0(t,x_1)ฯต(x_1x_2)J_3^0(t,x_2)ฯต(x_2x_3)J_3^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_2^0(t,x_1)ฯต(x_1x_2)J_2^0(t,x_2)ฯต(x_2x_3)J_3^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_2^0(t,x_1)ฯต(x_1x_2)J_3^0(t,x_2)ฯต(x_2x_3)J_2^0(t,x_3)๐x_1๐x_2๐x_3}`$
$``$ $`{\displaystyle \frac{1}{2}}{\displaystyle J_3^0(t,x_1)ฯต(x_1x_2)J_2^0(t,x_2)ฯต(x_2x_3)J_2^0(t,x_3)๐x_1๐x_2๐x_3},`$
$`Q^{(4)}`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle J_1^0(x_1)ฯต(x_1x_2)J_1^0(x_2)ฯต(x_2x_3)J_2^0(x_3)ฯต(x_3x_4)J_3^0(x_4)๐x_1๐x_2๐x_3๐x_4}`$ (4)
$`+`$ $`{\displaystyle \frac{1}{4}}{\displaystyle J_1^0(x_1)ฯต(x_1x_2)J_2^0(x_2)ฯต(x_2x_3)J_2^0(x_3)ฯต(x_3x_4)J_2^0(x_4)๐x_1๐x_2๐x_3๐x_4}`$
$``$ $`{\displaystyle \frac{1}{4}}{\displaystyle J_2^0(x_1)ฯต(x_1x_2)J_3^0(x_2)ฯต(x_2x_3)J_3^0(x_3)ฯต(x_3x_4)J_3^0(x_4)๐x_1๐x_2๐x_3๐x_4},`$
$`\mathrm{}`$
where the sum is over all orders of the indices of currents. The remaining terms are to be constructed taking into account the generic additive term
$$Q_i^{(n)}=\pm \frac{1}{2^{n2}}\left(\underset{k=1}{\overset{n1}{}}J_{\alpha _k}^0(t,x_k)ฯต(x_kx_{k+1})dx_k\right)J_{\alpha _n}^0(t,x_n)๐x_n,$$
(5)
with the indices $`\alpha _k`$ satisfying the constraint equation
$$\underset{k=1}{\overset{n}{}}\alpha _k=3mod4.$$
(6)
The sign has to be properly chosen in order to achieve conservation of the sum. The time derivative of the current can be exchanged by the space derivative, a commutator with a gauge-valued current and a nontrivial commutator. The space derivative is integrated by parts giving rise to a Dirac delta used to perform one integration, and leaving a lower order term in the integration variables, but with a nontrivial commutator. The commutator either cancels a similar one arising from a lower order derivative, or is gauge valued. Therefore, we claim that
$$_0Q^{(n)}=\delta _{der}Q^{(n)}+\delta _{com}Q^{(n)}+\text{gauge terms}$$
(7)
where the first term arises from the integration by parts of the space derivative of the current giving rise to a Dirac-delta term of the form of a commutator among two currents connected by the $`ฯต(x_kx_{k+1})`$ function, while the second term arises from the remaining term of the equation of motion, with exception of the gauge ($`๐_0`$) commutators. In the first term we leave aside the commutators which take values in the gauge sector, namely, $`[๐_2,๐_2]`$ and $`[๐_1,๐_3]`$, which are, together with the explicit gauge terms containing $`๐_0`$ left to the last term. Under these conditions, the definition of the charges leads us to the result
$$\delta _{der}Q^{(n+1)}=\delta _{com}Q^{(n}).$$
(8)
Therefore, we
Claim: the nonlocal charge
$$Q_3=\underset{n=1}{\overset{\mathrm{}}{}}Q^{(n)}$$
(9)
is classically conserved up to a gauge dressing, defined substituting some currents by $`_0`$ valued elements.
We also claim that analogous charges, whose first term are either obtained from $`J_2^0`$ or from $`J_1^0`$ are also conserved.
Acknowledgement: This work has been supported by CNPq and FAPESP, Brazil. We would like like to thank several discussions with Dr. Breno Vallilo.
|
warning/0506/math0506271.html
|
ar5iv
|
text
|
# Non-Emptiness of the Height Strata of the Moduli Stack of Polarized K3 Surfaces
## Introduction
In positive characteristic one can define interesting subvarieties of moduli spaces of abelian varieties and of curves. Such loci can be given by considering the collection of these objects having fixed certain discrete invariants, such as for instance filtrations on $`\mathrm{BT}_1`$-groups or Newton polygons (see \[Oor01a\] and \[Oor01b\]). A similar approach can be taken when studying moduli spaces of K3 surfaces.
To every K3 surface over a perfect field $`k`$ of characteristic $`p>0`$ one associates a Newton polygon. By definition it is the Newton polygon of the $`F`$-crystal $`H_{\mathrm{cris}}^2(X/W(k))`$. Denote by $`\alpha `$ the smallest slope of the Newton polygon of $`X`$. We define the height of $`X`$ to be infinite if $`\alpha =1`$ and $`1/(1\alpha )`$ otherwise. If finite, the height of a K3 surface takes integral values from 1 to 10. Denote by $`_{2d}`$ the moduli stack of K3 spaces with a polarization of degree $`2d`$ and suppose that $`p`$ does not divide $`2d`$. We look at the subspaces $`_{2d,๐ฝ_p}^{(h)}`$ of $`_{2d}๐ฝ_p`$ of K3 surfaces with height at least $`h`$. The collection of those 11 subspaces is called the height stratification of $`_{2d}๐ฝ_p`$. One further stratifies $`_{2d,๐ฝ_p}^{(11)}`$ by the Artin invariant (see for instance \[Art74\]). In this way we obtain a filtration of the moduli space $`_{2d}๐ฝ_p`$
$$_{2d}๐ฝ_p=_{2d,๐ฝ_p}^{(1)}_{2d,๐ฝ_p}^{(2)}\mathrm{}_{2d,๐ฝ_p}^{(11)}=\mathrm{\Sigma }_1\mathrm{}\mathrm{\Sigma }_{10}.$$
The following question rises naturally.
###### Question.
Are all the subspaces in the height stratification of $`_{2d}๐ฝ_p`$ non-empty?
This can be reformulated in the following way: For a given natural number $`d`$ and a prime $`p`$ determine all Newton polygons of polarized K3 surfaces of degree $`2d`$ over fields of characteristic $`p`$. This is an analogue of the Manin problem for Newton polygons of abelian varieties (\[Man63, Conj. 2, p. 76\]).
In this note we answer partially the question posed above by proving the following result.
###### Theorem.
For every $`d`$, large enough and prime to $`p>2`$, the subspaces in the height strata of $`_{2d}๐ฝ_p`$ are non-empty.
The idea of the proof is to start with a polarized abelian surface $`(A,\lambda )`$ over $`k`$ of certain degree and to use $`\lambda `$ to construct an ample line bundle on the Kummer surface $`X`$ associated to $`A`$. In this way we find $`\overline{k}`$-valued points of $`_{2d}๐ฝ_p`$. Making some appropriate choices of supersingular polarized abelian surfaces $`(A,\lambda )`$ we are able to show that the height strata of $`_{2d}๐ฝ_p`$ are non-empty if $`d`$ is large enough. The construction gives explicit bounds for $`d`$.
The organization of this note is the following. In Section 1 we recall some definitions and give an overview of some results on the height stratification of $`_{2d,๐ฝ_p}`$. Section 2 is devoted to Kummer surfaces. Starting with an ample line bundle on an abelian surface we describe a way of constructing ample line bundles on its associated Kummer surfaces. This allows us to find points in $`_{2d,๐ฝ_p}(\overline{๐ฝ}_p)`$ which belong to certain height strata of $`_{2d,๐ฝ_p}`$. In Section 3 we take this idea one step further and construct Kummer morphisms from moduli stacks of polarized abelian surfaces to moduli stacks of polarized K3 surfaces. We use these morphisms to give an affirmative answer to the question posed above in case $`d`$ is large enough.
Notations
We write $`_{2d}`$ for the Deligne-Mumford stack of K3 spaces with a polarization of degree $`2d`$. It is a smooth stack over $`\mathrm{Spec}([1/2d])`$. See \[Riz05b, ยง4.3, Thm. 4.7\] and \[Riz05a, Ch. 1, ยง1.4.3\]
If $`A`$ is a ring, $`AB`$ a ring homomorphism then for any $`A`$-module ($`A`$-algebra etc.) $`V`$ we will denote by $`V_B`$ the $`B`$-module ($`B`$-algebra etc.) $`V_AB`$.
For an algebraic stack $``$ over a scheme $`S`$ and a morphism of schemes $`S^{}S`$ we will denote by $`_S^{}`$ the product $`\times _SS^{}`$ and consider it as an algebraic stack over $`S^{}`$.
We denote by $`๐_{g,d,n}`$ the moduli stack of $`g`$-dimensional abelian varieties with a polarization of degree $`d^2`$ and a Jacobi level $`n`$-structure. It is a Deligne-Mumford stack which is smooth over $`[1/dn]`$. We will write $`๐_{g,d}`$ for $`๐_{g,d,1}`$.
Acknowledgments
This note contains the results of the second chapter of my Ph.D. thesis. I thank my advisors, Ben Moonen and Frans Oort for their help, their support and for everything I have learned from them. I thank the Dutch Organization for Research N.W.O. for the financial support with which my thesis was done.
## 1. The Height Stratification of $`_{2d,๐ฝ_p}`$
Let $`k`$ be a perfect field of characteristic $`p>0`$ and consider a K3 surface $`X`$ over $`k`$. Consider the contravariant functor
$$\mathrm{\Phi }^2:\underset{ยฏ}{\mathrm{Art}}\mathrm{Ab}$$
from the category of local artinian schemes to abelian groups defined by
$$\mathrm{\Phi }^2(S)=\mathrm{ker}\left(H_{\mathrm{et}}^2(X\times S,๐พ_m)H_{\mathrm{et}}^2(X,๐พ_m)\right).$$
This functor is representable by a formal Lie group, denoted by $`\widehat{B}r(X)`$ and called the *formal Brauer group* of $`X`$.
###### Proposition 1.1.
The formal group $`\widehat{B}r(X)`$ is $`1`$-dimensional and one has the following two possibilities for it:
1. The height of $`\widehat{B}r(X)`$ is infinite and then $`\widehat{B}r(X)\widehat{๐พ}_a`$.
2. The height is finite. Then $`\widehat{B}r(X)`$ is a $`p`$-divisible group. Moreover, its height satisfies $`1h\left(\widehat{B}r(X)\right)10`$.
For proofs we refer to \[AM77\]. From now on we will call the height of the formal Brauer group of $`X`$ simply the height of $`X`$ and denote it as $`h(X)`$.
The *Newton polygon* of $`X`$ is the Newton polygon of the $`F`$-crystal $`H_{\mathrm{cris}}^2(X/W)`$ where $`W`$ is the ring of Witt vectors $`W(k)`$ (see \[Ill95, ยง1.3 (c)\]). It is the lower convex polygon starting at $`(0,0)`$ and ending at $`(22,22)`$. The height of a K3 surface $`X`$ can be read off from its Newton polygon. If $`\alpha `$ is the smallest slope of the Newton polygon of $`X`$, then $`h(X)=1/(1\alpha )`$ if $`\alpha 1`$ and infinity otherwise. This follows from Corollary 3.3 in \[AM77, III\].
The *Hodge polygon in degree $`m`$* of a non-singular projective variety $`X`$ over $`k`$ is defined as the increasing convex polygon starting at $`(0,0)`$, having slope $`i`$ with multiplicity $`h^{i,mi}=dim_kH^{mi}(X,\mathrm{\Omega }_X^i)`$. For a K3 surface $`X`$ the Hodge polygon in degree $`2`$ will be called the Hodge polygon of $`X`$. The Newton polygon of a K3 surface lies on or above its Hodge polygon (\[Ill95, Thm. 1.3.9\]).
Recall that a K3 surface $`X`$ over $`k`$ is called *ordinary* if any of the following equivalent conditions is satisfied:
1. $`h(X)=1`$.
2. The Newton and the Hodge polygon of $`X`$ coincide i.e., the Newton slopes of $`X`$ are 0 and 2 with multiplicity one, 1 with multiplicity 20.
A K3 surface $`X`$ over $`k`$ is called *supersingular* if any of the following equivalent conditions is satisfied:
1. The height of $`X`$ is infinite.
2. The Newton polygon is a straight line i.e., all Newton slopes of $`X`$ are 1.
The fact that the two possible ways of defining ordinary and supersingular K3 surfaces are equivalent follows from \[AM77, III, Cor. 3.3\].
A K3 surface $`X`$ over $`k`$ is called *supersingular in the sense of Shioda* if the rank of $`\mathrm{NS}(X)`$ is 22. One easily sees that if a K3 surface is supersingular in the sense of Shioda, then it is supersingular. It is a conjecture of M. Artin that, conversely, a supersingular K3 surface has Nรฉron-Severi rank 22.
###### Example 1.2.
Let $`p3(mod4)`$ be a prime number. Then the Fermat K3 surface
$$x^4+y^4+z^4+w^4=0$$
in $`_{๐ฝ_p}^3`$ is supersingular in the sense of Shioda (see \[Shi75, Thm. 1\]).
Let $`d`$ be an integer and assume further that $`p`$ does not divide $`2d`$. Consider the moduli stack $`_{2d,๐ฝ_p}=_{2d}_{}๐ฝ_p`$ of K3 surfaces with a polarization of degree $`2d`$ over a basis in characteristic $`p`$. Define the height stratification of $`_{2d,๐ฝ_p}`$ as follows: For $`h1`$ let $`_{2d,๐ฝ_p}^{(h)}`$ be the full subcategory of $`_{2d,๐ฝ_p}`$
$$_{2d,๐ฝ_p}^{(h)}(S)=\left\{(XS,\lambda )_{2d,๐ฝ_p}(S)\right|h(X_{\overline{s}})h\mathrm{for}\mathrm{every}\mathrm{geometric}\mathrm{point}\overline{s}S\}.$$
It is known that $`_{2d,๐ฝ_p}^{(h)}`$ is a closed substack of $`_{2d,๐ฝ_p}`$ of codimension at most $`h1`$. One defines a stratification of $`_{2d,๐ฝ_p}^{(11)}`$ by the Artin invariant (see \[Art74\]). Let $`X`$ be a supersingular K3 surface and let $`\mathrm{\Delta }(\mathrm{NS}(X))`$ be the discriminant of the intersection pairing on $`\mathrm{NS}(X)`$
$$(,):\mathrm{NS}(X)\times \mathrm{NS}(X).$$
One can show that $`\mathrm{ord}_p(\mathrm{\Delta })=2\sigma _0`$ where $`\sigma _0`$ takes values $`1,\mathrm{},10`$. It is called the *Artin invariant* of $`X`$. Let $`\mathrm{\Sigma }_i`$ be the full subcategory of $`_{2d,๐ฝ_p}^{(11)}`$ defined by
$$\begin{array}{cc}\hfill \mathrm{\Sigma }_i(S)=\{(XS,\lambda )_{2d,๐ฝ_p}(S)|& h(X_{\overline{s}})=\mathrm{}\mathrm{and}\sigma _0(X_{\overline{s}})11i\hfill \\ & \mathrm{for}\mathrm{every}\mathrm{geometric}\mathrm{point}\overline{s}S\}.\hfill \end{array}$$
In this way we obtain a filtration of the moduli space
(1)
$$_{2d,๐ฝ_p}=_{2d,๐ฝ_p}^{(1)}_{2d,๐ฝ_p}^{(2)}\mathrm{}_{2d,๐ฝ_p}^{(11)}=\mathrm{\Sigma }_1\mathrm{}\mathrm{\Sigma }_{10}.$$
This is a chain of 20 closed substacks and the dimension drops with at least one at each step.
###### Theorem 1.3.
For $`h=1,\mathrm{},10,11`$ the locus $`_{2d,๐ฝ_p}^{(h)}`$, if non-empty, is of codimension $`h1`$ and for $`h11`$ is a local complete intersection.
###### Proof.
We refer to \[vdGK00, sect. 13,14 & 15\]. The statement presented above is Theorem 15.1. โ
###### Remark 1.4.
B. Moonen and T. Wedhorn (\[MW04\]) have a theory of *$`F`$-zips* which gives a scheme-theoretic and uniform definition of the filtration (1). For details we refer to Example 7.4 in loc. cit..
## 2. Kummer Surfaces
As we mentioned in the beginning of this chapter we will use Kummer surfaces to show that the height strata of $`_{2d,๐ฝ_p}`$ are non-empty for large enough $`d`$, prime to $`p`$. In Section 2.1 we will recall some basic facts about Kummer surfaces which we will need in the sequel. In the next section, starting with a polarized abelian surface $`(A,\lambda )`$ we describe a way for constructing polarizations on its associated Kummer surface $`X`$. For this we make use of Seshadri constants.
### 2.1. Kummer Surfaces
Recall that to an abelian surface $`A`$ over a field $`k`$ of characteristic different from 2 we associated a K3 surface $`X`$, called the Kummer surface of $`A`$. We do this in the following way: Let $`A[2]`$ be the kernel of the multiplication by-2-map, let $`\pi :\stackrel{~}{A}A`$ be the blow-up of $`A[2]`$ and let $`\stackrel{~}{E}`$ be the exceptional divisor. The automorphism $`[1]_A`$ lifts to an involution $`[1]_{\stackrel{~}{A}}`$ on $`\stackrel{~}{A}`$. Let $`X`$ be the quotient variety of $`\stackrel{~}{A}`$ by the group of automorphisms $`\{\mathrm{id}_{\stackrel{~}{A}},[1]_{\stackrel{~}{A}}\}`$ and denote by $`\iota :\stackrel{~}{A}X`$ the quotient morphism. It is a finite map of degree 2. The variety $`X`$ is the Kummer surface associated to $`A`$.
Assume further that all points in $`A[2](\overline{k})`$ are $`k`$-rational. Then the exceptional divisor $`\stackrel{~}{E}`$ on $`\stackrel{~}{A}`$ consists of 16 irreducible curves $`E_{}^{}{}_{j}{}^{}`$, $`j=1,\mathrm{},16`$, each corresponding to a point in $`A[2](\overline{k})`$. We have that $`(E_{}^{}{}_{j}{}^{},E_{}^{}{}_{l}{}^{})_{\stackrel{~}{A}}=\delta _{j,l}`$ where $`\delta _{j,l}`$ is the Kronecker $`\delta `$-function. Let us make the following notations:
$$\begin{array}{cc}\hfill _{}^{}{}_{j}{}^{}& :=๐ช_{\stackrel{~}{A}}(E_{}^{}{}_{j}{}^{})\text{a line bundle on}\stackrel{~}{A},\hfill \\ \hfill E_j& :=\iota (E_j^{})\text{a divisor on}X,\hfill \\ \hfill _j& :=๐ช_X(E_j^{})\text{the corresponding line bundle on}X.\hfill \end{array}$$
Then one has that
$$\iota ^{}_j_{j}^{}{}_{}{}^{2}\mathrm{and}(_j,_l)_X=2\delta _{j,l}.$$
Moreover the line bundle $`_{j=1}^{16}_j`$ is divisible by 2 in $`\mathrm{Pic}(X_{\overline{k}})`$.
We turn next to some $`p`$-adic discrete invariants of Kummer surfaces. From now on we will assume that $`k`$ is a field of positive characteristic different from $`2`$. Then $`X`$ is supersingular in the sense of Shioda if and only if $`A`$ is supersingular. Indeed, according to \[Shi79, ยง3, Prop. 3.1\], one has that $`\mathrm{NS}(X_{\overline{k}})_{}=\mathrm{NS}(A_{\overline{k}})_{}^{[1]_A}_{j=1}^{16}`$. Hence we have that $`\text{rk}_{}\mathrm{NS}(X)=22`$ if and only if $`\text{rk}_{}\mathrm{NS}(A)=6`$ which is equivalent to $`A`$ being supersingular. We will determine the Newton polygon of $`X`$ in term of the Newton polygon of $`A`$. To do that we shall need the following auxiliary result.
###### Lemma 2.1.
Let $`A`$ be an abelian surface and $`X`$ the associated Kummer surface over $`k`$. Then there is a natural isomorphism
$$H_{\mathrm{et}}^2(X_{\overline{k}},_l)H_{\mathrm{et}}^2(A_{\overline{k}},_l)_1^{16}_l(1).$$
###### Proof.
This follows directly from the construction of Kummer surfaces. See for instance the proof of Lemma 2.2 in \[Ito01\]. โ
###### Lemma 2.2.
Let $`k`$ be a finite field of characteristic different from 2. Then one has that
1. If $`A`$ is ordinary, then the Newton polygon slopes of $`X`$ are $`\mu _1=0;\mu _2=\mathrm{}=\mu _5=1;\mu _6=2;\mu _j=1`$ for $`j=7,\mathrm{}22`$. In this case $`X`$ is ordinary i.e., its height is $`1`$.
2. If the $`p`$-rank of $`A`$ is 1, then the Newton polygon slopes of $`X`$ are $`\mu _1=\mu _2=1/2;\mu _3=\mu _4=1;\mu _5=\mu _6=3/4;\mu _j=1`$ for $`j=7,\mathrm{}22`$. In this case $`X`$ has height is $`2`$.
3. If $`A`$ is supersingular, then $`X`$ is supersingular and all its Newton polygon slopes are $`1`$. In this case $`\mathrm{\Delta }(\mathrm{NS}(A))`$ and $`\mathrm{\Delta }(\mathrm{NS}(X))`$ differ only by a power of $`2`$, hence the Artin invariant of $`X`$ is $`1`$ or $`2`$. It is $`1`$ if and only if $`A`$ is a superspecial abelian surface.
###### Proof.
As $`k`$ is a finite field and $`X`$ and $`A`$ are projective varieties one can compute the Newton polygons of $`A`$ and $`X`$ using รฉtale cohomology instead of crystalline cohomology. We refer to \[Ill95, 1.3, Equality (1.3.5)\] for an explanation and details. We will use the relation between the รฉtale cohomology groups of $`A`$ and $`X`$ given in Lemma 2.1.
If the Newton polygon slopes of $`A`$ are $`\lambda _j`$ for $`j=1,\mathrm{},4`$, then those of $`X`$ satisfy $`\mu _1=\lambda _1+\lambda _2;\mu _2=\lambda _1+\lambda _3;\mu _3=\lambda _1+\lambda _4;\mu _4=\lambda _2+\lambda _3;\mu _5=\lambda _2+\lambda _4;\mu _6=\lambda _3+\lambda _4;\mu _i=1`$ for $`i=7,\mathrm{},22`$. The last statement follows from \[Shi79, ยง3, Prop. 3.1\]. โ
### 2.2. Ample Line Bundles on Kummer Surfaces
Let $`k`$ be an algebraically closed field of characteristic different from 2 and consider an abelian surface $`A`$ over $`k`$. Denote by $`X`$ the associated Kummer surface. In this section we will show how to construct ample line bundles on $`X`$ starting with an ample bundle on $`A`$. This will allow us to give explicitly points in $`_{2d,๐ฝ_p}`$ for some $`d`$.
Let $``$ be an ample line bundle on $`A`$ with $`\chi ()=d^{}`$. Then by Riemann-Roch we have that $`(,)_A=2d^{}`$. Let $`n`$ and fix 16 positive integers $`n_j`$. Consider the line bundle $`๐ฉ`$ on $`\stackrel{~}{A}`$ given by
$$๐ฉ=\pi ^{}\left(^n[1]_A^{}^n\right)\left(\underset{j=1}{\overset{16}{}}_{j}^{}{}_{}{}^{2n_j}\right).$$
We will compute its self-intersection and show that $`๐ฉ`$ is the pull-back of a line bundle on $`X`$.
###### Lemma 2.3.
With the notations as above one has:
1. $`(๐ฉ,๐ฉ)_{\stackrel{~}{A}}=8n^2d^{}4_{j=1}^{16}n_j^2`$;
2. There exists a line bundle $``$ on $`X`$ such that $`\iota ^{}๐ฉ`$. The line bundle $``$ is ample iff $`๐ฉ`$ is ample. Moreover if $`2d=(,)_X`$, then we have that
$$d=2n^2d^{}\underset{j=1}{\overset{16}{}}n_j^2.$$
###### Proof.
(a) Combining \[Har77, Ch. V, ยง3, Prop. 3.2\] and the fact that $`[1]_A^{}`$ and $``$ are algebraically equivalent we get
$$\begin{array}{cc}\hfill (๐ฉ,๐ฉ)_{\stackrel{~}{A}}& =2n^2(,)_A+2n^2(,[1]_A^{})_A4\underset{j=1}{}16n_j^2\hfill \\ & =8n^2d^{}4\underset{j=1}{}16n_j^2.\hfill \end{array}$$
(b) Take a divisor $`DA\backslash A[2]`$ such that $`๐ช_A(D)=`$. If $`U:=\stackrel{~}{A}_{j=1}^{16}E_j^{}`$ and $`V=X_{j=1}^{16}E_j`$, then the map $`\iota :UV`$ is รฉtale. Consider the divisor
$$D_1:=\iota \left(n\pi ^{}(D)+n\pi ^{}([1]_A^{}D)\right)$$
on $`X`$. As $`D_1V`$ we see that $`\iota ^{}D_1=n\pi ^{}(D)+n\pi ^{}([1]_A^{}D)`$. Hence if we set
$$๐ซ:=๐ช_X(D_1)$$
then we have that $`\iota ^{}๐ซ\pi ^{}\left(^n[1]_A^{}^n\right)`$ on $`\stackrel{~}{A}`$. Using the fact that $`\iota ^{}_j_{j}^{}{}_{}{}^{2}`$ one sees that the line bundle
$$=๐ซ\underset{j=1}{\overset{16}{}}_j^{n_j}$$
satisfies $`\iota ^{}๐ฉ`$ on $`\stackrel{~}{A}`$. Since $`\iota `$ is a finite morphism $``$ is ample on $`X`$ if and only if $`๐ฉ`$ is ample on $`\stackrel{~}{A}`$ (\[Har77, Ch. III, Exercise 5.7 (d)\]).
For the self-intersection number computation one has
$$2n^2(,)_A+2n^2(,[1]_A^{})4\underset{j=1}{\overset{16}{}}n_j^2=$$
$$=(๐ฉ,๐ฉ)_{\stackrel{~}{A}}=(\iota ^{},\iota ^{})_{\stackrel{~}{A}}=\mathrm{deg}(\iota )(,)_X=4d$$
which gives the formula from (b). โ
###### Remark 2.4.
Note that the line bundle $`^n[1]_A^{}^n`$ comes with a natural action of $`[1]_A`$. Hence its pull-back $`\pi ^{}\left(^n[1]_A^{}^n\right)`$ comes equipped with an action of $`[1]_{\stackrel{~}{A}}`$. Therefore one can apply \[Mum74, Ch. III ยง10, Thm. 1(B)\] to the morphism $`\iota :\stackrel{~}{A}X`$ and conclude that $`๐ซ=\iota _{}\left(\pi ^{}(^n[1]_A^{}^n)\right)^{[1]_{\stackrel{~}{A}}}`$ is the line bundle described in the proof of part (b).
Lemma 2.3 suggests a way to construct ample line bundles on the Kummer surface $`X`$. We will give sufficient conditions under which $`๐ฉ`$ is ample on $`\stackrel{~}{A}`$. To do this we will make use of *multiple Seshadri constants*. We will recall the definition below. For details we refer to \[Bau99\].
Seshadri constants. Let $`๐`$ be an ample line bundle on $`A`$ and let $`x_1,\mathrm{},x_{16}`$ be the points in $`A[2](k)`$ (recall that $`k=\overline{k}`$ and $`\mathrm{char}(k)2`$). We make this change of notations here to avoid any possible confusion as later we will compute Seshadri constants for the ample line bundle $`๐=[1]_A^{}^1`$. Let $`\mathrm{NS}(\stackrel{~}{A})_{}`$ denote $`\mathrm{NS}(\stackrel{~}{A})_{}`$ and let $`(,)_{\stackrel{~}{A},}`$ be the induced bilinear form. We will call an element $``$ of $`\mathrm{NS}(\stackrel{~}{A})_{}`$ *numerically effective*, or shortly *nef*, if for any irreducible curve $`\mathrm{\Gamma }`$ in $`\stackrel{~}{A}`$ we have that $`(,๐ช_{\stackrel{~}{A}}(\mathrm{\Gamma }))_{\stackrel{~}{A},}0`$. Further, for an element $`\mathrm{NS}(\stackrel{~}{A})_{}`$ and a real number $`ฯต`$ we will denote by $`^ฯต`$ the element $`ฯต\mathrm{NS}(\stackrel{~}{A})_{}`$.
One shows that
$$ฯต_๐=sup\{ฯต|\pi ^{}๐\underset{i=1}{\overset{16}{}}_{i}^{}{}_{}{}^{ฯต}\mathrm{is}\mathrm{nef}\mathrm{in}\mathrm{NS}(\stackrel{~}{A})_{}\}$$
exists. It is called the *multiple Seshadri constant* on $`A`$ for $`x_1,\mathrm{},x_{16}`$. An equivalent definition of the Seshadri constant $`ฯต`$ can be given in the following way:
$$ฯต_๐=inf\frac{(๐,๐ช_A(C))_A}{_{i=1}^{16}\mathrm{mult}_{x_i}C}$$
where $`\text{mult}_{x_i}C`$ is the multiplicity of $`C`$ at $`x_i`$ and the infimum is taken over all irreducible curves $`C`$ in $`A`$ which pass through at least one $`x_i`$.
###### Remark 2.5.
1. If $`0<\delta <ฯต_๐`$, then the line bundle $`\pi ^{}๐_{i=1}^{16}_{i}^{}{}_{}{}^{\delta }`$ is nef. Moreover, one has the strict inequality
$$(\pi ^{}๐\underset{i=1}{\overset{16}{}}_{i}^{}{}_{}{}^{\delta },๐ช_{\stackrel{~}{A}}(\mathrm{\Gamma }))_{\stackrel{~}{A},}>0$$
for any irreducible curve $`\mathrm{\Gamma }`$ on $`\stackrel{~}{A}`$.
2. If $`0<n_i<ฯต_๐`$, then $`\pi ^{}๐_{i=1}^{16}_{i}^{}{}_{}{}^{n_i}`$ is nef. Moreover, one has the strict inequality
$$(\pi ^{}๐\underset{i=1}{\overset{16}{}}_{i}^{}{}_{}{}^{n_i},๐ช_{\stackrel{~}{A}}(\mathrm{\Gamma }))_{\stackrel{~}{A}}>0$$
for any irreducible curve $`\mathrm{\Gamma }`$ on $`\stackrel{~}{A}`$.
These facts are clear from the second definition of $`ฯต`$.
Numerical estimates. We will apply the general results on Seshadri constants to our particular situation. To avoid confusion let us make the following convention: If $`A`$ is an abelian surface, then by *an elliptic curve $`E`$ in $`A`$* we shall mean an abelian subvariety $`E`$ of $`A`$ of dimension one.
###### Proposition 2.6.
Let $`A`$ be an abelian surface over an algebraically closed field $`k`$ of characteristic different from $`2`$. Let $`\{x_1,\mathrm{},x_{16}\}`$ be the set of two-torsion points on $`A`$. Then for an ample line bundle $`๐`$ on $`A`$ we are in one of the following cases:
1. The Seshadri constant satisfies the inequality
$$ฯต_๐\frac{\sqrt{2(๐,๐)_A}}{16}.$$
2. The abelian surface $`A`$ contains a curve $`E`$ of genus $`1`$ such that
(2)
$$ฯต_๐=\frac{(๐,๐ช_A(E))_A}{\mathrm{\#}\{i|x_iE(k)\}}.$$
###### Proof.
See \[Bau99, Prop. 8.3\]. Note that in this paper the assumption $`k=`$ is made. However, the proof of the above proposition uses only the Hodge index theorem, the Riemann-Roch theorem and some facts about blow-ups of curves. These results are valid over any algebraically closed field. โ
###### Remark 2.7.
Note that in (b) we may assume that $`E`$ is an elliptic curve in $`A`$. Indeed, we have that $`E`$ is a translate of an elliptic curve $`E^{}A`$ by a point $`aA`$. Since $`๐`$ is ample, the line bundles $`t_a^{}๐`$ and $`๐`$ are numerically equivalent. Therefore we have that
$$(๐,๐ช_A(E))=(t_a^{}๐,t_a^{}๐ช_A(E))=(๐,t_a^{}๐ช_A(E))=(๐,๐ช_A(E^{})).$$
We have further that $`\mathrm{\#}\{i|x_iE(k)\}4`$. Indeed, all these points correspond to points $`p_i=x_iaE^{}(k)`$ for which $`[2]p_i=[2]a`$ is a fixed point in $`E^{}(k)`$. As the isogeny $`[2]`$ is of degree $`4`$ (on $`E^{}`$) there are at most four such points. So we have that
$$\begin{array}{cc}\hfill ฯต_๐& =\frac{(๐,๐ช_A(E))_A}{_{i=1}^{16}\mathrm{mult}_{x_i}E}=\frac{(๐,๐ช_A(E))_A}{\mathrm{\#}\{i|x_iE(k)\}}=\hfill \\ & =\frac{(๐,๐ช_A(E^{}))_A}{\mathrm{\#}\{i|x_iE(k)\}}\frac{(๐,๐ช_A(E^{}))_A}{4}=\frac{(๐,๐ช_A(E^{}))_A}{_{i=1}^{16}\mathrm{mult}_{x_i}E^{}}ฯต_๐.\hfill \end{array}$$
Therefore we have equalities and we conclude that $`\mathrm{\#}\{i|x_iE(k)\}=4`$. We also see that $`aA[2](k)`$.
In what follows we will try to avoid case (b) of Proposition 2.6 as much as possible. The reason is that one has little control over the intersection $`(๐,๐ช_A(E))_A`$ in terms of the degree of $`๐`$. The bound in (a) increases with $`(๐,๐)_A`$, but $`A`$ can contain curves of genus $`1`$ of any given intersection index $`(๐,๐ช_A(E))_A`$, no matter how large $`(๐,๐)_A`$ is.
We will need the following auxiliary result which we shall apply to a line bundle $``$ defining the polarization $`\lambda `$ on $`A`$ (cf. the beginning of this section).
###### Lemma 2.8.
Let $``$ be an ample line bundle on an abelian surface $`A`$ and let $`EA`$ be an elliptic curve.
1. Suppose that $`(,๐ช_A(E))_A=1`$. Then there exists an elliptic curve $`E^{}A`$ such that $`AE\times E^{}`$. Moreover, if $`\pi _1:E\times E^{}E`$ and $`\pi _2:E\times E^{}E^{}`$ are the two projections, then there exists a point $`PE`$ and a line bundle $`๐ข`$ on $`E^{}`$ such that
$$\pi _1^{}๐ช_E(P)\pi _2^{}๐ข.$$
2. Suppose that $`(,๐ช_A(E))_A=m`$ for some $`m`$. Then there exist an elliptic curve $`E^{}A`$ and an isogeny $`f:E\times E^{}A`$ of degree at most $`m`$.
###### Proof.
(a): The proof can be found in \[Nak96, Lemma 2.6\].
(b): Consider the homomorphism
where $`\phi _{}`$ is the map $`at_a^{}^1`$ and the second map is the dual of the inclusion $`EA`$. Let $`E^{}`$ be the reduced subscheme of the zero component of $`\text{ker}(\varphi )`$. Then $`E^{}`$ is an elliptic curve in $`A`$. Note that $`|_E`$ is an invertible sheaf of degree at most $`m`$ hence $`(E,E^{})_Am`$. Define the homomorphism $`E\times E^{}A`$ to be $`(P,P^{})P+P^{}`$. It is surjective and its kernel is a finite group scheme hence it is an isogeny. Moreover its degree is exactly $`(E,E^{})_Am`$. โ
To get explicit conditions under which $`๐ฉ`$ is ample on $`\stackrel{~}{A}`$, one has to give some explicit estimates for $`ฯต_๐`$ for the ample line bundle $`๐=^n[1]_A^{}^n`$.
###### Lemma 2.9.
With the notations of Lemma 2.3 one has that
1. If $`d^{},n,n_1,\mathrm{}n_{16}`$ satisfy the following three inequalities
(3)
$$2n^2d^{}\underset{i=1}{\overset{16}{}}n_i>0$$
(4)
$$n_i<\frac{n}{4}$$
(5)
$$n_i<\frac{\sqrt{n^2d^{}}}{8},$$
then the line bundle $`๐ฉ`$ is ample on $`\stackrel{~}{A}`$.
2. Assume further that $`(A,)`$ is not isomorphic to a polarized product of elliptic curves, as in Lemma 2.8 (a). Then for the ampleness of $`๐ฉ`$ on $`\stackrel{~}{A}`$ it is enough to require $`n_i<n/2`$ instead of (4) along with the other two inequalities (3) and (5).
###### Proof.
(a): Suppose that the inequalities (3), (4) and (5) are fulfilled. The first one simply says that $`(๐ฉ,๐ฉ)_{\stackrel{~}{A}}>0`$. The second two are exactly the ones obtained from the explicit estimates for $`ฯต_๐`$.
Assume first that $`ฯต_๐`$ is computed by an elliptic curve $`E`$. Since $``$ is ample on $`A`$ one has that $`(,๐ช_A(E))_A1`$. Hence by Proposition 2.6 we have that
$$ฯต_๐=\frac{(^n[1]_A^{}^n,๐ช_A(E))_A}{4}=\frac{2n(,๐ช_A(E))_A}{4}\frac{n}{2}2n_i$$
for every $`i=1,\mathrm{},16`$.
If $`ฯต_๐`$ is not computed by by an elliptic curve, then case (a) of Proposition 2.6 and the fact that $`(๐,๐)_A=8n^2d^{}`$ give the estimate
$$ฯต_๐\frac{\sqrt{n^2d^{}}}{4}2n_i.$$
for every $`i=1,\mathrm{},16`$.
Thus if we impose these numerical conditions (3), (4) and (5) on $`n`$, $`d^{}`$ and $`n_i`$, then by Proposition 2.6 we have that $`2n_i<ฯต_๐`$. Hence by Remark 2.5 one has that $`(๐ฉ,๐ช_{\stackrel{~}{A}}(\mathrm{\Gamma }))_{\stackrel{~}{A}}>0`$ for any irreducible curve $`\mathrm{\Gamma }`$ on $`\stackrel{~}{A}`$. Therefore by the Nakai-Moishezon criterion (\[Har77, Ch. V, ยง1, Thm. 1.10\]) the line bundle $`๐ฉ`$ is ample.
(b): Suppose that $`(A,)`$ is not isomorphic to a polarized product of elliptic curves, then $`(,๐ช_A(E))_A2`$. If $`ฯต_๐`$ is computed by an elliptic curve $`E`$ we have that
$$ฯต_๐n2n_i$$
for all $`i=1,\mathrm{},16`$. Otherwise, just like in (a) one has that
$$ฯต_๐\frac{\sqrt{n^2d^{}}}{4}2n_i$$
for all $`i`$. Hence by the argument given in the proof of part (a) the line bundle $`๐ฉ`$ is ample. โ
## 3. Kummer Maps and Non-Emptiness of the Height Strata
In the preceding section we gave a way to construct points in $`_{2d}(\overline{๐ฝ}_p)`$ starting with points in $`๐_{2,d^{}}(\overline{๐ฝ}_p)`$ for some well-chosen integers $`d`$ and $`d^{}`$. Here we will show that this actually gives rise to morphisms between the stacks $`๐_{2,d^{},2,๐ฝ_p}`$ and $`_{2d,๐ฝ_p}`$. We call these maps Kummer morphisms and we give their construction in detail in Section 3.1. We will use them in Section 3.2 to produce supersingular points in $`_{2d,๐ฝ_p}(\overline{๐ฝ}_p)`$ for $`d`$ large enough. In this way we will conclude that the height strata of $`_{2d,๐ฝ_p}`$ are non-empty for these $`d`$.
### 3.1. The Kummer Morphisms
We already saw that starting with an ample line bundle $``$ on $`A`$ with $`\chi ()=d^{}`$ and fixing integers $`n,n_1,\mathrm{},n_{16}>0`$ one produces a K3 surface $`X`$ and a line bundle $``$ on it. This bundle is ample if further the numerical conditions from Lemma 2.9 are satisfied by $`d^{},n,n_1,\mathrm{},n_{16}`$. It turns out that the resulting line bundle $``$ depends only on the class of $``$ in $`\mathrm{NS}(A)`$. In other words, it depends only on the polarization $`\lambda _{}`$ defined by $``$. Indeed, the construction
$$\iota _{}\left(\pi ^{}([1]_A^{})\right)^{[1]_{\stackrel{~}{A}}}$$
gives a homomorphism of group schemes $`h:\mathrm{Pic}_{A/k}\mathrm{Pic}_{X/k}`$ and since $`\mathrm{Pic}_{X/k}^0`$ is trivial we see that $`h`$ vanishes on $`\mathrm{Pic}_{A/k}^0`$.
Suppose given numbers $`n,d^{}`$ and $`n_1,\mathrm{},n_{16}`$ satisfying the inequalities from Lemma 2.9 (a). Then using the remark made above one shows that starting with a polarized abelian surface $`(A,\lambda )`$ over an algebraically closed field $`k`$ one gets a polarized K3 surface $`(X,)`$. We will generalize this construction to a general base $`S`$. To do so let us first try to find a more intrinsic way of constructing the line bundle $``$.
Let $``$ be an ample line bundle on $`A`$ and let $`\lambda =\phi _{}`$. The polarization defined by $`[1]_A^{}`$ is $`2\lambda `$. Let $`๐ซ`$ be the Poincarรฉ sheaf on $`A\times A^t`$, where $`A^t`$ is the dual abelian surface. One has an isomorphism $`[1]_A^{}๐ซ๐ซ`$. Then $`๐=(\mathrm{id}_A\times \lambda )^{}๐ซ`$ is a symmetric ample line bundle on $`A`$ coming with an action of the group $`\{\mathrm{id}_A,[1]_A\}`$. Moreover, the polarization $`\phi _๐`$ is exactly $`2\lambda `$.
The line bundles $`๐`$ and $`[1]_A^{}`$ are isomorphic. Indeed, consider the composition
where $`\mathrm{\Delta }:AA\times A`$ is the diagonal. By construction, $`(\mathrm{id}_A\times \phi _{})^{}๐ซ`$ is the Mumford bundle $`\mathrm{\Lambda }()`$ on $`A\times A`$, which pulls-back to $`[2]^{}^2`$ under $`\mathrm{\Delta }`$. By Corollary 3 in \[Mum74, Ch. II ยง6\] we have that
$$[2]^{}=^3[1]_A^{}.$$
So we conclude that
$$[1]_A^{}=(\mathrm{id}_A\times \phi _{})^{}๐ซ.$$
We will use the bundle $`๐`$ to generalize the construction given in Section 2.2 in relative settings.
We need to make another observation in order to be able to define Kummer morphisms. In the previous section we worked over an algebraically closed field $`k`$. Then we made use of points in $`A[2](k)`$ which give rise to some exceptional divisors on the blow-up surface $`\stackrel{~}{A}`$. We will carry out the same idea in the relative case. In order to be able to consider these exceptional divisors in general, for instance if the field $`k`$ is not algebraically closed, we will be working with abelian surface with level $`2`$-structure.
Let $`(AS,\lambda ,\alpha )๐_{2,d^{},2}(S)`$ be a polarized abelian scheme over a base scheme $`S`$ with a Jacobi level $`2`$-structure $`\alpha `$. Let $`๐ซ`$ be the Poincarรฉ bundle on $`A\times _SA^t`$ where $`A^t`$ is the dual abelian scheme of $`A`$. Denote by $`๐`$ the symmetric relatively ample line bundle $`(\mathrm{id}_A\times \lambda )^{}๐ซ`$ on $`A`$ (see \[FC90, Ch. 1, ยง1, 1.6\]). Consider the blow-up $`\stackrel{~}{A}`$ of $`A`$ at $`A[2]`$. Then the automorphism $`[1]_A`$ extends to an involution $`[1]_{\stackrel{~}{A}}`$ on $`\stackrel{~}{A}`$. One forms the quotient $`X`$ of $`\stackrel{~}{A}`$ by the finite automorphism group $`\{\mathrm{id}_{\stackrel{~}{A}},[1]_{\stackrel{~}{A}}\}`$. Further we use the sheaf $`๐`$ to construct a polarization on $`๐ณ`$. We consider the sheaf
$$๐ฉ=\pi ^{}๐^n\left(\underset{j=1}{\overset{16}{}}_{j}^{}{}_{}{}^{2n_j}\right)$$
on $`\stackrel{~}{A}`$ where $`_j^{}`$ are the 16 exceptional sheaves. One uses then \[Mum74, Ch. III ยง10, Thm. 1(B)\] to conclude that $`๐ฉ`$ comes from a sheaf $``$ on $`X`$ as in Proposition 2.3 and Remark 2.4. Clearly this generalizes the construction we considered over an algebraically closed field $`k`$. The sheaf $``$ is then fiberwise ample and hence $`S`$-ample by Lemma 1.10 in \[Riz05b\] (Lemma 1.1.10 in \[Riz05a\]). This $`S`$-ample line bundle gives rise to a polarization of $`X`$. Isomorphisms of polarized abelian schemes with a Jacobi level $`2`$-structure are sent to isomorphisms of polarized K3 schemes in a natural way. In this way we get a morphism of stacks
$$K_{n,n_1,\mathrm{},n_{16}}:๐_{2,d^{},2}_{2d,[1/2]}$$
sending an object $`(AS,\lambda ,\alpha )๐_{2,d^{},2}`$ to the object $`(X,)_{2d,[1/2]}`$. We summarize this in the theorem below.
###### Theorem 3.1.
Let $`n,d^{},n_1,\mathrm{},n_{16}`$ and assume that they satisfy the numerical conditions (3), (4) and (5) of Lemma 2.9. Then there exists a morphism of algebraic stacks
$$K_{n,n_1,\mathrm{},n_{16}}:๐_{2,d^{},2}_{2d,[1/2]}$$
where $`d=2n^2d^{}_{j=1}^{16}n_j^2`$. The morphism sends a polarized abelian surface, to its associated Kummer surface with an ample line bundle.
###### Definition 3.2.
For any set of numbers $`n,d^{},n_1,\mathrm{},n_{16}`$ satisfying the inequalities (3), (4) and (5) we will call the morphism $`K_{n,n_1,\mathrm{},n_{16}}:๐_{2,d^{},2}_{2d,[1/2]}`$ constructed in Proposition 3.1 the *Kummer morphism* (or *Kummer map*) defined by $`n,d^{},n_1,\mathrm{},n_{16}`$.
Recall that there are some weaker conditions (Lemma 2.9 (b)) under which a polarized abelian surface, which is not isomorphic to a polarized product of elliptic curves, gives a polarized Kummer surface. We will deal with this case now. One has a natural map
$$p:๐_{1,1,2}\times ๐_{1,d^{},2}๐_{2,d^{},2}$$
sending a pair of polarized elliptic curves to their polarized product as in Lemma 2.8. Consider the open substack
$$๐ฐ_{2,d^{},2}=๐_{2,d^{},2}p(๐_{1,1,2}\times ๐_{1,d^{},2})$$
As we saw above one can construct a polarized Kummer surface out of any such abelian surface. In the same lines one gets
###### Proposition 3.3.
Let $`n,d^{},n_1,\mathrm{},n_{16}`$ satisfy the conditions of Lemma 2.9 (b). Then there exists a morphism of stacks
$$K_{n,n_1,\mathrm{},n_{16}}:๐ฐ_{2,d^{},2}_{2d,[1/2]}$$
as constructed in Theorem 3.1 where $`d=2n^2d^{}_{j=1}^{16}n_j^2`$.
###### Proof.
$`K_{n,n_1,\mathrm{},n_{16}}`$ maps a polarized abelian surface to the polarized Kummer surface and this time one has to impose the milder conditions of Lemma 2.9 due to the fact that the polarized products of elliptic curves are excluded. โ
###### Remark 3.4.
Let $`(A,\lambda ,\alpha )`$ be an a polarized abelian surface over a finite field $`k`$ of characteristic $`p>2`$. Then using Lemma 2.2 we see that the point $`K_{n,n_1,\mathrm{},n_{16}}((A,\lambda ,\alpha ))`$ in $`_{2d,๐ฝ_p}(k)`$ belongs to
1. $`_{2d,๐ฝ_p}^{(1)}_{2d,๐ฝ_p}^{(2)}`$ if $`A`$ is ordinary;
2. $`_{2d,๐ฝ_p}^{(2)}_{2d,๐ฝ_p}^{(3)}`$ if the $`p`$ rank of $`A`$ is 1;
3. $`\mathrm{\Sigma }_9\mathrm{\Sigma }_{10}`$ if $`A`$ is supersingular but not superspecial;
4. $`\mathrm{\Sigma }_{10}`$ if $`A`$ is superspecial.
### 3.2. Non-Emptiness of the Height Strata
Fix a prime number $`p>2`$. We will prove here that the height strata of $`_{2d,๐ฝ_p}`$ are non-empty for every large enough $`d`$ prime to $`p`$. The idea is to use the Kummer maps and show that $`_{2d,๐ฝ_p}`$ contains a supersingular Kummer surface. Then by Theorem 1.3 all strata are non-empty and so one has the claimed dimensions.
###### Theorem 3.5.
For every large enough $`d`$ prime to $`p`$ the height strata of $`_{2d,๐ฝ_p}`$ are non-empty.
We will need the following result first.
###### Lemma 3.6.
Every residue class modulo $`2\times 9^2`$ can be represented by an integer of the form $`_{j=1}^{16}n_j^2`$ with $`1n_j4`$ for all $`j`$.
###### Proof.
Explicit calculation. โ
###### Remark 3.7.
We believe that the statement of the preceding lemma remains valid for all $`n9`$. In other words, all residues modulo $`2n^2`$ can be represented by an integer of the form $`_{j=1}^{16}n_j^2`$ with $`1n_j<\frac{n}{2}`$. This is true for $`n[9,45]`$.
Proof of Theorem 3.5. First note that if for a given $`d`$ there exist numbers $`d^{}`$ and $`n_1,\mathrm{},n_{16}`$ giving a Kummer map
$$K_{d^{},n_1,\mathrm{},n_{16}}:๐ฐ_{2,d^{},2,๐ฝ_p}_{2d,๐ฝ_p},$$
as in Proposition 3.3, then the height strata of $`_{2d,๐ฝ_p}`$ are non-empty. This follows from Remark 3.4 as one can find a supersingular point in $`๐ฐ_{2,d^{},2,๐ฝ_p}`$.
Take $`n=9`$ and let $`d^{}26`$ so that the conditions of Lemma 2.9 (b) give $`n_j[1,4]`$. By Lemma 3.6 we can pick up 162 sets of numbers $`(n_1,\mathrm{},n_{16}),1n_j4`$ which define Kummer maps as above and such that $`F(n_1,\mathrm{},n_{16})=_{j=1}^{16}n_j^2`$ gives all possible resides modulo $`2\times 9^2`$. Hence the images of $`๐ฐ_{2,d^{},2,๐ฝ_p}`$ under those Kummer maps land in $`_{2d,๐ฝ_p}`$ where $`d=2\times 9^2d^{}_{j=1}^{16}n_j^2`$. Using this set of 162 sixteen-uples $`(n_1,\mathrm{},n_{16})`$ and letting $`d^{}26`$ vary we can construct Kummer maps for $`_{2d,๐ฝ_p}`$ for all $`d2\times 9^2\times 2616=4196`$. Therefore by the remark we started with the height strata of these moduli stacks are non-empty. This proves the assertion. โ
Using Kummer maps we saw that the height strata of $`_{2d,๐ฝ_p}`$ are non-empty if $`d4196`$. On the other hand one has that the Fermat K3 surface
$$x^4+y^4+z^4+w^4=0$$
in $`^3`$, which is a Kummer surface by \[Shi75, Thm. 1\], is supersingular if $`p3(mod4)`$. Hence using explicit Kummer surfaces one can show the non-emptiness of the height strata in lower polarization degrees. Using the same ideas as above we will โcut some more moduli pointsโ of abelian surfaces in order to improve the estimates in Lemma 2.9. In this way we will lower the bound for $`d`$.
First we will settle the case when $`d`$ is even. Let as before $`A`$ be an abelian surface and let $`X`$ be the associated Kummer surface. The invertible sheaf $`_{j=1}^{16}_j`$ is divisible by 2 in $`\mathrm{Pic}(X)`$. Hence $`_{j=1}^{16}_j^{}`$ comes from a line bundle on $`X`$ modulo 2 torsion in $`\mathrm{Pic}(\stackrel{~}{A})`$. Note that this torsion has to come from $`\mathrm{Pic}^0(A)`$. So it does not change neither our constructions nor the intersection indexes we were dealing with. Consider the following subset of $`๐_{2,d^{}}(\overline{๐ฝ}_p)`$
$$\begin{array}{cc}\hfill U_{2,d^{}}^3=\{(A,\lambda )๐_{2,d^{}}(\overline{๐ฝ}_p)|& \mathrm{there}\mathrm{does}\mathrm{not}\mathrm{exist}\mathrm{an}\mathrm{isogeny}\hfill \\ & E\times EA\mathrm{of}\mathrm{degree}<3\}.\hfill \end{array}$$
For any $`d^{}`$ the supersingular locus of $`๐_{2,d^{},๐ฝ_p}`$ remains non-empty because we exclude only finitely many points of it. Let $`(A,\lambda )U_{2,d^{}}^3`$ and let $``$ be any ample line bundle on $`A`$ defining the polarization $`\lambda `$. Then by Lemma 2.8 we have that $`(,๐ช_A(E))_A3`$ for every elliptic curve $`E`$ in $`A`$. Taking this into account and following the proofs of Lemma 2.9 and Theorem 3.1 one constructs a Kummer map of sets
$$K_d^{}:U_{2,d^{}}^3_{2d,๐ฝ_p}(\overline{๐ฝ}_p)$$
where $`n=n_1=\mathrm{}=n_{16}=1`$, $`d^{}32`$ and $`d=2d^{}16`$. Hence by Remark 3.4 we can conclude that
###### Corollary 3.8.
For all even $`d48`$ prime to $`p`$ the height strata of $`_{2d,๐ฝ_p}`$ are non-empty.
For the odd case we will construct Kummer maps with $`n_1=1,n_2=\mathrm{}=n_{16}=2`$. Define as before the set
$$\begin{array}{cc}\hfill U_{2,d^{}}^8=\{(A,\lambda )๐_{2,d^{}}(\overline{๐ฝ}_p)|& \mathrm{there}\mathrm{does}\mathrm{not}\mathrm{exist}\mathrm{an}\mathrm{isogeny}\hfill \\ & E\times EA\mathrm{of}\mathrm{degree}<8\}\hfill \end{array}$$
Take a point $`(A,\lambda )U_{2,d^{}}^8`$ and let $``$ be any ample line bundle on $`A`$ defining the polarization $`\lambda `$. Then according to Lemma 2.8 (b) we have that $`(,๐ช_A(E))_A9`$ for all elliptic curves $`E`$ in $`A`$. Just as above one constructs a Kummer map of sets
$$K_d^{}:U_{2,d^{}}^8_{2d,๐ฝ_p}(\overline{๐ฝ}_p)$$
where $`n=1,n_1=1,n_2=\mathrm{}n_{16}=2,d^{}512`$ and $`d=2d^{}15\times 41=2d^{}61`$. Using these maps we obtain the following result.
###### Corollary 3.9.
For every odd $`d963`$ prime to $`p`$ the height strata of $`_{2d,๐ฝ_p}`$ are non-empty.
|
warning/0506/astro-ph0506178.html
|
ar5iv
|
text
|
# Advances in Multi-Dimensional Simulation of Core-Collapse Supernovae 11footnote 1To appear in published proceedings of Open Issues in Core-Collapse Supernovae, which was conducted at The Institute for Nuclear Theory, University of Washington, Seattle, WA, USA, June, 2004.
## 1 Introduction: The Supernova โProblemโ
For over a decade researchers have struggled to model the convective post-bounce epoch of core collapse supernovae. The radiative-hydrodynamic flows that occur in the region below the stalled prompt shock have held both promise and pitfall for the supernova modeler. The promise of this phenomenon is that it might explain the long sought-after mechanism that converts the core bounce into the observed explosion. The pitfalls are legion, mostly involving a complex convective flow structure that is three-dimensional in nature and couples neutrinos to matter strongly. For this reason there remain many open issues in modeling the convective epoch of core collapse supernovae.
The supernova โproblemโ persists, despite more than four decades of concentrated research. The problem is this: We have no convincing explanation as to how the core collapse, which ends the evolution of a massive star, rebounds in such a way as to generate the explosion we observe in nature. The most realistic supernova models collapse and rebound, but create a shock wave that doesnโt eject matter, either on the hydrodynamic timescale ($`10`$ ms), or the diffusive timescale of the escaping neutrino radiation ($``$ 1โ10 s), or on any other timescale we can model.
This is not a problem of overall energetics. The gravitational energy released during core collapse and the subsequent neutron-star cooling phase is several factors of $`10^{53}`$ erg. In contrast, the kinetic energy of the explosion required for consistency with observation is only $`10^{51}`$ erg. Instead of insufficient energy, the problem is one of energy conversion and transportโhow a sufficient portion of the released gravitational energy is imparted to the material ejectus, giving it the requisite kinetic energy.
It has been understood for many years that neutrinos play a vital role in this process. In fact, essentially the entire remaining 99% of released energy (that which is not converted to kinetic energy of the matter) is radiated away as neutrinos. Thus, an accurate treatment of neutrino processes is a necessary component of any realistic model for a supernova.
In this article, we present what we currently regard as the most important issues in core-collapse supernova modeling. In Sec. 2, we discuss the major components that need to be part of any serious modeling endeavor. In Sec. 3, we present an outline of V2D, our new two-dimensional (2-D) supernova simulation code. Section 4 contains some preliminary results using this code. Our conclusions are in Sec. 5.
## 2 The Components of a Supernova Simulation
Broadly speaking, there are four main components to current supernova simulation models: (1) hydrodynamics, to track the collapse, rebound, and ejection of stellar material, (2) neutrino transport, to track the production of neutrino radiation and to follow its propagation and emission from the star, (3) nuclear microphysics, to describe the diverse states of matter encountered throughout a simulation, and (4) neutrino microphysics, to describe the reactions and interactions involving neutrinos and matter. It must be stressed that all of these components are tightly coupled to one another. Thus, the most effective models are designed with this coupling built in ab initio.
Hydrodynamics. For simplicity of implementation, it has been customary for supernova codes to employ explicit hydrodynamics, with either a Newtonian or a general relativistic formulation. Implicit algorithms have usually been avoided since they require the computationally expensive solution of large systems of non-linear equations.
Regardless of which approach is used, the hydrodynamic portion of the problem requires solution of some form the following equations, expressed here in Newtonian formalism:
$$\frac{\rho }{t}+\mathbf{}\left(\rho ๐ฏ\right)=0$$
(1)
$$\frac{\left(\rho Y_e\right)}{t}+\mathbf{}\left(\rho Y_e๐ฏ\right)=m_b\underset{f}{}๐ฯต\left(\frac{๐_ฯต}{ฯต}\frac{\overline{๐}_ฯต}{ฯต}\right)$$
(2)
$$\frac{E}{t}+\mathbf{}\left(E๐ฏ\right)+P\mathbf{}๐ฏ=\underset{f}{}๐ฯต\left(๐_ฯต+\overline{๐}_ฯต\right)$$
(3)
$$\frac{\left(\rho ๐ฏ\right)}{t}+\mathbf{}\left(\rho \mathrm{๐ฏ๐ฏ}\right)+\mathbf{}P+\rho \mathbf{}\mathrm{\Phi }+\mathbf{}\left\{\underset{f}{}๐ฯต\left(๐ฏ_ฯต+\overline{๐ฏ}_ฯต\right)\right\}=0.$$
(4)
Equation (1) is the continuity equation for mass, where $`\rho `$ is the mass density and $`๐ฏ`$ is the matter velocity, and where these quantities, and those in the following equations, are understood to be functions of position x and time $`t`$. Equation (2) expresses the evolution of electric charge, where $`Y_e`$ is the ratio of the net number electrons over positrons to the total number of baryons. In the presence of weak interactions, the right hand side is non-zero to account for reactions where the number of electrons can change. Here, we express the net emissivity of a neutrino flavor (of energy $`ฯต`$) and its antineutrino by $`๐_ฯต`$ and $`\overline{๐}_ฯต`$, respectively. This expression is integrated over all neutrino energies and summed over all neutrino flavors $`f`$. The mean baryonic mass is given by $`m_b`$. Evolution of the internal energy of the matter is given by the gas-energy equation, Eq. (3), where $`E`$ is the matter internal energy density and $`P`$ is the matter pressure. Again, the right hand side of this equation is non-zero whenever energy is transferred between matter and neutrino radiation as a result of weak interactions. We note that it is also possible to substitute for Eq. (3) an expression for the evolution of the total matter energy (internal plus kinetic plus potential). Finally, Eq. (4) expresses gas-momentum conservation, where $`\mathrm{\Phi }`$ is the gravitational potential, and $`๐ฏ_ฯต`$ and $`\overline{๐ฏ}_ฯต`$ are radiation-pressure tensors for each energy and flavor of neutrino and its anti-neutrino, respectively.
These equations must be discretized for solution within a computational framework. Traditionally, with one-dimensional models, it has been convenient to use Lagrangean methods, in which a computational mesh strictly co-moves with the mass elements of the fluid. With the advent of multi-dimensional models, however, it is common to use Eulerian hydrodynamics, where the mesh is fixed in an inertial frame of reference. This is because purely Lagrangean methods are difficult to implement in multi-dimensional schemes without the mesh suffering distortion and entanglement in convectively active regions.
For all the benefits of Eulerian meshes, they also present a number of thorny issues. This is especially true for spherical polar meshes, the most natural choice for supernova modeling. The most obvious issue is the coordinate singularity that exists when the polar angle, $`\theta 0`$. In addition, polar meshes exacerbate the problem of the timestep-restricting Courant-Friedrichs-Levy (CFL) condition at the center of the core. To deal with these issues, there are numerous resolutions and combinations of resolutions under active consideration. These include implicit methods, unstructured meshes, body-fitted meshes, and adaptive mesh refinement (AMR).
As mentioned above, it is also necessary to choose between a total energy and an internal energy formulation . For the supernova problem, an internal energy formulation, as given in Eq. (3), is preferred. This is because much of the energy is internal, as opposed to kinetic. Solving the gas-energy equation helps insure an accurate calculation of the entropy, which is critical in degenerate regimes where a small change in energy can lead to a large change in temperature.
The hydrodynamic algorithm must also have convergence properties that can deal with a realistic equation of state. This is particularly important in the regions of non-convex phase changes, such as the transition between nuclei and continuous nuclear matter.
Neutrino Transport. This component is the most difficult to implement in a supernova model and the most time-consuming computationally. This is because supernova neutrinos cannot, in general, be described by an equilibrium distribution function. A solution requires a complete phase-space description of each neutrinoโs position and momentum. To obtain such a solution, one must solve the six-dimensional Boltzmann Transport Equation or some reasonable approximation thereof. This extra dimensionality easily leads to the transport calculation completely dominating a simulation in terms of computer memory, execution time, and I/O requirements.
The Boltzmann Transport Equation (BTE) can be expressed in terms of the radiation intensity, $`I=I(ฯต,๐ฑ,๐,t),`$ where $`ฯต`$ is the energy of a neutrino, $`๐ฑ`$ its position, and $`๐`$ the solid angle into which the neutrino radiation is directed. In terms of $`I`$, the Newtonian BTE can be expressed as
$$\frac{1}{c}\frac{I}{t}+๐\mathbf{}I+\underset{i}{}a_i\frac{I}{p_i}=\left(\frac{f}{t}\right)_{\mathrm{coll}.},$$
(5)
where $`a_i`$ is the $`i^{th}`$ component of the matter acceleration and $`p_i`$ the $`i^{th}`$ component of the momentum of the neutrino. The right hand side of Eq. (5) lumps together the contributions from all interactions that a neutrino might experience and is collectively referred to as the collision integral.
A storm of issues faces one who implements a neutrino transport algorithm. Mezzacappa and Bruenn have the only โfullโ solution to the BTE implemented in supernova simulations and then only with one-dimensional hydrodynamics. Upon moving to multi-dimensional models, the full solution of the BTE becomes yet more challenging to implement and more time-consuming to compute. However, this is the way that the field must ultimately go. (Livne and colleagues purport to implement a two-dimensional $`S_n`$ solution of the BTE. However, this solution omits critical matter-radiation coupling terms and no numerical details of the method have been disclosed.)
In the meantime, a number of approximate transport moment methods have emerged. The most successful of these is use of a finite series of angular moments of the BTE. When this approach is taken, a limiting scheme is then required to close the resulting equations. Breunn implemented a $`P_1`$ scheme, with flux-limiting. Myra et al. closed the zeroth angular moment of the BTE by implementing the Levermore and Pomraning flux limiter. Bowers and Wilson also used a flux-limiting scheme of their own device. More recently, Rampp and Janka have implemented a variable-Eddington-factor approach to solve the first two angular moments of the BTE. In all the above cases, however, the implementation has been made in only one spatial dimension. (Janka et al. are developing a two-dimensional Boltzmann transport code (MuDBaTH), but the numerical details are as yet unpublished.)
Since all the schemes noted so far derive monochromatic transport equations, yielding a separate equation for each neutrino energy, they are referred to as multi-group schemes. Those that combine multi-group and flux-limiting are known as multi-group flux-limited diffusion (MGFLD) schemes.
A much simpler alternative to multi-group schemes is so-call โgreyโ transport, which is derived by integrating the BTE over both neutrino energy and angle. To perform these integrals, one must assume a spectral shape for the neutrino distribution. Typically, this requires defining an arbitrary neutrino โtemperature,โ and assuming that neutrinos can be parameterized by some kind of equilibrium Fermi-Dirac distribution. Among relatively recent models, this was first implemented by Cooperstein, van den Horn, and Baron, and later by Swesty and by Herant et al. This approximation is still in active use by the latter group.
Grey schemes have numerous shortcomings. First, work with multi-group schemes has shown that in areas where accurate neutrino transport is critical, neutrinos do not assume any kind of distribution that can parameterized once and for all as required by grey transport. Spectral distributions constantly evolve and, thus, a multi-group description is required to obtain even a qualitatively correct description. More troubling, Swesty has shown that by adjusting the grey parameterization within very small bounds, it is possible to โdialโ an explosion (or failed explosion) with the appropriate choices of these unknowable and unphysical tuning parameters. Hence, although grey codes have utility for making a sweeping exploration of parameter space, any scientific conclusions that rely on them should be viewed as highly suspicious, and not regarded as in any way definitive.
Regardless of which transport scheme is implemented, another critical issue that must be faced is matter-radiation coupling. Coupling occurs in Eqs. (2)โ(4) for lepton number, energy, and momentum evolution. Coupling that occurs on the right-hand side of these equations has a conceptually simple analytic structure. However, momentum transfer is more troublesome in approximate schemes. This is because the radiation momentum equation is often truncated in a way that makes the accuracy of the calculated momentum transfer less certain.
Matter-radiation coupling also enters implicitly in the neutrino transport equation through spectral rearrangement terms and in the dynamic diffusion term. Both these terms are frequently and erroneously neglected in supernova models, even though the dynamic diffusion term is the leading order contribution in optically-thick regions.
Equation of State. Adequate modeling of stellar-core collapse requires an equation of state (EOS) that handles a density range of roughly $`10^5`$$`10^{15}`$ g $`\mathrm{cm}^3`$, a temperature range of 0.1โ25 MeV, and an electron-fraction range of 0.0โ0.5. The EOS must also be able to handle different regimes of equilibrium states. Throughout most of the core, the material is in nuclear statistical equilibrium (NSE) and is usually modeled by one of the NSE equations of state. Although the gross features of nuclear matter are thought to be well-understood, there is still much open ground for investigation. Fertile regimes for such work include the EOS at supernuclear densities. In addition, little is known about the nature of nuclei at subnuclear densities when $`Y_e`$ is small.
Matter in the silcon shell and beyond does not attain NSE until the bounce-shock wave passes through it. Dealing with the transition between NSE and non-NSE EOSโs and with the network of nuclear reactions that joins them is a challenge that is only beginning to be addressed.
Neutrino Microphysics. Since the energetics of a core-collapse supernova is primarily a neutrino phenomenon, it is necessary to have correct opacities and rates for the various neutrino processes that are important. The collection of reactions that are important, or possibly important, to the supernova problem is rich and has evolved through the years. Arguably the most important development came with the discovery of weak neutral currents, from which it could be inferred that the dominant contribution to neutrino opacity in a collapsing stellar core is from coherent elastic scattering of neutrinos from nuclei.
The list of possible neutrino interactions is nearly endless, but those of demonstrated importance include the coherent scattering just mentioned, as well as conservative scattering from free nucleons. Also of undisputed importance are electron capture by protons (and protons bound in nuclei), neutrino production through electron-positron pair annihilation, and neutrino-electron scattering.
In recent years, with the experimental evidence pointing strongly to the existence of neutrino oscillations, it is also important to investigate the possible role of flavor-changing interactions to the supernova problem. Investigation into this has begun, but has yet to be incorporated in a detailed simulation.
## 3 V2D: A New Code for Two-Dimensional Radiation Hydrodynamics
Our new radiation-hydrodynamic simulation code, V2D, is a two-dimensional, Newtonian, pure Eulerian, staggered-mesh code based on a modified version of the algorithm for ZEUS-2D by Stone and Norman. Following Stone and Norman, it has been designed for use in a general orthogonal two-dimensional geometry, which makes its utility extend beyond the supernova problem.
V2D is an entirely new implementation, coded according to the Fortran 95 standard. It is a distributed-memory parallel code that uses calls to MPI-1 for message passing between processes. It has been designed for easy portability between computing platforms and currently runs on systems ranging from as small as a Linux-based laptop to as many as 2048 processors of an IBM SP. To aid in this portability, the input and output is formatted using parallel HDF5, which is built on the MPI-I/O portion of the MPI-2 standard.
One of the major design goals of V2D is componentization and, to adhere to this principle, we insist on completely separating microphysics from the numerical implementation of our radiation-hydrodynamics algorithm. This isolation of mathematics and computational science from physics has allowed significant contributions from applied mathematicians to enhancing the performance of our code.
The V2D algorithm relies on operator splitting, with advection steps split from source-term steps. Hydrodynamic and neutrino-transport source-term steps and coupling are interleaved. At the start of each simulation timestep, the gravitational mass interior to each point is calculated. Since the collapsed core is nearly spherically symmetric and highly condensed, we approximate the gravitational mass assuming that mass interior to the point of interest is in a spherically symmetric distribution. In future versions of our code, we will implement a more accurate Poisson solver to calculate the (slightly) non-spherical gravitational potential.
In a break from Stone and Normanโs method, V2D next performs the advection sweeps in the radiation-hydrodynamic quantities (mass density, matter internal energy, velocities and momenta, electron fraction, and neutrino distributions). Following this, a neutrino transport step is performed for each flavor and each matter-radiation energy exchange is calculated.
The matter pressure is next updated, upon which the gravitational and matter- and radiation-pressure forces are applied to the matter. Artificial viscosity is calculated next and its contributions applied to the fluid. Finally, the gas-energy equation is solved.
This procedure is repeated for each timestep in a simulation, with the provision that advection sweeps are ordered alternately according to timestep (i.e., $`x_1`$-direction first, followed by $`x_2`$, or vice versa).
### 3.1 Neutrino Transport Implementation
As an extension of earlier work by us, we implement neutrino transport by taking the zeroth angular moment of the BTE to yield the following neutrino monochromatic energy equation in the co-moving frame:
$$\frac{E_ฯต}{t}+\mathbf{}\left(E_ฯต๐ฏ\right)+\mathbf{}๐
_ฯตฯต\frac{}{ฯต}(๐ฏ_ฯต:\mathbf{}๐ฏ)=๐_ฯต,$$
(6)
where $`E_ฯต`$ is the neutrino energy density per unit energy interval at position $`๐ฑ`$ and time $`t`$, $`๐
_ฯต`$ is the neutrino energy flux per unit energy interval, and $`๐ฏ_ฯต`$ and $`๐_ฯต`$ are as defined earlier. The expression $`๐ฏ_ฯต:\mathbf{}๐ฏ`$ indicates contraction in both indices of the second-rank tensors $`๐ฏ_ฯต`$ and $`\mathbf{}๐ฏ`$. There is a corresponding equation to describe the antineutrinos. This pair of equations is repeated for each neutrino energy $`ฯต,`$ and neutrino flavor. We currently track electronic, muonic, and tauonic neutrinos.
Equation (6) is closed using Levermore and Pomraningโs prescription for flux-limited diffusion, which allows us to express $`๐
_ฯต`$ as
$$๐
_ฯต=D_ฯต\mathbf{}E_ฯต,$$
(7)
where $`D_ฯต`$ is a โvariableโ diffusion coefficient that varies in such a way as to yield the correct fluxes for the diffusion and free streaming limits and an approximate solution in the intermediate regime. This prescription also provides the elements of the radiation-pressure tensor $`๐ฏ_ฯต`$.
Presently, we employ the same prescriptions for electron capture and conservative scattering that we have used in the past. The rates for these process are calculated on the fly within the course of a simulation. Our model also implements neutrino production via electron-positron pair annihilation, as in Yueh and Buchler and Bruenn. Neutrino-electron scattering has been implemented, but is not currently turned on in the preliminary results we present here. These latter two sets of processes use tables of precomputed rates, which are interpolated via a tri-linear interpolation scheme over neutrino energy $`ฯต`$, temperature $`T`$, and electron chemical potential $`\mu _e`$.
Since the neutrino CFL restriction on a transport timestep is far too restrictive to permit an explicit solution, we use a purely implicit method to solve the transport. The equations comprising the description of each neutrino-antineutrino species are assembled in matrix form. We note that the second (advective) term in Eq. (6) is omitted from this process since it has been already treated during the operator splitting of the advective step described above. Blocking terms arising from Fermi-Dirac statistical restrictions on final neutrino states make this a system of non-linear equations. Fortunately, the system is sparse, which makes it amenable to solution by sparse iterative methods. A nested procedure is used, employing Newton-Krylov methods. In the innermost loop, a linearized system is solved using preconditioned Krylov-subspace methods. The outer loop uses a Newton-Raphson iterative scheme to resolve the non-linearity of the system. Besides being an effective general procedure for sparse systems, our implementation of parallel preconditioners also insures that is amenable to large-scale solution on parallel architectures. This is the chief reason our code exhibits its high degree of scalability across many platforms.
### 3.2 Equation of State
V2D is designed to use an arbitrary equation of state and we use several in the course of testing the code. For production runs, however, we use the Lattimer-Swesty EOS in tabular form. The thermodynamic quantities are tabulated in terms of independent variables, density, $`\rho `$, temperature, $`T`$, and electron fraction, $`Y_e`$. We have tabulated this EOS in a thermodynamically consistent way according to the prescription in Swesty. (We refer to this combination collectively as LS-TCT.) We note that although the Lattimer-Swesty EOS is commonly used, and tabulations of it are also common, most tabulations are not constructed in such a way as to guarantee thermodynamic consistency. When non-thermodynamically-consistent tables are used, spurious entropy can be generated or lost. Such problems have been sometimes incorrectly attributed to the Lattimer-Swesty EOS, rather than erroneous tabulation of the otherwise consistent EOS.
To guarantee tabular thermodynamic consistency, LS-TCT uses bi-quintic Hermite interpolation in the Helmholtz free energy $`F`$, as a function of $`T`$ and $`\rho `$. Functional dependence on $`Y_e`$ changes slowly enough to permit linear interpolation. This procedure is required since satisfaction of the Maxwell relations requires consistency among the second derivatives of $`F`$. In addition, we desire fidelity of the interpolation and continuity of derivatives to the underlying tabular data for both $`F`$ and its derivatives, $`(F/T)_\rho `$ and $`(F/\rho )_T`$.
Apart from thermodynamic considerations, we also want to insure that there are no discontinuities that might cause difficulties in the hydrodynamics. Hence, LS-TCT also insists that interpolations of the second derivatives of $`F`$ approach the correct tabulated values. (This is the equivalent of saying that we require the derivatives of pressure and internal energy with respect to temperature and density be continuous.)
A final requirement is that we wish the interpolation function and its first and second derivatives be continuous across table-cell boundaries. This insures that nothing untoward happens as a fluid element migrates from one thermodynamic regime of interest to another.
### 3.3 Software Validation and Verification
When engaged in a major project, such as V2D, it is important for software developers to be constantly vigilant about the quality of software being produced. It is important that the software meet the requirements that it is intended to address (validation) and that the code yields correct answers (verification).
We take these issues seriously and, in an effort to address them, have implemented strict source-code control and testing procedures to ensure that our software meets our rigorous standards. One of the most important elements of our program has been the implementation of a suite of regression tests, which we currently run four times daily. This is not a static suite, but is constantly growing. Our eventual aim is to cover every major element of code in V2D. Currently our suite consists of about two dozen separate problems that include tests of the hydrodynamics, neutrino transport, parallel solvers, message passing, and parallel I/O. Wherever possible, we try to include problems with analytic or at least verifiable solutions.
Although implementation of these procedures is labor intensive and time-consuming, we feel it is a justifiable investment. With current regression tests, our procedures have already been effective in finding errors in our code. They have also served as an effective safeguard against introducing new errors as we continually enhance V2Dโs functionality.
## 4 Initial Results
We are using V2D to carry out our first 2-D multigroup models of the post-bounce epoch. To date, simulations have not reached a sufficient time that would allow us to be certain about whether an explosion is obtained. Nevertheless, the results warrant some discussion as they reveal important features of the post-bounce epoch.
### 4.1 Initial Model
For a progenitor model we employ the widely-used Woosley and Weaver S15S7B2 $`15M_{}`$ progenitor. Much previous work has focused on the evolution of this progenitor through collapse, core bounce, and convective phases. The Fe core, the Si shell, and a portion of the O shell area are zoned into a 256 radial-mass-zone mesh with zoning that is tuned (by trial and error) so as to yield a high spatial resolution grid in the proto-neutron star and the inner 200 km of the collapsed core at bounce. This tuned zoning sets up a radial grid that is compatible with subsequent 2-D Eulerian simulations. The neutrino-energy spectrum, ranging from 0โ375 MeV, is discretized into 20 energy groups with group widths that increase geometrically with energy so as to resolve accurately the Fermi surface of the electrons and neutrinos in the proto-neutron star. The initial values for $`T`$, $`\rho `$, and $`Y_e`$, are interpolated from the original S15S7B2 data onto the Langragean mass grid and the initial radial coordinates of each mass shell are computed consistently with density. The neutrino energy densities $`E_\nu `$ are initialized to a small non-zero value that yield an initial neutrino luminosity that is many orders of magnitude below what the precollapse thermal pair-production luminosity would be. When the simulation is started in its pre-collapse quasi-static phase, the luminosity stabilizes within a light crossing time (5-10 ms).
### 4.2 Lagrangean Collapse Calculations
The initial model is collapsed using a 1-D Newtonian Lagrangean radiation hydrodynamics code RH1D that uses the mass and energy meshes described above. The evolution algorithm for each timestep utilizes operator splitting to first carry out a Lagrangean hydrodynamics step followed by Lagrangean neutrino evolution steps for each of the three neutrino flavors. After each Lgrangean neutrino evolution step the matter internal energy and electron fraction are corrected for any energy and lepton number exchange that has occurred.
The model contains neutrino microphysics as described in Bruenn with two exceptions. The neutrino-nuclei scattering opacity has been modified to take into account the form-factor introduced by Burrows, Mazurek, and Lattimer and we have neglected the effect of neutrino/anti-neutrino annihilation. Full neutrino-electron scattering as described in this paper is included in the code but is not turned on in the model described in this paper. Additional effects such as nucleon recoil, ion-ion correlations, etc. are being considered as follow-ons to this baseline model.
The EOS utilized is the TCT tabularized version of the nuclear EOS Lattimer-Swesty (LS-TCT) with the $`k=180`$ MeV parameter set, with the exception of the electron EOS that has been updated to improve accuracy and the range of applicability. The original LS EOS chose a value for the alpha particle binding energy $`B_\alpha =28.3`$ MeV that did not correctly account for the neutron-proton mass difference. This parameter has been subsequently corrected but in the model presented in this paper, the original value has been retained so as to allow comparison to other work.
Using RH1D with the microphysics and meshes described above the core collapses in approximately 233 ms. The central conditions at bounce are approximately $`T10.3`$ MeV, $`\rho 2.71\times 10^{14}\mathrm{g}\mathrm{cm}^3`$, $`Y_e0.306`$, and $`Y_{\mathrm{}}0.37`$. These conditions are in good agreement with those obtained in Lagrangean MGFLD and MGBT models .
### 4.3 Eulerian 2-D Calculations
Our 2-D models are carried out in spherical polar coordinates. The initial conditions for our 2-D simulations are taken from the 1-D Langrangean simulations as the central density of the core reaches the nuclear saturation density. The $`T`$, $`\rho `$, and $`Y_e`$ profiles at this point are shown in Figure 1 and are taken at approximate time of $`233.087`$ ms.
The choice of this epoch to begin our 2-D models was made for two reasons. First, the prompt shock propagates and stalls into a quasi-static equilibrium on the Eulerian mesh. We have found that if we attempt the transition from a Lagrangean algorithm to an Eulerian algorithm at the point where the shock has stalled, there will be noticeable differences in the quasi-static equilibrium caused by the presence of the nuclear โpastaโ phase transition. These differences in the equilibrium point can set off spurious unphysical shocks. By choosing to let the prompt shock propagate a stall on an Eulerian mesh, we avoid this problem. The second reason for choosing the initial point for our 2-D models near the moment of bounce is that the radial coordinates of the zones in the outer part have achieved desirable values for an Eulerian simulation.
The radial zoning for the 2-D models is taken directly from the radial coordinates of the 1-D Langrangean model zones. In this way we avoid any need for remapping of data between radial zones. This eliminates the introduction of any spurious forces into the initial conditions for the 2-D models. The initial data for $`T`$, $`\rho `$, $`Y_e`$, $`v_r`$, and the neutrino energy densities $`E_\nu `$ and $`\overline{E}_\nu `$ for each neutrino flavor are taken directly from the 1-D Langrangean model. The initial velocity in the $`\theta `$ direction is set to zero. The data are mapped in the polar-angular direction in a spherically symmetric fashion. The angular grid consists of 256 zones uniformly spaced in angle over the range of $`0\theta \pi `$. We place a small sinusoidal perturbation in the electron fraction $`Y_e`$ to seed convection in the region between 100 and 200 km of the form $`(Y_e)_{\mathrm{perturb}}=(Y_e)_{\mathrm{Lagrangean}}+C_p\mathrm{sin}(4\theta )`$ where $`C_p=10^6`$. The energy group structure is also left unchanged from the 1-D Langrangean runs so there is no need to remap the data in the energy dimension. In the remainder of this paper we shall refer to this model as Production Run 37 (PR37).
The use of spherical polar coordinates in combination with explicit hydrodynamic algorithms to model spatial domains that include the origin gives rise to a numerical stability problem. At $`r=0`$ the spherical coordinate system is degenerate and all zones that include the origin as a vertex should be in instantaneous sonic communication with one another. However, standard explicit numerical finite-difference, finite-volume, or finite-element techniques are limited to nearest-neighbor type spatial coupling and do not include numerical coupling between all zones containing a given vertex. In order to circumvent this problem we numerically introduce โbafflesโ into the center of the collapsed core, as though it were a tank of fluid, to prevent movement of fluid in the angular direction inside a certain radius. Since no fluid movement occurs in the $`\theta `$ direction inside the baffle radius, there is no CFL restriction based on the zone size in the $`\theta `$ direction for zones inside that radius. Nevertheless, the zones on either side of the baffle are sonically connected as sound waves flow around the outer edge of each baffle. We strive to keep the baffle radius small, so that the flow in the $`\theta `$ direction remains unimpeded in any region where convective instabilities may develop. For the mode described in this paper the baffle radius is approximately 8.5 km which yields an average CFL timestep of about $`5\times 10^7`$ s. As we will see, this baffle radius is well inside of any proto-neutron star (PNS) instability region.
The 2-D calculations have been carried out on the IBM-SP system at the National Energy Research Scientific Computing Center (NERSC). The models are run on 1024 processors with parallelism handled via message passing via calls to MPI libraries. Model PR37 required approximately 50,000 processor-hours of CPU time to reach a simulation time of 16 ms post-bounce.
### 4.4 The Onset of Convection
The 2-D models are carried out from the point of bounce. As expected the prompt shock weakens while propagating outward and finally stalls. In the 2-D models, the radius of the shock at 5 ms post-bounce is near 60 km and propagating outwards very slowly. This is in good agreement with our 1-D Lagrangean models.
One difference that we see from previous works including the grey models carried out by Swesty is that convective instabilities are born much earlier when the 2-D models are initialized near the point of core bounce. By 10 ms after bounce, the model has developed two separate unstable layers. The outer layer, shown in Fig. (2), is the classic entropy driven Rayleigh-Taylor convection.
The inner layer, shown in Fig. (3), seems to be an instability in the outer layers of the proto-neutron star similar to those seen in 2-D simulations carried out by by Keil et al. and Mezzacappa et al..
There is controversy about the existence of PNS instabilities. Originally, the work by Wilson and Mayle claimed to find doubly-diffusive instabilities in the region below the neutrinosphere. Later work by Breunn and collaborators cast doubt on the existence of such phenomena. The simulations of Keil, which utilized the grey flux-limited diffusion approximation to transport neutrinos along radial rays, found a PNS instability that grew over the time period of approximately one second to encompass the entire proto-neutron star. In contrast, the subsequent work of Mezzacappa et al. found a PNS instability that quickly damped out within a short time. The Mezzacappa et al. simulation utilized a multigroup flux-limited $`P_1`$ approximation to transport neutrinos along radial rays outward from an inner radius of r=20 km. The inner boundary conditions in this simulation were established as time-dependent data from the 1-D Langrangean code of Bruenn. It is important to note that neither the simulations of Keil et al. or Mezzacappa et al. took into account the fully radiation-hydrodynamic coupling via the compression and dynamic diffusion terms that are present in Eq. (6). Our simulations have confirmed the expected result that the effects of the dynamic diffusion term dominate the radiative diffusion term in the regions in which the optical depth is large.
Figure 3 shows the instantaneous structure of the velocity field in model PR37 at the same time as the data shown in Fig. (2). The velocity field is illustrated by means of a Lagrangean-Eulerian Advection (LEA) visualization technique that shows the direction of the vectors as streaks. One can clearly see eddies associated with the PNS instability layer at a radius of about 20โ25 km. This is well outside the baffle radius of 8.5 km. In fact, there are approximately a minimum of 40 radial zones separating the innermost of the vortices and the outer edge of the baffles. We do not believe that the baffles in any way impede the dynamics of the vortices. Nevertheless other simulations are underway where the baffle radius is made substantially smaller to verify this claim. We are also developing an implicit hydrodynamic algorithm that will avoid the need for baffles altogether.
Whether the PNS instability will grow or diminish with time is as yet unclear since we have only evolved model PR37 to a time of approximately 16 ms at the time of this writing.
The velocity structure at that time is shown in Fig. (4), which clearly reveals that much coherency has been lost in the vorticial structure of the PNS instability layer. This seems indicative of the decay of this PNS instability in a fashion similar to that described by Mezzacappa et al.. However, it is necessary to evolve this simulation substantially farther in time before any definitive statements can be made about the long-term behavior of this sector of the proto-neutron star. During this relatively short timescale the outer convective zone seen in Fig. (2) does not exhibit significant growth. With evolution to the next 30 ms, we should be able to make comparative statements regarding model PR37 and earlier grey models carried out by Swesty.
## 5 Conclusions and Future Directions
The issues and work described in this paper fall far short of offering complete coverage of the active issues that remain in the area of the explosion mechanism of core collapse supernovae. Indeed, we have ignored many important issues such as magnetic fields, rotation, and neutrino flavor mixing. Clearly there is a large gulf of unexplored physics incorporated in those subjects.
Our own future efforts will be focused in two areas in the near future. The first of these is understanding the effects of numerous microphysics enchancements that will be added to the models. The second of these efforts involves extending the models to 3-D where a more realistic convective flow structure can arise.
## Acknowledgments
The authors would like to thank Ed Bachta and Polly Baker of Indiana University at Indianapolis for their collaborative efforts in developing the visualization technology that went into Figures 3 and 4. We gratefully acknowledge the support of the U.S. Dept. of Energy, through SciDAC Award DE-FC02-01ER41185, by which this work was funded. We are also grateful to the National Energy Research Scientific Computing Center (NERSC) for computational support. Finally, we would like to thank the National Institute for Nuclear Theory at the University of Washington for its hospitality in hosting the workshop โOpen Issues in Core-Collapse Supernovae,โ at which this work was presented in June 2004.
|
warning/0506/cond-mat0506441.html
|
ar5iv
|
text
|
# References
Theory of spin-polarized transport in ferromagnet-semiconductor structures: Unified description of ballistic and diffusive transport
R. Lipperheide, U. Wille<sup>1</sup><sup>1</sup>1Corresponding author. Phone: ++49-30-80622685; FAX: ++49-30-80622098; E-mail: wille@hmi.de.
Abteilung Theoretische Physik, Hahn-Meitner-Institut Berlin,
Glienicker Str. 100, D-14109 Berlin, Germany
Abstract
A theory of spin-polarized electron transport in ferromagnet-semiconductor heterostructures, based on a unified semiclassical description of ballistic and diffusive transport in semiconductors, is outlined. The aim is to provide a framework for studying the interplay of spin relaxation and transport mechanism in spintronic devices. Transport inside the (nondegenerate) semiconductor is described in terms of a thermoballistic current, in which electrons move ballistically in the electric field arising from internal and external electrostatic potentials, and are thermalized at randomly distributed equilibration points. Spin relaxation is allowed to take place during the ballistic motion. For arbitrary potential profile and arbitrary values of the momentum and spin relaxation lengths, an integral equation for a spin transport function determining the spin polarization in the semiconductor is derived. For field-driven transport in a homogeneous semiconductor, the integral equation can be converted into a second-order differential equation that generalizes the spin drift-diffusion equation. The spin polarization in ferromagnet-semiconductor structures is obtained by matching the spin-resolved chemical potentials at the interfaces, with allowance for spin-selective interface resistances. Illustrative examples are considered.
PACS: 72.25.-b, 72.25.Hg, 73.40.Cg, 73.40.Sx
Keywords: Theory; Heterostructures; Spin-polarized transport; Transport mechanisms; Spintronics
1. Introduction
In spintronics research, particular emphasis is currently placed on the study of spin-polarized electron transport in heterostructures formed of a nonmagnetic semiconductor and two (metallic or semiconducting) ferromagnetic contacts . For the actual design of spintronic devices, a detailed theoretical understanding of this kind of transport problem is indispensable. Several pertinent studies have been performed so far, which mostly rely on the drift-diffusion model. A number of important results have emerged. (i) For an interface between a metallic ferromagnet and a semiconductor without spin-selective interface resistance, the injected current spin polarization is predicted to be very low owing to the large conductivity mismatch . (ii) Spin-selective interface resistances arising from tunnel barriers or Schottky barriers can greatly enhance the injection efficiency . (iii) A similar effect is to be expected when a sufficiently high electric field is applied across the semiconductor . (iv) Under conditions where, in the semiconductor, ballistic transport prevails over drift-diffusion, the injection efficiency is controlled by the Sharvin interface resistance unless spin-selective interface resistances are introduced.
While the theory of spin-polarized electron transport in ferromagnet-semiconductor heterostructures has reached a level of considerable sophistication, it appears that certain aspects of the semiconductor part of the transport problem require a more systematic, unified treatment, such as the interplay of spin relaxation and transport mechanism all the way from the diffusive to the ballistic regime, and the effect of the detailed shape of the electrostatic potential profile. In this paper, we outline the principal ideas of a theory that meets these requirements. For illustrative purposes, we present calculated results for the position dependence of the zero-bias current spin polarization along a heterostructure as well as for the injected polarization as a function of bias. A detailed account of the formal development as well as specific applications of our theory will be published elsewhere .
2. Thermoballistic current
Our treatment of spin-polarized electron transport relies on our previously formulated unified semiclassical description of (spinless) electron transport in parallel-plane semiconductor structures . The description is based on the concept of a โthermoballistic electron currentโ which combines elements of the ballistic and diffusive transport mechanisms. Here, we briefly summarize this concept.
Assuming a one-dimensional geometry, we consider a (nondegenerate) semiconducting sample enclosed between two plane-parallel ferromagnetic contacts at $`x=x_1`$ and $`x=x_2`$, respectively, so that $`S=x_2x_1`$ is the sample length (see Fig. 1). The electron current density $`J(x^{},x^{\prime \prime })`$ across the โballistic intervalโ $`[x^{},x^{\prime \prime }]`$ between two equilibration points $`x^{}`$ and $`x^{\prime \prime }`$ is given by
$$J(x^{},x^{\prime \prime })=v_eN_ce^{\beta E_c^m(x^{},x^{\prime \prime })}\left[e^{\beta \mu (x^{})}e^{\beta \mu (x^{\prime \prime })}\right]$$
(1)
($`x_1x^{}<x^{\prime \prime }x_2`$), which is the difference of the ballistic current injected into the interval at its left end at $`x^{}`$ and the analogous current injected at its right end at $`x^{\prime \prime }`$. The function $`\mu (x)`$ is the chemical potential at the equilibration point $`x`$. Furthermore, $`v_e=(2\pi m^{}\beta )^{1/2}`$ is the emission velocity, $`N_c=2(2\pi m^{}/\beta h^2)^{3/2}`$ is the effective density of states at the conduction band edge, $`m^{}`$ is the effective mass of the electrons, and $`\beta =(k_BT)^1`$. The current (1) contains only the transmitted electrons, i.e., those with sufficient energy to surmount the potential barrier
$$\widehat{E}_c^m(x^{},x^{\prime \prime })=E_c^m(x^{},x^{\prime \prime })E_c^0,$$
(2)
where $`E_c^m(x^{},x^{\prime \prime })`$ is the maximum value of the potential profile $`E_c(x)`$ in the interval $`[x^{},x^{\prime \prime }]`$, and $`E_c^0`$ is its overall minimum across the sample. The profile $`E_c(x)`$ comprises the (equilibrium) conduction band edge potential and the external electrostatic potential.
From the ballistic electron current $`J(x^{},x^{\prime \prime })`$, we construct the thermoballistic current $`J(x)`$ at position $`x`$ inside the sample by summing up the contributions from all intervals $`[x^{},x^{\prime \prime }]`$ for which $`x_1x^{}<x<x^{\prime \prime }x_2`$, each weighted with the probability of occurrence of the interval. For simplicity, we here take this probability in its one-dimensional form $`\mathrm{exp}(|x^{\prime \prime }x^{}|/l)`$, where $`l`$ is the mean free path for momentum relaxation (momentum relaxation length). We then have
$`J(x)=v_eN_ce^{\beta E_c^0}\{w(x_1,x_2;l)[e^{\beta \mu _1}e^{\beta \mu _2}]+{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}w(x^{},x_2;l)[e^{\beta \mu (x^{})}e^{\beta \mu _2}]`$
$`+`$ $`{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{}}{l}}w(x_1,x^{};l)[e^{\beta \mu _1}e^{\beta \mu (x^{})}]+{\displaystyle _{x_1}^x}{\displaystyle \frac{dx^{}}{l}}{\displaystyle _x^{x_2}}{\displaystyle \frac{dx^{\prime \prime }}{l}}w(x^{},x^{\prime \prime };l)[e^{\beta \mu (x^{})}e^{\beta \mu (x^{\prime \prime })}]\},`$
where
$$w(x^{},x^{\prime \prime };l)=e^{|x^{\prime \prime }x^{}|/l}e^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}.$$
(4)
The quantities $`\mu _{1,2}=\mu (x_{1,2})`$ are the chemical potentials on the contact side of the interfaces, i.e., immediately outside of the sample. The thermoballistic current $`J(x)`$ is not conserved by itself; however, when averaged over the sample, it yields the physical electron current $`J`$,
$$\frac{1}{x_2x_1}_{x_1}^{x_2}๐xJ(x)=J.$$
(5)
Furthermore, the thermoballistic current entering at one end of the sample must equal the current leaving at the other end ,
$$J(x_1^+)=J(x_2^{}).$$
(6)
Substituting expression (LABEL:eq:3) in condition (5), we can derive Volterra-type integral equations for two auxiliary functions $`\chi _1(x)`$ and $`\chi _2(x)`$, which are distinguished by discontinuities at $`x=x_1`$ and $`x=x_2`$, respectively. Using condition (6), we determine from $`\chi _1(x)`$ and $`\chi _2(x)`$ a unique chemical potential $`\mu (x)`$. It exhibits discontinuities at both interfaces, which are related to the Sharvin interface resistance . From $`\mu (x)`$, the thermoballistic current $`J(x)`$ is calculated via Eq. (LABEL:eq:3). The current-voltage characteristic is obtained for arbitrary values of $`l`$ and arbitrary $`E_c(x)`$ in terms of a โreduced resistanceโ $`\stackrel{~}{\chi }`$ composed of $`\chi _1(x_2)`$ and $`\chi _2(x_1)`$. The thermoballistic density $`n(x)`$ of electrons making up the current $`J(x)`$ is constructed in a similar way.
For electron transport in a homogeneous semiconductor driven by a (constant) electric field of strength $``$, the thermoballistic current $`J(x)`$ and density $`n(x)`$ can be expressed essentially in terms of the dimensionless parameters $`x/S`$, $`l/S`$, and $`ฯตS`$, where $`ฯต=\beta e||`$. In Fig. 2, we show the dependence of $`n(x)`$ on $`x/S`$ for $`ฯตS=1`$ and various values of $`l/S`$. The initial decrease of $`n(x)`$ with increasing $`x/S`$ is a ballistic effect that reflects the increase of the electron velocity in the electric field and becomes more pronounced as $`l/S`$ becomes larger. The corresponding thermoballistic current $`J(x)`$ turns out to differ only insignificantly from the physical current $`J`$.
3. Spin-polarized transport in semiconductors
We now extend the unified description by allowing spin relaxation to take place during the electron motion across the ballistic intervals. The off-equilibrium spin-polarized current $`J_{}(x)=J_{}(x)J_{}(x)`$ is connected with the off-equilibrium spin-polarized density $`n_{}(x)=n_{}(x)n_{}(x)`$ through the balance equation
$$\frac{dJ_{}(x)}{dx}+\frac{n_{}(x)}{\tau _s}=0,$$
(7)
where $`\tau _s`$ is the spin relaxation time. Using this equation in the ballistic transport regime, in which $`J_{}(x)=2v_en_{}(x)`$, we construct the off-equilibrium ballistic spin-polarized current $`J_{}(x^{},x^{\prime \prime };x)`$ at position $`x`$ in the ballistic interval $`[x^{},x^{\prime \prime }]`$, obtaining
$$J_{}(x^{},x^{\prime \prime };x)=v_eN_ce^{\beta \widehat{E}_c^m(x^{},x^{\prime \prime })}\left[A(x^{})e^{(xx^{})/l_s}A(x^{\prime \prime })e^{(x^{\prime \prime }x)/l_s}\right]$$
(8)
($`x_1x^{}<x<x^{\prime \prime }x_2`$), with the ballistic spin relaxation length $`l_s=2v_e\tau _s`$. The parameter $`l_s`$ comprises the overall effect of the underlying microscopic spin relaxation mechanisms. The function $`A(x)`$ is the โspin transport functionโ defined as
$$A(x^{})=e^{\beta [E_c^0\mu (x^{})]}\alpha _{}(x^{});$$
(9)
here, $`\alpha _{}(x^{})=\alpha _{}(x^{})\alpha _{}(x^{})`$ is the off-equilibrium โspin fraction excessโ at the equilibration point $`x^{}`$, which is defined in terms of the spin fractions $`\alpha _{}(x^{})`$, with $`\alpha _{}(x^{})+\alpha _{}(x^{})=1`$. The spin fractions are related to the spin-resolved thermoballistic chemical potentials $`\mu _{}(x^{})`$ via
$$e^{\beta \mu _{}(x^{})}=e^{\beta \mu (x^{})}\alpha _{}(x^{}).$$
(10)
Proceeding as in the spinless case, we now sum the (weighted) contributions of the ballistic spin-polarized current (8) over all randomly distributed intervals $`[x^{},x^{\prime \prime }]`$. The resulting expression for the off-equilibrium thermoballistic spin-polarized current $`J_{}(x)`$ is of the form (LABEL:eq:3), but with the terms in brackets therein replaced with those obtained by evaluating the bracketed term in Eq. (8) for the different cases. A similar expression is found for the off-equilibrium thermoballistic spin-polarized density $`n_{}(x)`$.
Substituting the thermoballistic expressions for $`J_{}(x)`$ and $`n_{}(x)`$ in Eq. (7), we arrive at a linear, Fredholm-type integral equation for the spin transport function $`A(x)`$,
$`W(x_1,x;l,l_s)A_1`$ $`+`$ $`W(x,x_2;l,l_s)A_2`$ (11)
$``$ $`W_0(x_1,x_2;x;l)A(x)+{\displaystyle _{x_1}^{x_2}}{\displaystyle \frac{dx^{}}{l}}W(x^{},x;l,l_s)A(x^{})=0,`$
where
$$W(x^{},x^{\prime \prime };l,l_s)=w(x^{},x^{\prime \prime };l)e^{|x^{\prime \prime }x^{}|/l_s},$$
(12)
$$W_0(x_1,x_2;x;l)=w(x_1,x;l)+w(x,x_2;l)+_{x_1}^{x_2}\frac{dx^{}}{l}w(x^{},x;l),$$
(13)
and $`A_{1,2}=A(x_{1,2})`$. The solution of the fundamental equation (11) for $`x_1<x<x_2`$ determines the spin-polarized electron transport inside the semiconducting sample, and is obtained in terms of the values of $`A_1`$ and $`A_2`$ on the contact side of the interfaces at the ends of the sample. The latter are given by the spin fraction excesses $`\alpha _{1,2}=\alpha _{}(x_{1,2})`$ and the chemical potentials $`\mu _{1,2}=\mu (x_{1,2})`$ in the contacts via Eq. (9).
For field-driven transport in a homogeneous semiconductor, Eq. (11) can be converted, by twofold differentiation and elimination of the quantities $`A_1`$ and $`A_2`$, into a homogeneous integrodifferential equation. Within a judicious approximation, the latter equation can be reduced to a generalized spin drift-diffusion equation of the form
$$b_0(x)\frac{d^2A(x)}{dx^2}+b_1(x)\frac{dA(x)}{dx}+b_2(x)A(x)=0,$$
(14)
with coefficient functions $`b_i(x)`$ depending linearly on the function $`\mathrm{exp}[(ฯต+1/l)(xx_1)]`$. In the diffusive regime $`l/l_s1`$, where $`b_0(x)=1`$, $`b_1(x)=ฯต`$, $`b_2(x)=1/(ll_s)`$, Eq. (14) reduces to the standard spin drift-diffusion equation if there the intrinsic spin diffusion length $`L`$ is identified with $`\sqrt{ll_s}`$.
With $`A(x)`$ determined by solving the integral equation (11) or, in special cases, simplified equations like Eq. (14), we can evaluate the thermoballistic spin-polarized current $`J_{}(x)`$ from the analogue of expression (LABEL:eq:3). Dividing by the (total) thermoballistic current $`J(x)`$ given by Eq. (LABEL:eq:3), we obtain the current spin polarization
$$P_J(x)=\frac{J_{}(x)}{J(x)}.$$
(15)
The density spin polarization is calculated analogously.
4. Ferromagnet-semiconductor heterostructures
We now consider spin-polarized transport in heterostructures formed of a homogeneous, nonmagnetic semiconductor and two ferromagnetic contacts, which are treated as fully degenerate Fermi systems. Equating the splitting $`\mu _{}(x)=\mu _{}(x)\mu _{}(x)`$ of the spin-up and spin-down chemical potentials in the ferromagnets at $`x=x_{1,2}`$, respectively, with that of the semiconductor on the contact side of the interfaces, we find, using Eq. (10),
$$[\mu _{}(x_{1,2})]_{\mathrm{ferromagnet}}=[\mu _{}(x_{1,2})]_{\mathrm{semiconductor}}=\frac{1}{\beta }\mathrm{ln}\left(\frac{1+\alpha _{1,2}}{1\alpha _{1,2}}\right).$$
(16)
This condition allows the current spin polarizations $`P_J(x_{1,2})`$ on the ferromagnet sides of the interfaces to be expressed through the spin fraction excesses $`\alpha _{1,2}`$. Neglecting spin-flip scattering at the interfaces, we invoke continuity of the polarization at the interfaces. We then obtain two coupled nonlinear equations from which $`\alpha _1`$ and $`\alpha _2`$, and hence $`J_{}(x)`$ and $`P_J(x)`$, can be calculated in terms of the material parameters characterizing the ferromagnets and the semiconductor, the bulk polarizations $`P_1`$ and $`P_2`$ in the left and right ferromagnet, respectively, and the physical current $`J`$. Spin-selective interface resistances can be introduced via discontinuities of the chemical-potential splitting $`\mu _{}(x)`$ at the interfaces .
In the case of zero bias, $`ฯต0`$, the solutions of Eq. (14) are $`A(x)\mathrm{exp}(\pm x/L)`$, and the position dependence of the current spin polarization inside the semiconductor is given by
$$P_J(x)=\frac{2v_eN_c\overline{l}}{LJ}\left[C_1e^{(xx_1)/L}C_2e^{(x_2x)/L}\right],$$
(17)
where $`L=\sqrt{\overline{l}l_s}`$ is the generalized spin diffusion length (or โpolarization decay lengthโ), and $`\overline{l}=ll_s/(l+l_s)`$. The coefficients $`C_{1,2}`$ are simple functions of the spin fraction excesses $`\alpha _{1,2}`$ which, in turn, are given explicitly in terms of the material parameters, the polarizations $`P_{1,2}`$, and the current $`J`$.
In Fig. 3, the zero-bias current spin polarization $`P_J(x)`$ for a symmetric ferromagnet-semiconductor-ferromagnet heterostructure with $`S=1`$ $`\mu `$m at $`T=300`$ K is shown as a function of $`x`$ for various values of the momentum relaxation length $`l`$ and for zero as well as nonzero interface resistances. For the parameters of the ferromagnets, the values $`10^3`$ $`\mathrm{\Omega }^1`$ cm<sup>-1</sup> for the bulk conductivities, $`60`$ nm for the spin diffusion lengths, and $`0.5`$ for the bulk polarizations $`P_{1,2}`$ have been adopted from Ref. . With a look at recent experiments on the spin dynamics in n-doped GaAs , the values 0.067 $`m_e`$ for the effective electron mass $`m^{}`$, $`2.0\times 10^{18}`$ cm<sup>-3</sup> for the equilibrium electron density, and $`1`$ $`\mu `$m for the ballistic spin relaxation length $`l_s`$ have been chosen for the material parameters of the semiconductor. We are aware of the fact that, by considering a specific semiconducting material with fixed doping concentration, one essentially fixes the value of the momentum relaxation length $`l`$. Therefore, when varying $`l`$ in a fairly broad range, we assume the above parameter values (or, at least, their order of magnitude) to be representative for a class of semiconducting materials that differ in the strength of the impurity scattering and hence in the magnitude of $`l`$.
The momentum relaxation length $`l`$ affects the results shown in Fig. 3 in a twofold way. (i) It determines the conduction in the semiconductor and thus the conductivity mismatch with the ferromagnets. For small values of $`l`$, this mismatch is large, leading to a small injected current spin polarization $`P_J(0)`$. (ii) It determines the polarization decay length $`L`$, so that for small $`l`$ the polarization dies out rapidly inside the semiconductor. A substantial degree of polarization all along the semiconductor is achieved when the value of $`l`$ is increased up to a length of the order of the sample length, in which case the ballistic component becomes prevalent. Figure 3 also shows that, by introducing appropriately chosen spin-selective interface resistances, one may offset the suppression of the injected polarization due to the conductivity mismatch for small $`l`$; however, the rapid decay of the polarization inside the semiconductor cannot be prevented in this way.
For nonzero, constant electric field and $`S/L\mathrm{}`$, i.e., disregarding the effect of the right ferromagnet, we have for the current spin polarization $`P_J(x_1)`$ injected at the left interface
$$P_J(x_1)=\stackrel{~}{\chi }\mathrm{\Gamma }_J\alpha _1,$$
(18)
where $`\stackrel{~}{\chi }`$ is the reduced resistance entering the current-voltage characteristic, and the quantity $`\mathrm{\Gamma }_J`$ involves the function $`A(x)`$, which is obtained by numerically solving Eq. (14). Since only the interface at $`x=x_1`$ enters into consideration, a single nonlinear equation has to be solved to determine the spin fraction excess $`\alpha _1`$ as a function of the polarization $`P_1`$.
In Fig. 4, we show the dependence of $`P_J(x_1)`$ on the electric-field parameter $`ฯต`$ for various values of the momentum relaxation length $`l`$, the remaining parameter values being the same as in Fig. 3. In conformity with the drift-diffusion results of Ref. , the injected polarization generally rises with increasing $`ฯต`$; however, as in Fig. 3, the main effect is due to the variation of $`l`$.
5. Concluding remarks
We have outlined the principal ideas of a theory of spin-polarized electron transport in ferromagnet-semiconductor heterostructures. It generalizes previous theoretical treatments based on the drift-diffusion model by introducing the momentum relaxation length in the semiconductor as a new degree of freedom, thus allowing a systematic study of the interplay of spin relaxation and transport mechanism. By considering illustrative examples, we have shown that the momentum relaxation length has a significant influence both on the polarization injected at a ferromagnet-semiconductor interface and on the decay of the polarization inside the semiconductor. To study in detail the influence of the transport mechanism on spin-polarized transport (in particular, when ballistic effects take over), the identification and design of classes of novel semiconducting materials is called for. In this way, new possibilities to improve the efficiency of spintronic devices may open up.
Figure Captions
Schematic diagram showing a semiconducting sample of length $`S`$ enclosed between two plane-parallel ferromagnetic contacts. Illustrated are expression (LABEL:eq:3) for the thermoballistic current $`J(x)`$ and its analogue for the thermoballistic spin-polarized current $`J_{}(x)`$.
The thermoballistic density $`n(x)`$ inside a homogeneous semiconductor for constant electric field, plotted versus $`x/S`$ (assuming $`x_1=0`$) for $`ฯตS=1`$ and the indicated values of $`l/S`$. The density is normalized to the constant value it assumes in the diffusive limit $`l/S0`$.
The zero bias ($`ฯต0`$) current spin polarization $`P_J(x)`$ (assuming $`x_1=0`$) along a symmetric ferromagnet-semiconductor-ferromagnet structure with $`S=1`$ $`\mu `$m for the indicated values of the momentum relaxation length $`l`$. The solid curves correspond to zero interface resistance, the dashed curves to interface resistances of $`10^7`$ $`\mathrm{\Omega }`$ cm<sup>2</sup> for spin-up electrons and $`2\times 10^7`$ $`\mathrm{\Omega }`$ cm<sup>2</sup> for spin-down electrons, respectively. For the remaining parameter values, see text.
The injected current spin polarization $`P_J(x_1)`$ for $`S/L\mathrm{}`$ as a function of the electric-field parameter $`ฯต`$ for the indicated values of the momentum relaxation length $`l`$. The solid curves correspond to zero interface resistance, the dashed curves to interface resistances of $`10^7`$ $`\mathrm{\Omega }`$ cm<sup>2</sup> for spin-up electrons and $`2\times 10^7`$ $`\mathrm{\Omega }`$ cm<sup>2</sup> for spin-down electrons, respectively. For the remaining parameter values, see text.
FIGURE 1
FIGURE 2
FIGURE 3
FIGURE 4
|
warning/0506/physics0506061.html
|
ar5iv
|
text
|
# Simultaneous cooling of axial vibrational modes in a linear ion-trap
## I Introduction
Atomic ions trapped in an electrodynamic cage allow for preparation and measurement of individual quantum systems, and represent an ideal system to investigate fundamental questions of quantum physics, for instance, related to decoherence Myatt00 ; Roos99 , the measurement process Hannemann02 ; Wunderlich03 , or multiparticle entanglement Entangle . Also, trapped ions satisfy all criteria necessary for quantum computing. Two internal states of each ion represent one elementary quantum mechanical unit of information (a qubit). The quantized vibrational motion of the ions (the โbus-qubitโ) is used as means of communication between individual qubits to implement conditional quantum dynamics with two or more qubits Cirac95 . In recent experiments quantum logic operations with two trapped ions were realized QGate and teleportation of an atomic state has been demonstrated Teleport .
These implementations of quantum information processing (QIP) with trapped ions require that the ion string is cooled to low vibrational collective excitations Cirac95 ; QGate ; Sorensen00 ; Jonathan00 . In particular, this condition should be fulfilled by all collective vibrational modes Wineland98 . Therefore, in view of the issue of scalable QIP with ion traps, it is important to find efficient cooling schemes that allow to prepare vibrationally cold ion chains.
Cooling of the vibrational motion of two ions in a common trap potential has been demonstrated experimentally King98 ; Peik99 ; Roos00 ; Reiss02 (see also Eschner03 for a recent review). This is deemed to be sufficient for a quantum information processor which utilizes two ions at a time for quantum logic operations with additional ions stored in spatially separated regions Kielpinski02 . If more than two ions reside in a common trap potential and shall be used simultaneously for quantum logic operations, however, the task of reducing the ionsโ motional thermal excitation becomes increasingly challenging with a growing number of ions and represents a severe obstacle on the way towards scalable QIP with an ion chain. Straightforward extensions of laser-cooling schemes for one particle to many ions, like sequentially applying sideband cooling Eschner03 to each one of the modes, becomes inefficient as the number of ions increases, since after having cooled the last mode, the first one may already be considerably affected by heating due to photon scattering and/or due to fluctuations of the trap potential. Therefore, it is desirable to find a method that allows for simultaneous and efficient cooling of many vibrational modes of a chain of ions.
In this article we propose a scheme that allows for simultaneous sideband cooling of all collective modes of an ion chain to the ground state. This is achieved by inducing position dependent Zeeman shifts through a suitably designed magnetic field, thereby shifting the spectrum of each ion in such a way that the red-sideband transitions of each mode may occur at the same frequency. Thus, by irradiating the ion string with monochromatic radiation all axial modes are cooled. We investigate numerically the efficiency and explore implementations of simultaneous sideband excitation by means of laser light, and alternatively, by using long-wavelength radiation in the radio-frequency or microwave regime Mintert01 ; Wunderlich02 ; McHugh05 .
The remainder of this article is organized as follows: In section II the cooling scheme is outlined. Numerical investigations of the cooling efficiency are presented for implementations using an optical Raman transition (section III) and a microwave transition (section IV). In section V possible experimental implementations are discussed and the cooling scheme is studied under imperfect experimental conditions. The paper is concluded in section VI.
## II The concept of simultaneous sideband cooling
### II.1 Axial vibrational modes
We consider $`N`$ crystallized ions each of mass $`m`$ and charge $`e`$ in a harmonic trap. The trap potential has cylindrical symmetry around the $`z`$axis providing strong radial confinement such that the ions are aligned along this axis Footnote:Example . We denote by $`\nu _r`$, $`\nu _z`$ the radial and axial frequencies of the resulting harmonic potential, where $`\nu _r\nu _z`$, and by $`z_j^{(0)}`$ the ions classical equilibrium positions along the trap axis. The typical axial distance $`\delta z`$ between neighboring ions scales like $`\delta z\zeta _02N^{0.57}`$ with $`\zeta _0(e^2/(4\pi ฯต_0m\nu _1^2))^{1/3}`$Steane97 ; James98 . For brevity, in the remainder of this article, the ion at the classical equilibrium position $`z_j^{(0)}`$ is often referred to as โion $`j`$โ.
At sufficiently low temperatures the ions vibrations around their respective equilibrium positions are harmonic and the axial motion is described by $`N`$ harmonic oscillators according to the Hamiltonian
$$\stackrel{~}{H}_{\mathrm{mec}}=\underset{\alpha =1}{\overset{N}{}}\mathrm{}\nu _\alpha (a_\alpha ^{}a_\alpha +1/2),$$
(1)
where $`\nu _\alpha `$ are the frequencies of the chain collective modes and $`a_\alpha ^{}`$ and $`a_\alpha `$ the creation and annihilation operators of a phonon at energy $`\mathrm{}\nu _\alpha `$. We denote with $`Q_\alpha ,P_\alpha `$ the corresponding quadratures, such that $`[Q_\alpha ,P_\alpha ]=\mathrm{i}\mathrm{}`$, and choose the labelling convention $`\nu _1<\nu _2<\mathrm{}<\nu _N`$, whereby $`\nu _1=\nu _z`$ (in this article we often refer to the collective vibrational mode characterized by $`\nu _\alpha `$ as โmode $`\alpha `$โ). The local displacement $`q_j=z_jz_j^{(0)}`$ of the ion $`j`$ from equilibrium is related to the coordinates $`Q_\alpha `$ by the transformation
$$q_j=\underset{\alpha }{}S_j^\alpha Q_\alpha $$
(2)
where $`S_j^\alpha `$ are the elements of the unitary matrix $`S`$ that transforms the dynamical matrix $`A`$, characterizing the ions potential, such that $`S^1AS`$ is diagonal. The frequencies, $`\nu _\alpha `$ of the vibrational modes are given by $`\sqrt{\upsilon _\alpha }\times \nu _1`$ where $`\upsilon _\alpha `$ are the eigenvalues of $`A`$ James98 . The normal modes are excited by displacing an ion from its equilibrium position $`z_j^{(0)}`$ by an amount $`q_j`$. Thus, the coefficients $`S_j^\alpha `$ describe the strength with which a displacement $`q_j`$ from $`z_j^{(0)}`$ couples to the collective mode $`\alpha `$.
Excitation of a vibrational mode can be achieved through the mechanical recoil associated with the scattering of photons by the ions. This excitation is scaled by the Lamb-Dicke parameter (LDP) Stenholm86 , which for a single ion corresponds to $`\sqrt{\omega _R/\nu }`$, where $`\omega _R=\mathrm{}k^2/2m`$ is the recoil frequency and $`\mathrm{}k`$ the linear momentum of a photon. In an ion chain we associate a Lamb-Dicke parameter $`\eta _\alpha `$ with each mode according to the equation
$$\eta _\alpha =\sqrt{\frac{\omega _R}{\nu _\alpha }}.$$
(3)
Hence, if a photon is scattered by the ion at $`z_j^{(0)}`$, the ion recoil couples to the mode $`\alpha `$ according to the relation Morigi01
$$\eta _j^\alpha =S_j^\alpha \eta _\alpha .$$
(4)
In the remainder of this article we will assume that the ions are in the Lamb-Dicke regime, corresponding to the fulfillment of condition $`\sqrt{a_\alpha ^{}a_\alpha }\eta _\alpha 1`$. In this regime the scattering of a photon does not couple to the vibrational excitations at leading order in this small parameter, while changes of one vibrational quantum $`\mathrm{}\nu _\alpha `$ occur with probability that scales as $`|\eta _\alpha |^2`$. Changes by more than one vibrational quantum are of higher order and are neglected here.
### II.2 Sideband cooling of an ion chain
In this section we consider a schematic description of sideband cooling of an ion chain, in order to introduce the concepts relevant for the following discussion. We denote by $`|0`$ and $`|1`$ the internal states of the ion transition at frequency $`\omega _0`$, in absence of external fields, and linewidth $`\gamma `$. A spatially inhomogeneous magnetic field is applied that shifts the transition frequency of each ion individually such that for the ion at position $`z_j^{(0)}`$ the value $`\omega _j`$ is assumed. Each ion transition couples to radiation at frequency $`\omega _L`$, which drives it well below saturation. In this limit, the contributions of scattering from each ion to the excitation of the modes add up incoherently Morigi99 ; Morigi01 .
For this system, the equations describing the dynamics of laser sideband cooling of an ion chain can be reduced to rate equations of the form
$`{\displaystyle \frac{\mathrm{d}}{\mathrm{d}t}}P_\alpha (n^{(\alpha )})`$ $`=`$ $`(n^{(\alpha )}+1)\left[A_{}^\alpha P_\alpha (n^{(\alpha )}+1)A_+^\alpha P_\alpha (n^{(\alpha )})\right]`$ (5)
$`n^{(\alpha )}\left[A_{}^\alpha P_\alpha (n^{(\alpha )})A_+^\alpha P_\alpha (n^{(\alpha )}1)\right]`$
where $`P_\alpha (n^{(\alpha )})`$ is the average occupation of the vibrational number state $`|n^{(\alpha )}`$ of the mode $`\alpha `$, and $`A_+^\alpha `$ ($`A_{}^\alpha `$) characterizes the rate at which the mode is heated (cooled). Equation (5) is valid in the Lamb-Dicke regime, i.e. when the LDP is sufficiently small to allow for a perturbative expansion in this parameter. Denoting by $`\mathrm{\Omega }_j`$ the Rabi frequency, the heating and cooling rate takes the form Morigi01
$$A_\pm ^\alpha =\underset{j=1}{\overset{N}{}}|\eta _j^\alpha |^2\frac{\mathrm{\Omega }_j^2}{2\gamma }\left[\frac{\gamma ^2}{4(\delta _j\nu _\alpha )^2+\gamma ^2}+\varphi \frac{\gamma ^2}{4\nu _\alpha ^2+\gamma ^2}\right]$$
(6)
where the detuning $`\delta _j\omega _L\omega _j`$. The coefficient $`\varphi `$ emerges from the integral over the angles of photon emission, according to the pattern of emission of the given transition Stenholm86 . For $`A_{}^\alpha >A_+^\alpha `$ a steady state exists, it is approached at the rate
$$\mathrm{\Gamma }_{\mathrm{cool}}^{(\alpha )}=A_{}^\alpha A_+^\alpha $$
(7)
and the average number of phonons of mode $`\alpha `$ at steady state is given by the expression
$$n^{(\alpha )}=\frac{A_+^\alpha }{A_{}^\alpha A_+^\alpha }.$$
(8)
Sideband cooling reaches $`n^{(\alpha )}1`$ through $`A_{}^\alpha A_+^\alpha `$. This condition is obtained by selectively addressing the motional resonance at $`\omega _0\nu _\alpha `$. This is accomplished for a single collective mode when $`\gamma \nu _\alpha `$ and $`\delta _\alpha =\nu _\alpha `$.
In this work, we show how the application of a suitable magnetic field allows for simultaneous sideband cooling of all modes. In particular, the field induces space-dependent frequency shifts that suitably shape the excitation spectrum of the ions. Simultaneous cooling is then achieved when for each mode $`\alpha `$ there is one ion $`j`$ with the matching resonance frequency, that is, such that $`\delta _j=\omega _L\omega _j=\nu _\alpha `$. This procedure is outlined in detail in the following subsection.
### II.3 Shaping the spectrum of an $`N`$ ion chain
Assume the ion transition $`|0|1`$ and that a magnetic fieldโwhose magnitude varies as a function of $`z`$โis applied to the linear ion trap, Zeeman shifting this resonance. As a result, the ions resonance frequencies $`\omega _j`$ are no longer degenerate. The field gradient is designed such that all ions share a common motion-induced resonance. This resonance corresponds to one of the transitions $`|0,n^{(\alpha )}|1,n^{(\alpha )}1`$, namely to the red sideband of the modes $`\alpha `$. The resonance frequency of each ion is shifted such that the red sidebands of all modes can be resonantly and simultaneously driven by monochromatic radiation at frequency $`\omega =\omega _1\nu _1=\mathrm{}=\omega _N\nu _N`$. Ionic resonances and the associated red sideband resonancesโoptimally shifted for simultaneous coolingโare illustrated in Fig. 1 for the case of 10 ions.
Sideband excitation can be accomplished by either laser light or microwave radiation according to the scheme discussed in Mintert01 . With appropriate recycling schemes this leads to sideband cooling on all $`N`$ modes simultaneously. A discussion on how a suitable field gradient shifting the ionic resonances in the desired fashion can be generated is deferred to section V.
### II.4 Theoretical model
As an example, we discuss simultaneous sideband cooling of the collective axial modes of a chain composed of <sup>171</sup>Yb<sup>+</sup>ions with mass $`m=171`$ a.m.u.. The ions are crystallized along the axis of a linear trap characterized by $`\nu _1=1\times 2\pi `$MHz. A magnetic field $`B(z)`$ along the axis is applied that Zeeman-shifts the energy of the internal states. The value of the field along $`z`$ is such that it shifts the red-sidebands of all modes into resonance along the chain, while at the same time its gradient is sufficiently weak to negligibly affect the frequencies of the normal modes Wunderlich02 .
The selective drive of the motional sidebands can be implemented on a magnetic dipole transition in <sup>171</sup>Yb<sup>+</sup>close to $`\omega _0=12.6\times 2\pi `$ GHz between the hyperfine states $`|0=|S_{1/2},F=0`$ and $`|1=|S_{1/2},F=1,m_F=1`$. The magnetic field gradient lifts the degeneracy between the resonances of individual ions, and the transition frequency $`\omega _j`$ of ion $`j`$ is proportional to $`B(z_j)`$ in the weak field limit $`\mu _BB/\mathrm{}\omega _01`$, where $`\mu _B`$ is the Bohr magneton. For strong magnetic fields the variation of $`\omega _j`$ with $`B`$ is obtained from the Breit-Rabi formula Wunderlich03 .
We investigate two cases, corresponding to two different implementations of the excitation of the sideband transition between states $`|0`$ and $`|1`$. In the first case, discussed in section III, the sideband transition is driven by two lasers with appropriate detuning, namely a Raman transition is implemented with intermediate state $`|2=|P_{1/2}`$. In the second case, presented in section IV, microwave radiation drives the magnetic dipole.
Since spontaneous decay from state $`|1`$ back to $`|0`$ is negligible on this hyperfine transition, laser light is used to optically pump the ion into the $`|0`$ state via excitation of the $`|1|2`$ electric dipole transition. This laser light is close to 369nm and serves at the same time for state selective detection by collecting resonance fluorescence on this transition, and for initial Doppler cooling of the ions. The state $`|2`$ decays with rates $`\mathrm{\Gamma }_{21}=11\times 2\pi `$MHz and $`\mathrm{\Gamma }_{20}=5.5\times 2\pi `$MHz into the states $`|1`$ and $`|0`$, respectively FootnoteB . The considered level scheme is illustrated in Fig. 2, and the corresponding model is described in the appendix.
We evaluate the efficiency of the cooling procedure by neglecting the coupling between different vibrational modes by photon scattering, which is reasonable when the system is in the Lamb-Dicke regime. In this case, the dynamics reduce to solving the equations for each mode $`\alpha `$ independently, and the contributions from each ion to the dynamics of the mode are summed up incoherently Morigi01 , as outlined in Sec. II.2. The steady state and cooling rates for each mode are evaluated using the method discussed in Marzoli94 and extended to a chain of $`N`$ ions. The extension of this method to a chain of ions is presented in the appendix. The numerical calculations were carried out for this scheme and chains of $`N`$ ions with $`1<N10`$ and for some values $`N>10`$. Since the qualitative conclusions drawn from these calculations did not depend on $`N`$, we therefore restrict the discussion in sections III, IV, and V to the case $`N=10`$.
## III Raman sideband cooling of an ion chain
We consider sideband cooling of an ion chain when the red sideband transition is driven by a pair of counter-propagating lasers, which couple resonantly the levels $`|0`$ and $`|1`$. The two counter-propagating light fields couple with frequency $`\omega _{R1}`$, $`\omega _{R2}`$ to the optical dipole transitions $`|0|2`$ and $`|1|2`$, respectively. The two lasers are far detuned from the resonance with level $`|2`$ such that spontaneous Raman transitions are negligible compared to the stimulated process. We denote by $`\mathrm{\Delta }_{01}=[(\omega _{R1}\omega _{R2})\omega _1]`$ the Raman detuning, such that $`\mathrm{\Delta }_{01}=0`$ corresponds to driving resonantly the transition $`|0|1`$ at the first ion in the chain, and by $`\mathrm{\Omega }_{01}`$ the Rabi frequency describing the effective coupling between the two states. A third light field with Rabi frequency $`\mathrm{\Omega }_{12}`$ is tuned close to the resonance $`|1|2`$ and serves as repumper into state $`|0`$ (compare Fig. 2). The frequencies $`\omega _{Ri}`$ are close to the <sup>171</sup>Yb<sup>+</sup>resonance at 369nm, and the trap frequency is $`\nu =1\times 2\pi `$MHz. Hence, from Eq. (3) the Lamb-Dicke parameter takes the value $`\eta _10.0926`$.
### III.1 Sequential cooling
In absence of external field gradients shifting inhomogeneously the ions transition frequencies (namely, when $`\omega _1=\mathrm{}=\omega _N=\omega _0`$), cooling of an ion chain could be achieved by applying sideband cooling to each mode sequentially. In each step of the sequence all ions are illuminated simultaneously by laser light with detuning $`\mathrm{\Delta }_{01}=\nu _\alpha `$, thereby achieving sideband cooling of a particular mode $`\alpha `$. Since all ions are illuminated, they all contribute to the cooling of mode $`\alpha `$.
In Fig. 3a) the steady state vibrational number of each mode at the end of the cooling dynamics is displayed as a function of the relative detuning $`\mathrm{\Delta }_{01}`$. Each mode $`\nu _\alpha `$ reaches its minimal excitation at values of the detuning $`\mathrm{\Delta }_{01}=\nu _\alpha `$. Therefore, in order to cool all modes close to their ground state, the detuning of the laser light has to be sequentially set to the optimal value for each mode $`\alpha `$.
The cooling rates $`\mathrm{\Gamma }_{\mathrm{cool}}^{(\alpha )}`$ of mode $`\alpha `$ at $`\mathrm{\Delta }_{01}=\nu _\alpha `$, as defined in Eq. (7), are displayed in Fig. 3b). They are different for each mode and vary between 1kHz and 100kHz for the parameters chosen here. Even though these cooling rates would, in principle, allow for cooling sequentially all modes in a reasonably short time, this scheme may not be effective, since while a particular mode $`\alpha `$ is cooled all other modes are heated (i) by photon recoil, and, (ii) by coupling to the environment. As external source of heating we consider here the coupling of the ions charges to the fluctuating patch fields at the electrodes Turchette00 . The effects of these processes on the efficiency of cooling are discussed in what follows.
The consequences of heating due to photon scattering are visible in Fig. 3a). Here, one can see that while cooling one mode, others can be simultaneously heated, such that their average phonon number at steady state is very large. These dynamics are due to the form of the resonances in a three-level configuration Marzoli94 ; EIT00 . In general, however, the time scale of heating processes due to photon scattering is considerably longer than the time scale at which a certain mode is optimally sideband cooled, since the transitions leading to heating are out of resonance. In the case discussed in Fig. 3, for instance, the heating rates of these modes at $`\mathrm{\Delta }_{01}=\nu _1`$ are orders of magnitude smaller than the cooling rate of mode 1, and their dynamics can be thus neglected while mode 1 is sideband cooled. Similar dynamics are found for $`\mathrm{\Delta }_{01}=\nu _\alpha `$. Thus, in general one may neglect photon scattering as source of unwanted heating of modes that are not being efficiently cooled.
Nevertheless, heating by fluctuating electric fields occurs with appreciable rates ranging between 5s<sup>-1</sup> and $`10^4`$s<sup>-1</sup> Turchette00 . The heating rate is different for each mode and was observed to be considerably larger for the COM mode (here denoted as mode 1) than for modes that involve differential relative displacements of individual ions. Obviously, cooling can only be achieved, if the cooling rate $`\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\alpha )}`$ of each mode exceeds in magnitude the corresponding trap heating rate denoted by $`\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}`$:
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\alpha )}\alpha =1,\mathrm{},10.$$
(9)
In addition, one must consider that after a particular mode $`\alpha `$ has been cooled, it might heat up again while all other modes, $`\beta \alpha `$ are being cooled. This imposes a second condition on the cooling rate. In order to quantify this second condition, we first evaluate the time, $`T_{\mathrm{cool}}^{(\alpha )}`$ it takes to cool one particular mode $`\alpha `$ from an initial thermal distribution, obtained by means of Doppler cooling and characterized by the average occupation number $`n^{(\alpha )}_i`$, to a final distribution characterized by $`n^{(\alpha )}_f`$. This time can be estimated to be Stenholm86
$$T_{\mathrm{cool}}^{(\alpha )}\mathrm{ln}\frac{n^{(\alpha )}_i}{n^{(\alpha )}_f}/(\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\alpha )}\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}).$$
(10)
where $`n^{(\alpha )}_f1`$ at steady state was assumed.
From this relation one obtains the total time, $`T_{\mathrm{seq}}^{(\alpha )}`$ needed to cool all modes except mode $`\alpha `$, or, in other words the time during which mode $`\alpha `$ is not cooled and could get heated. This time, $`T_{\mathrm{seq}}^{(\alpha )}`$ needed to sideband cool all modes with $`\beta \alpha `$ is
$$T_{\mathrm{seq}}^{(\alpha )}=\underset{\beta ,\beta \alpha }{}T_{\mathrm{cool}}^{(\beta )}=\underset{\beta \alpha }{}\mathrm{ln}\frac{n^{(\beta )}_i}{n^{(\beta )}_f}1/(\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\beta )}\mathrm{\Gamma }_{\mathrm{heat}}^{(\beta )}).$$
(11)
If mode $`\alpha `$ is to stay cold during this time, the heating rate affecting it must be small enough. Hence the condition for efficient sequential sideband cooling is derived,
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}\times T_{\mathrm{seq}}^{(\alpha )}1$$
(12)
namely, during time $`T_{\mathrm{seq}}^{(\alpha )}`$, necessary for cooling the modes $`\beta \alpha `$, the heating of mode $`\alpha `$ has to be negligible. Clearly, this condition is stronger than the one derived in relation (9), and its fulfillment becomes critical as the number of vibrational modes (ions) is increased.
A rough estimate of the time $`T_{\mathrm{seq}}^{(\alpha )}`$ to be inserted in (12) can be obtained from eq. (11) under the assumption that all modes start out with the same mean excitation $`n_i`$ (usually determined by initial Doppler cooling) and are cooled to the same final excitation $`n_f`$. Using condition (9), one obtains
$$T_{\mathrm{seq}}^{(\alpha )}\mathrm{ln}\frac{n_i}{n_f}\underset{\beta \alpha }{}1/\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\beta )}.$$
(13)
Substituting this expression into (12) gives
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}\left(\mathrm{ln}\frac{n_i}{n_f}\underset{\beta \alpha }{}1/\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\beta )}\right)^1\mathrm{\Gamma }_{<,\mathrm{seq}}$$
(14)
which places a stronger restriction than (9) on the trap heating rate that can be tolerated, if sequential cooling is to work. This relation has to hold true for $`\alpha =1,\mathrm{},N`$. Expression (14) will be used for a comparison with simultaneous sideband cooling (see Sec. III.3).
### III.2 Simultaneous cooling
We consider now the case, when a magnetic field gradient is applied to the ion chain, such that the situation shown in Fig. 1 is realized. The axial modes of the chain can then be simultaneously cooled.
Figure 4 displays the steady state mean vibrational excitations that are obtained when the effective Rabi frequency for the Raman coupling $`\mathrm{\Omega }_{01}=5\times 2\pi `$kHz, the Rabi frequency of the repumper $`\mathrm{\Omega }_{12}=100\times 2\pi `$kHz, and the detuning $`\mathrm{\Delta }_{12}=10\times 2\pi `$MHz. Fig. 4a) displays the mean vibrational quantum number $`n^{(1)}`$ of the COM mode as a function of the detuning $`\mathrm{\Delta }_{01}`$. Here three minima are visible. The leftmost minimum occurs at $`\mathrm{\Delta }_{01}=\nu _1`$ and corresponds to resonance with the red sideband of the COM in the spectrum of the first ion. The minimum in the middle stems from the resonant drive of the red COM-sideband in the spectrum of the second ion while the one on the right is caused by the spectrum of the third ion in the chain. The location of these resonances correspond to the ones shown in Fig. 1. Heating of the COM mode occurs if the blue sideband of the COM mode is driven resonantly. In Fig. 4a) the heating at the blue sideband of the first ion, i.e. $`\mathrm{\Delta }_{01}=\nu _1`$, is visible.
Figure 4b) displays $`n^{(1)}`$, $`n^{(2)}`$, $`n^{(3)}`$, and $`n^{(4)}`$ as a function of $`\mathrm{\Delta }_{01}`$. These mean excitations have been calculated using the same parameters as in Fig. 4a). The minima visible in this figure can be identified with the corresponding resonances in the spectra of the ions by comparison with Fig. 1. A common minimum occurs at $`\mathrm{\Delta }_{01}=\nu _1`$ where all four vibrational modes are simultaneously cooled to low excitation numbers.
The mean vibrational quantum number $`n^{(\alpha )}`$ of all ten axial modes is displayed in Fig. 4c) as a function of the detuning $`\mathrm{\Delta }_{01}`$ in the neighborhood of the value $`\mathrm{\Delta }_{01}=\nu _1`$. At this value of $`\mathrm{\Delta }_{01}`$ the mean excitation $`n^{(\alpha )}`$ reaches its minimum for all modes. The mode at frequency $`\nu _{10}`$ displays a relatively narrow minimum and its mean vibrational number, although very small at exact resonance, is orders of magnitude larger than the ones of the other modes. In fact, ion 10 participates only little in the vibrational motion of mode 10. This is described by the small matrix element $`S_{10}^{10}=0.0018`$ that scales the corresponding Lamb-Dicke parameter as shown in Eq. (4).
Fig. 4d) displays the steady state temperature of each mode when the detuning of the Raman beams is set close to $`\nu _1`$. At this detuning the average excitation reaches its minimum for each mode, which is $`n^{(\alpha )}<10^3`$.
#### III.2.1 Cooling rates for simultaneous Raman cooling
We now turn to the cooling rates that are achieved when simultaneously cooling all modes, that is, the field gradient leading to the spectrum in Fig.1 is applied, and $`\mathrm{\Delta }_{01}=\nu _1`$. These rates are indicated by black bars in Fig. 5, and have been evaluated with the same parameters used for the simulation of sequential cooling in Fig. 3, that is, $`\mathrm{\Omega }_{01}=30\times 2\pi `$kHz, $`\mathrm{\Omega }_{12}=100\times 2\pi `$kHz, and $`\mathrm{\Delta }_{12}=10\times 2\pi `$MHz.
Figure 5 shows that the COM mode characterized by $`\nu _1`$ and the mode characterized by $`\nu _2`$ are cooled most efficiently, while modes 9 and 10 display much smaller cooling rates. In particular, $`\mathrm{\Gamma }_{\mathrm{cool}}^{(10)}10^1`$s<sup>-1</sup> which will make cooling of this mode very slow at $`\mathrm{\Delta }_{01}=\nu _1`$ <sup>2</sup><sup>2</sup>2It is expected that mode 10 is much less susceptible to heating by stray fields (the dominant heating mechanism as discussed in section III.1) than the COM mode Turchette00 . Therefore, such a low cooling rate will suffice once this mode is close to its ground state. However, in order to initially bring it close to the ground state the rate $`\mathrm{\Gamma }_{\mathrm{cool}}^{10}`$ needs to be larger..
The origin of this behaviour can be understood as follows. The scheme of simultaneous sideband cooling requires that each ion of the chain is employed to cool one of the axial modes. This is achieved by applying a suitable magnetic field gradient. The simplest experimental implementation uses a monotonically increasing magnetic field, such that the first ion of the chain is used for cooling mode 1, the second ion mode 2, etc. (compare Fig. 1). However, the coupling of a certain ion displacement to a certain mode can be very small. It occurs, for instance, that the largest axial excitations couple weakly to the ions at the edges of the ion string, instead they are mainly characterized by oscillations of ions in the center of the chain PRL04 . A manifestation of this behavior is the small value of the matrix element $`S_{10}^{10}`$. Thus, mode 10 is not efficiently cooled by illuminating ion 10, rather it is optimally cooled by addressing an ion closer to the center of the chain. This is evident by inspection of the grey bars in Fig. 5 that indicate the optimal cooling rate for each individual mode, obtained by employing that particular ion in the chain which has the largest coupling to the mode to be cooled. In presence of the magnetic field gradient, this optimal cooling rate is achieved, if the detuning, $`\mathrm{\Delta }_{01}`$ is set such that the appropriate red-sideband transition of this particular ion is driven resonantly. In this way, the cooling rate is maximal for that particular mode (i.e., each grey bar corresponds to a different detuning).
For the case of mode 10 the difference between the cooling rates depending on which ion is addressed is particularly striking: the cooling rate of mode 10 at detuning $`\mathrm{\Delta }_{01}=\nu _1`$ (black bar) is very small, as noted above, whereas the optimal rate $`\mathrm{\Gamma }_{\mathrm{cool}}^{(10)}=2.63\times 10^3`$s<sup>-1</sup> (given the parameters used here) is achieved at $`\mathrm{\Delta }_{01}=3.28\times \nu _1`$. At this detuning ion 6 is used for cooling mode 10, as can be seen from Fig. 1.
Efficient cooling of all modes, given the field gradient that gives rise to the spectrum illustrated in Fig. 1, can be obtained by combining sequential and simultaneous cooling as follows: First the ion string is illuminated with radiation such that $`\mathrm{\Delta }_{01}=3.28\times \nu _1`$ which will efficiently cool mode 10 and will have little effect on all other modes. Then, we apply radiation with $`\mathrm{\Delta }_{01}=\nu _1`$ (cooling rates indicated by black bars in Fig. 5). This will simultaneously cool all other modes. This is a specific recipe to cool 10 ions. For an arbitrary number of ions, first one cools the modes which cannot be efficiently simultaneously cooled with all other modes, and then, in a second step, as many modes as possible are cooled simultaneously (in our case modes 1 through 9).
A necessary condition for the efficiency of this scheme is
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}\mathrm{\Gamma }_{\mathrm{cool},\mathrm{sim}}^{(\alpha )}\alpha =1,\mathrm{},10.$$
(15)
where $`\mathrm{\Gamma }_{\mathrm{cool},\mathrm{sim}}^{(\alpha )}`$ is the cooling rate of mode $`\alpha `$ when simultaneously cooling it with the other modes. Condition (15) is the analogue to (9) which we derived for sequential cooling. Moreover, since in the scheme proposed here modes 1 through 9 are simultaneously cooled, a condition analogous to the one in Eq. (12) has to be fulfilled for mode 10 only: After mode 10 has been cooled, this mode may not heat up again appreciably during the time for which modes 1 through 9 are simultaneously cooled. This condition is expressed as
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(10)}\times T_{\mathrm{sim}}^{(10)}1.$$
(16)
where $`T_{\mathrm{sim}}^{(10)}`$ is the time during which modes 1 through 9 are cooled simultaneously to the desired final values, that is, the time during which mode 10 is not cooled and could heat up. Here, it is given by
$$T_{\mathrm{sim}}^{(10)}=\mathrm{max}_{\beta =1\mathrm{}9}\left[\mathrm{ln}\frac{n^{(\beta )}_i}{n^{(\beta )}_f}/(\mathrm{\Gamma }_{\mathrm{cool}}^{(\beta )}\mathrm{\Gamma }_{\mathrm{heat}}^{(\beta )})\right],$$
(17)
This relation is analogous to relation (11) for sequential cooling.
The times $`T_{\mathrm{seq}}^{(10)}`$ (eq. (11)) and $`T_{\mathrm{sim}}^{(10)}`$ (eq. (17)) give an upper limit for the tolerable trap heating rate of mode 10 for the cases of sequential and simultaneous cooling, respectively. It turns out that the relevant time scales $`T_{\mathrm{seq}}^{(10)}`$ and $`T_{\mathrm{sim}}^{(10)}`$ are of the same order of magnitude: Using eq. 17 to compute $`T_{\mathrm{sim}}^{(10)}`$ for the parameters used in Fig. 5 gives $`T_{\mathrm{sim}}^{(10)}\mathrm{ln}(n_i/n_f)\times 2.4\times 10^3`$s, while $`T_{\mathrm{seq}}^{(10)}\mathrm{ln}(n_i/n_f)\times 1.3\times 10^3`$s is obtained from eq. (11) for the same set of parameters footnoteC . Thus, the cooling scheme proposed here does not increase the admissible trap heating rate of mode 10. Since this rate is expected to be low anyway, this will not be a restriction prohibiting the successful cooling of a long ion chain using either sequential or simultaneous cooling.
### III.3 Comparison of sequential and simultaneous cooling
So far we have stated the general conditions that have to be met in order to efficiently cool an ion chain with sequential cooling and with a scheme combining simultaneous and sequential sideband cooling. In this section we discuss their efficiencies.
We note that when using sequential sideband cooling, one may utilize all ions in the chain in order to cool one mode, where the cooling rates of each ion add up incoherently. In the case of simultaneous sideband cooling, on the other hand, only one ion is employed in order to cool a particular mode. Assuming that the coupling of this ion to the mode is sufficiently large to allow for efficient cooling, the following expression for the average simultaneous cooling rate is deduced
$$\mathrm{\Gamma }_{\mathrm{cool},\mathrm{sim}}\frac{1}{N}\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}$$
(18)
where $`\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}`$ is the cooling rate for sequential cooling averaged over all modes. Thus, for simultaneous cooling relation (15) yields
$$\mathrm{\Gamma }_{\mathrm{heat}}^{(\alpha )}\frac{1}{N}\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}\mathrm{\Gamma }_{<,\mathrm{sim}},$$
(19)
From (18), since the total cooling rate is $`\mathrm{\Gamma }_{<,\mathrm{seq}}=\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}/N`$, one obtains that the efficiencies of simultaneous and sequential cooling are comparable. However, it should be remarked that this estimate corresponds to the worst case for simultaneous cooling. In fact, estimate (18) is correct for long-wavelength vibrational excitations, which correspond to low-frequency axial modes, where practically all ions of the chain participate in the mode oscillation. Short-wavelength excitations, on the other hand, are characterized by large displacements of the central ions, while the ions at the edges practically do not move PRL04 . Due to this property, for these modes one may find a magnetic field configuration such that $`\mathrm{\Gamma }_{\mathrm{cool},\mathrm{sim}}^{(\alpha )}\mathrm{\Gamma }_{\mathrm{cool},\mathrm{seq}}^{(\alpha )}`$.
On the basis of these qualitative considerations, one may in general state that simultaneous cooling of the chain is at least as efficient as sequential cooling. For the configuration discussed in this paper the two methods are comparable. Differently from sequential cooling, however, the efficiency of simultaneous cooling can be substantially improved by choosing a suitable magnetic field configuration that maximizes coupling of each mode to one ion of the chain.
## IV Sideband cooling using microwave fields
### IV.1 Effective Lamb-Dicke parameter
We investigate now sideband cooling of the ion chainโs collective motion using microwave radiation for driving the sideband transition. It should be remarked that in this frequency range sideband excitation cannot be achieved by means of photon recoil which for long wavelengths is negligible. This is evident from eq. 4 using a typical trap frequency, $`\nu _\alpha `$ of the order $`2\pi \times 1`$MHz. Nevertheless, in Mintert01 it was shown that with the application of a magnetic field gradient an additional mechanical effect can be produced accompanying the absorption/emission of a photon. This is achieved by realizing different mechanical potentials for the states $`|1`$ and $`|0`$ Mintert01 ; Wunderlich02 ; Wunderlich03 . Hence, by changing an ionโs internal state by stimulated absorption or emission of a microwave photon the ion experiences a mechanical force.
An effective LDP can be associated with this force, that is defined as Wunderlich02 :
$$\eta _{j\alpha }^{\mathrm{eff}}e^{i\phi _j}\eta _j^\alpha +i\epsilon _j^\alpha $$
(20)
where
$$\epsilon _j^\alpha =S_j^\alpha \frac{_z\omega _j\mathrm{\Delta }z_\alpha }{\nu _\alpha },$$
(21)
Here $`_z\omega _j`$ is the spatial derivative of the resonance frequency of the ion at $`z_j^{(0)}`$ with respect to $`z_j`$, and $`\mathrm{\Delta }z_\alpha =\sqrt{\mathrm{}/(2m\nu _\alpha )}`$. The reader is referred to the appendix for the theoretical description of the ion chain dynamics in presence of a spatially-varying magnetic field.
The term $`\eta _j^\alpha `$ appearing in Eq. (20) is the LDP due to photon recoil and defined in Eq. (4), while the term $`\epsilon _j^\alpha `$ is the LPD arising from the mechanical effect induced by the magnetic field gradient. Their ratio is given by
$$\frac{\epsilon _j^\alpha }{\eta _j^\alpha }=\frac{\kappa _j^\alpha }{k\mathrm{\Delta }z_\alpha }=\frac{1}{2\pi }\frac{\lambda }{\mathrm{\Delta }z_\alpha }\kappa _j^\alpha $$
(22)
where $`\lambda `$ is the wavelength of the considered transition and $`\kappa _j^\alpha =\mathrm{\Delta }z_\alpha _z\omega _j/\nu _\alpha `$ is the rescaled frequency gradient Mintert01 ; Wunderlich02 .
For a transition in the microwave frequency range, like the transition between the states $`|0`$ and $`|1`$ in <sup>171</sup>Yb<sup>+</sup>, using typical values, like $`\kappa _j^\alpha 10^3`$, $`\lambda 10^2`$m, $`\mathrm{\Delta }z_\alpha 10^8`$m, one finds that the Lamb-Dicke parameter due to the recoil of a microwave photon is at least two orders of magnitude smaller than the one due to the mechanical effect induced by the magnetic field, and thus $`\eta _{j\alpha }^{\mathrm{eff}}\epsilon _j^\alpha `$ in the microwave region.
### IV.2 Steady state population and cooling rates
We study sideband cooling using microwave radiation using the model outlined in the appendix. For the Rabi frequencies the same parameters as in section III.2 are employed (i.e., $`\mathrm{\Omega }_{01}=5\times 2\pi `$kHz, $`\mathrm{\Omega }_{12}=100\times 2\pi `$kHz, $`\mathrm{\Delta }_{12}=10\times 2\pi `$MHz). The effective LDP is determined by the magnitude of the magnetic field gradient that is used to superimpose the motional sidebands of all vibrational modes, and is given by $`\eta _{j\alpha }^{\mathrm{eff}}\epsilon _j^\alpha =S_j^\alpha \kappa _j^\alpha S_j^\alpha \times 2\times 10^3`$.
It must be remarked that these parameters are outside the range of validity for the application of the numerical method we use. In fact, by applying it one neglects the fourth and sixth orders in the LDP expansion of the optical transition of the repumping cycle, which are of the same order of magnitude as the microwave sideband excitation. Nevertheless, given the complexity of the problem, characterized by a large number of degrees of freedom, we have chosen to use this simpler method in order to get an indicative estimate of the efficiency as compared to the case in which the sidebands are driven by optical radiation. Therefore, the results we obtain in this section are indicative. In fact, neglecting higher orders in the LDP of the optical transition corresponds to underestimate heating effects due to diffusion.
Fig. 6a) shows the steady state mean vibrational quantum number $`n^{(\alpha )}_f`$ of the 10 axial vibrational modes as a function of the detuning $`\mathrm{\Delta }_{01}`$ of the microwave radiation, where $`\mathrm{\Delta }_{01}=0`$ when the microwave field is at frequency $`\omega _1`$. Figure 6b) displays the final excitation number of all vibrational modes as a function of $`\mathrm{\Delta }_{01}`$ around the value $`\mathrm{\Delta }_{01}=\nu _1`$. Figure 6c) shows the steady state mean excitation of all 10 modes at $`\mathrm{\Delta }_{\mathrm{M}W}=\nu _1`$.
All vibrational modes are cooled close to their ground state. However, the average occupation number of the highest vibrational frequency $`\nu _{10}`$ is orders of magnitude larger than the ones of the other modes. The origin of this behavior is the small value of the coefficient $`S_{10}^{10}=0.0018`$ as is discussed in Sec. III.2. A comparison of Fig. 6c) with the results obtained by simultaneously cooling using an optical Raman processes (Fig. 4d) shows that the mean vibrational numbers that are achieved here are considerably larger.
In Fig. 6d) the cooling rates are displayed that are obtained when driving the sideband transition with microwave radiation. These rates are much smaller than the ones shown in Fig. 5 that are obtained using an optical Raman process.
From this comparison it is evident that simultaneous cooling is more effective by using an optical Raman process than microwave radiation. In particular, the cooling rates obtained with simultaneous microwave sideband cooling can be comparable to the heating rates in some experimental situations, resulting in inefficient cooling.
This striking difference in the efficiency can already be found, if comparing the two methods when they are applied to cooling a single ion. Its origin lies in the fact that in the optical case the LDP accounting for the photon recoil, namely $`\eta _j^\alpha `$ in Eq. (20), is considerably larger than the LDP, $`\epsilon _j^\alpha `$ caused by the magnetic field gradient used here to superimpose the motional sidebands. This can be verified by using an optical wavelength in Eq. (22) which gives $`\eta _{j\alpha }^{\mathrm{eff}}\eta _j^\alpha `$. On the other hand, driving the $`|0|1`$ transition directly by microwaves results in $`\eta _{j\alpha }^{\mathrm{eff}}\epsilon _j^\alpha \eta _j^\alpha `$. For the model system considered here, the optical LDP is at least one order of magnitude larger than the microwave LDP for simultaneous cooling.
In the microwave sideband cooling scheme optical transitions are used for repumping, thus leading to an enhanced diffusion rate during the dynamics and thus to lower efficiencies. The fundamental features of these dynamics can be illustrated by a rate equation of the form Eq. (5), describing sideband cooling of a single ion with microwave radiation, where the rates are
$`A_{}={\displaystyle \frac{\mathrm{\Omega }^2}{2\gamma }}\left[|\epsilon |^2+\varphi |\eta _{\mathrm{opt}}|^2{\displaystyle \frac{\gamma ^2}{4\nu ^2+\gamma ^2}}\right]`$ (23)
$`A_+={\displaystyle \frac{\mathrm{\Omega }^2}{2\gamma }}\left[|\epsilon |^2{\displaystyle \frac{\gamma ^2}{16\nu ^2+\gamma ^2}}+\varphi |\eta _{\mathrm{opt}}|^2{\displaystyle \frac{\gamma ^2}{4\nu ^2+\gamma ^2}}\right],`$ (24)
$`\eta _{\mathrm{opt}}`$ accounts for the recoil due to the spontaneous emission when the ion is optically pumped to the state $`|0`$ Morigi01 , and $`\epsilon `$ is the LDP for the microwave transition due to the field gradient. The latter multiplies the terms where a sideband transition occurs by microwave excitation, whereas $`\eta _{\mathrm{opt}}`$ multiplies the terms where sideband excitation occurs by means of spontaneous emission, which thus describe the diffusion during the cooling process. Efficient ground state cooling is achieved when the rate of cooling is much larger than the rate of heating, which corresponds to the condition $`A_{}/A_+1`$. For $`\nu _1\gamma _\mu `$, whereby $`\gamma _\mu `$ is the linewidth of the $`|0|1`$ transition, this ratio scales as
$$\frac{A_{}}{A_+}\frac{|\epsilon |^2}{|\eta _{\mathrm{opt}}|^2}\frac{4\nu ^2}{\gamma ^2}$$
(25)
This result differs for the ratio obtained in the all-optical case, where $`A_{}/A_+4\nu ^2/\gamma ^2`$. For typical parameters, corresponding to a magnetic field gradient that superposes all sidebands in an ion trap, $`|\epsilon ||\eta _{\mathrm{opt}}|`$. Hence, for a given value of the ratio $`\gamma /\nu `$ the cooling efficiency in the optical case is considerably larger than in the microwave case.
Note that the parameter $`\epsilon `$ can be made larger by increasing the magnitude of the magnetic field gradient, which in Eq. (22) corresponds to increasing $`\kappa _j^\alpha `$. However, if the modes of an ion chain are to be cooled simultaneously, the choice of the magnetic field gradient is fixed by the distance between neighboring ions, and the efficiency is thus limited by this requirement.
If an individual ion (or a neutral atom confined, for example, in an optical dipole trap) is to be sideband cooled, or sequential cooling is applied to a chain of ions, then the above mentioned restriction on the magnitude of $`\epsilon `$ is not present and microwave sideband cooling can be as efficient as Raman cooling. For this case, method Marzoli94 may be implemented, provided the Lamb-Dicke regime applies and $`\eta _{\mathrm{opt}}`$ and $`ฯต`$ are of the same order of magnitude.
## V Experimental considerations
In this section we discuss how the magnetic field gradient for simultaneous sideband cooling can be generated and how cooling is affected, if the red motional sidebands of different vibrational modes are not perfectly superposed. In order to demonstrate the feasibility of the proposed scheme it is sufficient to restrict the discussion to very simple arrangements of magnetic field generating coils.
The use of a position dependent ac-Stark shift has been proposed in Staanum02 to modify the spectrum of a linear ion chain. This may be another way of appropriately shifting the sideband resonances but will not be considered here.
### V.1 Required magnetic field gradient
If the vibrational resonances and the ions were equally spaced in frequency and position space, respectively, then a constant field gradient, appropriately chosen, could make all $`N`$ modes overlap and let them be cooled at the same time. Since $`\nu _\alpha \nu _{\alpha 1}`$ decreases monotonically with growing $`\alpha `$ and the ionsโ mutual distances vary with $`z_j`$, the magnetic field gradient has to be adjusted along the $`z`$axis. The field gradient needed to shift the ionsโ resonances by the desired amount is obtained from
$`{\displaystyle \frac{B}{z}}|_{(z_j+z_{j1})/2}`$ $``$ $`{\displaystyle \frac{B(z_j)B(z_{j1})}{z_jz_{j1}}}`$ (26)
$`\stackrel{!}{=}`$ $`{\displaystyle \frac{\upsilon _j\upsilon _{j1}}{\zeta _j\zeta _{j1}}}\zeta _0\nu _1{\displaystyle \frac{\mathrm{}}{\mu _B}},j=2,\mathrm{},N`$
where $`\zeta _jz_j/\zeta _0`$ is the scaled equilibrium position of ion $`j`$, and $`\upsilon _j`$ is the square root of the $`j`$th eigenvalue of the dynamical matrix. Eq. 26 describes the situation for moderate magnetic fields (the Zeeman energy is much smaller than the hyperfine splitting), such that $`_z\omega _j=1/2g_J\mu _B_zB`$ with $`g_Jg_s=2`$ (state $`|0`$ does not depend on $`B`$). As an example, we consider again a string of $`N=10`$ <sup>171</sup>Yb<sup>+</sup>ions in a trap characterized by $`\nu _z=1\times 2\pi `$MHz (thus, $`\zeta _0=2.7\mu `$m).
The markers in Fig. 7 indicate the values of the required field gradient according to eq. 26 whereas the solid line shows the gradient generated by 3 single windings of diameter 100$`\mu `$m, located at $`z=100,50,`$ and $`100`$ mm $`36\zeta _0`$, respectively (the trap center is chosen as the origin of the coordinate system) Fortagh98 . Running the currents -5.33A, -6.46A, and 4.29A,respectively, through these coils produces the desired field gradient at the location of the ions. Micro electromagnets with dimensions of a few tens of micrometers and smaller are now routinely used in experiments where neutral atoms are trapped and manipulated Drndic98 . Current densities up to $`10^8`$A/cm<sup>2</sup> have been achieved in such experiments. A current density more than two orders of magnitude less than was achieved in atom trapping experiments would suffice in the above mentioned example.
This configuration of magnetic field coils shall serve as an example to illustrate the feasibility of the proposed cooling scheme in what follows. It will be shown that with such few current carrying elements in this simple arrangement one may obtain good results when simultaneously sideband cooling all axial modes. More sophisticated structures for generating the magnetic field gradients can of course be employed, making use of more coils, different diameters, variable currents, or completely different configurations of current carrying structures.
Ideally, all 10 sideband resonances would be superimposed for optimal cooling. The resonances shown in Fig. 8 result from the field gradient calculated using the simple field generating configuration described above, and do not all fall on top of each other. Nevertheless, Fig. 8 shows how well all 10 sideband resonances are grouped around $`\omega _1\nu _1`$. Vertical bars indicate the location relative to $`\omega _1`$ of the red sideband resonance, $`\omega _j\nu _j`$ of the $`j`$th ion, with $`j=1,\mathrm{},10`$. These resonances all lie within a frequency interval of about $`0.015\times \nu _1/2\pi =15`$kHz.
The height of the bars in Fig. 8 indicates the strength of the coupling between the driving radiation and the respective sideband transition relative to the COM sideband of ion number 1. The relevant coupling parameter is the LDP. For optical transitions $`|\eta _{j\alpha }^{\mathrm{eff}}|/|\eta _{11}^{\mathrm{eff}}||\eta _j^\alpha |/|\eta _1^1|=S_j^\alpha \nu _\alpha ^{(1/2)}/S_1^1\nu _1^{(1/2)}`$ with $`j=\alpha `$. The ratio of these parameters for the highest vibrational mode ($`\alpha =10`$) is about 3 orders of magnitude smaller than for mode 1, since ion 10 is only slightly displaced from its equilibrium position when mode 10 is excited (compare the discussion in section III.2).
We will now investigate how well simultaneous cooling can be done with the sideband resonances not perfectly superposed.
### V.2 Simultaneous Raman cooling with non-ideal gradient
In Fig. 9a) the steady state vibrational excitation, $`n^{(\alpha )}_f`$ of a string of 10 ions is displayed as a function of the detuning of the Raman beams relative to the resonance frequency $`\omega _1`$ of ion 1. The Rabi frequencies and detuning, too, are the same as have been used to generate Fig. 4. However, the field gradient that shifts the ionsโ resonances is not assumed ideal as in Fig. 4, instead the one generated by three single windings as described above (Fig. 7) has been used. Despite the imperfect superposition of the cooling resonances, low temperatures of all modes close to their ground state can be achieved as can be seen in Fig. 9b). Here, the value of $`n^{(\alpha )}_f`$ for each mode has been plotted at that detuning $`\mathrm{\Delta }_{01}=1.008\nu _1`$ where the sum of all excitations is minimal.
Fig. 10a displays the excitation of each mode over a wide range of the detuning such that all first order red sideband resonances are visible. Here, the Rabi frequency $`\mathrm{\Omega }_{12}`$ of the repump laser has been increased to $`1\times 2\pi `$MHz as compared to $`100\times 2\pi `$kHz in the previous figures. This results i) in higher final temperatures, and, ii) in broader resonances as is evident in Fig. 10b and thus makes cooling less susceptible to errors in the relative detuning between laser light and ionic resonances.
A higher steady state vibrational excitation due to the larger intensity of the repump laser is evident in Fig. 10c where $`n^{(\alpha )}_f`$ for each mode is plotted with the same parameters as in Fig. 9b, however with $`\mathrm{\Omega }_{12}=1\times 2\pi `$MHz.
Errors and fluctuations in the relative detuning of the Raman laser beams driving the sideband transition are expected to be small and not to affect the efficiency of simultaneous cooling, if the two light fields inducing the stimulated Raman process are derived from the same laser source using, for example, acousto-optic or electro-optic modulators. This is feasible by translating into the optical domain the microwave or radio frequency that characterizes the splitting of states $`|0`$ and $`|1`$. Microwave or rf signals can be controlled with high precision and display low enough drift to ensure efficient cooling. If a large enough intensity of the repump laser is employed, then the steady state vibrational excitation varies slowly as a function of $`\mathrm{\Delta }_{01}`$ as is visible in Fig. 10b. Thus, the requirements regarding both the precision of adjustment and the drift of the source generating the Raman difference frequency are further relaxed.
It should be noted that efficient cooling does not only occur around the resonance $`\mathrm{\Delta }_{01}=\nu _1`$ but also at other values of $`\mathrm{\Delta }_{01}`$ as can be seen in Fig. 10a. As an example, Fig. 10d shows $`n^{(\alpha )}_f`$ of all modes at that detuning, $`\mathrm{\Delta }_{01}=3.91\nu _1`$ where mode 10 reaches its absolute minimum. At this resonance the red sideband of the 5th ion corresponding to the 10th mode is driven by the Raman beams (compare Fig. 1). Note that all other vibrational modes are also cooled at the same time. Therefore, an efficient procedure for cooling all vibrational modes close to their ground state would be to first tune the Raman beams such that mode 10 is optimally cooled (i.e., $`\mathrm{\Delta }_{01}=3.91\nu _1`$, Fig. 10d), and subsequently set $`\mathrm{\Delta }_{01}=1.022\nu _1`$ (Fig. 10c) in order to simultaneously cool modes 1 through 9. This approach is discussed in more detail in section III.2.1.
Initial cooling of vibrational modes often is prerequisite for subsequent coherent manipulation of internal and motional degrees of freedom of an ion chain, for example, quantum logic operations. When implementing quantum logic operations it may not be advantageous to address all motional sidebands with a single frequency as is done here for simultaneous sideband cooling. The magnetic field gradient that superposes the sideband resonances for cooling may then be turned off adiabatically after initial Raman cooling of all vibrational modes. This should be done fast enough not to allow for appreciable heating of the ion string, for example, by patch fields, and slow enough not to excite vibrational modes in the process. A lower limit for the time it takes to ramp up the gradient seems to be $`2\pi /\nu _1`$. Thus, with $`\nu _1=1\times 2\pi `$MHz the additional time needed to change the field gradient is negligible compared to the time needed to sideband cool the ion string which for typical parameters takes between a few hundred $`\mu `$s and a few ms (compare Fig. 5).
If microwave radiation is used to coherently manipulate internal and motional degrees of freedom and for quantum logic operations, it is useful not to turn off the magnetic field gradient, but instead to ramp up the field gradient to a value where all coincidences between internal and motional resonances are removed Mintert01 ; Wunderlich03 . Also, a larger field gradient for quantum logic operations is desirable in this case to have stronger coupling between internal and external states (i.e., a larger effective LDP $`\eta _{j\alpha }^{\mathrm{eff}}`$ \[eq. 20\]). The considerations in the previous paragraph regarding the time scale of change of the field gradient apply here, too.
For the cooling scheme introduced here to work, the field gradient has to vary in the axial direction and it remains to be shown in what follows that this variation is compatible with the neglect of higher-order terms in the local displacement $`q_j`$ of ion $`j`$ and in $`_zB`$ in the derivation of the effective Lamb-Dicke parameter induced by the magnetic field gradient Mintert01 ; Wunderlich02 . Neglecting higher order terms is justified as long as $`|q_j^2_z^2B||q_j_zB|`$. Using 26 and $`q_j\mathrm{\Delta }z=\sqrt{\mathrm{}/2m\nu _1}`$ this condition can be written as
$$\frac{2}{\zeta _{j+1}\zeta _{j1}}\left|\frac{(\upsilon _{j+1}\upsilon _j)(\zeta _j\zeta _{j1})}{(\upsilon _j\upsilon _{j1})(\zeta _{j+1}\zeta _j)}1\right|\frac{\zeta _0}{\mathrm{\Delta }z}.$$
(27)
Considering the region where the second derivative of the magentic field is maximal ($`j=9`$) and inserting numbers into relation 27 gives for ten ions $`0.242.9\times 10^3(m/\nu _1)^{\frac{1}{6}}`$. The right-hand side of this inequality is dimensionless if $`m`$ is inserted in a.m.u., and yields $`500`$ for <sup>171</sup>Yb<sup>+</sup>ions and $`\nu _1=1\times 2\pi `$MHz. Hence, relation 27 is fulfilled. This can be understood by considering that the typical distance, $`\zeta _0`$ over which the gradient has to vary is much larger than the range of motion, $`q_j\mathrm{\Delta }z`$ of an individual ion, and the approximation of a linear field gradient is a good one.
## VI Conclusions
We have proposed a scheme for cooling the vibrational motion of ions in a linear trap configuration. Axial vibrational modes are simultaneously cooled close to their ground state by superimposing the red motional sidebands in the absorption spectrum of different ions such that the red sideband of each mode is excited when driving an internal transition of the ions with monochromatic radiation. This spectral property is achieved by applying a magnetic field gradient along the trap axis shifting individually the internal ionic resonances by a desired amount.
Exemplary results of numerical simulations for the case of an ion chain consisting of $`N=10`$ ions are presented and extensively discussed. Detailed simulations have also been carried out with $`1<N<10`$ and for some values $`N>10`$. They lead to the same qualitative conclusion as presented for the case of $`N=10`$.
Numerical studies show that simultaneously Raman cooling all axial modes is effective for realistic sets of parameters. These studies also reveal that using microwave radiation to drive the sideband transition is not as efficient, due to the relatively small mechanical effect associated with the excitation of this transition. The mechanical effect could be enhanced by applying a larger magnetic field gradients. For simultaneous sideband cooling of all vibrational modes, however, the gradient is fixed by the requirement of superimposing the sidebands. Usual sideband cooling on a hyperfine or Zeeman transition using microwave radiation becomes possible, if a larger field gradient is used. This may be particularly useful, if a single ion (or an atom in an optical dipole trap) is to be sideband cooled using a microwave transition, or, if the vibrational modes of an ion chain are to be cooled sequentially using microwave radiation. Moreover, these techniques could also be implemented for sympathetic cooling of an ion chain, whereby some modes are simultaneously cooled by addressing ions of other species embedded in the chain Morigi01 ; Kielpinski01 ; Barrett03 .
In conclusion, we have shown that simultaneous sideband cooling with optical radiation can be efficiently implemented taking into account experimental conditions and even for a simple arrangement of magnetic field generating elements.
## VII Acknowledgements
We acknowledge financial support by the Deutsche Forschungsgemeinschaft, Science Foundation Ireland under Grant No. 03/IN3/I397, and the European Union (QGATES,QUIPROCONE).
## VIII Appendix
In this appendix we introduce the hamiltonian and master equation describing the dynamics discussed in Sec. IV.2. We consider a chain consisting of $`N`$ identical ions aligned along the $`z`$-axis, and in presence of a magnetic field $`B(z)`$. The internal electronic states of each ion which are relevant for the dynamics are the stable states $`|0`$ and $`|1`$ and the excited state $`|2`$. The transitions $`|0|1`$, $`|1|2`$ are respectively a magnetic and an optical dipole transition. We assume that the magnetic moments of $`|0`$ and $`|2`$ vanish, while $`|1`$ has magnetic moment $`\mu `$. Thus its energy with respect to $`|0`$ is shifted proportionally to the field, $`\mathrm{}\omega _0(z)|B(z)|`$. The Hamiltonian describing the internal degrees of freedom has the form:
$$H_{\mathrm{int}}=\mathrm{}\underset{j}{}\left(\omega _0(z_j)|1_j1|+\omega _2|2_j2|\right)$$
(28)
where the index $`j`$ labels the ions along the chain. The collective excitations of the chain are described by the eigenmodes at frequency $`\nu _1,\mathrm{},\nu _N`$, which are independent of the internal states. Denoting with $`Q_\alpha ,P_\alpha `$ the normal coordinates and conjugate momenta of the oscillator at frequency $`\nu _\alpha `$, the Hamiltonian for the external degrees of freedom then has the form
$`H_{\mathrm{mec}}`$ $`=`$ $`{\displaystyle \frac{1}{2m}}{\displaystyle \underset{\alpha =1}{\overset{N}{}}}P_\alpha ^2`$
$`+{\displaystyle \frac{m}{2}}{\displaystyle \underset{\alpha =1}{\overset{N}{}}}\nu _\alpha ^2\left[Q_\alpha +{\displaystyle \frac{\mathrm{}}{2m\nu _\alpha ^2}}{\displaystyle \underset{j}{}}{\displaystyle \frac{\omega _j}{z_j}}|_{z_{0,j}}|11|S_j^\alpha \right]^2`$
where we have neglected the higher spatial derivatives of the magnetic field.
Thus, the coupling of the excited state $`|1`$ to a spatially varying magnetic field shifts the center of the oscillators for the ions in the electronic excited state.
The states $`|0`$ and $`|1`$ are coupled by radiation according to the Hamiltonian
$$W_{01}=\underset{j}{}\frac{\mathrm{}\mathrm{\Omega }_{01}}{2}[|1_j0|\mathrm{e}^{\mathrm{i}(\omega _{01}tk_{01}z_j+\psi )}+\mathrm{H}.\mathrm{c}.]$$
(30)
where the coupling can be generated either by microwave radiation driving the magnetic dipole, or by a pair of Raman lasers. In the first case, $`\mathrm{\Omega }_{01}`$ is the Rabi frequency, $`\omega _{01}`$ the frequency of radiation and $`k_{01}`$ the corresponding wave vector. In the case of coupling by Raman lasers, $`\mathrm{\Omega }_{01}`$ is the effective Rabi frequency, $`\omega _{01}`$ the frequency and $`k_{01}`$ the resulting wave vector describing the two-photon process.
The transition $`|1|2`$ is driven below saturation by a laser at Rabi frequency $`\mathrm{\Omega }_{12}`$, frequency $`\omega _{12}`$, and wave vector $`k`$. The interaction term reads:
$$W_{12}=\underset{j}{}\frac{\mathrm{}\mathrm{\Omega }_{12}}{2}[|2_j1|\mathrm{e}^{\mathrm{i}(\omega _{12}tkz_j+\psi )}\mathrm{e}^{\mathrm{i}kq_j}+\mathrm{H}.\mathrm{c}.]$$
(31)
where $`q_j`$ is the displacement of the ion $`j`$ from the classical equilibrium position $`z_j`$.
The master equation for the density matrix $`\rho `$, describing the internal and external degrees of freedom of the ions, reads:
$$\frac{}{t}\rho =\frac{1}{\mathrm{i}\mathrm{}}[H,\rho ]+\rho $$
(32)
where $`H=H_0+H_{\mathrm{mec}}+W_{01}+W_{12}`$, and $`\rho `$ is the Liouvillian describing the spontaneous emission processes, i.e. the decay from the state $`|2`$ into the states $`|0`$ and $`|1`$ at rates $`\mathrm{\Gamma }_{20}`$, $`\mathrm{\Gamma }_{21}`$, respectively, where $`\mathrm{\Gamma }=\mathrm{\Gamma }_{20}+\mathrm{\Gamma }_{21}`$ is the total decay rate. The Liouville operator for the spontaneous decay is
$`\rho ={\displaystyle \frac{1}{2}}\mathrm{\Gamma }{\displaystyle \underset{j}{}}\left[|2_j2|\rho +\rho |2_j2|\right]`$ (33)
$`+\mathrm{\Gamma }_{20}{\displaystyle _1^1}du๐ฉ(u)\mathrm{e}^{\mathrm{i}kuq_j}|0_j2|\rho |2_j0|\mathrm{e}^{\mathrm{i}kuq_j}`$
$`+\mathrm{\Gamma }_{21}{\displaystyle _1^1}du๐ฉ(u)\mathrm{e}^{\mathrm{i}kuq_j}|1_j2|\rho |2_j1|\mathrm{e}^{\mathrm{i}kuq_j}`$
with $`๐ฉ(u)`$ the dipole pattern for spontaneous emission and $`u`$ the projection of the direction of photon emission onto the trap axis. In order to study the dynamics, it is convenient to move to the inertial frames rotating at the field frequencies. Moreover, we apply the unitary transformation Wunderlich02
$$U=\mathrm{exp}\left[\mathrm{i}\underset{\alpha }{}\left(\frac{1}{2m\nu _\alpha ^2}\underset{j}{}\frac{\omega _{0,j}}{z_j}|_{z_{0,n}}|1_j1|S_j^\alpha \right)P_\alpha \right]$$
(34)
We denote with $`\stackrel{~}{\rho }`$ the density matrix in the new reference frame. The master equation now reads
$$\frac{}{t}\stackrel{~}{\rho }=\frac{1}{\mathrm{i}\mathrm{}}[\stackrel{~}{H},\stackrel{~}{\rho }]+\stackrel{~}{\rho }$$
(35)
where $`\stackrel{~}{H}=\stackrel{~}{H}_0+\stackrel{~}{H}_{\mathrm{mec}}+\stackrel{~}{W}_{01}+\stackrel{~}{W}_{12}`$, and the individual terms have the form:
$$\stackrel{~}{H_0}=\mathrm{}\underset{j}{}\left[\delta (z_j)|0_j0|+\mathrm{\Delta }(z_j)|22|\right]$$
(36)
with $`\delta (z_j)=\omega _{01}\omega _0(z_j)`$ and $`\mathrm{\Delta }(z_j)=\omega _2\omega _0(z_j)\omega _{12}`$. The mechanical energy is
$`\stackrel{~}{H}_{\mathrm{mec}}`$ $`=`$ $`{\displaystyle \frac{1}{2m}}{\displaystyle \underset{\alpha =1}{\overset{N}{}}}P_\alpha ^2+{\displaystyle \frac{m}{2}}{\displaystyle \underset{\alpha =1}{\overset{N}{}}}\nu _\alpha ^2Q_\alpha ^2`$ (37)
$`=`$ $`{\displaystyle \underset{\alpha }{}}\mathrm{}\nu _\alpha \left(a_\alpha ^{}a_\alpha +{\displaystyle \frac{1}{2}}\right)`$
where $`a_\alpha ^{}`$, $`a_\alpha `$ are the creation and annihilation operator, respectively of a quantum of energy $`\mathrm{}\nu _\alpha `$. The interaction term between the states $`|0`$ and $`|1`$ now reads
$$\stackrel{~}{W}_{01}=\underset{j}{}\frac{\mathrm{}\mathrm{\Omega }_{01}}{2}[|1_j0|\mathrm{e}^{\mathrm{i}(k_{01}z_j\psi )}\underset{\alpha }{}\mathrm{e}^{\mathrm{i}\stackrel{~}{k}_j^\alpha P_\alpha }+\mathrm{H}.\mathrm{c}]$$
(38)
where
$$\stackrel{~}{k}_j^\alpha =\frac{\omega _{0,j}}{z_j}\frac{1}{2m\nu _\alpha ^2}S_j^\alpha $$
(39)
Thus, the excitation between two states, where the mechanical potential in one is shifted with respect to the other, corresponds to an effective recoil, here described by the effective Lamb-Dicke parameter $`\stackrel{~}{\eta }_j^\alpha =\stackrel{~}{k}_j^\alpha \sqrt{\mathrm{}m\nu _\alpha /2}`$. Finally, the optical pumping between the states $`|1`$ and $`|2`$ is given by
$$\stackrel{~}{W}_{12}=\underset{j}{}\frac{\mathrm{}\mathrm{\Omega }_{12}}{2}[|2_j1|\mathrm{e}^{\mathrm{i}\chi _j}\underset{\alpha }{}\mathrm{e}^{\mathrm{i}(\eta _{j\alpha }^{\mathrm{eff}}a_\alpha ^{}+\eta _{j\alpha }^{\mathrm{eff}}a_\alpha )}+\mathrm{H}.\mathrm{c}.]$$
(40)
where $`\chi _j`$ is a constant phase, that depends on the position according to
$$\chi _j=kz_j\psi \mathrm{}k\underset{\alpha }{}\stackrel{~}{k}_j^\alpha S_j^\alpha /2$$
(41)
and $`\eta _{j\alpha }^{\mathrm{eff}}`$ is an effective Lamb-Dicke parameter defined in Eq. (20). The dynamics of the individual modes are effectively decoupled from the others in the Lamb-Dicke regime, which holds when the condition $`|\eta _{j\alpha }^{\mathrm{eff}}|\sqrt{n^\alpha }1`$ is fulfilled.
The cooling rate and steady state occupation are evaluated for each mode by using the procedure outlined in Marzoli94 . The rates $`A_\pm ^\alpha `$, which determine the cooling rate and the steady state mean occupation according to Eqs. (7) and (8), are given by
$$A_\pm ^\alpha =2\mathrm{R}\mathrm{e}\{S_\alpha (\nu _\alpha )+D_\alpha \}$$
(42)
where $`S_\alpha (\nu _\alpha )`$ is the fluctuation spectrum of the dipole force $`F_{j,\alpha }`$,
$$S_\alpha (\pm \nu _\alpha )=\frac{1}{M\nu _\alpha }\underset{j}{}_0^{\mathrm{}}dt\mathrm{e}^{\pm \mathrm{i}\nu _\alpha t}\mathrm{Tr}\{F_{j,\alpha }(t)F_{j,\alpha }(0)\rho _{\mathrm{St}}\}$$
(43)
and $`D_\alpha `$ is the diffusion coefficient due to spontaneous emission,
$$D_\alpha =\varphi \underset{j}{}\left(|\eta _{20}^{}{}_{\alpha }{}^{j}|^2\mathrm{\Gamma }_{20}+|\eta _{12}^{}{}_{\alpha }{}^{j}|^2\mathrm{\Gamma }_{21}\right)2|\rho _{\mathrm{St}}|2$$
(44)
where $`\varphi =๐u๐ฉ(u)u^2`$ and here $`\varphi =2/5`$. Here, $`\rho _{\mathrm{St}}`$ is the stationary solution of Eq. (35) at zero order in the Lamb-Dicke parameter, and the dipole force $`F_{j,\alpha }`$ is defined as
$$F_{j,\alpha }=\mathrm{i}\eta _{01}^{}{}_{\alpha }{}^{j}\frac{\mathrm{\Omega }_{01}}{2}|10|+\mathrm{i}\eta _{12}^{}{}_{\alpha }{}^{j}\frac{\mathrm{\Omega }_{12}}{2}|21|+\mathrm{H}.\mathrm{c}.$$
(45)
The fluctuation spectrum and the diffusion are found by evaluating numerically the steady state density matrix. The two-time correlation function in (43) is found by applying the quantum regression theorem according to the master equation (35) at zero order in the Lamb-Dicke parameter.
|
warning/0506/astro-ph0506364.html
|
ar5iv
|
text
|
# Neutron Diffusion and Nucleosynthesis in an Inhomogeneous Big Bang Model
## I Introduction
Big Bang Nucleosynthesis ( BBN ) is the primary mechanism of the creation of the lightest isotope species Smith et al. (1993); Steigman (2003). At temperatures of the universe $`T`$ 100 GK baryonic matter consisted mostly of free neutrons and protons in thermal equilibrium with each other via weak interconversion reactions. Weak freeze-out occurs when the temperature falls to $`T`$ 13 GK and the interconversion reactions fall out of equilibrium. Between $`T`$ 13 GK and $`T`$ 0.9 GK only neutron decay changes the neutron and proton abundances. Then nuclear reactions become significant, forming heavier and heavier nuclei. Nearly all free neutrons at the time of nucleosynthesis are incorporated into <sup>4</sup>He nuclei because of the large binding energy of that nuclei. The amount of free neutrons at that time depends on the neutron lifetime $`\tau _\mathrm{n}`$. BBN is also the only source of deuterium production, and a significant source of <sup>7</sup>Li production.
The nuclear reaction rates depend on the baryon energy density $`\rho _b`$, equivalently the baryon to photon ratio $`\eta `$. The abundance results of <sup>4</sup>He, deuterium and <sup>7</sup>Li can then be compared with measurements to put observational constraints on the value of $`\eta `$. BBN constraints on $`\eta `$ can be compared with constraints derived from Cosmic Microwave Background measurements Lee and et al. (2001); Jaffe and et al. (2001); Netterfield and et al. \[ BOOMERanG Collaboration \] (2002); Halverson and et al. (2002); Melchiorri (2002); Bennett and et al. (2003) Acoustic oscillations in the CMB angular power spectrum are fitted with spherical harmonic functions that depend on several cosmological parameters, including the density factor $`\mathrm{\Omega }_bh^2`$. The most recent CMB measurements set $`\mathrm{\Omega }_Bh^2=0.0224\pm 0.0009`$ Bennett and et al. (2003), corresponding to $`\eta =(5.96.4)\times 10^{10}`$.
The Standard Big Bang Nucleosynthesis ( SBBN ) model is the simplest BBN model. In SBBN all constituents are homogeneously and isotropically distributed. Parameters that define the SBBN model are $`\eta `$, $`\tau _\mathrm{n}`$, and the number of neutrino species $`N_\nu `$. But a variety of models alternative to SBBN can be fashioned by adding in other parameters. The ability for BBN to constrain the value of $`\eta `$ depends on the reliability of isotope measurements. Isotope observational constraints on $`\eta `$ have frequently appeared not to be in concordance with each other when applied to the SBBN model. The possibility of alternative BBN models resolving discrepencies has then been considered Skillman et al. (1994); Hata et al. (1995); Copi et al. (1995); Kernan and Sarker (1996); Hata et al. (1997); Kajino and Orito (1998); Kainulainen et al. (1999); Esposito et al. (2000); Pagel (2000); Jedamzik and Rehm (2001); Kurki-Suonio and Sihvola (2001); Olive and Skillman (2001); Esposito et al. (2001); Kajino (2002); Nollett and Lopez (2002); Ichiki et al. (2002); Bratt et al. (2002). A good understanding of alternative models should then be maintained.
Figure 1 shows graphs for the mass fraction $`X_{{}_{}{}^{4}\mathrm{He}}`$ of <sup>4</sup>He and the abundance ratios $`Y(\mathrm{d})/Y(\mathrm{p})`$ and $`Y(^7\mathrm{Li})/Y(\mathrm{p})`$ of deuterium and <sup>7</sup>Li, all as functions of $`\eta `$. These graphs correspond to an SBBN model. The SBBN code used for Figure 1 has been used by this author in previous articles Lara (1998).
<sup>4</sup>He is measured in low metallicity extragalactic HII regions. There is disagreement over how to extrapolate data points to zero metallicity. In some studies extrapolations have led to a higher mass fraction value of around 0.244 Izotov et al. (1994, 1997); Izotov and Thuan (1998), while in other studies the value is a lower 0.234 Olive and Steigman (1995); Olive et al. (1997); Peimbert et al. (2000). The most recent measurements of $`X_{{}_{}{}^{4}\mathrm{He}}`$ have been a lower $`0.239\pm 0.002`$ Luridiana et al. (2003) and a higher $`0.242\pm 0.002`$ Izotov and Thuan (2004). But the extent of systematic errors in these results is controversial. Olive et al Olive et al. (2000) have used a compromise value $`0.238\pm 0.005`$ combining both high and low measurements due the uncertainty in systematic error. Recently Olive and Skillman Olive and Skillman (2004) try to quantify uncertainties due to systematic error, reporting a large range 0.232 $`X_{{}_{}{}^{4}\mathrm{He}}`$ 0.258. This range should eventually go down as the quantification of systematic errors improves. Figure 1 shows the $`2\sigma `$ ranges by Luridinia et al Luridiana et al. (2003) and Izotov and Thuan Izotov and Thuan (2004) (IT04) combined, corresponding to a range of $`\eta =(2.26.1)\times 10^{10}`$
The deuterium measurement shown in Figure 1 is the weighted mean of five Quasi-Stellar Objects ( QSO )โs done by Kirkman et al Kirkman et al. (2003). This abundance ratio $`Y(\mathrm{d})/Y(\mathrm{p})=2.78_{0.38}^{+0.44}\times 10^5`$ is in good agreement with many previous measurements Tytler et al. (1996); Burles and Tytler (1998a, b); Steigman (2001); OโMeara et al. (2001). But Rugers and Hogan Rugers and Hogan (1996) measured $`Y(\mathrm{d})/Y(\mathrm{p})`$ an order of magnitude greater, at $`(1.9\pm 0.4)\times 10^4`$. The abundance ratio by Kirkman et al corresponds to $`\eta =(5.66.7)\times 10^{10}`$, which is in good agreement with the CMB results.
Ryan et al Ryan et al. (2000) measure <sup>7</sup>Li by looking at a group of very metal-poor stars and accounting for various systematic errors to derive a value $`Y(^7\mathrm{Li})/Y(\mathrm{p})=1.23_{0.32}^{+0.68}\times 10^{10}`$. This measurement has a smaller magnitude and value of $`\sigma `$ than preceding measurements Pinsonneault et al. (1999, 2002). The largest uncertainty in the calculation of this abundance range is the uncertainty in determining the effective temperature $`T_{\mathrm{eff}}`$ of the stars. Melendez & Ramirez Melendez and Ramirez (2004) make new calculations of $`T_{\mathrm{eff}}`$ and get higher temperatures than Ryan et al for lower metallicity stars. Melendez & Ramirez then derive a larger value $`Y(^7\mathrm{Li})/Y(\mathrm{p})=2.34_{0.96}^{+1.64}\times 10^{10}`$, also with a larger $`2\sigma `$ error. Figure 1 shows the measurements by both Ryan et al and Melendez & Ramirez. Ryan et alโs measurement corresponds to $`\eta =(1.64.2)\times 10^{10}`$ while Melendez & Ramirezโs measurement can correspond to two ranges, $`\eta =(1.12.0)\times 10^{10}`$ and $`\eta =(3.36.0)\times 10^{10}`$.
This measurement of <sup>4</sup>He by IT04 is in concordance with the deuterium measurement of Kirkman et al only at its $`2\sigma `$ range, for a narrow range $`\eta =(5.66.1)\times 10^{10}`$. The <sup>7</sup>Li constraints by Melendez & Ramirez is in concordance with the deuterium measurement also only at its $`2\sigma `$ range, while the <sup>7</sup>Li constraints by Ryan et al have no region of concordance at all. A depletion factor from stellar evolution could improve concordance between the deuterium constraints and the <sup>7</sup>Li constraints by Melendez & Ramirez. A factor of 2.8 would resolve the discrepency in the case of Ryan et alโs constraints. But models for <sup>7</sup>Li depletion in stars and measurements of a depletion factor remain controversial.
This article focuses on the particular alternative model of Big Bang Nucleosynthesis with an Inhomogeneous baryon distribution ( IBBN ). The IBBN code used in this article is an original code written by this author Lara (2001a), hereafter known as the Texas IBBN code. Upon publication of this article the code will be made publically available at the authorโs website Lara . The Texas IBBN code can serve as a consistency check against other IBBN codes, and against SBBN codes as well when run in its small distance scale limit.
Section II is a summary of the history of IBBN research, emphasizing developments that are significant to the way the Texas IBBN code is constructed. Section III lists the specific details of the IBBN model used for this article. Section IV shows the final abundance results of the IBBN code for a range of distance scale $`r_i`$ and baryon to photon ratio $`\eta `$. This section discusses how the time of neutron diffusion relative to weak freeze-out and nucleosynthesis significantly affects the final isotope abundances the code produces. The description of this relation in this article is a useful guide for how baryonic matter flows and is processed in an IBBN model. In Section V the IBBN model will be compared with the most recent constraints on <sup>4</sup>He, deuterium and <sup>7</sup>Li. For certain IBBN parameter values the acceptable range of $`\eta `$ from <sup>4</sup>He and deuterium constraints is widened. The IBBN model also permits a large range of <sup>7</sup>Li depletion factor that is of particular interest.
## II Development of the IBBN Code
Various theories of baryogenesis lead to inhomogeneous distributions of free neutrons and protons by the time of nucleosynthesis. The distributions can be modelled with many different symmetries. Baryon inhomogeneities may arise from a first order quark hadron phase transition Witten (1984); Kurki-Suonio (1988); Ignatius and Schwarz (2001). Transport of baryon number between quark gluon phase and hadronic phase is inefficient, leading to concentration of baryon number in the last remaining regions regions of quark gluon plasma Malaney and Mathews (1993). The magnitude of the bubble surface tension determines if the quark gluon plasma regions form in centrallly condensed spherical bubbles Thomas et al. (1994); Kainulainen et al. (1999) or cylindrical filaments Orito et al. (1997). A cosmic string moving through matter during the quark hadron phase transition can also leave wakes of matter that remain in the quark gluon plasma phase longer than in the regions outside the wakes, forming sheets of planar symmetric inhomogeneity Layek et al. (2001, 2003). Baryon inhomogeneity may also form during the earlier electroweak phase transition Fuller et al. (1994); Heckler (1995); Kainulainen et al. (1999). A first order phase transition would proceed by bubble nucleation. Particles in the plasma interact with the bubble walls in a CP violating matter, leading to a baryon asymmetry forming along the walls Rubakov and Shaposhnikov (1996); Cline et al. (1998). The baryon density is in the form of high density shells, spherical or cylindrical Orito et al. (1997); Kainulainen et al. (1999). Baryon inhomogeneities can also arise from phase transitions involving inflation-generated isocurvature fluctuations Dolgov and Silk (1993), or kaon condensation phase Nelson (1990).
The earliest articles on inhomogeneous codes Zeldovich (1975); Epstein and Petrosian (1975); Barrow and Morgan (1983) treated regions of different density as separate SBBN models. They would run a model with a high value of $`\rho _b`$, then a model with a low value, and then average the mass fractions from each model together, weighting each on how large each density region was. Applegate, Hogan and Scherrer Applegate et al. (1987) considered the possibility of nucleons diffusing from high density regions to low density regions. Neutrons diffuse by scattering off of electrons and protons. Protons scatter off of neutrons and Coulomb scatter off of electrons, but the mean free path of protons is about $`10^6`$ times smaller than that for neutrons because of the Coulomb scattering. Diffusion of other isotopes is negligible compared to neutron scattering because the isotopes are much more massive.
In early IBBN codes that featured neutron diffusion Applegate et al. (1987); Alcock et al. (1987); Kajino and Boyd (1990) the diffusion part is run first, at early times and high temperatures. Then nucleosynthesis within the regions is allowed to run. In their IBBN code Kurkio-Suonio et al Kurki-Suonio et al. (1988) (KMCRW88) made the significant innovation of having neutron diffusion occur both before and during nucleosynthesis. This code was for planar symmetric baryon inhomogeneity, and was split in a uniform grid of 20 zones. Kurki-Suonio and Matzner Kurki-Suonio and Matzner (1989) (KM89) and Kurki-Suonio et al Kurki-Suonio et al. (1990) (KMOS90) looked at cylindrical and spherical models, using uniform grids as well. But for larger ratios between high and low densities, or lower volume fractions of high density region, the number of zones needed for the code to run accurately increased considerably. The codes used by Kurki-Suonio and Matzner Kurki-Suonio and Matzner (1990) (KM90) and Mathews et al Mathews et al. (1990, 1996) instead use nonuniform grids, with a greater number of narrower zones around the boundary between high and low density regions, where they are needed. Mathews et al Mathews et al. (1990) halve the width of a zone the closer the zone is to the boundary. KM90 use a stretching function to make a grid of 64 zones that get very narrow around the boundary.
## III The IBBN Model
In an IBBN model the universe is represented as a lattice of baryon inhomogeneous regions. An IBBN code models one region in that lattice. The inhomogeneity can have planar, cylindrical or spherical symmetry. The Texas IBBN code has been used to model condensed spheres Lara (2001b) and cylinders ( a high density core ) and spherical and cylindrical Lara (2004) shells ( a high density outer layer ). The parameters that define an IBBN model are the baryon to photon ratio $`\eta `$, the distance scale $`r_i`$, the density contrast $`R_\rho `$, and the volume fraction $`f_v`$. The distance scale is the initial size of the model at a chosen time. In this article that time is the starting time of a run, when the temperature $`T=`$ 100 GK. The density contrast is the initial ratio of high baryon density to low density. The volume fraction is the fraction of the model occupied by the high density region. $`f_v`$ is parametrized such that it corresponds to a specific radius.
A cylindrical shell model will be used in this article. This symmetry has been used by Orito et al Orito et al. (1997) and Lara 2004 Lara (2004). The isotope abundance results are represented as contour maps in a parameter space defined by $`\eta `$ and $`r_i`$. The values of the remaining parameters are taken from Orito et al Orito et al. (1997).
$`R_\rho `$ $`=`$ $`10^6`$
$`1\sqrt{1f_v}`$ $`=`$ $`0.075`$
The contour lines of abundance values to be discussed in Sections IV and V are most greatly exaggerated in a cylindrical shell model with the parameter values from Orito et al, meaning that observational constraints will be satisfied for the highest possible values of $`\mathrm{\Omega }_Bh^2`$ ( $`\eta `$ ). The parametrization of $`f_v`$ means that the thickness of the high density outer shell equals 0.075 the radius of the whole model. For the neutron lifetime the most recent world average $`\tau _\mathrm{n}=`$ 885.7 seconds Eidelman and et al \[ Particle Data Group \] (2004) is used.
The model is divided into a core and 63 cylindrical shells. These zones need to be thin at the boundary radius $`r_b`$ between high and low density to accurately model neutron diffusion. The Texas IBBN code uses the stretching function from KM90 Kurki-Suonio and Matzner (1990) to set the radii of the shells.
$`\xi (r)`$ $`=`$ $`\xi (r_b)+{\displaystyle \frac{1}{C_1}}\left(1{\displaystyle \frac{1}{C_3}}\right)\sqrt{{\displaystyle \frac{C_2}{C_3}}}\mathrm{arctan}\left[(rr_b)\sqrt{{\displaystyle \frac{C_3}{C_2}}}\right]+{\displaystyle \frac{rr_b}{C_1C_2}}`$ (1)
$`\xi (r)`$ is the shell number out from the center, with a radius $`r`$ in normalized units that range from 0 to 64. The boundary radius $`r_b`$ = 59.2 as determined by the value of $`f_v`$ Figure 2 shows how $`\xi (r)`$ maps onto $`r`$
Appendix A describes in detail the method the calculations are made for each timestep in the run.
## IV Results
Figures 3-5 are contour maps of the overall mass fraction $`X_{{}_{}{}^{4}\mathrm{He}}`$ and abundance ratios $`Y(\mathrm{d})/Y(\mathrm{p})`$ and $`Y(^7\mathrm{Li})/Y(\mathrm{p})`$ at the end of the Texas IBBN codeโs run, drawn in a parameter space defined by $`\eta `$ and $`r_i`$. In Figure 5 the abundance ratios of both <sup>7</sup>Li and <sup>7</sup>Be are shown combined, as all the <sup>7</sup>Be has decayed to <sup>7</sup>Li by now. Neutron diffusion starts at the boundary between the high density outer region and the low density inner region, and then progresses outwards to the outermost shell and inwards to the core. The time neutron diffusion takes to homogenize neutrons determines the shapes of the contour lines shown in Figures 3-5. The two milestone times in element synthesis are the times of weak freeze-out and nucleosynthesis. The contour lines can be described in terms of whether neutron diffusion occurs before weak freeze-out, between weak freeze-out and nucleosynthesis, or after nucleosynthesis.
### IV.1 Before Weak Freeze-Out
For the smallest distance scales $`r_i`$ neutron diffusion homogenizes neutrons very early in a run. Protons are still coupled with neutrons via the interconversion reactions. In the high density outers shells these interconversion reactions run in the direction of converting protons to neutrons, to keep up with neutron diffusion. The protons converted to neutrons diffuse to the low density inner shells, where the the interconversion reactions run in the opposite direction, converting neutrons to protons. Protons are then homogenized along with neutrons, and the final abundances are the same as the abundances from an SBBN model.
For larger $`r_i`$ neutron diffusion takes longer to affect all shells of the model. At a distance scale of around 1600 cm the time when diffusion ends coincides with weak freeze-out. Protons are not as coupled with neutrons as with smaller distance scales, and so are not completely homogenized by the time when neutrons have been homogenized. A larger proton density makes nucleosynthesis occur earlier in the outer shells. For a given value of $`\eta `$ then the final abundance results are the results from an SBBN model with earlier nucleosynthesis: greater <sup>4</sup>He, lesser deuterium, and greater <sup>7</sup>Li and <sup>7</sup>Be production. That corresponds to the shift in the contour lines of Figures 3-5 to lower $`\eta `$ for distance scales $`r_i`$ from 1600 cm to 25000 cm.
### IV.2 Between Weak Freeze-Out and Nucleosynthesis: <sup>7</sup>Li and <sup>7</sup>Be
In models with $`r_i`$ from $``$ 25000 cm to 3.2 $`\times 10^5`$ cm neutrons are homogenized at a time in between weak freeze-out and nucleosynthesis.
If $`r_i`$ 25000 cm, neutron diffusion becomes significant everywhere right around the time of weak freeze-out. Figures 6-7 show an example of how the various reactions in the code interact with one another. Figures 6-7 correspond to shell number 62, a high density outer shell that is two shells away from the outer edge of the model. The reaction rates are normalized to the average baryon number density $`n_{b0}`$ and the expansion rate of the universe $`\dot{\alpha _R}`$.
In Figure 6 the neutron diffusion rates peak at around $`T=`$ 10.0 GK and remain large up to $`T=`$ 3.0 GK. The diffusion rate from shell 62 out to shell 61 is larger than the rate from shell 63 into shell 62 all through that time. The net effect is outflow of neutrons from shell 62, as it is happening in all high density shells at this time. The peak temperature $`T=`$ 10.0 GK is just after the temperature $`T`$ 13 GK of weak freeze-out. Figure 6 shows the rates for the reactions that convert neutrons to protons ( n $``$ p ) and the rates that convert protons to neutrons ( p $``$ n ) in short dashed lines. These rates are the same as they would be in the SBBN model. So no proton redistribution via these reactions is possible, and the proton number density in shell 62 remains high.
Figure 7 shows the nuclear reaction rate n + p $``$ d + $`\gamma `$ in short dashed lines. This reaction falls out of Nuclear Statistical Equilibrium ( NSE ) at $`T=`$ 0.9 GK, starting off the chain of nucleosynthesis. Because the proton number density in the outer shells is high the nuclear reactions go at faster rates than they would in the SBBN model. Nucleosynthesis then occurs slightly earlier in the outer shells, depleting neutrons there. This deficit of neutrons leads to back diffusion. Figure 7 shows the rates of diffusion from shell 61 into shell 62, and from shell 62 out to shell 61. The net effect is now a concentration of neutrons in the high density shells. Nearly all nucleosynthesis is concentrated in the outer shells.
<sup>7</sup>Li is created primarily by the nuclear reactions t + <sup>4</sup>He $`^7`$Li + $`\gamma `$ and n + <sup>7</sup>Be $``$ p + <sup>7</sup>Li, and destroyed primarily by the reactions p + <sup>7</sup>Li $``$ 2( <sup>4</sup>He ) and d + <sup>7</sup>Li $``$ n + 2( <sup>4</sup>He ). The depletion reaction p + <sup>7</sup>Li $``$ 2( <sup>4</sup>He ) dominates over other reactions involving <sup>7</sup>Li. <sup>7</sup>Be is created primarily by <sup>3</sup>He + <sup>4</sup>He $`^7`$Be + $`\gamma `$ and destroyed primarily by n + <sup>7</sup>Be $``$ p + <sup>7</sup>Li. In contrast to <sup>7</sup>Li the creation reaction of <sup>7</sup>Be dominates over the destruction reaction, and greater <sup>4</sup>He production in the high density shells magnifies the dominance even further. Figure 8 shows the number densities of <sup>7</sup>Li and <sup>7</sup>Be as functions of radius. The number density of <sup>7</sup>Be is considerably larger in the high density outer shells than in the rest of the model. Due to this greater <sup>7</sup>Be production the contour lines in Figure 5 have a larger shift to lower $`\eta `$ than the contour lines in Figures 3-4.
### IV.3 Between Weak Freeze-Out and Nucleosynthesis: <sup>4</sup>He and deuterium
In models with $`r_i`$ from $``$ 25000 cm to $`10^5`$ cm the proton number density is unchanged from the time of weak freeze-out to nucleosynthesis, except for a slight increase due to neutron decay. For this range of $`r_i`$ the contour lines in Figures 3-5 lie along nearly constant values of $`\eta `$.
In models with $`r_i=10^5`$ cm the amount of time needed for back diffusion to affect all shells is the same as the duration time of nucleosynthesis. For larger distance scales the shells furthest from the boundary are not as well coupled by back diffusion to the boundary shells. Nucleosynthesis becomes concentrated in the shells immediately around the boundary. This concentration leads to an overall drop in <sup>4</sup>He production. For $`r_i`$ from $`10^5`$ cm to 3.2 $`\times 10^5`$ cm the contour lines in Figures 3-5 shift to higher $`\eta `$. Figure 9 shows the final number density of <sup>4</sup>He as a function of radius for $`r_i3.2\times 10^5`$ cm, with <sup>4</sup>He very concentrated around the boundary.
For models $`r_i>3.2\times 10^5`$ cm diffusion cannot homogenize neutrons before nucleosynthesis. A larger neutron number density remains in the outermost high density shells, and a lower density in the low density core and innermost shells. The larger neutron number density leads to greater <sup>4</sup>He production in the outermost shells. Figure 9 shows the final number density of <sup>4</sup>He for $`r_i=2.0\times 10^6`$ cm. There is greater <sup>4</sup>He production around the boundary and the outermost shells and a trough of lower production in between. The overall <sup>4</sup>He mass fraction increases again, For $`r_i>3.2\times 10^5`$ cm the contour lines in Figure 3 and Figure 5 shift to lower $`\eta `$.
Decreased <sup>4</sup>He production tends to be accompanied by increased deuterium production. Figure 10 shows the final number density of deuterium for $`r_i3.2\times 10^5`$ cm and $`r_i=2.0\times 10^6`$ cm. In the radii corresponding to the trough of <sup>4</sup>He production in Figure 9 Figure 10 has a peak in deuterium production. The contour lines in Figure 4 shift to higher $`\eta `$ for $`r_i`$ from $`10^5`$ cm to 3.2 $`\times 10^5`$ cm, just as in Figure 3 and Figure 5. But for $`r_i`$ from $`3.2\times 10^5`$ cm to 2.0 $`\times 10^6`$ cm the deuterium contour lines still shift to higher $`\eta `$ because of the increased deuterium production shown in Figure 10.
### IV.4 After Nucleosynthesis
At $`r_i2.0\times 10^6`$ cm neutron diffusion peaks at the same time as nucleosynthesis. For models with larger $`r_i`$ neutron diffusion becomes less significant. More neutrons initially in the high density outer region remain there, increasing <sup>4</sup>He production. The trough in Figure 9 disappears and so deuterium production decreases. In Figure 4 the deuterium contour lines shift to lower $`\eta `$ to coincide with the contour shifts in Figure 3 and Figure 5. The largest models behave as two separate SBBN models; a high density SBBN model with considerable <sup>4</sup>He and <sup>7</sup>Li$`+^7`$Be production and minimal deuterium production, and a low density SBBN model with minimal <sup>4</sup>He and <sup>7</sup>Li$`+^7`$Be production and substantial deuterium production. Final results are the average results from the two models.
### IV.5 Generalization
The contour maps shown in Figures 3-5 are for a specific IBBN model. If the model geometry is changed or if the values of the other parameters, the density contrast $`R_\rho `$ and the volume fraction $`f_v`$, are changed the shifts in the contour lines become more or less exaggerated. But the basic shapes of the contour lines persist. For all geometries and values of $`R_\rho `$ and $`f_v`$ there will be a range of distance scale where neutron homogenization occurs in the interim between weak freeze-out and nucleosynthesis, leading to the shift to lower $`\eta `$ as shown in this articleโs model for $`r_i`$ 25000 cm. There will also be a range of $`r_i`$ where neutron diffusion coincides with nucleosynthesis. A trough of lower <sup>4</sup>He production between the boundary and the high density shells furthest from the boundary develops in this range, like the trough shown in Figure 9. IBBN models will then have a distance scale where the contour lines of <sup>4</sup>He and deuterium diverge. For a talk at the Sixth ResCEU International Symposium Lara (2004) this author looked at models with the geometries of condensed cylinders, condensed spheres, and spherical shells as well as cylindrical shells. The values of $`R_\rho `$ and $`f_v`$ used by Orito et al Orito et al. (1997) were used in those runs. The contour maps in all the models showed the same features as seen in Figures 3-5.
## V Observational Constraints
Figure 11 shows the observational constraints from Figure 1 applied to the contour maps of Figures 3-5. The maximum $`X_{{}_{}{}^{4}\mathrm{He}}`$ 0.246 constraint from IT04 Izotov and Thuan (2004) and the <sup>7</sup>Li constraints from Ryan et al Ryan et al. (2000) are shown in Figure 11.
Regions of concordance between the IT04 <sup>4</sup>He maximum constraint and the deuterium constraints are shown in yellow. A concordance region exists for distance scales $`r_i`$ 5000 cm and $`\eta =(5.66.1)\times 10^{10}`$. These limits on $`\eta `$ are the same limits as seen in SBBN models. The maximum limit of $`r_i`$ is set by the shift to lower $`\eta `$ as neutron diffusion occurs closer to weak freeze-out. The $`X_{{}_{}{}^{4}\mathrm{He}}=`$ 0.246 contour have a greater shift than the contour lines for the deuterium constraints, because of increased <sup>4</sup>He production in the outer region.
Another region of concordance appears for $`r_i=(1.36.0)\times 10^5`$ cm, when the contour lines shift to higher $`\eta `$ due to the concentration of nucleosynthesis along the boundary. The upper cutoff of $`r_i`$ is determined by the condition when a trough as shown in Figure 9 exists in the <sup>4</sup>He abundance distribution. Greater <sup>4</sup>He production in the outermost shells cause the $`X_{{}_{}{}^{4}\mathrm{He}}=`$ 0.246 contour to shift to lower $`\eta `$ while greater deuterium production in the trough cause the deuterium contour lines to remain shifted to higher $`\eta `$. The acceptable range of $`\eta `$ is $`(4.312.0)\times 10^{10}`$, larger than in the SBBN case.
The <sup>7</sup>Li constraints from Ryan et al Ryan et al. (2000) are shown in darkest green in Figure 11. The contour lines for <sup>7</sup>Li tend to shift in the same direction with the contour lines of <sup>4</sup>He and deuterium. So the <sup>7</sup>Li constraints do not have a region of concordance with the <sup>4</sup>He and deuterium constraints for this IBBN model, and the lack of a region of concordance persists for other geometries and parameter values. Figure 11 also shows the region of the <sup>7</sup>Li constraints with a depletion factor of 2.8. That depletion factor would bring the <sup>7</sup>Li constraints in concordance with the other isotopes for distances scales $`r_i`$ 5000 cm. For the region of concordance corresponding to $`r_i=(1.36.0)\times 10^5`$ cm a larger depletion factor of 5.9 is needed. The greater production of <sup>7</sup>Be shown in Figure 8 leads to the larger shift to lower $`\eta `$ in the <sup>7</sup>Li contour lines compared to the <sup>4</sup>He and deuterium contour lines, and the larger depletion factor. Figure 11 shows the region of the <sup>7</sup>Li constraints with the depletion factor of 5.9. The benefit of IBBN models then is to allow for a larger range of <sup>7</sup>Li depletion factor than permitted by the SBBN model.
Figure 11 is similar to Figure 2 from the proceedings article Lara 2004 Lara (2004). Differences between the figures include use of the newer $`X_{{}_{}{}^{4}\mathrm{He}}`$ 0.246 constraint Izotov and Thuan (2004) in place of the $`X_{{}_{}{}^{4}\mathrm{He}}`$ 0.248 constraint Olive et al. (2000). The method of calculating the diffusion coefficients Jedamzik and Rehm (2001) is also newer than the method Kurki-Suonio et al. (1992) used in Lara 2004. The neutron lifetime $`\tau _\mathrm{n}=`$ 885.7 seconds Eidelman and et al \[ Particle Data Group \] (2004) was also updated for this article.
Figure 12 shows the <sup>7</sup>Li constraints from Melendez & Ramirez Melendez and Ramirez (2004). The outermost edge of these $`2\sigma `$ constraints has concordance with most of the concordance region between <sup>4</sup>He and deuterium for $`r_i`$ 5000 cm. With a small depletion factor of 1.35 these <sup>7</sup>Li constraints cover the whole region of concordance. A larger depletion factor of 2.8 is needed to cover the region of concordance corresponding to $`r_i=(1.36.0)\times 10^5`$ cm. The regions of concordance between <sup>4</sup>He and deuteurium are controversial because of considerable disagreement regarding <sup>4</sup>He constraints. Nonetheless both Figure 11 and Figure 12 show that Inhomogeneous Big Bang Nucleosynthesis allows for a larger range of acceptable <sup>7</sup>Li depletion factor to bring deuterium and <sup>7</sup>Li in concordance with each other, due to the greater shift in <sup>7</sup>Li contour lines to lower $`\eta `$ for distance scales $`r_i`$ from $``$ 1600 cm to $`10^5`$ cm.
## VI Conclusions
The Texas IBBN code is an original code written such that the weak and nuclear reactions of element synthesis are coupled with neutron diffusion. The time of neutron diffusion relative to the times of weak freeze-out and nucleosynthesis have a significant influence on the final production amounts of <sup>4</sup>He, deuterium, and <sup>7</sup>Li. Because diffusion is coupled to the reaction network the code correctly accounts for neutron back diffusion, wherein neutrons flow back into regions with higher proton density due to earlier nucleosynthesis in those regions. Back diffusion has an influence over the results especially when the time of neutron diffusion is close to the time of nucleosynthesis. Of most interest in the results is the larger range of depletion factor for <sup>7</sup>Li that the IBBN model permits over the SBBN model.
In models where diffusion homogenizes the neutron distribution before weak freeze-out protons are coupled with the neutrons via the weak interconversion reactions. Protons are then redistributed. Proton redistribution is less effective in models with the time of diffusion closer to the time of weak freeze-out, leaving a higher proton density in the outer shells. Increasing proton density leads to earlier nucleosynthesis in the outer shells. Neutrons then back diffuse into the outer shells, concentrating nucleosynthesis there. Nucleosynthesis in the high density shells produces decreasing amounts of deuterium and increasing amounts of <sup>4</sup>He and especially <sup>7</sup>Be. The increased production of <sup>7</sup>Be is significant in the determination of the depletion factor of <sup>7</sup>Li.
For models with the time of diffusion close to the time of nucleosynthesis neutron back diffusion becomes less effective. Nucleosynthesis is concentrated in the volume fimmediately around the boundary. This concentration leads to decreasing <sup>4</sup>He and <sup>7</sup>Li+<sup>7</sup>Be production, and increasing deuterium production. In models where the time of neutron diffusion coincides with nucleosynthesis neutrons are not homogenized during nucleosynthesis. An increasing neutron number density remains in the outermost shells as well as a decreasing number density in the innermost shells. <sup>4</sup>He, <sup>7</sup>Li and <sup>7</sup>Be production jumps in the high density outermost shells, and overall production of these isotopes increases. But between the boundary and the outermost shells are shells with a trough of low <sup>4</sup>He production. Deuterium is produced in large amounts in that trough. The deuterium contour lines in Figure 4 diverge from the contour lines in Figures 3 and 4.
For models where neutron diffusion peaks at the same time as nucleosynthesis the trough in <sup>4</sup>He production has disappeared. Deuterium production decreases and the deuterium contour lines in Figure 4 are in line with the lines in Figures 3 and 4. The divergence in the directions of contour lines is significant in setting constraints on $`\eta `$ and $`r_i`$ in the IBBN model.
Application of observational constraints to this IBBN model found slivers of concordance between the most recent deuterium constraints Kirkman et al. (2003) and <sup>4</sup>He constraints by IT04 Izotov and Thuan (2004). Concordance occurs for $`\eta =(5.66.1)\times 10^{10}`$ and $`r_i`$ 5000 cm, and for $`\eta =(4.312.3)\times 10^{10}`$ and $`r_i=(1.36.0)\times 10^5`$ cm. The point of divergence between the <sup>4</sup>He and deuterium contour lines sets the maximum limit of acceptable $`\eta `$. The reliability of <sup>4</sup>He constraints remains controversial Olive and Skillman (2004).
Contour lines between <sup>4</sup>He, deuterium, and <sup>7</sup>Li run roughly parallel to each other. The region Figure 11 marked by the <sup>7</sup>Li constraints by Ryan et al Ryan et al. (2000) then does not have an overlap with the slivers of concordance of <sup>4</sup>He and deuterium. A depletion factor of 2.8 would bring concordance in both the cases of SBBN and the first region of concordance. But because of the larger shift of the <sup>7</sup>Li contour lines to lower $`\eta `$ a larger depletion factor of 5.9 is needed to bring the <sup>7</sup>Li constraints in agreement with the second region of concordance. Recent <sup>7</sup>Li constraints by Melendez & Ramirez Melendez and Ramirez (2004) have weak concordance with <sup>4</sup>He and deuterium constraints in the SBBN case. But an IBBN model still allows for a larger range of depletion factor, up to 2.8, to have <sup>7</sup>Li be in concordance with <sup>4</sup>He and deuterium.
The IBBN abundance results for <sup>7</sup>Li will be compared with new measurements of the <sup>7</sup>Li primordial abundance derived from the ratio $`(^7\mathrm{Li}/^6\mathrm{Li})`$ measured in the InterStellar Medium Kawanomoto et al. (2003); Kawanomoto and et al. \[ SUBARU/HDS Collaboration \] (2005). A new neutron lifetime $`\tau _\mathrm{n}=878.5\pm 0.7\pm `$ 0.3 seconds has recently been measured Serebrov and et al (2005). Constraints on $`\eta `$ in an SBBN model have been reassessed with the new lifetime Mathews et al. (2005), and the constraints on $`\eta `$ and $`r_i`$ in Figures 11 and 12 will also be reassessed with the new lifetime in an upcoming article Lara and et al. (2005). Additionally, this article will be followed up by articles applying an original solution of the neutrino heating effect Lara (2001a) to both SBBN and IBBN models.
## VII Acknowledgements
This work was partially funded by National Science Foundation grants PHY 9800725, PHY 0102204, and PHY 035482. This author thanks the Center for Relativity at the University of Texas at Austin for the opportunity to work on this research, and Professor Richard Matzner in particular for his help in preparing this article. This author also thanks Professor Toshitaka Kajino of the National Astronomical Observatory of Japan for his advice on what to focus in this article.
## Appendix A Trace of the Texas IBBN Code
The stretching function that sets the radii of the zones in the cylindrical shell model
$`\xi (r)`$ $`=`$ $`\xi (r_b)+{\displaystyle \frac{1}{C_1}}\left(1{\displaystyle \frac{1}{C_3}}\right)\sqrt{{\displaystyle \frac{C_2}{C_3}}}\mathrm{arctan}\left[(rr_b)\sqrt{{\displaystyle \frac{C_3}{C_2}}}\right]+{\displaystyle \frac{rr_b}{C_1C_2}}`$ (2)
was used by KM90 Kurki-Suonio and Matzner (1990). The radius $`r`$ is normalized to equal 64 for the full radius of the model. Eq. 2 maps radii $`r`$ of the zone boundaries to unit values of $`\xi `$. Zones near $`r_b`$ have a width around $`C_1`$. Zones far from $`r_b`$ have widths determined by $`C_3`$ and a rate of zone-size change controlled by $`C_2`$. $`r_b`$ always corresponds to a unit value of $`\xi `$. For the model in this article there are 20 zones covering the high density outer shell and 44 covering the low density inner region. The baryon number densities $`n_{bhigh}`$ in the outer shellโs zones and $`n_{blow}`$ in the inner regionโs zones are set
$`n_{blow}`$ $`=`$ $`{\displaystyle \frac{n_{b0}}{f_vR_\rho +(1f_v)}}`$ (3)
$`n_{bhigh}`$ $`=`$ $`R_\rho n_{blow}`$ (4)
such that the number density averages out to $`n_{b0}`$ over the whole model.
At any given timestep the code solves the differential equation Mathews et al. (1990)
$`{\displaystyle \frac{n(i,s)}{t}}`$ $`=`$ $`n_b(s){\displaystyle \underset{j,k,l}{}}N_i\left({\displaystyle \frac{Y^{N_i}(i,s)Y^{N_j}(j,s)}{N_i!N_j!}}[ij]+{\displaystyle \frac{Y^{N_k}(k,s)Y^{N_l}(l,s)}{N_k!N_l!}}[kl]\right)`$ (5)
$`3\dot{\alpha _R}n(i,s)+{\displaystyle \frac{1}{r^p}}{\displaystyle \frac{}{r}}\left(r^pD_n{\displaystyle \frac{\xi }{r}}{\displaystyle \frac{n(i,s)}{\xi }}\right)`$
for the number density $`n(i,s)`$ of isotope $`i`$ in zone $`s`$. The first two terms correspond to the weak and nuclear reactions that destroy ( $`[ij]`$ ) or create ( $`[kl]`$ ) isotope $`i`$ within zone $`s`$. $`n_b(s)`$ is the total baryon number density in zone $`s`$ and $`Y(i,s)`$ is the abundance $`Y(i,s)`$ of isotope $`i`$ in zone $`s`$.
$`Y(i,s)`$ $`=`$ $`{\displaystyle \frac{n(i,s)}{n_b(s)}}`$
The $`3\dot{\alpha _R}n(i,s)`$ term corresponds for the expansion of the universe, where $`R`$ is the expansion coefficient of the universe and $`\alpha _R=\mathrm{ln}R`$. This term can be eliminated by transforming to comoving coordinates. From here on $`r`$ will be in comoving coordinates. The last term corresponds to diffusion of isotope $`i`$ into and out of zone $`s`$. The factor $`p`$ depends on the geometry of the model. $`p=`$ 0 for planar symmetry, 1 for cylindrical symmetry and 2 for spherical symmetry. Currently only neutrons can diffuse in the Texas IBBN code. The neutron diffusion coefficient $`D_\mathrm{n}`$ is calculated from the coefficients $`D_{\mathrm{ne}}`$ for neutron electron scattering and $`D_{\mathrm{np}}`$ for neutron proton scattering.
$`{\displaystyle \frac{1}{D_\mathrm{n}}}`$ $`=`$ $`{\displaystyle \frac{1}{D_{\mathrm{ne}}}}+{\displaystyle \frac{1}{D_{\mathrm{np}}}}`$ (6)
Banerjee and Chitre Banerjee and Chitre (1991) derived a master equation for the diffusion coefficient between two particles scattering off of each other, based on the first order Chapman-Enskog approximation de Groot et al. (1980). Kurki-Suonio et al (KAGMBCS92) Kurki-Suonio et al. (1992), and Jedamzik and Rehm Jedamzik and Rehm (2001) derive the same equation for the diffusion coefficient $`D_{\mathrm{ne}}`$ for neutron-electron scattering.
$`D_{\mathrm{ne}}`$ $`=`$ $`{\displaystyle \frac{3}{8}}\sqrt{{\displaystyle \frac{\pi }{2}}}{\displaystyle \frac{c}{n_\mathrm{e}\sigma _{\mathrm{ne}}}}{\displaystyle \frac{K_2(z)}{\sqrt{z}K_{5/2}(z)}}(1{\displaystyle \frac{n_\mathrm{n}}{n_t}})`$ (7)
$`K_2(z)`$ and $`K_{5/2}(z)`$ are modified Bessel functions of order 2 and 5/2, $`\sigma _{\mathrm{ne}}`$ is the transport cross section of the scattering and $`z=m_\mathrm{e}/kT`$. $`n_\mathrm{n}/n_t`$ is the neutron fraction of the total number of ALL particles. This fraction is of the order $`10^{10}`$ and so can be ignored. For neutron-proton scattering Jedamzik and Rehm Jedamzik and Rehm (2001) derive an updated expression for the diffusion coefficient $`D_{\mathrm{np}}`$
$`D_{\mathrm{np}}`$ $`=`$ $`{\displaystyle \frac{3}{8\sqrt{\pi }}}{\displaystyle \frac{c}{a_s^2}}{\displaystyle \frac{1}{n_\mathrm{p}}}\sqrt{{\displaystyle \frac{k_BT}{m_Nc^2}}}{\displaystyle \frac{1}{I(a_1,b_1)+\frac{3a_t^2}{a_s^2}I(a_2,b_2)}}`$ (8)
$`I(a,b)`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _0^{\mathrm{}}}๐x{\displaystyle \frac{x^2e^x}{ax+\left(1\frac{bx}{2}\right)^2}}`$
$`a_1`$ $`=`$ $`a_s^2{\displaystyle \frac{m_Nc^2}{\mathrm{}^2c^2}}k_bT`$
$`b_1`$ $`=`$ $`r_sa_s{\displaystyle \frac{m_Nc^2}{\mathrm{}^2c^2}}k_bT`$
$`a_2`$ $`=`$ $`a_t^2{\displaystyle \frac{m_Nc^2}{\mathrm{}^2c^2}}k_bT`$
$`b_2`$ $`=`$ $`r_ta_t{\displaystyle \frac{m_Nc^2}{\mathrm{}^2c^2}}k_bT`$
$`m_N`$ is the nucleon mass. The parameters $`a_s=`$ -23.71 fm, $`r_s=`$ 2.73 fm, $`a_t=`$ 5.432 fm, and $`r_t=`$ 1.749 fm come from singlet and triplet scattering.
The Texas IBBN code progresses a timestep $`\mathrm{\Delta }t_m`$ for each step $`m`$. Eq. 5 is evolved using a implicit second order Runge-Kutta method Kawano (1992). To use this method, Eq. 5 has to be linearized. The weak-nuclear reaction terms can be linearized in a manner similar to the linearization of abundances $`Y`$ used by Wagoner Wagoner (1969) and this author Lara (1998).
$`{\displaystyle \frac{n_{mA}(i,s)n_{m1}(i,s)}{\mathrm{\Delta }t_{m1}}}`$ $`=`$ $`{\displaystyle \underset{j,k,l}{}}{\displaystyle \frac{N_i[ij]}{N_i!N_j!(N_i+N_j)}}[N_iY_m^{N_i1}(i,s)Y_m^{N_j}(j,s)n_{mA}(i,s)+`$ (9)
$`N_jY_m^{N_i}(i,s)Y_m^{N_j1}(j,s)n_{mA}(j,s)]`$
$`+{\displaystyle \frac{N_i[kl]}{N_k!N_l!(N_k+N_l)}}[N_kY_m^{N_k1}(k,s)Y_m^{N_l}(l,s)n_{mA}(k,s)+`$
$`N_lY_m^{N_k}(k,s)Y_m^{N_l1}(l,s)n_{mA}(l,s)]+\mathrm{}`$
$`\mathrm{\Delta }t_{m1}`$ is the time difference between step $`m1`$ and step $`m`$. For the diffusion term the zones are defined on a grid whose points $`r(s)`$ correspond to the outer radii of zones $`s`$. Number densities $`n(i,s)`$ are considered the number densities at the midpoint radius between the inner and outer radii of zone $`s`$. The points $`r(s)`$ correspond to points in $`\xi (s)`$ space a distance of one unit between each other. The first space derivative in the diffusion term in Eq. 5 can be discretized as:
$`{\displaystyle \frac{n}{t}}`$ $`=`$ $`\mathrm{}{\displaystyle \frac{1}{r^p}}{\displaystyle \frac{}{r}}\left(r^pD{\displaystyle \frac{\xi }{r}}{\displaystyle \frac{n}{\xi }}\right)`$
$`{\displaystyle \frac{n}{t}}`$ $`=`$ $`\mathrm{}{\displaystyle \frac{1}{r^p}}{\displaystyle \frac{}{r}}\left[\left(r^pD{\displaystyle \frac{\xi }{r}}\right)_{s1/2}{\displaystyle \frac{n[r(s)]n[r(s1)]}{1}}\right]`$
Note that the coefficient $`[r^pD(\xi )/(r)]`$ depends on $`r`$. The $`(1/r^p)(/r)`$ can be rewritten as a partial derivative of $`r^{p+1}`$. One can then write the discretization of the second space derivative as:
$`{\displaystyle \frac{n}{t}}`$ $`=`$ $`\mathrm{}(p+1){\displaystyle \frac{}{(r^{p+1})}}\left[\left(r^pD{\displaystyle \frac{\xi }{r}}\right)_{s1/2}{\displaystyle \frac{n[r(s)]n[r(s1)]}{1}}\right]`$
$`{\displaystyle \frac{n}{t}}`$ $`=`$ $`\mathrm{}(p+1)({\displaystyle \frac{\left(r^pD\frac{\xi }{r}\right)_s\{n[r(s+\frac{1}{2})]n[r(s\frac{1}{2})]\}}{r^{p+1}(s)r^{p+1}(s1)}}`$
$`{\displaystyle \frac{\left(r^pD\frac{\xi }{r}\right)_{s1}\{n[r(s\frac{1}{2})]n[r(s\frac{3}{2})]\}}{r^{p+1}(s)r^{p+1}(s1)}})`$
For an isotope $`i`$ the densities $`n[r(s+1/2)]`$, $`n[r(s1/2)]`$ and $`n[r(s3/2)]`$ are defined as $`n(i,s+1)`$, $`n(i,s)`$ and $`n(i,s1)`$ respectively. The coefficient $`[r^2D(\xi )/(r)]_s`$ is calculated at $`r(s)`$. One can apply this discretization to an implicit version of the diffision part of Eq. 5.
$`{\displaystyle \frac{n_{mA}(i,s)n_{m1}(i,s)}{\mathrm{\Delta }t_{m1}}}`$ $`=`$ $`\mathrm{}{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_s}{r^{p+1}(s)r^{p+1}(s1)}}n_{mA}(i,s+1)`$ (10)
$`+(p+1){\displaystyle \frac{\left(r^pD\frac{\xi }{r}\right)_s+\left(r^pD\frac{\xi }{r}\right)_{s1}}{r^{p+1}(s)r^{p+1}(s1)}}n_{mA}(i,s)`$
$`{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_{s1}}{r^{p+1}(s)r^{p+1}(s1)}}n_{mA}(i,s1)`$
Any baryons that flow out beyond the distance scale are assumed to be replenished by baryons flowing in from other sets of shells, and $`r(0)=`$ 0 is the center of the shells. The code uses reflective boundary conditions at the endpoints of the grid.
$`{\displaystyle \frac{n_{mA}(i,1)n_{m1}(i,1)}{\mathrm{\Delta }t_{m1}}}`$ $`=`$ $`\mathrm{}{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_1}{r^{p+1}(1)r^{p+1}(0)}}n_{mA}(i,2)`$
$`+{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_1}{r^{p+1}(1)r^{p+1}(0)}}n_{mA}(i,1)`$
$`{\displaystyle \frac{n_{mA}(i,64)n_{m1}(i,64)}{\mathrm{\Delta }t_{m1}}}`$ $`=`$ $`\mathrm{}+{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_{64}}{r^{p+1}(64)r^{p+1}(63)}}n_{mA}(i,64)`$
$`{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_{63}}{r^{p+1}(64)r^{p+1}(63)}}n_{mA}(i,63)`$
where $`r(64)=`$ distance scale $`r_i`$. The above equations can be applied to the diffusion of any isotope $`i`$, but only neutrons ( $`i=`$ 1 ) diffuse for the results of this article.
Eq. 9 and Eq. 10 combined together can be rewritten as a matrix equation for a new number density value $`n_{mA}(i,s)`$. The matrix consists of a 68 $`\times `$ 68 matrix for each of the 64 zones, built from the terms in Eq. 9. From Eq. 10 come terms that couple $`n(1,s)`$ with $`n(1,s+1)`$ and $`n(1,s1)`$ due to neutron diffusion. $`n_{mA}(i,s)`$ is then used in the following equation
$`\stackrel{~}{n}_m(i,s)`$ $`=`$ $`n_m(i,s)+\left[{\displaystyle \frac{n_{mA}(i,s)n_{m1}(i,s)}{\mathrm{\Delta }t_{m1}}}\right]\mathrm{\Delta }t_m`$
to calculate an interim value $`\stackrel{~}{n}_m(i,s)`$ of the number densities. This is the first step of the Runge-Kutta method, with $`\stackrel{~}{n}_m(i,s)`$ the first estimate of the values of $`n_m(i,s)`$ at time $`t_m+\mathrm{\Delta }t_m`$. Using $`\stackrel{~}{Y}_m(i,s)=\stackrel{~}{n}_m(i,s)/\stackrel{~}{n}_b(s)`$ the code solves a second matrix equation
$`{\displaystyle \frac{n_{mB}(i,s)n_m(i,s)}{\mathrm{\Delta }t_m}}`$ $`=`$ $`{\displaystyle \underset{j,k,l}{}}{\displaystyle \frac{N_i[ij]}{N_i!N_j!(N_i+N_j)}}[N_i\stackrel{~}{Y}_m^{N_i1}(i,s)\stackrel{~}{Y}_m^{N_j}(j,s)n_{mB}(i,s)+`$
$`N_j\stackrel{~}{Y}_m^{N_i}(i,s)\stackrel{~}{Y}_m^{N_j1}(j,s)n_{mB}(j,s)]`$
$`+{\displaystyle \frac{N_i[kl]}{N_k!N_l!(N_k+N_l)}}[N_k\stackrel{~}{Y}_m^{N_k1}(k,s)\stackrel{~}{Y}_m^{N_l}(l,s)n_{mB}(k,s)+`$
$`N_l\stackrel{~}{Y}_m^{N_k}(k,s)\stackrel{~}{Y}_m^{N_l1}(l,s)n_{mB}(l,s)]`$
$`{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_s}{r^{p+1}(s)r^{p+1}(s1)}}n_{mB}(i,s+1)`$
$`+(p+1){\displaystyle \frac{\left(r^pD\frac{\xi }{r}\right)_s+\left(r^pD\frac{\xi }{r}\right)_{s1}}{r^{p+1}(s)r^{p+1}(s1)}}n_{mB}(i,s)`$
$`{\displaystyle \frac{(p+1)\left(r^pD\frac{\xi }{r}\right)_{s1}}{r^{p+1}(s)r^{p+1}(s1)}}n_{mB}(i,s1)`$
for new number density values $`n_{mB}(i,s)`$. $`\mathrm{\Delta }t_m`$ is the time difference between step $`m`$ and step $`m+1`$. Final new values for $`n_{m+1}(i,s)`$ at timestep $`m+1`$ can then be calculated.
$`n_{m+1}(i,s)`$ $`=`$ $`n_m(i,s)+{\displaystyle \frac{1}{2}}\left[{\displaystyle \frac{n_{mA}(i,s)n_{m1}(i,s)}{\mathrm{\Delta }t_{m1}}}+{\displaystyle \frac{n_{mB}(i,s)n_m(i,s)}{\mathrm{\Delta }t_m}}\right]\mathrm{\Delta }t_m`$ (11)
This is the second step ( โ$`B`$โ ) of the Runge-Kutta method. At the same time as with $`n_m(i,s)`$ the Texas IBBN evolves $`\mathrm{ln}R`$ and the electromagnetic plasma energy density $`\rho _{\mathrm{e}+\gamma }`$
$`{\displaystyle \frac{d(\mathrm{ln}R)}{dt}}`$ $`=`$ $`\sqrt{{\displaystyle \frac{8}{3}}\pi G(\rho _\gamma +\rho _\mathrm{e}+\rho _\nu )}`$ (12)
$`{\displaystyle \frac{d\rho _{\mathrm{e}+\gamma }}{dt}}`$ $`=`$ $`4{\displaystyle \frac{\dot{R}}{R}}\rho _\gamma 3{\displaystyle \frac{\dot{R}}{R}}(p_\mathrm{e}+\rho _\mathrm{e})`$ (13)
also by the Runge-Kutta method.
After both Runge-Kutta steps have been done the code determines the new baryon number density $`n_b(s)`$ of each zone using
$`n_b(s)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{68}{}}}A_in(i,s)`$
where $`A_i`$ is the atomic weight of isotope $`i`$. From $`n_b(s)`$ and $`n(i,s)`$ the code can calculate $`Y(i,s)`$. At any given time the abundance $`Y_{av}(i)`$ and mass fraction $`X_i`$ of isotope $`i`$ in the entire model can be calculated from $`Y(i,s)`$ using
$`Y_{av}(i)`$ $`=`$ $`{\displaystyle \frac{_{s=1}^{64}n(i,s)[r^{p+1}(s)r^{p+1}(s1)]}{_{s=1}^{64}n_b(s)[r^{p+1}(s)r^{p+1}(s1)]}}`$
$`X_i`$ $`=`$ $`A_iY_{av}(i)`$
These overall abundances and mass fractions can be shown as contour maps of the IBBN codeโs parameters, and compared to observational constraints.
|
warning/0506/math0506469.html
|
ar5iv
|
text
|
# 1 Introduction
### 1 Introduction
This paper studies the asymptotics of certain โ$`\beta `$โPlancherelโ measures on subsets of partitions. Let $`Y_n=\{\lambda \lambda n\}`$ denote the partitions of $`n`$, and let $`f^\lambda `$ denote the number of standard Young tableaux of shape $`\lambda =(\lambda _1,\lambda _2,\mathrm{})`$. For general references regarding partitions, Young diagrams and Young tableaux โ see . Let $`0<\beta `$. Baik and Rains consider the following โ$`\beta `$Plancherelโ measure $`M_n^\beta `$ on $`Y_n`$:
$`M_n^\beta (\lambda ):={\displaystyle \frac{(f^\lambda )^\beta }{_{\mu n}(f^\mu )^\beta }}.`$ (1)
Indeed, $`M_n^2`$ is the so called Plancherel measure on $`Y_n`$. We generalize to subsets $`\mathrm{\Gamma }_nY_n`$, considering the subset of the partitions (i.e. diagrams) in the $`(k,\mathrm{})`$ hook: Let $`k,\mathrm{}0`$ be integers and let $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ denote the following subset of $`Y_n`$:
$$H(k,\mathrm{};n)=\{\lambda n\lambda _{k+1}\mathrm{}\}.$$
These subsets arise in the representation theory of Lie groups, algebras and superalgebras, see for example . The measures $`\rho _n^{(\beta ;k,\mathrm{})}`$ below are the $`(k,\mathrm{})`$โhook restrictions of the above measures $`M_n^\beta `$.
###### Definition 1.1
Let $`\lambda H(k,\mathrm{};n)`$ and $`\beta >0`$, then
$$\rho ^{(\beta ;k,\mathrm{})}(\lambda )=\rho _n^{(\beta ;k,\mathrm{})}(\lambda ):=\frac{(f^\lambda )^\beta }{_{\mu H(k,\mathrm{};n)}(f^\mu )^\beta }.$$
#### 1.1 Expected shape
Given $`\mathrm{\Gamma }_nY_n,n=1,2,\mathrm{}`$ and the probability measures $`\rho =\{\rho _n\}_{n=1}^{\mathrm{}}`$ on the $`\mathrm{\Gamma }_n`$โs, one studies the asymptotics of the expected value (i.e. average length) of the first row $`\lambda _1`$, denoted $`\lambda _{1,E}`$, and similarly for the second row $`\lambda _{2,E}`$, etc. Similarly for the columns. Explicitly, when $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$, expected values are given by the following definition.
###### Definition 1.2
If $`\lambda n`$, we write $`\lambda =(\lambda _{1,n},\lambda _{2,n},\mathrm{})`$. Also, $`\lambda ^{}`$ is the conjugate partition of $`\lambda `$. Let $`1pk`$ and $`1q\mathrm{}`$. The expected value of the $`p`$-th row is $`E(\lambda _p)=\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)`$, where
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)=\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }\text{and}\lambda _E^{(\beta ;k,\mathrm{})}(n)=(\lambda _{1,E}^{(\beta ;k,\mathrm{})}(n),\lambda _{2,E}^{(\beta ;k,\mathrm{})}(n),\mathrm{})$$
Similarly for the expected $`q`$-th column
$$\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)=\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{q,n}^{}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }\text{and}\lambda _E^{{}_{}{}^{}(\beta ;k,\mathrm{})}(n)=(\lambda _{1,E}^{{}_{}{}^{}(\beta ;k,\mathrm{})}(n),\lambda _{2,E}^{{}_{}{}^{}(\beta ;k,\mathrm{})}(n),\mathrm{}).$$
Of course, one can replace $`H(k,\mathrm{};n)`$ in the above definition by other subsets $`\mathrm{\Gamma }_nY_n`$.
The case $`\mathrm{\Gamma }_n=Y_n`$ and $`\beta =2`$ (Plancherel) has a long history. Let
$$w(n)=\frac{_{\lambda n}\lambda _{1,n}(f^\lambda )^2}{_{\lambda n}(f^\lambda )^2}=\frac{_{\lambda n}\lambda _{1,n}(f^\lambda )^2}{n!}$$
be the expected value of the first row โ for the Plancherel measure $`M_n^2`$. Hammersley showed that the limit $`c=lim_n\mathrm{}w(n)/\sqrt{n}`$ exists. Vershik and Kerov proved that $`c=2`$ (independently, Logan and Shepp proved that $`c2`$). Vershik and Kerov โ and Logan and Shepp also determined the asymptotics of the expected shape $`\lambda `$ in this case.
Recently, in a major breakthrough paper, Baik, Deift and Johansson determined the distribution function of the asymptotics of the first row, relating it to the Tracy-Widom distribution , see also and . The distribution function for the second row is given, by these same authors, in . The distribution functions for the general rows are given in , and ; see also for the analogue results for colored permutations. The above results also establish deep connections with the theory of random matrices . For detailed reviews of these results โ see and .
The main objective of the present paper is to compute the asymptotics of the above expected values (i.e. shapes) $`\lambda _E^{(\beta ;k,\mathrm{})}(n)`$, as well as the corresponding distribution functions. The first term approximation is relatively simple, as we show that for each $`1pk`$ and $`1q\mathrm{}`$
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n),\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}},$$
see Theorem 4.1. Second term approximations of $`\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)`$ are introduced and studied in Sections 5, 6, and they have different values for different rows and for different columns. These second-term-approximations are given as ratios of certain Selberg-type integrals, see Theorems 5.3 and 6.3 below.
#### 1.2 Distribution functions
In Section 7 we introduce and study the asymptotics of the Distribution functions $`\lambda _p^{(\beta ;k,\mathrm{})}(n,z),\lambda _q^{(\beta ;k,\mathrm{})}(n,z)`$ for the lengths of the rows and the columns in $`H(k,\mathrm{};n)`$ โ with respect to the above measures. We are able to calculate, asymptotically, a firstโterm approximation of these functions, but only a conjecture is given, about the second term approximations. That firstโterm approximation is
$$\lambda _p^{(\beta ;k,\mathrm{})}(n,z),\lambda _q^{(\beta ;k,\mathrm{})}(n,z)\frac{n}{k+\mathrm{}}r_{(k,\mathrm{}),\beta }(z),$$
where $`r_{(k,\mathrm{}),\beta }(z)`$ is given by Equation (14), see Theorem 7.3.
#### 1.3 Comparison with maximal shape
Given a subset of partitions $`\mathrm{\Gamma }_nY_n`$, one looks for $`\lambda \mathrm{\Gamma }_n`$ with maximal degree $`f^\lambda `$. Call it maximal shape (with respect to $`\mathrm{\Gamma }_n`$) and denote it by $`\lambda _{max}`$. In Sections 89 and 10 the expected shapes for $`\beta =1,2`$ are compared with the maximal shape. When $`\mathrm{\Gamma }_n=Y_n`$, the asymptotics of $`\lambda _{max}`$ was calculated by Vershik and Kerov , , and by Logan and Shepp . In particular, they proved that asymptotically, the expected shape for $`\beta =2`$ and the maximal shape are the same, and that shape is given by the two axes and by the curve
$`y=1+\left({\displaystyle \frac{2}{\pi }}\right)[x\sqrt{1x^2}arccosx].`$ (2)
A comparison with the case $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ is intriguing.
When $`\mathrm{\Gamma }_n`$ is the $`k`$โstrip $`\mathrm{\Gamma }_n=H(k,0;n)`$, the asymptotics of $`\lambda `$ with maximal $`f^\lambda `$ was calculated in , and is given by the curve
$`y=\left({\displaystyle \frac{2}{\pi }}\right)[xarcsinx+\sqrt{1x^2}].`$ (3)
When $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$, the maximal $`\lambda `$ was given in . These results are reviewed in Section 8.2. Consider for example the โstripโ case $`\mathrm{}=0`$, and denote the maximal $`\lambda `$ by $`\lambda _{max}^{(k,0)}`$. Comparing it with $`\lambda _E^{(2;k,0)}`$, the โPlanchereleโ expected $`\lambda `$ in $`H(k,0;n)`$, we show that these asymptotic shapes are not equal โ even in their first raw. Nevertheless, numerically $`\lambda _{max}^{(k,0)}`$ and $`\lambda _E^{(2;k,0)}`$ are remarkably close, at least in the few special cases we check below, see Section 8 .
Also, the asymptotics of $`\lambda _{max}`$ for $`\mathrm{\Gamma }_n=Y_n`$ is not the limit case of $`\lambda _{max}^{(k,0)}`$ as $`k\mathrm{}`$, but the similarity between (2) and (3) is intriguing. It should be interesting to see if the ratios of the Selberg-type integrals, which give the expected shapes and the distribution functions for $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$, are in any way related to the TracyโWidom distributions , which give the (Plancherel and the โinvolutionโ) distribution functions in $`Y_n`$.
#### 1.4 RSK
In the case of $`\rho _n^{(1;k,\mathrm{})}`$ and $`\rho _n^{(2;k,\mathrm{})}`$, the RSK correspondence provides an interesting interpretation of the above asymptotics. The RSK (RobinsonโSchenstedโKnuth) correspondence $`\sigma (P_\lambda ,Q_\lambda )`$ corresponds $`\sigma S_n`$ with a pair of standard Young tableaux of shape $`\lambda `$ . In the Plancherel case $`\beta =2`$ it relates the above expected values of the first row to the statistics of the longest increasing (and decreasing) subsequences in permutation. For example, when $`\sigma (P_\lambda ,Q_\lambda )`$, $`\lambda _1`$ is the length of a longest increasing subsequence in $`\sigma `$, while $`\lambda _1^{}`$ is the length of a longest decreasing subsequence in $`\sigma `$. By C. Greenโs theorem there are similar interpretations for $`\lambda _2,\lambda _3,`$ etc. For a detailed account of the RSK see . Thus the results in etc. can also be stated in terms of longest increasing subsequences in permutations.
It is well known that $`\sigma `$ is an involution ifโf $`\sigma (P_\lambda ,P_\lambda )`$. The analogue Probability theory of longest increasing subsequences in involutions in $`S_n`$ is done in .
Denote by $`S_{k,\mathrm{};n}S_n`$ the subset of the permutations $`\sigma S_n`$ such that under the RSK correspondence $`\sigma (P_\lambda ,Q_\lambda )`$, we have $`\lambda H(k,\mathrm{};n)`$. For example, $`S_{k,0;n}`$ is the subset of those permutations in $`S_n`$ where any descending subsequence has length $`k`$. Thus, $`\lambda _{1,E}^{(1;k,\mathrm{})}`$ is the expected value of the longest increasing subsequence in the involutions in $`S_{k,\mathrm{};n}`$.
### 2 Selberg type integrals
As mentioned above, the main results in this paper involve Selbergโtype integrals, hence we briefly review these type of multiโintegrals. In A. Selberg proved the following formula:
$$_0^1\mathrm{}_0^1(u_1\mathrm{}u_n)^{x1}[(1u_1)\mathrm{}(1u_n)]^{y1}\underset{1i<jn}{}|u_iu_j|^{2z}du_1\mathrm{}du_n=$$
$$=\underset{k=1}{\overset{n}{}}\frac{\mathrm{\Gamma }(1+kz)\mathrm{\Gamma }(x+(k1)z)\mathrm{\Gamma }(y+(k1)z)}{\mathrm{\Gamma }(1+z)\mathrm{\Gamma }(x+y+(n+k2)z)}.$$
Various integral formulas can be deduced from Selberโs integra, see for example for the MacdonaldโMehta integras. For example Mehtaโs integral formula (which was a conjecture for some time)
$$_^ke^{(1/2)({\scriptscriptstyle x_i^2})}\underset{1i<jk}{}|x_ix_j|^{2z}dx_1\mathrm{}dx_k=(\sqrt{2\pi })^k\underset{j=1}{\overset{k}{}}\frac{\mathrm{\Gamma }(1+jz)}{\mathrm{\Gamma }(1+z)}$$
can be deduced from Selbergโs formula, see for details. We call these and related integrals โSelbergโtype integralsโ. A connection between the RSK and these integrals, as well as with random matrices, appears in , . Since the formulas from , are needed later, we record it here, together with certain variations of these asymptotics and integrals that are also needed below. Let
$$\mathrm{\Omega }_k=\{(x_1,\mathrm{},x_k)^kx_1x_2\mathrm{}x_k\text{and}x_1+\mathrm{}+x_k=0\},$$
and more generally,
$$\mathrm{\Omega }_{(k,\mathrm{})}=\{(x_1,\mathrm{},x_k,y_1,\mathrm{},y_{\mathrm{}})x_1\mathrm{}x_k;y_1\mathrm{}y_{\mathrm{}};x_i+y_j=0\}.$$
###### Theorem 2.1
(Theorem 2.10 in ) Let $`\gamma _k=(1/\sqrt{2\pi })^{k1}k^{k^2/2}`$ and $`D_k(x)=_{1i<jk}(x_ix_j)`$, then
$$\underset{\lambda H(k,0;n)}{}(f^\lambda )^\beta \left[\gamma _k\left(\frac{1}{n}\right)^{(k1)(k+2)/4}k^n\right]^\beta (\sqrt{n})^{k1}I(k,0,\beta ),$$
where
$$I(k,0,\beta )=_{\mathrm{\Omega }_k}\left[D_k(x)e^{\frac{k}{2}({\scriptscriptstyle x_i^2})}\right]^\beta d^{(k1)}x.$$
Note that by certain symmetry properties of the above integrand, $`\mathrm{\Omega }_k`$ is transformed in into $`^k`$.
Given $`\lambda =(\lambda _{1,n},\lambda _{2,n},\mathrm{})n`$, write $`\lambda _{p,n}=n/k+c_{p,n}\sqrt{n}`$ and denote $`c_{p,n}=c_{p,n}(\lambda )`$. The same arguments in prove the following theorem.
###### Theorem 2.2
$$\underset{\lambda H(k,0;n)}{}c_{p,n}(\lambda )(f^\lambda )^\beta \left[\gamma _k\left(\frac{1}{n}\right)^{(k1)(k+2)/4}k^n\right]^\beta (\sqrt{n})^{k1}I^{}(k,0,\beta ),$$
where
$$I^{}(k,0,\beta )=_{\mathrm{\Omega }_k}x_p\left[D_k(x)e^{\frac{k}{2}({\scriptscriptstyle x_i^2})}\right]^\beta d^{(k1)}x.$$
The $`(k,\mathrm{})`$โhook analogue of Theorem 2.1 is proved in :
###### Theorem 2.3
(See Theorem 7.18 in ) Let
$$\gamma _{k,\mathrm{}}=(1/\sqrt{2\pi })^{k+\mathrm{}1}(k+\mathrm{})^{(k^2+\mathrm{}^2)/2}(1/2)^k\mathrm{},$$
then
$$\underset{\lambda H(k,\mathrm{};n)}{}(f^\lambda )^\beta \left[\gamma _{k,\mathrm{}}\left(\frac{1}{n}\right)^{(k(k+1)+\mathrm{}(\mathrm{}+1)2)/4}(k+\mathrm{})^n\right]^\beta (\sqrt{n})^{k+\mathrm{}1}I(k,\mathrm{},\beta ),$$
where
$$I(k,\mathrm{},\beta )=_{\mathrm{\Omega }_{k,\mathrm{}}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y).$$
The same arguments also prove
###### Theorem 2.4
Let $`1pk`$, let $`\lambda =(\lambda _{1,n},\lambda _{2,n},\mathrm{})H(k,\mathrm{};n)`$ and define $`c_{p,n}(\lambda )`$
$$via:\lambda _{p,n}=\frac{n}{k+\mathrm{}}+c_{p,n}(\lambda )\sqrt{n}.\text{Similarly for}1q\mathrm{}\text{and}c_{q,n}^{}(\lambda ):=c_{q,n}(\lambda ^{}).$$
Then
$$\underset{\lambda H(k,\mathrm{};n)}{}c_{p,n}(\lambda )(f^\lambda )^\beta $$
$$\left[\gamma _{k,\mathrm{}}\left(\frac{1}{n}\right)^{(k(k+1)+\mathrm{}(\mathrm{}+1)2)/4}(k+\mathrm{})^n\right]^\beta (\sqrt{n})^{k+\mathrm{}1}I^{}(k,\mathrm{},\beta ),$$
where
$$I^{}(k,\mathrm{},\beta )=_{\mathrm{\Omega }_{k,\mathrm{}}}x_p\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y).$$
Similarly for the sum
$$\underset{\lambda H(k,\mathrm{};n)}{}c_{q,n}^{}(\lambda )(f^\lambda )^\beta ,$$
with the corresponding integral
$$I^{{}_{}{}^{}}(k,\mathrm{},\beta )=_{\mathrm{\Omega }_{k,\mathrm{}}}y_q\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y).$$
Generalizations of Theorem 2.4 โ with more general functions of the $`c_{p,n}(\lambda ),c_{q,n}^{}(\lambda )`$ โ are rather obvious, but will not be given here.
The same arguments of Section 7 of , applied to the asymptotics of both numerator and denominator, prove the following theorem โ which is needed later.
###### Theorem 2.5
Let $`z>0`$,
$$H(k,\mathrm{};n,z)=\{\lambda H(k,\mathrm{};n)\lambda _{1,n},\lambda _{1,n}\frac{n}{k+\mathrm{}}+z\sqrt{n}\}$$
and let
$$\mathrm{\Omega }_{(k,\mathrm{}),z}=\{(x_1,\mathrm{},x_k;y_1,\mathrm{},y_{\mathrm{}})\mathrm{\Omega }_{(k,\mathrm{})}x_1,y_1z\}.$$
Then
$$\frac{_{\lambda H(k,\mathrm{};n,z)}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }$$
$$\frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}.$$
We call all the above โSelbergโtype integralsโ, and remark that the above expected values and distribution functions are given below as ratio of such integrals.
### 3 The main results
#### 3.1 The expected values
We study the expected values of the row and of the column lengths in $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ with respect to the measures $`\rho _n^{(\beta ;k,\mathrm{})}`$ introduced in Definition 1.1. The first term asymptotics is given by
###### Theorem 3.1
(See Theorem 4.1) Let $`\beta >0`$ and $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$. For each $`1pk`$,
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}},\text{ namely}\underset{n\mathrm{}}{lim}\left(\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)\right)/\left(\frac{n}{k+\mathrm{}}\right)=1.$$
Similarly, for each $`1q\mathrm{}`$,
$$\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}}.$$
The second term approximations are given as follows. Define $`c_{p,E}^{(\beta ;k,\mathrm{})}(n)`$ and $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ via
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)=\frac{n}{k+\mathrm{}}+c_{p,E}^{(\beta ;k,\mathrm{})}(n)\sqrt{n},\text{and}c_{p,E}^{(\beta ;k,\mathrm{})}=\underset{n\mathrm{}}{lim}c_{p,E}^{(\beta ;k,\mathrm{})}(n).$$
Similarly for the columns. Then
###### Theorem 3.2
(see Theorem 6.3) Let $`1pk`$, then the limit $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ exists, and is given as follows:
$`c_{p,E}^{(\beta ;k,\mathrm{})}={\displaystyle \frac{_{\mathrm{\Omega }_k}x_p\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}(x)}{_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}(x)}}.`$ (4)
Similarly for the columns.
Equation (4) (or (9)) is a consequence of the seemingly more symmetric Equation (11).
In Sections 89 and 10 the expected shapes for $`\beta =1,2`$ are compared with the maximal shape $`\lambda _{max}`$ in few special cases. As mentioned in Section 1.3, this shows a different behavior in the hook case $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ compared with the general case $`\mathrm{\Gamma }_n=Y_n`$.
#### 3.2 The distribution function for the first row
In Section 7 we study the distribution functions of the length on the rows and the columns. We calculate the first term approximations and conjecture the second term approximations. Given $`0<z`$, denote
$$H(k,\mathrm{};n,z)=\{\lambda H(k,\mathrm{};n)\lambda _{1,n},\lambda _{1,n}^{}\frac{n}{k+\mathrm{}}+z\sqrt{n}\}.$$
Let $`1pk,1q\mathrm{}`$. The distribution of the length of the $`p`$โth row as a function of $`z`$ is defined as
$$\lambda _p^{(\beta ;k,\mathrm{})}(n,z)=\frac{_{\lambda H(k,\mathrm{};n,z)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta },$$
and similarly for the columns. Recall $`\mathrm{\Omega }_{(k,\mathrm{})}`$ from Section 2 and denote
$$\mathrm{\Omega }_{(k,\mathrm{}),z}=\{(x_1,\mathrm{},x_k;y_1,\mathrm{},y_{\mathrm{}})\mathrm{\Omega }_{(k,\mathrm{})}x_1,y_1z\}.$$
Then
###### Theorem 3.3
(see Theorem 7.3) Let $`\beta ,z>0`$ and denote
$`r_{(k,\mathrm{}),\beta }(z)={\displaystyle \frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}}.`$ (5)
Then
$$\lambda _p^{(\beta ;k,\mathrm{})}(n,z),\lambda _q^{(\beta ;k,\mathrm{})}(n,z)\frac{n}{k+\mathrm{}}r_{(k,\mathrm{}),\beta }(z).$$
In Section 7.2 we make some conjectures about the second term approximations of $`\lambda _p^{(\beta ;k,\mathrm{})}(n,z)`$ and $`\lambda _q^{(\beta ;k,\mathrm{})}(n,z)`$, both in terms of ratios of Selbergโtype integrals. In the last three sections (Sections 7, 8 and 9), we calculate some special cases and include also some computer calculations.
## Part I Expected shape, the $`\beta `$โPlancherel probability
### 4 First term approximation
Recall the notation $``$: Let $`a_n,b_n`$ be two sequences of, say, real numbers, and assume $`b_n0`$ if $`n`$ is large enough. Then $`a_nb_n`$ if $`lim_n\mathrm{}a_n/b_n=1`$. Extend $``$ to vectors as follows: $`(a_{1,n},\mathrm{},a_{r,n})(b_{1,n},\mathrm{},b_{r,n})`$ ifโf $`a_{i,n}b_{i,n}`$ for $`i=1,\mathrm{}r`$.
Following the techniques and arguments in Section 7 of , (see also ) we show below the following first approximation of the expected shape $`\lambda _E=\lambda _E^{(\beta ;k,\mathrm{})}`$.
###### Theorem 4.1
Let $`\beta >0`$ and $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ and let $`\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)`$ and $`\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)`$ be given by Definition 1.2. For each $`1pk`$,
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}},\text{ namely}\underset{n\mathrm{}}{lim}\left(\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)\right)/\left(\frac{n}{k+\mathrm{}}\right)=1.$$
Similarly, for each $`1q\mathrm{}`$,
$$\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}}.$$
Proof. We sketch the proof for the expected row length in the case $`\mathrm{}=0`$, thus showing that
$$\lambda _{p,E}^{(\beta ;k,0)}(n)\frac{n}{k}.$$
The main point is that since $`\beta >0`$, both sums in the numerator and the denominator of definition 1.2 are dominated by the summands corresponding to the partitions $`\lambda `$, such that $`\lambda _i=\frac{n}{k}+c_i\sqrt{n}`$ and with $`c_i`$โs in a bounded interval. In other words, let $`a>0`$ and denote
$$H_a(k,0;n)=\{\lambda H(k,0;n)\lambda _i=\frac{n}{k}+c_i\sqrt{n},\text{where}|c_i|a,i=1,2,\mathrm{},k\}.$$
Then
$$\underset{n\mathrm{}}{lim}\lambda _{p,E}^{(\beta ;k,0)}(n)=\underset{n\mathrm{}}{lim}\frac{_{\lambda H(k,0;n)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H(k,0;n)}(f^\lambda )^\beta }=\underset{a\mathrm{}}{lim}\left[\underset{n\mathrm{}}{lim}\frac{_{\lambda H_a(k,0;n)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H_a(k,0;n)}(f^\lambda )^\beta }\right].$$
Writing $`\lambda _{p,n}=\frac{n}{k}+c_{p,n}\sqrt{n}`$, the expression in the brackets equals
$$\frac{_{\lambda H_a(k,0;n)}(\frac{n}{k})(f^\lambda )^\beta }{_{\lambda H_a(k,0;n)}(f^\lambda )^\beta }+\frac{_{\lambda H_a(k,0;n)}c_{p,n}\sqrt{n}(f^\lambda )^\beta }{_{\lambda H_a(k,0;n)}(f^\lambda )^\beta }.$$
The first summand equals $`\frac{n}{k}`$ while the absolute value of the second summand is bounded by $`a\sqrt{n}`$ since all $`|c_p(n)|a`$. Since $`\frac{n}{k}+a\sqrt{n}\frac{n}{k}`$, it follows that
$$\lambda _{p,E}^{(\beta ;k,0)}(n)\frac{n}{k},$$
which completes the proof in the case $`\mathrm{}=0`$. q.e.d.
### 5 Second term approximation, the โstripโ case ($`\mathrm{}=0`$)
Because of Theorem 4.1, we look for a more subtle approximation of $`\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)`$, namely, we look for the expected deviation โ of the form $`c\sqrt{n}`$ โ from $`\frac{n}{k+\mathrm{}}`$. This leads us to introduce the asymptotic expected value $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ below. We begin with the โstripโ case $`\mathrm{}=0`$. The general $`(k,\mathrm{})`$โhook case is given in the next section.
###### Definition 5.1
Let $`\mathrm{\Gamma }_n=H(k,0;n)`$ and let $`1pk`$, with $`\lambda _{p,E}^{(\beta ;k,0)}(n)`$ given by Definition 1.2. Define $`c_{p,E}^{(\beta ;k,0)}(n)`$ via the equation
$$\lambda _{p,E}^{(\beta ;k,0)}(n)=\frac{n}{k}+c_{p,E}^{(\beta ;k,0)}(n)\sqrt{n},\text{and}c_{p,E}^{(\beta ;k,0)}=\underset{n\mathrm{}}{lim}c_{p,E}^{(\beta ;k,0)}(n).$$
Thus, when $`n`$ goes to infinity,
$$\lambda _{p,E}^{(\beta ;k,0)}(n)\frac{n}{k}+c_{p,E}^{(\beta ;k,0)}\sqrt{n}.$$
###### Remark 5.2
It is not obvious that the limit $`c_{p,E}^{(\beta ;k,0)}=lim_n\mathrm{}c_{p,E}^{(\beta ;k,0)}(n)`$ exists. However, Theorem 5.3 asserts that in fact, this limit does exist.
Our aim is to calculate $`c_{p,E}^{(\beta ;k,0)}`$, thus calculating โthe second termโ in the approximation of the expected value of the $`p`$-th rowโlength $`\lambda _{p,E}^{(\beta ;k,0)}`$. Definitions 1.2 and 5.1 obviously imply the equation
$`c_{p,E}^{(\beta ;k,0)}(n)=\left({\displaystyle \frac{_{\lambda H(k,0;n)}\lambda _{p,n}\left(f^\lambda \right)^\beta }{_{\lambda H(k,0;n)}\left(f^\lambda \right)^\beta }}{\displaystyle \frac{n}{k}}\right){\displaystyle \frac{1}{\sqrt{n}}}.`$ (6)
If $`k=1`$, Equation (6) implies that for any $`\beta >0`$, $`c_{1,E}^{(\beta ;1,0)}=0`$.
###### Theorem 5.3
Let $`k2`$. The limit $`c_{p,E}^{(\beta ;k,0)}=lim_n\mathrm{}c_{p,E}^{(\beta ;k,0)}(n)`$ exists, and is given by
$`c_{p,E}^{(\beta ;k,0)}={\displaystyle \frac{_{\mathrm{\Omega }_k}x_p\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}x}{_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}x}}.`$ (7)
Proof. In Equation (6) write $`\lambda _{i,n}=\frac{n}{k}+c_{i,n}\sqrt{n}`$, and consider $`\lambda `$โs with $`c_i`$ bounded in some interval: $`|c_i|a`$ for some $`a>0`$. By an argument similar to the proof of Theorem 4.1, it follows that as $`n\mathrm{}`$, $`c_{p,E}^{(\beta ;k,0)}(n)`$ is approximated by the ratio
$`{\displaystyle \frac{_{\lambda H_a(k,0;n)}c_{p,n}\left(f^\lambda \right)^\beta }{_{\lambda H_a(k,0;n)}\left(f^\lambda \right)^\beta }}.`$ (8)
The proof now follows by applying Theorem 2.1 to the denominator, Theorem 2.2 to the numerator, then cancelling equal terms. q.e.d.
###### Remark 5.4
Note that the denominator of (7) is actually a โSelbergโโor a MacdonaldโMehta โ integral, which can be evaluated for any $`\beta `$. For example, let $`\beta =2`$. By comparing (F.2.10) with (F.4.5.2) of , deduce that
$$_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^2d^{(k1)}x=$$
$$=\left(\sqrt{2\pi }\right)^{k1}\left(\frac{1}{\sqrt{2}}\right)^{k^21}\left(\frac{1}{\sqrt{k}}\right)^{k^2}1!2!\mathrm{}(k1)!.$$
Similarly, by comparing (F.2.10) with (F.4.5.1) of , deduce that when $`\beta =1`$,
$$_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]d^{(k1)}x=$$
$$=\left(\frac{1}{\sqrt{k}}\right)^{k(k+1)}\frac{1}{k!}\left(\sqrt{2}\right)^{3k1}\frac{1}{\sqrt{\pi }}\underset{j=1}{\overset{k}{}}\mathrm{\Gamma }\left(1+\frac{1}{2}j\right).$$
For explicit values, recall that $`\mathrm{\Gamma }\left(\frac{3}{2}\right)=\frac{\pi }{2}`$, that $`\mathrm{\Gamma }(1)=1`$, and that $`\mathrm{\Gamma }(z+1)=z\mathrm{\Gamma }(z)`$.
### 6 Second term approximation, the $`(k,\mathrm{})`$โhook case
We turn now to the general $`(k,\mathrm{})`$โhook case.
###### Definition 6.1
Let $`\beta >0`$ and let $`1pk`$ and $`1q\mathrm{}`$, with $`\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)`$ given by Definition 1.2. Define $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ via the equation
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)=\frac{n}{k+\mathrm{}}+c_{p,E}^{(\beta ;k,\mathrm{})}(n)\sqrt{n},\text{and}c_{p,E}^{(\beta ;k,\mathrm{})}=\underset{n\mathrm{}}{lim}c_{p,E}^{(\beta ;k,\mathrm{})}(n).$$
Similarly for the columns:
$$\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)=\frac{n}{k+\mathrm{}}+c_{q,E}^{(\beta ;k,\mathrm{})}(n)\sqrt{n},\text{and}c_{q,E}^{(\beta ;k,\mathrm{})}=\underset{n\mathrm{}}{lim}c_{q,E}^{(\beta ;k,\mathrm{})}(n).$$
The existence of the limits $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ and $`c_{q,E}^{(\beta ;k,\mathrm{})}`$ is asserted by Theorem 6.3 below.
Definitions 1.2 and 6.1 obviously imply
###### Remark 6.2
$$c_{p,E}^{(\beta ;k,\mathrm{})}(n)=\left(\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{p,n}\left(f^\lambda \right)^\beta }{_{\lambda H(k,\mathrm{};n)}\left(f^\lambda \right)^\beta }\frac{n}{k+\mathrm{}}\right)\frac{1}{\sqrt{n}},$$
and
$$c_{q,E}^{(\beta ;k,\mathrm{})}(n)=\left(\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{q,n}^{}\left(f^\lambda \right)^\beta }{_{\lambda H(k,\mathrm{};n)}\left(f^\lambda \right)^\beta }\frac{n}{k+\mathrm{}}\right)\frac{1}{\sqrt{n}}.$$
###### Theorem 6.3
Let $`1pk`$, $`1q\mathrm{}`$, then the limits $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ and $`c_{q,E}^{(\beta ;k,\mathrm{})}`$ exist, and are given as follows.
$`c_{p,E}^{(\beta ;k,\mathrm{})}={\displaystyle \frac{_{\mathrm{\Omega }_k}x_p\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}(x)}{_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^\beta d^{(k1)}(x)}},`$ (9)
and
$`c_{q,E}^{(\beta ;k,\mathrm{})}={\displaystyle \frac{_\mathrm{\Omega }_{\mathrm{}}y_q\left[D_{\mathrm{}}(y_1,\mathrm{},y_{\mathrm{}})e^{\frac{k+\mathrm{}}{2}(y_1^2+\mathrm{}+y_{\mathrm{}}^2)}\right]^\beta d^{(\mathrm{}1)}(y)}{_\mathrm{\Omega }_{\mathrm{}}\left[D_{\mathrm{}}(y_1,\mathrm{},y_{\mathrm{}})e^{\frac{k+\mathrm{}}{2}(y_1^2+\mathrm{}+y_{\mathrm{}}^2)}\right]^\beta d^{(\mathrm{}1)}(y)}}.`$ (10)
Thus, as $`n`$ goes to infinity,
$$\lambda _{p,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}}+c_{p,E}^{(\beta ;k,\mathrm{})}\sqrt{n}\text{and}\lambda _{q,E}^{(\beta ;k,\mathrm{})}(n)\frac{n}{k+\mathrm{}}+c_{q,E}^{(\beta ;k,\mathrm{})}\sqrt{n}.$$
Proof. We prove for $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ โ in two steps.
Step 1. We claim that the limit $`c_{p,E}^{(\beta ;k,\mathrm{})}`$ exists, and is given by
$`c_{p,E}^{(\beta ;k,\mathrm{})}={\displaystyle \frac{_{\mathrm{\Omega }_{(k,\mathrm{})}}x_p\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}}.`$ (11)
The proof of Equation (11) is essentially the same as that of Equation (7). Its starting point is Remark 6.2 (instead of Equation (6)). Here we write, for $`1ik`$ and for $`1j\mathrm{}`$,
$$\lambda _{i,n}=\frac{n}{k+\mathrm{}}+c_{i,n}\sqrt{n}\text{ and}\lambda _{j,n}^{}=\frac{n}{k+\mathrm{}}+c_{j,n}^{}\sqrt{n},$$
and consider $`\lambda `$โs with $`c_i`$ and $`c_j^{}`$ bounded in some interval. Now follow a โhookโโgeneralization of the proof of Theorem 5.3, applying Theorems 2.3 and 2.4, and complete the proof of Equation (11).
Step 2. We now transform (11) into (9). Let $`I_p^{(\beta ;k,\mathrm{})}`$ denote the numerator of (11):
$$I_p^{(\beta ;k,\mathrm{})}=_{\mathrm{\Omega }_{(k,\mathrm{})}}x_p\left[D_k(x)D_{\mathrm{}}(y)e^{((k+\mathrm{})/2)({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y).$$
Setting $`x_i=u`$ we have $`y_j=u`$, and $`I_p^{(\beta ;k,\mathrm{})}=_{\mathrm{}}^{\mathrm{}}K(u)L(u)๐u`$, where
$$K(u)=_{M_k(x,u)}x_p\left[D_k(x)e^{((k+\mathrm{})/2)({\scriptscriptstyle x_i^2})}\right]^\beta d^{(k1)}x$$
and
$$L(u)=_{M_{\mathrm{}}(y,u)}\left[D_{\mathrm{}}(y)e^{((k+\mathrm{})/2)({\scriptscriptstyle y_j^2})}\right]^\beta d^{(k1)}x.$$
Here $`M_k(x,u)=\{(x_1,\mathrm{},x_k)x_1\mathrm{}x_k\text{and}x_i=u\}`$ and similarly, $`M_{\mathrm{}}(y,u)=\{(y_1,\mathrm{},y_{\mathrm{}})y_1\mathrm{}y_{\mathrm{}}\text{and}y_j=u\}`$.
To evaluate $`I_p^{(\beta ;k,\mathrm{})}`$, proceed as follows. In $`K(u)`$ and $`L(u)`$ substitute $`x_i^{}=x_i(u/k)`$, and $`y_j^{}=y_j+(u/\mathrm{})`$. The Jacobians are $`=1`$, $`x_t=x_t^{}+(u/k);D_k(x^{})=D_k(x);D_{\mathrm{}}(y^{})=D_{\mathrm{}}(y);x_i^2=x_i^2+(u^2/k)`$ and $`y_j^2=y_j^2+(u^2/\mathrm{})`$. Replacing $`x_i^{}`$ by $`x_i`$ and $`y_j^{}`$ by $`y_j`$, it follows that
$$I_p^{(\beta ;k,\mathrm{})}=J_1(x)J_3(y)A(u)+J_2(x)J_3(y)B(u).$$
Here
$$A(u)=_{\mathrm{}}^{\mathrm{}}e^{\frac{(k+\mathrm{})^2u^2\beta }{2k\mathrm{}}}๐u,B(u)=_{\mathrm{}}^{\mathrm{}}\frac{u}{k}e^{\frac{(k+\mathrm{})^2u^2\beta }{2k\mathrm{}}}๐u,$$
$$J_1(x)=_{M_k(x,0)}x_p\left[D_k(x)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2})}\right]^\beta d^{(k1)}x,(J_1(x)=1\text{if}k=1),$$
$$J_2(x)=_{M_k(x,0)}\left[D_k(x)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2})}\right]^\beta d^{(k1)}x,$$
and
$$J_3(y)=_{M_{\mathrm{}}(y,0)}\left[D_{\mathrm{}}(x)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle y_j^2})}\right]^\beta d^{(\mathrm{}1)}x,(J_3(y)=1\text{if}\mathrm{}=1),$$
$`(M_r(x,0)=\{(x_1,\mathrm{},x_r)x_1\mathrm{}x_r\text{and}x_1+\mathrm{}+x_r=0\}`$, etc.). Since, trivially, $`B(u)=0`$, deduce that $`I_p^{(\beta ;k,\mathrm{})}=J_1(x)J_3(y)A(u)`$.
Let $`\overline{I}^{(\beta ;k,\mathrm{})}`$ denote the denominator in Equation (11). By exactly the same arguments it follows that $`\overline{I}^{(\beta ;k,\mathrm{})}=J_2(x)J_3(y)A(u)`$. By (11)
$$c_{p,E}^{(\beta ;k,\mathrm{})}=\frac{I_p^{(\beta ;k,\mathrm{})}}{\overline{I}^{(\beta ;k,\mathrm{})}}=\frac{J_1(x)}{J_2(x)},$$
which is the right-hand-side of Equation (9). This completes the proof. q.e.d.
###### Remark 6.4
The denominator-integral in (11) is
$$\overline{I}^{(\beta ;k,\mathrm{})}=_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x_1,\mathrm{},x_k)D_{\mathrm{}}(y_1,\mathrm{},y_{\mathrm{}})e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2+y_1^2+\mathrm{}+y_{\mathrm{}}^2)}\right]^\beta d^{(k+\mathrm{}1)}(x;y),$$
and is calculated explicitly in \[ , sec 7\] (here $`\beta =2z`$):
$$I^{}(k,\mathrm{})=I(k,\mathrm{},2)=\frac{1}{k!\mathrm{}!}\sqrt{2\pi ^{k+\mathrm{}1}}\sqrt{\frac{\beta }{2\pi }}\left(\frac{1}{\beta (k+\mathrm{})}\right)^{\frac{1}{2}[(k(k1)+\mathrm{}(\mathrm{}1))(\beta /2)+k+\mathrm{}]}$$
$$\times \frac{_{i=1}^k\mathrm{\Gamma }(i\beta /2+1)_{j=1}^{\mathrm{}}\mathrm{\Gamma }(j\beta /2+1)}{\mathrm{\Gamma }(\beta /2+1)},$$
where $`\mathrm{\Gamma }`$ is the Gamma function $`(\mathrm{\Gamma }(n+1)=n!)`$.
Theorems 5.3 and 6.3 imply
###### Corollary 6.5
Let $`1pk`$ and $`1q\mathrm{}`$. Then
$$c_{p,E}^{(\beta ;k,\mathrm{})}=\sqrt{\frac{k}{k+\mathrm{}}}c_{p,E}^{(\beta ;k,0)},\text{and similarly}c_{q,E}^{(\beta ;k,\mathrm{})}=\sqrt{\frac{\mathrm{}}{k+\mathrm{}}}c_{q,E}^{(\beta ;0,\mathrm{})}.$$
Proof. Let $`\alpha =\frac{k}{k+\mathrm{}}`$ and in Equation (9) substitute $`x=\sqrt{\alpha }v`$. By routine calculations, this substitution transforms the ration of integrals (9) into the ratio in (7) โ multiplied by the factor $`\sqrt{\alpha }`$, which completes the proof.
### 7 The distribution functions
#### 7.1 First term approximation
###### Definition 7.1
Let $`z>0`$. Denote
$$H(k,\mathrm{};n,z)=\{\lambda H(k,\mathrm{};n)\lambda _{1,n},\lambda _{1,n}^{}\frac{n}{k+\mathrm{}}+z\sqrt{n}\}.$$
Let $`\beta >0`$, $`1pk`$ and $`1q\mathrm{}`$. The distribution of the length of the $`p`$โth row as a function of $`z`$ is defined as
$`\lambda _p^{(\beta ;k,\mathrm{})}(n,z)={\displaystyle \frac{_{\lambda H(k,\mathrm{};n,z)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }}.`$ (12)
Similarly, the distribution of the length of the $`q`$โth column as a function of $`z`$ is defined as
$`\lambda _q^{(\beta ;k,\mathrm{})}(n,z)={\displaystyle \frac{_{\lambda H(k,\mathrm{};n,z)}\lambda _{q,n}^{}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }}.`$ (13)
Let $`\lambda H(k,\mathrm{};n,z)`$ with $`n`$ large, then necessarily $`\lambda _1,\mathrm{},\lambda _k\mathrm{}`$ and $`\lambda _1^{},\mathrm{},\lambda _{\mathrm{}}^{}k`$ (in fact, Lemma 7.2 proves a much stronger property) so that $`\lambda _1,\mathrm{},\lambda _k+\lambda _1^{},\mathrm{},\lambda _{\mathrm{}}^{}=n+k\mathrm{}`$. Write
$$\lambda _p=\frac{n+k\mathrm{}}{k+l}+c_p\sqrt{n},p=1,\mathrm{},k\text{ and}\lambda _q^{}=\frac{n+k\mathrm{}}{k+l}+c_q^{}\sqrt{n},q=1,\mathrm{},\mathrm{},$$
and notice that $`c_p+c_q^{}=0`$.
###### Lemma 7.2
With the above notations (and $`n`$ large),
$$\frac{\mathrm{}+p1}{kp+1}zc_pz,p=1,\mathrm{},k\text{and}\frac{k+q1}{\mathrm{}q+1}zc_q^{}z,q=1,\mathrm{},\mathrm{}.$$
Proof. Clearly, all $`c_p,c_q^{}z`$. Assume for example that
$$c_p<\frac{\mathrm{}+p1}{kp+1}z,\text{hence also}c_k,c_{k1},\mathrm{},c_p<\frac{\mathrm{}+p1}{kp+1}z.\text{Thus}$$
$$0=c_1+\mathrm{}+c_k+c_1^{}+\mathrm{}+c_{\mathrm{}}^{}<(kp+1)\frac{\mathrm{}+p1}{kp+1}z+c_1+\mathrm{}+c_{p1}+c_1^{}+\mathrm{}+c_{\mathrm{}}^{}$$
$$(\mathrm{}+p1)z+\mathrm{}+p1)z=0,$$
a contradiction. Similarly for $`c_q^{}`$. This proves the lemma.
Recall that
$`\mathrm{\Omega }_{(k,\mathrm{})}=\{(x_1,\mathrm{},x_k,y_1,\mathrm{},y_{\mathrm{}})x_1\mathrm{}x_k;y_1\mathrm{}y_{\mathrm{}};x_i+y_j=0\}`$ and $`\mathrm{\Omega }_{(k,\mathrm{}),z}=\{(x_1,\mathrm{},x_k;y_1,\mathrm{},y_{\mathrm{}})\mathrm{\Omega }_{(k,\mathrm{})}x_1,y_1z\}`$. For example, $`\mathrm{\Omega }_{(2,0),z}=\{(x,x)0xz\}`$.
###### Theorem 7.3
Let $`\beta ,z>0`$ and denote
$`r_{(k,\mathrm{}),\beta }(z)={\displaystyle \frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}}.`$ (14)
Let $`1pk`$, $`1q\mathrm{}`$ and let $`n\mathrm{}`$, then
$$\lambda _p^{(\beta ;k,\mathrm{})}(n,z),\lambda _q^{(\beta ;k,\mathrm{})}(n,z)\frac{n}{k+\mathrm{}}r_{(k,\mathrm{}),\beta }(z).$$
In other words,
$$\underset{n\mathrm{}}{lim}\frac{k+\mathrm{}}{n}\lambda _p^{(\beta ;k,\mathrm{})}(n,z)=\underset{n\mathrm{}}{lim}\frac{k+\mathrm{}}{n}\frac{_{\lambda H(k,\mathrm{};n,z)}\lambda _{p,n}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }=$$
$`={\displaystyle \frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}}.`$ (15)
Similarly for
$$\underset{n\mathrm{}}{lim}\frac{k+\mathrm{}}{n}\lambda _q^{(\beta ;k,\mathrm{})}(n,z).$$
Proof is similar to the proof of Theorem 4.1: Let $`\lambda _{p,n}=\frac{n}{k+\mathrm{}}+c_{p,n}\sqrt{n}`$, then
$$\frac{k+\mathrm{}}{n}\lambda _p^{(\beta ;k,\mathrm{})}(n,z)=\frac{_{\lambda H(k,\mathrm{};n,z)}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }+\frac{k+\mathrm{}}{\sqrt{n}}\left(\frac{_{\lambda H(k,\mathrm{};n,z)}c_{p,n}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }\right).$$
By Theorem 2.5, the first summand approaches $`r_{(k,\mathrm{}),\beta }(z)`$ as $`n`$ goes to infinity, and by Lemma 7.2, $`|c_{p,n}|bz`$ for an appropriate constant $`b>0`$, therefore the second summand obviously goes to zero as $`n`$ goes to infinity. q.e.d.
Theorem 7.3 is a firstโterm approximation of the distribution function.
#### 7.2 Conjectures about the second term approximation
Let
$`s_{(k,\mathrm{}),\beta }(n,z)={\displaystyle \frac{_{\lambda H(k,\mathrm{};n,z)}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }}.`$ (16)
As in the proof of Theorem 7.3, $`lim_n\mathrm{}s_{(k\mathrm{}),\beta }(n,z)=r_{(k,\mathrm{}),\beta }(z)`$. Numerical evidence suggest the following (vague) conjecture.
###### Conjecture 7.4
For all $`k,\mathrm{}0`$ and $`\beta ,z>0`$, as $`n`$ goes to infinity the expression
$`{\displaystyle \frac{\sqrt{n}}{k+\mathrm{}}}\left[s_{(k,\mathrm{}),\beta }(n,z)r_{(k,\mathrm{}),\beta }(z)\right]=`$ (17)
$$=\frac{\sqrt{n}}{k+\mathrm{}}\left[\frac{_{\lambda H(k,\mathrm{};n,z)}(f^\lambda )^\beta }{_{\lambda H(k,\mathrm{};n)}(f^\lambda )^\beta }\frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x;y)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x)}\right]$$
oscillates in some symmetric bounded interval centered at zero. We denote that interval as $`(L(k,\mathrm{},\beta ,z),L(k,\mathrm{},\beta ,z))`$, where (we conjecture that) $`L(k,\mathrm{},\beta ,z)`$ and $`L(k,\mathrm{},\beta ,z)`$ are the respective infimum and supremum of the values in Equation (17).
###### Definition 7.5
Let $`\beta ,z>0`$, $`1pk`$ and $`1q\mathrm{}`$. Then $`\lambda _p^{(\beta ;k,\mathrm{})}(n,z)`$ is given by Equation (12). Similarly for the columns $`\lambda _q^{(\beta ;k,\mathrm{})}(n,z)`$. Now define $`c_p^{(\beta ;k,\mathrm{})}(n,z)`$ via the equation
$$\lambda _p^{(\beta ;k,\mathrm{})}(n,z)=\frac{n}{k+\mathrm{}}r_{(k,\mathrm{}),\beta }(z)+c_p^{(\beta ;k,\mathrm{})}(n,z)\sqrt{n}.$$
Similarly for $`c_q^{(\beta ;k,\mathrm{})}(n,z)`$. We would like to understand the behavior of $`c_p^{(\beta ;k,\mathrm{})}(n,z)`$ and $`c_q^{(\beta ;k,\mathrm{})}(n,z)`$ as $`n`$ goes to infinity. We consider $`c_p^{(\beta ;k,\mathrm{})}(n,z)`$.
Note that Definition 7.1, Theorem 7.3 and Definition 7.5 imply
###### Proposition 7.6
$`c_p^{(\beta ;k,\mathrm{})}(n,z)=\left({\displaystyle \frac{_{\lambda H(k,\mathrm{};n,z)}\lambda _{p,n}\left(f^\lambda \right)^\beta }{_{\lambda H(k,\mathrm{};n)}\left(f^\lambda \right)^\beta }}{\displaystyle \frac{n}{k+\mathrm{}}}r_{(k,\mathrm{}),\beta }(z)\right){\displaystyle \frac{1}{\sqrt{n}}}.`$ (18)
###### Conjecture 7.7
Recall the interval $`(L(k,\mathrm{},\beta ,z),L(k,\mathrm{},\beta ,z))`$ from Conjecture 7.4. As $`n`$ goes to infinity, $`c_p^{(\beta ;k,\mathrm{})}(n,z)`$ oscillates in the interval $`(L(k,\mathrm{},\beta ,z),L(k,\mathrm{},\beta ,z))+s_p^{(k,\mathrm{}),\beta }(z)`$, where
$`s_p^{(k,\mathrm{}),\beta }(z)={\displaystyle \frac{_{\mathrm{\Omega }_{(k,\mathrm{}),z}}x_p\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x)}{_{\mathrm{\Omega }_{(k,\mathrm{})}}\left[D_k(x)D_{\mathrm{}}(y)e^{\frac{k+\mathrm{}}{2}({\scriptscriptstyle x_i^2}+{\scriptscriptstyle y_j^2})}\right]^\beta d^{(k+\mathrm{}1)}(x)}}.`$ (19)
The proof is based on Conjecture 7.4 as follows. In (18) write $`\lambda _{p,n}=\frac{n}{k+\mathrm{}}+c_{p,n}\sqrt{n}`$, then $`c_p^{(\beta ;k,\mathrm{})}(n,z)=A(n)+B(n)`$, where
$$A(n)=\frac{\sqrt{n}}{k+\mathrm{}}\left(\frac{_{\lambda H(k,\mathrm{};n,z)}\left(f^\lambda \right)^\beta }{_{\lambda H(k,\mathrm{};n)}\left(f^\lambda \right)^\beta }r_{(k,\mathrm{}),\beta }(z)\right)$$
and
$$B(n)=\frac{_{\lambda H(k,\mathrm{};n,z)}c_{p,n}\left(f^\lambda \right)^\beta }{_{\lambda H(k,\mathrm{};n)}\left(f^\lambda \right)^\beta }.$$
By arguments similar to those in previous proofs, $`lim_n\mathrm{}B(n)=s_{(k,\mathrm{}),\beta }(z)`$, and by Conjecture 7.4, $`A(n)`$ oscillates in the interval $`(L(k,\mathrm{},\beta ,z),L(k,\mathrm{},\beta ,z))`$.
## Part II Some special cases
### 8 $`\beta =2`$, comparison of expected and maximal shapes
#### 8.1 Expected shape
Recall the RSK bijection $`\sigma (P_\lambda ,Q_\lambda )`$ and the subsets
$$S_{k,\mathrm{};n}=\{\sigma S_n\text{under the RSK}\sigma (P_\lambda ,Q_\lambda ),\lambda H(k,\mathrm{};n)\}$$
from Section 1.4. For example if $`\mathrm{}=0`$, it follows from well known properties of the RSK that $`S_{k,0;n}`$ are the permutations in $`S_n`$ with longest decreasing subsequence having length $`k`$. In general, if $`\sigma S_n`$ is of shape $`\lambda =(\lambda _1,\lambda _2,\mathrm{})`$, then $`\lambda _1`$ is the length of a maximal increasing subsequence in $`\sigma `$. Thus, for example, $`\lambda _{1,E}^{(2;k,\mathrm{})}`$ (see Definition 1.2) is the expected length of the longest increasing subsequences in $`S_{k,\mathrm{};n}`$.
Consider first the case $`\mathrm{}=0`$. Since the case $`k=1`$ is trivial, assume $`k2`$ (and $`\mathrm{}=0`$). In that case, by Theorem 5.3, the expected shape in $`S_{k,0;n}`$ is
$`\lambda _E^{(2;k,0)}(\frac{n}{k}+c_1\sqrt{n},\frac{n}{k}+c_2\sqrt{n},\mathrm{},\frac{n}{k}+c_k\sqrt{n})`$, where for $`1pk`$,
$`c_p=c_{p,E}^{(2;k,0)}={\displaystyle \frac{_{\mathrm{\Omega }_k}x_p\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^2d^{(k1)}x}{_{\mathrm{\Omega }_k}\left[D_k(x_1,\mathrm{},x_k)e^{\frac{k}{2}(x_1^2+\mathrm{}+x_k^2)}\right]^2d^{(k1)}x}}.`$ (20)
In general, the expected shape of the permutations in $`S_{k,\mathrm{};n}`$ is given by Definition 1.2, and the asymptotic shape as $`n\mathrm{}`$ is given by Theorem 6.3, both with $`\beta =2`$.
#### 8.2 Maximal $`f^\lambda `$
Given a subset of partitions $`\mathrm{\Gamma }_nY_n`$, one looks for $`\lambda =\lambda _{max}\mathrm{\Gamma }_n`$ with maximal degree $`f^\lambda `$, see Section 1.3. In the case $`\mathrm{\Gamma }_n=H(k,0;n)`$, the asymptotics of $`\lambda _{max}`$ is given in , which we briefly describe here. The analogue result from , for the case $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$, is also described below.
Let $`H_k(x)`$ denote the $`k`$-th Hermit polynomial. It is defined via the equation
$$\frac{d^k}{dx^k}\left(e^{x^2}\right)=(1^k)H_k(x)e^{x^2}.$$
Thus $`H_0(x)=1,H_1(x)=2x,H_2(x)=4x^22`$, $`H_3(x)=4x(2x^23)`$, $`H_4(x)=16x^448x^2+12`$, etc. $`H_k(x)`$ is of degree $`k`$ and its roots are real and distinct, denoted
$$x_1^{(k)}<x_2^{(k)}<\mathrm{}<x_k^{(k)}.$$
Also, $`x_1^{(k)}+x_2^{(k)}+\mathrm{}+x_k^{(k)}=0`$ . The following theorem is proved in :
###### Theorem 8.1
As $`n\mathrm{}`$, the maximum $`\mathrm{max}\{f^\lambda \lambda H(k,0;n)\}`$ occurs when
$$\lambda \lambda _{max}^{(k,0)}=(\frac{n}{k}+x_k^{(k)}\sqrt{\frac{k}{n}},\mathrm{},\frac{n}{k}+x_1^{(k)}\sqrt{\frac{k}{n}}).$$
The analogue result for $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$ is given in :
###### Theorem 8.2
Let $`\lambda =\lambda _{max}^{(k,\mathrm{})}H(k,\mathrm{};n)`$ maximize $`f^\lambda `$ in $`H(k,\mathrm{};n)`$ and write $`\lambda _{max}^{(k,\mathrm{})}=(\lambda _1,\mathrm{},\lambda _k,\mathrm{})`$ and $`\lambda _{max}^{{}_{}{}^{}(k,\mathrm{})}=(\lambda _1^{},\mathrm{},\lambda _{\mathrm{}}^{},\mathrm{})`$, and assume $`n\mathrm{}`$, then
$$(\lambda _1,\mathrm{},\lambda _k)(\frac{n}{k+\mathrm{}}+x_k^{(k)}\sqrt{\frac{n}{k+\mathrm{}}},\mathrm{},\frac{n}{k+\mathrm{}}+x_1^{(k)}\sqrt{\frac{n}{k+\mathrm{}}})$$
and
$$(\lambda _1^{},\mathrm{},\lambda _{\mathrm{}}^{})(\frac{n}{k+\mathrm{}}+x_{\mathrm{}}^{(\mathrm{})}\sqrt{\frac{n}{k+\mathrm{}}},\mathrm{},\frac{n}{k+\mathrm{}}+x_1^{(\mathrm{})}\sqrt{\frac{n}{k+\mathrm{}}}).$$
Here $`x_1^{(k)}<\mathrm{}<x_k^{(k)}`$ are the roots of the $`k`$โth Hermit polynomial and similarly for the $`x_j^{(\mathrm{})}`$โs.
#### 8.3 Examples of some $`(k,0)`$ cases
Let $`\mathrm{\Gamma }_n=H(k,0;n)`$ and $`\beta =2`$ ($`_{RSK}S_{k,0;n}`$) and compare the expected shape $`\lambda _E^{(2;k,0)}`$ with the maximizing shape $`\lambda _{max}^{(k,0)}`$. We begin with
The case $`k=2`$.
Here the expected shape is given by $`\lambda _E^{(2;2,0)}(\frac{n}{2}+c_{1,E}^{(2;2,0)}\sqrt{n},\frac{n}{2}+c_{2,E}^{(2;2,0)}\sqrt{n})`$, where $`c_{2,E}^{(2;2,0)}=c_{1,E}^{(2;2,0)}`$, and by Equation (20), $`c_{1,E}^{(2;2,0)}=I_1/I_2`$ is the ratio of the following integrals:
$$I_1=_0^{\mathrm{}}x\left[2xe^{2x^2}\right]^2๐x=\frac{1}{8},\text{ and}I_2=_0^{\mathrm{}}\left[2xe^{2x^2}\right]^2๐x=\frac{\sqrt{\pi }}{8}.$$
Thus $`c_{1,E}^{(2;2,0)}=\frac{1}{\sqrt{\pi }},`$ hence
$`\lambda _E^{(2;2,0)}({\displaystyle \frac{n}{2}}+{\displaystyle \frac{1}{\sqrt{\pi }}}\sqrt{n},{\displaystyle \frac{n}{2}}{\displaystyle \frac{1}{\sqrt{\pi }}}\sqrt{n})=({\displaystyle \frac{n}{2}}+0.56419\sqrt{n},{\displaystyle \frac{n}{2}}0.56419\sqrt{n}).`$ (21)
Compare $`\lambda _E^{(2;2,0)}`$ with $`\lambda _{max}^{(2,0)}`$: Here $`x_2^{(2,0)}=1/\sqrt{2},x_1^{(2,0)}=1/\sqrt{2}`$, so
$$\lambda _{max}^{(2,0)}=(\frac{n}{2}+\frac{1}{2}\sqrt{n},\frac{n}{2}\frac{1}{2}\sqrt{n})=(\frac{n}{2}+0.5\sqrt{n},\frac{n}{2}0.5\sqrt{n}).$$
Note: working with Equation (6) and with โMathematicaโ we calculated $`c_{1,E}^{(2;2,0)}(n)`$. For $`n=100,200,300,400,500,600,700,800,900`$, the corresponding values of $`c_{1,E}^{(2;2,0)}(n)`$ are:
$`0.517699,0.530593,0.536496,0.54007,0.542533,0.544364,0.545795,0.546952`$ and $`0.547915`$, agreeing with Theorem 5.3.
The case $`k=3`$
Here $`\lambda _E^{(2;3,0)}(\frac{n}{3}+c_{1,E}^{(2;3,0)}\sqrt{n},\frac{n}{3}+c_{2,E}^{(2;3,0)}\sqrt{n},\frac{n}{3}+c_{3,E}^{(2;3,0)}\sqrt{n})`$, so we calculate the $`c`$โs. By Equation (20), $`c_{1,E}^{(2;3,0)}=J_1/J_2`$ is the ratio of the following integrals:
$$J_1=_{\mathrm{\Omega }_3}x_1\left[D_3(x)e^{\frac{3}{2}(x_1^2+x_2^2+x_3^2)}\right]^2d^{(2)}x\text{and}J_2=_{\mathrm{\Omega }_3}\left[D_3(x)e^{\frac{3}{2}(x_1^2+x_2^2+x_3^2)}\right]^2d^{(2)}x.$$
By Remark 5.4 with $`\beta =2`$ and $`k=3`$, $`J_2=\frac{\pi }{324\sqrt{3}}`$. We calculate the numerator $`J_1`$. The domain $`\mathrm{\Omega }_3`$ of integration is defined by: $`x_1x_2x_3`$ and $`x_3=(x_1+x_2)`$, so $`x_1x_2(x_1+x_2).`$ When $`x_20`$, that condition is equivalent to $`x_2x_1/2`$. When $`x_20`$, that condition is equivalent to $`x_2x_1`$. It follows that
$$J_1=_0^{\mathrm{}}\left[_{\frac{x_1}{2}}^{x_1}\left(x_1\left[(x_1x_2)(2x_1+x_2)(x_1+2x_2)e^{3(x_1^2+x_2^2+x_1x_2)}\right]^2\right)๐x_2\right]๐x_1.$$
After some routine calculations (โMathematicaโ was used here) we obtain $`J_1=\frac{\sqrt{\pi }}{288\sqrt{2}}`$.
It follows that when $`k=3`$,
$`\underset{n\mathrm{}}{lim}c_{1,E}^{(2;3,0)}(n)=c_{1,E}^{(2;3,0)}={\displaystyle \frac{9\sqrt{3}}{8\sqrt{2\pi }}}=0.777362\mathrm{}`$ (22)
By similar calculations it follows that $`c_{2,E}^{(2;3,0)}=0`$, hence $`c_{3,E}^{(2;3,0)}=c_{1,E}^{(2;3,0)}`$. Thus
$$\lambda _E^{(2;3,0)}(\frac{n}{3}+\frac{9\sqrt{3}}{8\sqrt{2\pi }}\sqrt{n},\frac{n}{3},\frac{n}{3}\frac{9\sqrt{3}}{8\sqrt{2\pi }}\sqrt{n})=$$
$$=(\frac{n}{3}+0.777362\sqrt{n},\frac{n}{3},\frac{n}{3}0.777362\sqrt{n}).$$
Compare now with $`\lambda _{max}^{(3,0)}`$. Since $`H_3(x)=4x(2x^23),x_3^{(3,0)}=\frac{\sqrt{3}}{\sqrt{2}},x_2^{(3,0)}=0`$ and $`x_1^{(3,0)}=\frac{\sqrt{3}}{\sqrt{2}}`$. Thus
$$\lambda _{max}^{(3,0)}(\frac{n}{3}+\frac{1}{\sqrt{2}}\sqrt{n},\frac{n}{3},\frac{n}{3}\frac{1}{\sqrt{2}}\sqrt{n})=(\frac{n}{3}+0.707107\sqrt{n},\frac{n}{3},\frac{n}{3}0.707107\sqrt{n}).$$
Note: For $`n=200,300,400,500,700,850`$ and $`1100`$, โMathematicaโ and Equation (6) give the following corresponding values of $`c_{1,E}^{(2;3,0)}(n)`$: $`0.719084,\mathrm{\hspace{0.33em}0.729014},\mathrm{\hspace{0.33em}0.735096},\mathrm{\hspace{0.33em}0.739317},\mathrm{\hspace{0.33em}0.74494},\mathrm{\hspace{0.33em}0.747816}`$ and $`0.751261`$, in accordance with Theorem 5.3.
### 9 Examples of some $`(k,\mathrm{})`$โhook cases, $`\beta =2`$
#### 9.1 The $`(1,1)`$ case
By Equation (9) $`c_{1,E}^{(2;1,1)}=0`$. \[Alternatively, calculate $`c_{1,E}^{(\beta ;1,1)}`$ by Equation (11). Here $`\mathrm{\Omega }_{1,1}=\{(x,y)x+y=0\}`$, so $`y=x`$ and $`\mathrm{}x\mathrm{}`$. Thus $`c_{1,E}^{(\beta ;1,1)}=I_1(\beta )/I_2(\beta )`$ where $`I_1(\beta )=_{\mathrm{}}^{\mathrm{}}x\left[e^{2x^2}\right]^\beta ๐x=0`$.\] Similarly $`c_{1,E}^{(2;1,1)}=0`$ by Equation (10). It follows that (for any $`\beta >0`$)
$$\lambda _E^{(\beta ;1,1)}(\frac{n}{2},1^{\frac{n}{2}}).$$
#### 9.2 The $`(2,1)`$ case
By Corollary 6.5
$`\underset{n\mathrm{}}{lim}c_{1,E}^{(2;2,1)}(n)=c_{1,E}^{(2;2,1)}=\sqrt{{\displaystyle \frac{2}{3}}}c_{1,E}^{(2;2,0)}=\sqrt{{\displaystyle \frac{2}{3}}}{\displaystyle \frac{1}{\sqrt{\pi }}}=0.460659.`$ (23)
Since $`c_{2,E}^{(2;2,0)}=c_{1,E}^{(2;2,0)}`$, deduce that
$`c_{2,E}^{(2;2,1)}=\sqrt{{\displaystyle \frac{2}{3}}}{\displaystyle \frac{1}{\sqrt{\pi }}}=0.460659.`$ (24)
Since the sum of the coordinates in $`\lambda _E^{(2;2,1)}(n)`$ is $`n`$, it follows that $`c_{1,E}^{(2;2,1)}=0`$ (this can also be deduced directly from Theorem 6.3). Therefore
$$\lambda _E^{(2;2,1)}(\frac{n}{3}+0.460659\sqrt{n},\frac{n}{3}0.460659\sqrt{n},1^{n/3}).$$
Note: for $`n=100,\mathrm{\hspace{0.33em}200},\mathrm{\hspace{0.33em}300},\mathrm{\hspace{0.33em}400},\mathrm{\hspace{0.33em}700}`$ and $`1100`$, Remark (6.2) and โMathematicaโ give the following values of $`c_{1,E}^{(2;2,1)}(n)`$: $`0.45423,\mathrm{\hspace{0.33em}0.45571},\mathrm{\hspace{0.33em}0.456475},\mathrm{\hspace{0.33em}0.456962},\mathrm{\hspace{0.33em}0.457777}`$ and $`0.458317`$, agreeing with Theorem 6.3.
### 10 Expected shape, $`\beta =1`$
#### 10.1 The general case
Since $`\sigma S_n`$ is an involution ifโf the RSK yields $`\sigma (P_\lambda ,P_\lambda )`$, therefore $`\lambda _E^{(1;k,\mathrm{})}(n)`$ is the expected shape of the involutions in $`S_{k,\mathrm{};n}`$. Note that when $`\mathrm{\Gamma }_n=Y_n`$ (and $`\beta =1`$) Baik and Rains showed that the expected length of the first row (or of the longest increasing subsequence in involutions in $`S_n`$) is again $`2\sqrt{n}`$, i.e. the same as the first row of $`\lambda _{max}`$, see (1.5). As we show below, this is not the case when $`\mathrm{\Gamma }_n=H(k,\mathrm{};n)`$.
Denote $`\stackrel{~}{\lambda }^{(k,\mathrm{})}(n)=\lambda _E^{(1;k,\mathrm{})}(n)`$. We summarize:
1. Let $`1pk,1q\mathrm{}`$, then
$$\stackrel{~}{\lambda }_{p,n}^{(k,\mathrm{})}=\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{p,n}f^\lambda }{_{\lambda H(k,\mathrm{};n)}f^\lambda }\text{and}\stackrel{~}{\lambda }_{q,n}^{(k,\mathrm{})}=\frac{_{\lambda H(k,\mathrm{};n)}\lambda _{q,n}^{}f^\lambda }{_{\lambda H(k,\mathrm{};n)}f^\lambda }.$$
2. Define $`\stackrel{~}{c}_{p,n}^{(k,\mathrm{})}`$ via
$$\stackrel{~}{\lambda }_{p,n}^{(k,\mathrm{})}=\frac{n}{k+\mathrm{}}+\stackrel{~}{c}_{j,n}^{(k,\mathrm{})}\sqrt{n},\text{and}\stackrel{~}{c}_p^{(k,\mathrm{})}=\underset{n\mathrm{}}{lim}\stackrel{~}{c}_{p,n}^{(k,\mathrm{})}.$$
Similarly for $`\stackrel{~}{c}_{q,n}^{(k,\mathrm{})}`$ and $`\stackrel{~}{c}_q^{(k,\mathrm{})}`$. Thus, when $`n\mathrm{}`$,
$$\stackrel{~}{\lambda }_{p,n}^{(k,\mathrm{})}\frac{n}{k+\mathrm{}}+\stackrel{~}{c}_p^{(k,\mathrm{})}\sqrt{n}\text{and}\stackrel{~}{\lambda }_{q,n}^{(k,\mathrm{})}\frac{n}{k+\mathrm{}}+\stackrel{~}{c}_q^{(k,\mathrm{})}\sqrt{n}.$$
3. (Theorem 6.3, $`\beta =1`$) Let $`1pk`$, $`1q\mathrm{}`$, then the limits $`\stackrel{~}{c}_p^{(k,\mathrm{})}`$ and $`\stackrel{~}{c}_q^{(k,\mathrm{})}`$ exist, and are given as follows.
$`\stackrel{~}{c}_p^{(k,\mathrm{})}={\displaystyle \frac{_{\mathrm{\Omega }_k}x_pD_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}d^{(k1)}(x)}{_{\mathrm{\Omega }_k}D_k(x_1,\mathrm{},x_k)e^{\frac{k+\mathrm{}}{2}(x_1^2+\mathrm{}+x_k^2)}d^{(k1)}(x)}}.`$ (25)
Similarly for the expected columns:
$`\stackrel{~}{c}_q^{(k,\mathrm{})}={\displaystyle \frac{_\mathrm{\Omega }_{\mathrm{}}y_qD_{\mathrm{}}(y_1,\mathrm{},y_{\mathrm{}})e^{\frac{k+\mathrm{}}{2}(y_1^2+\mathrm{}+y_{\mathrm{}}^2)}d^{(\mathrm{}1)}(y)}{_\mathrm{\Omega }_{\mathrm{}}D_{\mathrm{}}(y_1,\mathrm{},y_{\mathrm{}})e^{\frac{k+\mathrm{}}{2}(y_1^2+\mathrm{}+y_{\mathrm{}}^2)}d^{(\mathrm{}1)}(y)}}.`$ (26)
4. (Corollary 6.5, $`\beta =1`$) We have
$$\stackrel{~}{c}_p^{(k,\mathrm{})}=\sqrt{\frac{k}{k+\mathrm{}}}\stackrel{~}{c}_p^{(k,0)},\text{and similarly}\stackrel{~}{c}_q^{(k,\mathrm{})}=\sqrt{\frac{\mathrm{}}{k+\mathrm{}}}\stackrel{~}{c}_q^{(0,\mathrm{})}.$$
#### 10.2 Examples for involutions in $`S_{k,0;n}`$
(i.e. $`\beta =1`$ and $`\mathrm{}=0`$).
When $`k=2`$, we obtain $`_0^{\mathrm{}}2xe^{2x^2}=\frac{1}{2}\text{and}_0^{\mathrm{}}x2xe^{2x^2}=\frac{\sqrt{\pi }}{4\sqrt{2}}`$.
Thus $`\stackrel{~}{c}_1^{(2,0)}=\frac{\sqrt{\pi }}{2\sqrt{2}}`$ and
$$\stackrel{~}{\lambda }^{(2,0)}=(\stackrel{~}{\lambda }_1^{(2,0)},\stackrel{~}{\lambda }_2^{(2,0)})(\frac{n}{2}+\frac{\sqrt{\pi }}{2\sqrt{2}}\sqrt{n},\frac{n}{2}\frac{\sqrt{\pi }}{2\sqrt{2}}\sqrt{n})=$$
$`=({\displaystyle \frac{n}{2}}+0.626657\sqrt{n},{\displaystyle \frac{n}{2}}0.626657\sqrt{n}).`$ (27)
Note: for $`n=200,\mathrm{\hspace{0.17em}400},\mathrm{\hspace{0.17em}600},\mathrm{\hspace{0.17em}800},\mathrm{\hspace{0.17em}1000},\mathrm{\hspace{0.17em}1200}`$ and $`1400`$, Remark 6.2 and โMathematicaโ give the following values of $`\stackrel{~}{c}_1^{(2,0)}(n)`$: $`0.592086,\mathrm{\hspace{0.17em}0.602049},\mathrm{\hspace{0.17em}0.606506}`$, $`0.609175`$, $`0.611002`$, $`0.612354`$ and $`0.613406`$, agreeing with Equation (25).
When $`k=3`$ we have
By Equation (25), $`c_{1,E}^{(1;3,0)}=\stackrel{~}{J}_1/\stackrel{~}{J}_2`$ is the ratio of the following integrals:
$$\stackrel{~}{J}_1=_{\mathrm{\Omega }_3}x_1D_3(x)e^{\frac{3}{2}(x_1^2+x_2^2+x_3^2)}d^{(2)}x\text{and}\stackrel{~}{J}_2=_{\mathrm{\Omega }_3}D_3(x)e^{\frac{3}{2}(x_1^2+x_2^2+x_3^2)}d^{(2)}x$$
By Remark 5.4 ($`\beta =1`$, $`k=3`$), $`\stackrel{~}{J}_2=\frac{\sqrt{\pi }}{27}`$. We calculate the numerator $`\stackrel{~}{J}_1`$. Similar to the evaluation of $`J_1`$ in Section 8.3, here
$$\stackrel{~}{J}_1=_0^{\mathrm{}}\left[_{\frac{x_1}{2}}^{x_1}\left(x_1(x_1x_2)(2x_1+x_2)(x_1+2x_2)e^{3(x_1^2+x_2^2+x_1x_2)}\right)๐x_2\right]๐x_1=1/18.$$
It follows that when $`k=3`$, $`\stackrel{~}{c}_1^{(3,0)}=c_{1,E}^{(1;3,0)}=\frac{3}{2\sqrt{\pi }}=0.846284\mathrm{}`$ Again, $`\stackrel{~}{c}_2^{(3,0)}=0`$ so $`\stackrel{~}{c}_3^{(3,0)}=\stackrel{~}{c}_1^{(3,0)}`$. Thus, when $`k=3`$,
$$\stackrel{~}{\lambda }^{(3,0)}=\lambda _E^{(1;3,0)}=(\frac{n}{3}+\frac{3}{2\sqrt{\pi }}\sqrt{n},\frac{n}{3},\frac{n}{3}\frac{3}{2\sqrt{2\pi }}\sqrt{n})=$$
$`=({\displaystyle \frac{n}{3}}+0.846284\sqrt{n},{\displaystyle \frac{n}{3}},{\displaystyle \frac{n}{3}}0.846284\sqrt{n}).`$ (28)
Note: for $`n=200,\mathrm{\hspace{0.17em}400},\mathrm{\hspace{0.17em}600},\mathrm{\hspace{0.17em}800},\mathrm{\hspace{0.17em}1000},`$ and $`1200`$, Remark 6.2 and โMathematicaโ give the following values of $`\stackrel{~}{c}_1^{(3,0)}(n)=c_{1,E}^{(1;3,0)}`$:
$`0.789051,\mathrm{\hspace{0.17em}0.80486},\mathrm{\hspace{0.17em}0.812115},\mathrm{\hspace{0.17em}0.816513},\mathrm{\hspace{0.17em}0.819547}`$ and $`0.821802`$, as predicted by Equation (25)
Acknowledgement:
I would like to thank G. Olshanski for some very fruitful discussions and suggestions.
|
warning/0506/hep-ph0506297.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The discovery of neutrino masses and mixing angles, arguably the greatest advance in physics over the past decade, has provided new clues in the search for a theory of quark and lepton masses and mixings. For example, it is interesting to compare the observed or bounded lepton mixing angles :
$$\theta _{12}=33.2^{}\pm 5^{},\theta _{23}=45^{}\pm 10^{},\theta _{13}<13^{},$$
(1)
to the observed quark mixing angles and phase:
$$\theta _{12}^{\mathrm{CKM}}=13.0^{}\pm 0.1^{},\theta _{23}^{\mathrm{CKM}}=2.4^{}\pm 0.1^{},\theta _{13}^{\mathrm{CKM}}=0.2^{}\pm 0.1^{},\delta ^{\mathrm{CKM}}=60^{}\pm 14^{}.$$
(2)
The quest to understand the relation between the very different lepton and quark mixing angles has led to a great deal of theoretical model building . The poorly determined lepton parameters (especially the neutrino oscillation phase $`\delta `$ which is completely undetermined) as compared to the quark mixing angles, presents an opportunity to make testable predictions in the lepton sector by relating the lepton mixing parameters to the quark ones. This can provide theoretical motivation for making high precision measurements in the neutrino sector. For instance the empirical relation between the leptonic mixing angle $`\theta _{12}`$ (the solar angle) and the Cabibbo angle $`\theta _\mathrm{C}=\theta _{12}^{\mathrm{CKM}}`$
$`\theta _{12}+\theta _\mathrm{C}{\displaystyle \frac{\pi }{4}}`$ (3)
has recently been the subject of much speculation . The interest arises from the possibility that this so-called quark-lepton complementarity (QLC) relation could be a signal of some high scale quark-lepton unification. All the attempts to reproduce the QLC relation in the literature so far start from some kind of maximal or bi-maximal mixing in either the neutrino or the charged lepton sectors, then consider the corrections to maximal mixing coming from the other sector.
For example, in the context of inverted hierarchy models with a pseudo-Dirac structure, it was observed some time ago that the predictions for the neutrino mixing angles of $`\theta _{12}^\nu =\pi /4`$, $`\theta _{13}^\nu =0`$, may receive corrections from the charged lepton mixing angle of order the Cabibbo angle, $`\theta _{12}^e\theta _\mathrm{C}`$, resulting in $`\theta _{12}`$ being in the LMA MSW range, and $`\theta _{13}`$ close to its current experimental limit. Recently it was shown that such a scheme, when combined with a Pati-Salam symmetry, could lead to approximate QLC. However the way that this was achieved was quite non-trivial. The contribution to $`\theta _{12}`$ coming from the charged lepton mixing angle $`\theta _{12}^e`$ is suppressed by a factor of $`1/\sqrt{2}`$ , due to the approximately maximal atmospheric mixing angle, and an approximate QLC relation was achieved by selecting operators which give rise to $`\theta _{12}^e(3/2)\theta _C`$, enhancing the charged lepton mixing angle by a Clebsch factor of $`3/2`$, in order to approximately cancel the suppression factor of $`1/\sqrt{2}`$ . This approach leads to the predictions $`m_\mu /m_s=2`$ at the GUT scale, and the โreactorโ leptonic mixing angle $`\theta _{13}\theta _C`$, both of which are on the edge of current experimental limits. The traditional expectation from unified models that $`\theta _{12}^e\theta _\mathrm{C}/3`$, corresponding to the Georgi-Jarlskog (GJ) relation $`m_\mu /m_s=3`$ at the GUT scale, while being more consistent with data, is clearly inconsistent with the above approach to QLC. This motivates the search for alternative models of QLC which would be consistent with the GJ relations.
In this paper we discuss QLC from a completely different starting point, namely tri-bimaximal neutrino mixing. We emphasise that, unlike , tri-bimaximal mixing in the neutrino sector is merely a staging point in our considerations, and the final form of the lepton mixing matrix, after charged lepton mixing angles have been taken into account, will not have the tri-bimaximal form. We therefore refer to this approach as tri-bimaximal complementarity to distinguish it from the usual tri-bimaximal neutrino mixing. To be precise we shall derive from the see-saw mechanism a neutrino mixing matrix of the tri-bimaximal form:
$`V_{\nu _\mathrm{L}}^{}\left(\begin{array}{ccc}\sqrt{\frac{2}{3}}& \frac{1}{\sqrt{3}}& 0\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{2}}\\ \frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}}& \frac{1}{\sqrt{2}}\end{array}\right)`$ (4)
Then, in the conventions of Appendix A, the MNS matrix is given by $`U_{\mathrm{MNS}}=V_{e_\mathrm{L}}V_{\nu _\mathrm{L}}^{}`$, and so the MNS matrix will not be of the tri-bimaximal form but will involve a left multiplication by the charged lepton mixing matrix $`V_{e_\mathrm{L}}`$. Whereas the neutrino mixing angles arising from tri-bimaximal mixing take the approximate values $`\theta _{12}^\nu =\mathrm{sin}^1(1/\sqrt{3})=35.26^{}`$, $`\theta _{23}^\nu =45^{}`$, $`\theta _{13}^\nu =0^{}`$, the physical lepton mixing angles arising from tri-bimaximal complementarity will differ from these values due to the charged lepton mixing angle corrections, which in turn are related to the quark mixing angles. The atmospheric angle is predicted to be approximately maximal $`\theta _{23}=45^{}`$, corrected by the quark mixing angle $`\theta _{23}^{\mathrm{CKM}}2.4^{}`$, with the correction controlled by an undetermined phase in the quark sector. The reactor angle is predicted to be $`\theta _{13}\frac{1}{\sqrt{2}}\frac{\theta _\mathrm{C}}{3}3.06^{}`$. <sup>1</sup><sup>1</sup>1This prediction for the reactor angle also follows from bi-maximal neutrino mixing, in which $`\theta _{13}^\nu =0`$, and $`\theta _{13}`$ originates from charged lepton mixing angle of order one third of the Cabibbo angle, $`\theta _{12}^e\theta _\mathrm{C}/3`$, typical of the GJ correction . However the solar angle cannot be accounted for by such charged lepton corrections in bi-maximal neutrino mixing . The solar angle is predicted from the tri-bimaximal complementarity relation,
$`\theta _{12}+{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{\theta _\mathrm{C}}{3}}\mathrm{cos}(\delta \pi )35.26^{}.`$ (5)
In Eq.5 the factor of $`1/3`$ arises from the GJ relations, the factor of $`1/\sqrt{2}`$ arises from the atmospheric angle as discussed previously, and $`\delta `$ is the MNS oscillation phase. The tri-bimaximal complementarity relation in Eq.5 may be compared to the bimaximal complementarity relation Eq.3. The two relations are approximately numerically equivalent in the case that $`\delta =\pi `$ since $`\frac{1}{\sqrt{2}}\frac{\theta _\mathrm{C}}{3}3.06^{}`$. <sup>2</sup><sup>2</sup>2Typically the bi-maximal complementarity relation in Eq.3 will also involve a similar phase which is often neglected without good reason. Note that the terminology โtri-bimaximal complementarityโ as introduced here is a short-hand for โcharged lepton corrections to tri-bimaximal neutrino mixing in which the charged lepton mixing angles are related to the quark mixing anglesโ. In particular it refers to the relation in Eq.5.โ The tri-bimaximal complementarity relation has the twin advantages that it incorporates the GJ relations, as well as the factor of $`1/\sqrt{2}`$ which proves troublesome for bi-maximal complementarity. In addition Eq.5 may be used to predict the neutrino oscillation phase $`\delta `$ from a future accurate measurement of the solar angle $`\theta _{12}`$, $`\mathrm{cos}(\delta \pi )(35.26^{}\theta _{12}^{})/3.06^{}`$. These predictions can be tested by future high precision neutrino oscillation experiments.
There has recently been some progress with achieving tri-bimaximal neutrino mixing from the see-saw mechanism using vacuum alignment with various family symmetries such as $`SU(3)`$ or the discrete symmetry $`A_4`$ . Here we shall show how tri-bimaximal neutrino mixing can emerge in a natural and general way from the see-saw mechanism using sequential dominance , with certain simple constraints imposed on the Yukawa couplings, independently of any particular choice of family symmetry. We refer to this general approach as constrained sequential dominance (CSD). CSD can arise from the vacuum alignment some non-Abelian family symmetry, and here we focus on $`SO(3)`$. A potential advantage of using $`SO(3)`$ family symmetry is that it is possible to โup-gradeโ any resulting model of hierarchical neutrino masses to a type II see-saw model with a quasi-degenerate spectrum of neutrino masses , and improved prospects for leptogenesis , although in this paper we shall restrict ourselves to hierarchical neutrino masses. We shall subsequently present an explicit model based on $`SO(3)`$ family symmetry and Pati-Salam unification which is consistent with all quark and lepton masses and mixings, and gives rise to tri-bimaximal complementarity using the Georgi-Jarlskog relations. All types of complementarity are crucially dependent on the plethora of complex phases which are generally present in the Yukawa matrices. Here we shall keep careful track of all the phases and in our approach show how tri-bimaximal complementarity may be linked to the MNS CP violating phase.
This paper has been organized as follows: in Sec. 2, we discuss how to achieve tri-bimaximal mixing using CSD, and briefly show how CSD could arise from vacuum alignment with an $`SO(3)`$ family symmetry. In Sec. 3 we present an explicit model based on $`SO(3)`$ famly symmetry and Pati-Salam unification which has all the necessary ingredients that we require. In Sec. 4 we discuss the predictions arising from the model, including a careful discussion of the complex phases which appear in tri-bimaximal complementarity. Sec. 5 concludes the paper. In Appendix A we specify our conventions, while in Appendix B we discuss vacuum alignment in the model.
## 2 Tri-bimaximal neutrino mixing from the see-saw mechanism with constrained sequential dominance
The fact that the tri-bimaximal neutrino mixing matrix in Eq.4 involves square roots of simple ratios motivates models in which the mixing angles are independent of the mass eigenvalues. One such class of models are see-saw models with sequential dominance (SD) of right-handed neutrinos . In SD, the atmospheric and solar neutrino mixing angles are determined in terms of ratios of Yukawa couplings involving the dominant and subdominant right-handed neutrinos, respectively. If these Yukawa couplings are simply related in some way, then it is possible for simple neutrino mixing angle relations, such as appear in tri-bimaximal neutrino mixing, to emerge in a simple and natural way, independently of the neutrino mass eigenvalues.
To see how tri-bimaximal neutrino mixing could emerge from SD, we begin by writing the right-handed neutrino Majorana mass matrix $`M_{\mathrm{RR}}`$ in a diagonal basis as
$`M_{\mathrm{RR}}\left(\begin{array}{ccc}Y& 0& 0\\ 0& X& 0\\ 0& 0& X^{}\end{array}\right),`$ (6)
where we shall assume
$$YXX^{}.$$
(7)
Then in this basis we write the neutrino (Dirac) Yukawa matrix $`Y_{LR}^\nu `$ in terms of the complex Yukawa couplings $`a,b,c,d,e,f,a^{},b^{},c^{}`$ as
$`Y_{\mathrm{LR}}^\nu =\left(\begin{array}{ccc}d& a& a^{}\\ e& b& b^{}\\ f& c& c^{}\end{array}\right).`$ (8)
in the convention where the Yukawa matrix corresponds to the Lagrangian coupling $`\overline{L}H_uY_{LR}^\nu \nu _R`$, where $`L`$ are the left-handed lepton doublets, $`H_u`$ is the Higgs doublet coupling to up-type quarks and neutrinos, and $`\nu _R`$ are the right-handed neutrinos. The Dirac neutrino mass matrix is then given by $`m_{\mathrm{LR}}^\nu =Y_{LR}^\nu v_\mathrm{u}`$, where $`v_\mathrm{u}`$ is the vacuum expectation value (VEV) of $`H_u`$.
For simplicity we shall henceforth assume that $`d=0`$, although this is not strictly necessary . Then the condition for sequential dominance (SD) is that the right-handed neutrino of mass $`Y`$ gives the dominant contribution to the see-saw mechanism, while the right-handed neutrino of mass $`X`$ gives the leading sub-dominant contribution
$$\frac{|e^2|,|f^2|,|ef|}{Y}\frac{|xy|}{X}\frac{|x^{}y^{}|}{X^{}}$$
(9)
where $`x,ya,b,c`$ and $`x^{},y^{}a^{},b^{},c^{}`$, and all Yukawa couplings are assumed to be complex. The combination of Eqs.7,9 is called light sequential dominance (LSD) since the lightest right-handed neutrino makes the dominant contribution to the see-saw mechanism. LSD is motivated by unified models in which only small mixing angles are present in the Yukawa sector, and implies that the heaviest right-handed neutrino of mass $`X^{}`$ is irrelevant for both leptogenesis and neutrino oscillations (for a discussion of all these points see ). In addition many realistic models in the literature (see for example ) involve an approximate texture zero in the 11 position, corresponding to our simplifying assumption $`d=0`$. This will have the effect of removing one of the see-saw phases.
Assuming Eq.9 the neutrino masses are given to leading order in $`m_2/m_3`$ by the results in , summarized as:
$`m_1`$ $``$ $`O({\displaystyle \frac{x^{}y^{}}{X^{}}})v_u^2`$ (10)
$`m_2`$ $``$ $`{\displaystyle \frac{|a|^2}{X(s_{12}^\nu )^2}}v_u^2`$ (11)
$`m_3`$ $``$ $`{\displaystyle \frac{(|e|^2+|f|^2)}{Y}}v_u^2`$ (12)
where $`s_{12}^\nu =\mathrm{sin}\theta _{12}^\nu `$ may be obtained from the further results given below. Note that with SD each neutrino mass is generated by a separate right-handed neutrino, and the sequential dominance condition naturally results in a neutrino mass hierarchy $`m_1m_2m_3`$. The neutrino mixing angles are given to leading order as ,
$`\mathrm{tan}\theta _{23}^\nu `$ $``$ $`{\displaystyle \frac{|e|}{|f|}}`$ (13)
$`\mathrm{tan}\theta _{12}^\nu `$ $``$ $`{\displaystyle \frac{|a|}{c_{23}^\nu |b|\mathrm{cos}(\varphi _b^{})s_{23}^\nu |c|\mathrm{cos}(\varphi _c^{})}}`$ (14)
$`\theta _{13}^\nu `$ $``$ $`e^{i(\varphi _2^\nu +\varphi _a\varphi _e)}{\displaystyle \frac{|a|(e^{}b+f^{}c)}{[|e|^2+|f|^2]^{3/2}}}{\displaystyle \frac{Y}{X}}`$ (15)
where we have written some (but not all) complex Yukawa couplings as $`x=|x|e^{i\varphi _x}`$. The phase $`\chi ^\nu `$ is fixed to give a real angle $`\theta _{12}^\nu `$ by,
$$c_{23}^\nu |b|\mathrm{sin}(\varphi _{}^{}{}_{b}{}^{})s_{23}^\nu |c|\mathrm{sin}(\varphi _{}^{}{}_{c}{}^{})$$
(16)
where
$`\varphi _b^{}`$ $``$ $`\varphi _b\varphi _a\varphi _2^\nu \chi ^\nu ,`$
$`\varphi _c^{}`$ $``$ $`\varphi _c\varphi _a+\varphi _e\varphi _f\varphi _2^\nu \chi ^\nu .`$ (17)
The phase $`\varphi _2^\nu `$ is fixed to give a real angle $`\theta _{13}^\nu `$ by ,
$$\varphi _2^\nu \varphi _e\varphi _a\varphi _{\mathrm{COSMO}}$$
(18)
where
$$\varphi _{\mathrm{COSMO}}=\mathrm{arg}(e^{}b+f^{}c)$$
(19)
is the leptogenesis phase corresponding to the interference diagram involving the lightest and next-to-lightest right-handed neutrinos . The auxiliary phases appearing above are defined in Appendix A.
We can now ask what are the conditions for achieving tri-bimaximal neutrino mixing as in Eq.4, in which $`\mathrm{tan}\theta _{23}^\nu =1`$, $`\mathrm{tan}\theta _{12}^\nu =1/\sqrt{2}`$ and $`\theta _{13}^\nu =0`$ in the framework of sequential dominance? Note that in sequential dominance the mixing angles are determined by ratios of Yukawa couplings, and are independent of the neutrino masses. We propose the following set of conditions which are sufficient to achieve tri-bimaximal mixing within the framework of sequential dominance:
$`|a|`$ $`=`$ $`|b|=|c|,`$ (20a)
$`|d|`$ $`=`$ $`0,`$ (20b)
$`|e|`$ $`=`$ $`|f|,`$ (20c)
$`\varphi _b^{}`$ $`=`$ $`0,`$ (20d)
$`\varphi _c^{}`$ $`=`$ $`\pi .`$ (20e)
Eqs.20a, 20b, 20c are conditions on the magnitudes of the Yukawa couplings, while Eqs.20d, 20e are generic phase conditions which can be satisfied by several different types of phase structure in the Yukawa matrix. The condition in Eq.20c clearly gives rise to $`\mathrm{tan}\theta _{23}^\nu =1`$, as can be seen from Eq.13. The remaining conditions in Eq.20 result in $`\mathrm{tan}\theta _{12}^\nu =1/\sqrt{2}`$ as can be seen from Eq.14. Eqs.20d and 20e, together with the definitions in Eq.17, imply the condition on the phases of the Yukawa couplings:
$$\varphi _c\varphi _b+\varphi _e\varphi _f=\pi .$$
(21)
Eq.21, together with Eqs.20c,20a, then implies that:
$$e^{}b+f^{}c=0.$$
(22)
Eq.22 implies from Eq.15 that $`\theta _{13}^\nu =0`$. It also implies that leptogenesis is zero at leading order, independently of the choice of charged lepton basis . We conclude that the conditions in Eq.20, together with the conditions for sequential dominance, are sufficient to result in tri-bimaximal neutrino mixing as in Eq.4. We shall refer to this as constrained sequential dominance (CSD). Note that, with the conditions in Eq.20 satisfied, the angle $`\theta _{12}^\nu `$ is automaticaly real, so the phase $`\chi ^\nu `$ is undetermined, and will be expected to play no part in physics. The phase $`\varphi _2^\nu `$ is similarly undetermined and unphysical, since $`\theta _{13}^\nu =0`$.
Since there are undetermined phases above, it is instructive to consider tri-bimaximal mixing as a limiting case of a known example where the phases are determined. The example will also serve as an introduction to the model of quark and lepton masses and mixings discussed in the next section based on $`SO(3)`$ family symmetry and Pati-Salam unification, in which a neutrino Yukawa matrix which satisfies the conditions of CSD, will arise. It should be emphasised that other examples based on $`SU(3)`$ or discrete family symmetries may also give rise to CSD, the general conditions for which are given in Eqs.20,21.
Consider a supersymmetric theory in which the lepton doublets $`L`$ are triplets of an $`SO(3)`$ family symmetry, but the (CP conjugates of) right-handed neutrinos $`\nu _i^c`$ and Higgs doublets $`H_u`$ are singlets under the family symmetry . Yukawa couplings arise from the superpotential terms of the form:
$$|y_1|e^{i\delta _1}LH_u\nu _1^c\frac{\varphi _{23}}{M}+|y_2|e^{i\delta _2}LH_u\nu _2^c\frac{\varphi _{123}}{M}+|y_3|e^{i\delta _3}LH_u\nu _3^c\frac{\varphi _3}{M}$$
(23)
where $`\varphi _{23},\varphi _{123},\varphi _3`$ are $`SO(3)`$ triplet flavon fields whose vacuum expectation values (VEVs) break the $`SO(3)`$ family symmetry, and allow Dirac neutrino mass terms to be generated. We have written the Yukawa couplings in terms of magnitudes and phases $`|y_i|`$ and $`e^{i\delta _i}`$, and $`M`$ is a real positive mass scale. Each term in Eq.23 only involves a particular flavon superfield coupling together with a particular right-handed neutrino superfield. This may readily be enforced by symmetries, as we shall discuss later in the framework of the Pati-Salam theory.
In it was also shown how to generate real flavon VEVs:
$$\begin{array}{c}\frac{|y_2|\varphi _{123}}{M}=\left(\begin{array}{c}a\\ b\\ c\end{array}\right),\frac{|y_1|\varphi _{23}}{M}=\left(\begin{array}{c}0\\ e\\ f\end{array}\right),\frac{|y_3|\varphi _3}{M}=\left(\begin{array}{c}0\\ 0\\ c^{}\end{array}\right).\hfill \end{array}$$
(24)
When these VEVs are inserted into the couplings in Eq.23 this results in a neutrino Yukawa matrix:
$`Y_{\mathrm{LR}}^\nu =\left(\begin{array}{ccc}0& ae^{i\delta _2}& 0\\ ee^{i\delta _1}& be^{i\delta _2}& 0\\ fe^{i\delta _1}& ce^{i\delta _2}& c^{}e^{i\delta _3}\end{array}\right),`$ (25)
where here $`a,b,c,e,f,c^{}`$ are real (positive or negative) numbers. In order to satisfy the CSD constraints in Eqs.20,21 it is sufficient to show that it is possible to arrange for the real VEVs in Eq.24 to be aligned such that:
$`e`$ $`=`$ $`f,`$ (26a)
$`a`$ $`=`$ $`b=c.`$ (26b)
The phases required to ensure positive neutrino mixing angles are then given in the limit of such a vacuum alignment by: <sup>3</sup><sup>3</sup>3 The undetermined phases $`\varphi _2^\nu ,\chi ^\nu `$ are specified above by slightly relaxing the conditions in Eqs.26a,26b, while keeping $`|b|>|c|`$. However these phases are unphysical in the case of tri-bimaximal neutrino mixing and they could equally well be set to zero.
$`\varphi _b^{}`$ $`=`$ $`\varphi _2^\nu \chi ^\nu =0,`$
$`\varphi _c^{}`$ $`=`$ $`\pi \varphi _2^\nu \chi ^\nu =\pi ,`$ (27a)
$`\varphi _2^\nu `$ $`=`$ $`2(\delta _1\delta _2),`$ (27b)
$`\varphi _3^\nu `$ $`=`$ $`\varphi _2^\nu +\pi ,`$ (27c)
$`\omega _1^\nu `$ $`=`$ $`\delta _3,`$ (27d)
$`\omega _2^\nu `$ $`=`$ $`\delta _2,`$ (27e)
$`\omega _3^\nu `$ $`=`$ $`\delta _1,`$ (27f)
where the phases are defined in Appendix A. Such a vacuum alignment would then satisfy the constraints in Eqs.20,21. In order to arrange for the VEVs in Eqs.26a,26b, we need to introduce additional flavon superfields and superpotential terms involving these and other superfields, as discussed in Appendix B. It is clear that $`SO(3)`$ family symmetry and vacuum alignment can provide a realization of CSD and hence tri-bimaximal neutrino mixing. To obtain tri-bimaximal complementarity we also require quark-lepton unification which incorporates the GJ relations, and we now turn to the construction of such a model.
## 3 $`SO(3)`$ family symmetry and Pati-Salam unification
We now construct a realistic model based on a family symmetry $`SO(3)`$ and Pati-Salam unification. The model will incorporate both the vacuum alignment in $`SO(3)`$ necessary to achieve tri-bimaximal neutrino mixing via constrained sequential dominance, and also will involve the GJ relations necessary to relate the charged lepton mixing angle to the Cabibbo angle. The explicit model will demonstrate that all these features can be achieved together within a single consistent framework, and will lead to further relations between the lepton and quark mixing angles.
The model is a supersymmetric theory based on the family symmetry $`SO(3)`$ together with the Pati-Salam gauge group
$$G_{PS}=\text{SU}(4)_{\mathrm{PS}}\times \text{SU}(2)_\mathrm{L}\times \text{SU}(2)_\mathrm{R}.$$
(28)
Assuming the Pati-Salam symmetry to start with has the advantage that it explicitly exhibits $`SU(4)_{PS}`$ quark-lepton and $`SU(2)_R`$ isospin symmetry, allowing Georgi-Jarlskog factors to be generated and isospin breaking to be controlled, while avoiding the Higgs doublet-triplet splitting problem . Quarks and leptons are unified in the $`\text{SU}(4)_{\mathrm{PS}}`$-quartets $`F_i`$ and $`F_i^c`$ of $`\text{SU}(4)_\mathrm{C}`$, which are doublets of $`\text{SU}(2)_\mathrm{L}`$ and $`\text{SU}(2)_\mathrm{R}`$, respectively,
$`F_i=\left(\begin{array}{cccc}u_i& u_i& u_i& \nu _i\\ d_i& d_i& d_i& e_i\end{array}\right),F_j^c=\left(\begin{array}{cccc}u_j^c& u_j^c& u_j^c& \nu _j^c\\ d_j^c& d_j^c& d_j^c& e_j^c\end{array}\right),`$ (33)
where $`i`$ and $`j`$ are family indices. In addition the left-handed quarks and leptons are assigned to be triplets, while the CP-conjugates of the right-handed quarks and leptons are singlets under an $`SO(3)`$ family symmetry,
$$F_i\mathrm{๐},F_j^c\mathrm{๐}.$$
(34)
This implies in particular that the right-handed neutrinos $`\nu _j^cF_j^c`$ are singlets under $`SO(3)`$, and the lepton electroweak doublets $`L_iF_i`$ are triplets under $`SO(3)`$, as assumed previously. The usual SUSY Higgs doublets $`H_u,H_d`$ are contained in a single PS bi-doublet $`h`$, and further heavy Higgs superfields $`H,\overline{H}`$ are introduced to break the Pati-Salam symmetry group down to the Standard Model . As in the $`SU(3)`$ model in , we include an adjoint $`\mathrm{\Sigma }`$ field which develops vevs in the $`SU(4)_{PS}\times SU(2)_R`$ direction which preserves the hypercharge generator $`Y=T_{3R}+(BL)/2`$. This implies that any coupling of the $`\mathrm{\Sigma }`$ to a fermion and a messenger such as $`\mathrm{\Sigma }_{b\beta }^{a\alpha }F_{a\alpha }^c\chi ^{b\beta }`$, where the $`SU(2)_R`$ and $`SU(4)_{PS}`$ indices have been displayed explicitly, is proportional to the hypercharge $`Y`$ of the particular fermion component of $`F^c`$ times the vev $`\sigma `$. For example the coupling of $`\mathrm{\Sigma }`$ to right-handed neutrinos gives zero. In addition to $`SO(3)\times G_{PS}`$, the flavour symmetry group also includes $`R\times Z_4^2\times Z_3^2\times U(1)`$ symmetries in order to restrict the form of the mass matrices, where the continuous R-symmetry may be alternatively be replaced by a discrete $`Z_{2R}`$ symmetry. The superfields transform under the full symmetry group of the model as shown in Table 1.
We need spontaneous breaking of the family symmetry
$$SO(3)SO(2)\mathrm{Nothing}$$
(35)
To achieve this symmetry breaking we introduce additional flavon fields $`\varphi _i`$, $`\varphi _{23},\varphi _{123}`$ in the representations given in Table 1. The vacuum alignment of the flavon VEVs plays a crucial role in this model. In the $`SO(3)`$ model the flavon VEVs are all real and, as discussed in Appendix B, may be aligned in the following way:
$$\begin{array}{c}\varphi _1\left(\begin{array}{c}1\\ 0\\ 0\end{array}\right),\varphi _2\left(\begin{array}{c}0\\ 1\\ 0\end{array}\right),\varphi _3\left(\begin{array}{c}0\\ 0\\ 1\end{array}\right),\varphi _{23}\left(\begin{array}{c}0\\ 1\\ 1\end{array}\right),\varphi _{123}\left(\begin{array}{c}1\\ 1\\ 1\end{array}\right).\hfill \end{array}$$
(36)
The leading operators allowed by the symmetries are
$`W_{\mathrm{Yuk}}`$ $`=`$ $`{\displaystyle \frac{y_1}{M^3}}(F.\varphi _{23})\varphi _{23}^2F_1^ch`$ (37)
$`+`$ $`{\displaystyle \frac{y_2}{M^3}}(F.\varphi _{123})\varphi _{123}^2F_2^ch+{\displaystyle \frac{y_2^{}\mathrm{\Sigma }}{M^2}}(F.\varphi _{23})F_2^ch`$ (38)
$`+`$ $`{\displaystyle \frac{y_3}{M}}(F.\varphi _3)F_3^ch+{\displaystyle \frac{y_3^{}}{M^3}}(F.\varphi _2)\varphi _2^2F_3^ch+{\displaystyle \frac{y_3^{\prime \prime }}{M^3}}(F.\varphi _1)\varphi _1^2F_3^ch`$ (39)
$`W_{\mathrm{Maj}}`$ $``$ $`{\displaystyle \frac{1}{M}}(F_3^cH)^2`$ (40)
$`+`$ $`{\displaystyle \frac{1}{M^7}}(F_2^cH)^2(\varphi _{123}^2\varphi _{23}^4+\varphi _{123}^6)`$ (41)
$`+`$ $`{\displaystyle \frac{1}{M^7}}(F_1^cH)^2(\varphi _{23}^6+\varphi _{23}^2\varphi _{123}^4)`$ (42)
$`+`$ $`{\displaystyle \frac{1}{M^5}}(F_2^cH)(F_3^cH)\varphi _{123}^3\varphi _3`$ (43)
$`+`$ $`{\displaystyle \frac{1}{M^5}}(F_1^cH)(F_3^cH)\varphi _{23}^3\varphi _3`$ (44)
$`+`$ $`{\displaystyle \frac{1}{M^7}}(F_1^cH)(F_2^cH)\varphi _{23}^3\varphi _{123}^3,`$ (45)
where we have included complex Yukawa couplings $`y_i=|y_i|e^{i\delta _i}`$ in the Yukawa superpotential, but have suppressed similar Yukawa couplings which would appear multiplying the Majorana operators.
In order to obtain the Yukawa matrices from $`W_{\mathrm{Yuk}}`$ and the Majorana matrix from $`W_{\mathrm{Maj}}`$ requires some discussion of the messenger sector that is responsible for the operators above. This was fully discussed in , and we shall only briefly repeat the essential points here. The operators arise from Froggat-Nielsen diagrams and the scale $`M`$ represents the right-handed up and down messenger mass scales $`M^{u,d}`$, corresponding to the dominance of right-handed messengers over left-handed messengers, which applies if $`M<M^L`$ where $`M^L`$ represents the left-handed messenger mass scale. Specifically it was assumed that
$$M^d\frac{1}{3}M^uM^L.$$
(46)
It was further assumed that the right-handed lepton messenger scales satisfy the approximate $`SU(4)_{PS}`$ relations $`M^\nu M^u`$, and $`M^eM^d`$. The splitting of the messenger mass scales relies on left-right and $`SU(2)_R`$ breaking effects which was assumed to be due to a Wilson line symmetry breaking mechanism . Eq.46 then allows the expansion parameters associated with $`\varphi _{23}`$ to take the numerical values :
$$ฯต\frac{\varphi _{23}}{M^u}0.05,\overline{ฯต}\frac{\varphi _{23}}{M^d}0.15$$
(47)
where here and henceforth we assume that the fields have been replaced by their VEVs e.g. $`\varphi _{23}<\varphi _{23}>`$, etc. We shall also assume that the flavons $`\varphi _{123}`$ take similar VEVs:
$$\frac{\varphi _{123}}{M^u}ฯต,\frac{\varphi _{123}}{M^d}\overline{ฯต}$$
(48)
In the present model the flavons $`\varphi _{1,2}`$ lead to independent expansion parameters which we will assume to satisfy:
$$\frac{\varphi _1}{M^d}\overline{ฯต},\frac{\varphi _2}{M^d}\overline{ฯต}^{2/3}.$$
(49)
Eqs.47, 49 then imply:
$$\frac{\varphi _1}{M^u}ฯต,\frac{\varphi _2}{M^u}\frac{\overline{ฯต}^{2/3}}{3}0.7ฯต^{2/3}.$$
(50)
The flavon $`\varphi _3`$ transforms under $`SU(2)_R`$ as $`\mathrm{๐}\mathrm{๐}`$, and develops isospin breaking vevs in the up and down $`SU(2)_R`$ directions, and we assume as in :
$$\frac{\varphi _3^u}{M^u}=\frac{\varphi _3^d}{M^d}\sqrt{\overline{ฯต}}.$$
(51)
It remains to specify the expansion parameter associated with $`\sigma `$, the vev of $`\mathrm{\Sigma }`$. This was determined purely by phenomenological considerations in , and here we assume the same value:
$$Y(d)\frac{\sigma }{M^d}\overline{ฯต}.$$
(52)
Note that the operators involving $`\mathrm{\Sigma }`$ must be multiplied by the hypercharge of the relevant right-handed fermion, where $`Y(d)=1/3`$ is the hypercharge of $`d^c`$, $`Y(u)=2/3`$ is the hypercharge of $`u^c`$, $`Y(e)=1`$ is the hypercharge of $`e^c`$, and $`Y(\nu )=0`$ is the hypercharge of $`\nu ^c`$.
The operators in Eqs.37,38,39 when combined with the messenger sector just described, then leads to the Yukawa matrices:
$`Y_{LR}^U`$ $``$ $`\left(\begin{array}{ccc}0& y_2ฯต^3\hfill & \hfill y_3^{\prime \prime }ฯต^3\\ y_1ฯต^3& y_2ฯต^32y_2^{}ฯต^2\hfill & \hfill 0.34y_3^{}ฯต^2\\ y_1ฯต^3& y_2ฯต^3+2y_2^{}ฯต^2\hfill & \hfill y_3\overline{ฯต}^{\frac{1}{2}}\end{array}\right),`$ (56)
$`Y_{LR}^D`$ $``$ $`\left(\begin{array}{ccc}0& y_2\overline{ฯต}^3\hfill & \hfill y_3^{\prime \prime }\overline{ฯต}^3\\ y_1\overline{ฯต}^3& y_2\overline{ฯต}^3+y_2^{}\overline{ฯต}^2\hfill & \hfill y_3^{}\overline{ฯต}^2\\ y_1\overline{ฯต}^3& y_2\overline{ฯต}^3y_2^{}\overline{ฯต}^2\hfill & \hfill y_3\overline{ฯต}^{\frac{1}{2}}\end{array}\right),`$ (60)
$`Y_{LR}^E`$ $``$ $`\left(\begin{array}{ccc}0& y_2\overline{ฯต}^3\hfill & \hfill y_3^{\prime \prime }\overline{ฯต}^3\\ y_1\overline{ฯต}^3& y_2\overline{ฯต}^3+3y_2^{}\overline{ฯต}^2\hfill & \hfill y_3^{}\overline{ฯต}^2\\ y_1\overline{ฯต}^3& y_2\overline{ฯต}^33y_2^{}\overline{ฯต}^2\hfill & \hfill y_3\overline{ฯต}^{\frac{1}{2}}\end{array}\right),`$ (64)
$`Y_{LR}^\nu `$ $``$ $`\left(\begin{array}{ccc}0& y_2ฯต^3\hfill & \hfill y_3^{\prime \prime }ฯต^3\\ y_1ฯต^3& y_2ฯต^3\hfill & \hfill 0.34y_3^{}ฯต^2\\ y_1ฯต^3& y_2ฯต^3\hfill & \hfill y_3\overline{ฯต}^{\frac{1}{2}}\end{array}\right).`$ (68)
The leading corrections are given by additional operators similar to those displayed but with insertions of powers of $`\varphi _3^2/M^2\overline{ฯต}`$.
The leading heavy right-handed neutrino Majorana mass arises from the operator of Eq.40 which gives,
$$M_3\frac{<H>^2}{M^\nu },$$
(69)
to the third family, where $`M^\nu =M^u`$ is the same messenger mass scale as in the up sector due to $`SU(4)_{PS}`$. Operators involving $`\mathrm{\Sigma }`$ do not contribute since it does not couple to right-handed neutrinos which have zero hypercharge. However the other Majorana operators fill out the Majorana mass matrix, and after small angle right-handed rotations the Majorana matrix takes the form:
$$M_{RR}=\left(\begin{array}{ccc}pฯต^6& 0& \hfill 0\\ 0& qฯต^6& \hfill 0\\ 0& 0& \hfill 1\end{array}\right)M_3,$$
(70)
where $`p,q`$ are complex couplings.
## 4 Predictions for Neutrino Parameters
The neutrino Yukawa matrix in Eq.68 has the CSD form considered in Eqs. 25,26 and, provided the SD conditions in Eq.9 are satisfied, it will lead to tri-bimaximal neutrino mixing. The complex phases in $`M_{RR}`$ in Eq.70 may be removed by rotations on the right-handed neutrino fields, which only results in a redefinition of the Yukawa phases appearing in the complex Yukawa couplings $`y_i=|y_i|e^{i\delta _i}`$. Effectively, then, the Majorana masses in $`M_{RR}`$ may be taken to be real without loss of generality. The SD conditions in Eq.9 are then satisfied providing:
$$\frac{|y_1^2|}{p}\frac{|y_2^2|}{q}|y_3^2|\overline{ฯต}.$$
(71)
Since the Yukawa couplings are expected to be of order unity, the model predicts a rather mild hierarchy in physical neutrino masses, from Eqs.10, 11,12:
$`m_1`$ $``$ $`{\displaystyle \frac{|y_3^2|\overline{ฯต}}{M_3}}v_u^2`$ (72)
$`m_2`$ $``$ $`{\displaystyle \frac{3|y_2^2|}{qM_3}}v_u^2`$ (73)
$`m_3`$ $``$ $`{\displaystyle \frac{2|y_1^2|}{pM_3}}v_u^2`$ (74)
The neutrino mixing angles take the tri-bimaximal values:
$`\mathrm{tan}\theta _{23}^\nu `$ $``$ $`1`$ (75)
$`\mathrm{tan}\theta _{12}^\nu `$ $``$ $`{\displaystyle \frac{1}{\sqrt{2}}}`$ (76)
$`\theta _{13}^\nu `$ $``$ $`0`$ (77)
since the Yukawa matrix in Eq.68 has the CSD form considered in Eqs. 25,26, as already discussed. The neutrino auxiliary phases then take the values given in Eq.27, where the phases $`\delta _i`$ refer to the phases of the Yukawa couplings $`y_i=|y_i|e^{i\delta _i}`$ (assuming without loss of generality real $`p,q`$). From Eq.27 and Eqs. A.33-A.35 we obtain the neutrino phases:
$`\delta _{12}^{\nu _L}`$ $`=`$ $`(\delta _3\delta _2)`$ (78)
$`\delta _{13}^{\nu _L}`$ $`=`$ $`(\delta _3\delta _1)`$ (79)
$`\delta _{23}^{\nu _L}`$ $`=`$ $`3(\delta _1\delta _2).`$ (80)
The lepton mixing angles receive corrections from the charged lepton sector which in this model are completely derivable from the charged lepton Yukawa matrix in Eq.64, using the results in Appendix A. The charged lepton Yukawa matrix in Eq.64 leads to charged lepton masses in the ratios:
$`m_e:m_\mu :m_\tau `$ $``$ $`{\displaystyle \frac{|y_1||y_2|}{3|y_2^{}|}}\overline{ฯต}^4:3|y_2^{}|\overline{ฯต}^2:y_{33}`$ (81)
In a small charged lepton angle approximation,
$`\theta _{23}^{E_L}`$ $``$ $`{\displaystyle \frac{|y_3^{}|\overline{ฯต}^2}{y_{33}}}`$ (82)
$`\theta _{13}^{E_L}`$ $``$ $`{\displaystyle \frac{|y_3^{\prime \prime }|\overline{ฯต}^3}{y_{33}}}`$ (83)
$`\theta _{12}^{E_L}`$ $``$ $`{\displaystyle \frac{|y_2|\overline{ฯต}}{3|y_2^{}|}}`$ (84)
where we have written $`y_{33}=|y_3|\overline{ฯต}^{\frac{1}{2}}`$. The auxiliary charged lepton phases, used to make the charged lepton mixing angles real and positive, are:
$`\varphi _2^{E_L}`$ $`=`$ $`\delta _3^{}\delta _3^{\prime \prime },`$ (85a)
$`\varphi _3^{E_L}`$ $`=`$ $`\delta _3\delta _3^{\prime \prime },`$ (85b)
$`\chi ^{E_L}`$ $`=`$ $`\delta _2^{}\delta _2\delta _3^{}+\delta _3^{\prime \prime }`$ (85c)
$`\omega _i^{E_L}`$ are undetermined and are used to remove phases from the MNS matrix. From Eq.27,85 and Eqs. A.36-A.38 we obtain the charged lepton phases:
$`\delta _{23}^{E_L}`$ $`=`$ $`(\delta _3\delta _3^{})\pi 3(\delta _1\delta _2)`$ (86)
$`\delta _{13}^{E_L}`$ $`=`$ $`(\delta _3\delta _3^{\prime \prime })+(\delta _3\delta _1)\pi 2(\delta _1\delta _2)`$ (87)
$`\delta _{12}^{E_L}`$ $`=`$ $`(\delta _2^{}\delta _2)+(\delta _3\delta _2)`$ (88)
The leading charged lepton corrections to the MNS angles and phases are given from Eqs.A.30-A.32:
$`s_{23}e^{i\delta _{23}}`$ $``$ $`e^{i\delta _{23}^{\nu _L}}\left[s_{23}^{\nu _L}\theta _{23}^{E_L}c_{23}^{\nu _L}e^{i(\delta _{23}^{E_L}\delta _{23}^{\nu _L})}\right]`$ (89)
$`\theta _{13}e^{i\delta _{13}}`$ $``$ $`\theta _{12}^{E_L}s_{23}^{\nu _L}e^{i(\delta _{23}^{\nu _L}+\delta _{12}^{E_L})}`$ (90)
$`s_{12}e^{i\delta _{12}}`$ $``$ $`e^{i\delta _{12}^{\nu _L}}\left[s_{12}^{\nu _L}\theta _{12}^{E_L}c_{23}^{\nu _L}c_{12}^{\nu _L}e^{i(\delta _{12}^{E_L}\delta _{12}^{\nu _L})}\right]`$ (91)
where we have kept the leading charged lepton correction in each term. From Eqs.89-91, we see that the lepton phases are approximately given by:
$`\delta _{23}`$ $``$ $`\delta _{23}^{\nu _L}`$ (92)
$`\delta _{13}`$ $``$ $`\delta _{23}^{\nu _L}+\delta _{12}^{E_L}+\pi `$ (93)
$`\delta _{12}`$ $``$ $`\delta _{12}^{\nu _L}`$ (94)
and hence $`\delta `$, the MNS CP phase relevant for neutrino oscillations, is given by
$$\delta =\delta _{13}\delta _{23}\delta _{12}\delta _{12}^{E_L}\delta _{12}^{\nu _L}+\pi \delta _2^{}\delta _2+\pi .$$
(95)
It is remarkable to observe that the phase appearing in the leading charged lepton correction to the solar angle in Eq.91 is just equal to $`\delta \pi `$, where $`\delta `$ is the MNS phase. From Eqs.89-91 and the phases in Eqs. 78-80 and Eqs. 86-88 and the tri-bimaximal neutrino mixing angles in Eqs. 75-77, the lepton mixing angles are given by:
$`s_{23}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(1+\theta _{23}^{E_L}\mathrm{cos}(\delta _3\delta _3^{})\right)`$ (96)
$`\theta _{13}`$ $``$ $`{\displaystyle \frac{\theta _{12}^{E_L}}{\sqrt{2}}}`$ (97)
$`s_{12}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{3}}}\left(1\theta _{12}^{E_L}\mathrm{cos}(\delta \pi )\right)`$ (98)
We now turn to the quark sector. The quark Yukawa matrices in Eqs.56,60 lead to quark masses in the ratios:
$`m_d:m_s:m_b`$ $``$ $`{\displaystyle \frac{|y_1||y_2|}{|y_2^{}|}}\overline{ฯต}^4:|y_2^{}|\overline{ฯต}^2:y_{33}`$ (99)
$`m_u:m_c:m_t`$ $``$ $`{\displaystyle \frac{|y_1||y_2|}{2|y_2^{}|}}ฯต^4:2|y_2^{}|ฯต^2:y_{33}`$ (100)
Comparing the down masses in Eq.99 to the charged lepton masses in Eq.81 we see the expected GJ relations (valid at the GUT scale):
$$\frac{m_e}{m_d}=\frac{1}{3},\frac{m_\mu }{m_s}=3,\frac{m_\tau }{m_b}=1.$$
(101)
Using the conventions in Appendix A, the quark Yukawa matrices in Eqs.56,60 lead to mixing angles:
$`\theta _{23}^{U_L}`$ $``$ $`{\displaystyle \frac{0.34|y_3^{}|ฯต^2}{y_{33}}}`$ (102)
$`\theta _{13}^{U_L}`$ $``$ $`{\displaystyle \frac{|y_3^{\prime \prime }|ฯต^3}{y_{33}}}`$ (103)
$`\theta _{12}^{U_L}`$ $``$ $`{\displaystyle \frac{|y_2|ฯต}{2|y_2^{}|}}`$ (104)
$`\theta _{23}^{D_L}`$ $``$ $`{\displaystyle \frac{|y_3^{}|\overline{ฯต}^2}{y_{33}}}`$ (105)
$`\theta _{13}^{D_L}`$ $``$ $`{\displaystyle \frac{|y_3^{\prime \prime }|\overline{ฯต}^3}{y_{33}}}`$ (106)
$`\theta _{12}^{D_L}`$ $``$ $`{\displaystyle \frac{|y_2|\overline{ฯต}}{|y_2^{}|}}`$ (107)
where we have written $`y_{33}=|y_3|\overline{ฯต}^{\frac{1}{2}}`$, and we have used a small quark angle approximation. The auxiliary quark phases, used to make the quark mixing angles real and positive, are exactly the same as the charged lepton phases in Eq.85, except that $`\chi ^{U_L}`$ has an additional phase of $`\pi `$ resulting from the negative 22 element of the up quark Yukawa matrix. Using the results in Appendix A we find the CKM angles and phase:
$`\theta _{23}^{\mathrm{CKM}}`$ $``$ $`\theta _{23}^{D_L}\theta _{23}^{U_L}`$ (108)
$`\theta _{13}^{\mathrm{CKM}}`$ $``$ $`\theta _{13}^{D_L}`$ (109)
$`\theta _{12}^{\mathrm{CKM}}`$ $``$ $`\theta _{12}^{D_L}+\theta _{12}^{U_L}`$ (110)
$`\delta ^{\mathrm{CKM}}`$ $``$ $`(\delta _2^{}\delta _2)+(\delta _3^{}\delta _3^{\prime \prime })`$ (111)
The CKM phase is equal to the MNS phase in Eq.95 plus a second independent phase determined by elements in the third column of the Yukawa matrix which are irrelevant for neutrino mixing.
Since the CKM angles are approximately given by the down quark mixing angles, using Eqs.108-110, Eqs.105-107, Eqs.82-84, we may relate the charged lepton mixing angles to the CKM angles,
$`\theta _{23}^{E_L}`$ $``$ $`\theta _{23}^{D_L}\theta _{23}^{\mathrm{CKM}}`$ (112)
$`\theta _{13}^{E_L}`$ $``$ $`\theta _{13}^{D_L}\theta _{13}^{\mathrm{CKM}}`$ (113)
$`\theta _{12}^{E_L}`$ $``$ $`{\displaystyle \frac{1}{3}}\theta _{12}^{D_L}{\displaystyle \frac{1}{3}}\theta _{12}^{\mathrm{CKM}}{\displaystyle \frac{1}{3}}\theta _C`$ (114)
where the factor of $`1/3`$ in Eq.114 originates from the GJ structure.
Using Eqs.112-114, the lepton mixing angle relations in Eqs.96-98 become
$`\theta _{13}`$ $``$ $`{\displaystyle \frac{\theta _C}{3\sqrt{2}}}`$ (115)
$`s_{12}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{3}}}\left(1{\displaystyle \frac{1}{3}}\theta _C\mathrm{cos}(\delta \pi )\right)`$ (116)
$`s_{23}`$ $``$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(1+\theta _{23}^{\mathrm{CKM}}\mathrm{cos}(\delta _3\delta _3^{})\right).`$ (117)
Eq.115 gives a prediction for the reactor angle:
$$\theta _{13}3.06^{},\mathrm{sin}\theta _{13}0.052,\mathrm{sin}^2\theta _{13}2.7\times 10^3,\mathrm{sin}^22\theta _{13}1.1\times 10^2$$
(118)
From Eqs.116,117 the deviations from tri-bimaximal mixing may be expressed as:
$`\left|s_{12}^21/3\right|`$ $``$ $`\left|(2/9)\theta _C\mathrm{cos}\delta \right|<0.050`$ (119)
$`\left|s_{23}^21/2\right|`$ $``$ $`\left|\theta _{23}^{\mathrm{CKM}}\mathrm{cos}(\delta _3\delta _3^{})\right|<0.042`$ (120)
Eq.117 may also be expressed as
$$\theta _{23}^{}45^{}+\theta _{23}^{\mathrm{CKM}}{}_{}{}^{}\mathrm{cos}(\delta _3\delta _3^{})$$
(121)
which shows that the atmospheric angle is maximal $`\theta _{23}=45^{}`$ up to a correction no larger than $`\theta _{23}^{\mathrm{CKM}}2.4^{}`$
$$\theta _{23}=45^{}\pm 2.4^{}.$$
(122)
Eq.116 leads to the tri-bimaximal complementarity relation in Eq.5:
$`\theta _{12}^{}+{\displaystyle \frac{\theta _C^{}}{3\sqrt{2}}}\mathrm{cos}(\delta \pi )`$ $``$ $`35.26^{}`$ (123)
Eq.123 shows that the solar angle takes its tri-bimaximal value $`\theta _{12}=35.26^{}`$ up to a correction no larger than $`\frac{1}{\sqrt{2}}\frac{\theta _\mathrm{C}}{3}3.06^{}`$,
$$\theta _{12}=35.26^{}\pm 3.06^{}.$$
(124)
From Eq.115 and Eq.123 we find the sum rule:
$`\theta _{12}^{}+\theta _{13}^{}\mathrm{cos}(\delta \pi )`$ $``$ $`35.26^{}`$ (125)
Using Eq.116 we can predict the neutrino oscillation phase $`\delta `$ from a future accurate measurement of the solar angle $`\theta _{12}`$:
$`\mathrm{cos}(\delta \pi )13.3(1\sqrt{3}s_{12})`$ (126)
or alternatively from Eq.125,
$`\mathrm{cos}(\delta \pi )`$ $``$ $`{\displaystyle \frac{35.26^{}\theta _{12}^{}}{3.06^{}}}`$ (127)
For example, from an accurate measurement of the solar angle of $`\theta _{12}=33^{}`$ we predict $`\mathrm{cos}(\delta \pi )=0.74`$ or $`\delta =222^{}`$.
The above results are subject to some theoretical corrections as follows. The renormalisation group running corrections in running from the GUT scale $`M_X`$ to $`M_Z`$ depend strongly on $`\mathrm{tan}\beta `$ but may be typically estimated for $`\mathrm{tan}\beta 40`$ as <sup>4</sup><sup>4</sup>4These estimates have been made using the software packages REAP/MPT introduced in :
$`\theta _{12}(M_Z)\theta _{12}(M_X)`$ $``$ $`1^{}`$ (128)
$`\theta _{13}(M_Z)\theta _{13}(M_X)`$ $``$ $`0.5^{}`$ (129)
$`\theta _{23}(M_Z)\theta _{23}(M_X)`$ $``$ $`2^{}`$ (130)
In addition there is some theoretical error in the predictions of a similar magnitude due to the analytic formulae used, the small angle approximations, and the subleading operator corrections.
## 5 Conclusions
The poorly determined MNS parameters, when compared to the accuracy of the measured quark mixing angles, presents an opportunity to make testable predictions in the lepton sector by relating the lepton mixing parameters to the quark ones. In this paper we have shown how the neutrino mixing angles and oscillation phase can be predicted from tri-bimaximal neutrino mixing, corrected by charged lepton mixing angles which we relate to quark mixing angles via quark-lepton unification. The resulting predictions provide a probe of the high energy structure of unified theories.
We have shown how tri-bimaximal neutrino mixing can originate from the see-saw mechanism using sequential dominance. We gave the conditions for tri-bimaximal neutrino mixing to originate from sequential dominance, thereby providing a general and natural framework for this approach called constrained sequential dominance (CSD). We discussed a realisation of CSD based on $`SO(3)`$ family symmetry and vacuum alignment, although there are other examples that are possible. We then constructed a realistic model of quark and lepton masses and mixings based on the $`SO(3)`$ family symmetry and vacuum alignment, together with quark-lepton unification arising from a Pati-Salam gauge group. With the ingredients of tri-bimaximal complementarity in place, the MNS parameters were then predicted in terms of the CKM parameters. Although these predictions have been derived for the specific model presented, we would expect them to apply to a more general class of models based on real vacuum alignment which lead to tri-bimaximal complementarity.
The atmospheric angle is predicted to be approximately maximal $`\theta _{23}=45^{}`$, corrected by the quark mixing angle $`\theta _{23}^{\mathrm{CKM}}2.4^{}`$, with the correction controlled by an undetermined phase in the quark sector. The solar angle is predicted by the tri-bimaximal complementarity relation: $`\theta _{12}+\frac{1}{\sqrt{2}}\frac{\theta _\mathrm{C}}{3}\mathrm{cos}(\delta \pi )35.26^{}`$, where $`\theta _\mathrm{C}`$ is the Cabibbo angle and $`\delta `$ is the neutrino oscillation phase. The reactor angle is predicted to be $`\theta _{13}\frac{1}{\sqrt{2}}\frac{\theta _\mathrm{C}}{3}3.06^{}`$. The neutrino oscillation phase $`\delta `$ is predicted in terms of the solar angle to be $`\mathrm{cos}(\delta \pi )(35.26^{}\theta _{12}^{})/3.06^{}`$. These predictions can all be tested by future high precision neutrino oscillation experiments. Indeed the link between low energy neutrino parameters and quark-lepton unification provides a powerful theoretical motivation for performing high precision neutrino oscillation experiments. In particular the prediction of the neutrino oscillation phase in terms of the solar angle is a remarkable result, which motivates an accurate measurement of the solar angle. The theoretical prediction for the reactor angle could be tested with the next generation of superbeam or reactor experiments, and the prediction for the oscillation phase could be accurately tested at a Neutrino Factory.
## Acknowledgements
I would like to thank Stefan Antusch, for helpful discussions.
## Appendix
## Appendix A Conventions and Mixing Formalism
We shall use the conventions defined in . The Dirac mass matrices of the charged leptons and neutrinos are given by $`m_{LR}^\mathrm{E}=Y_{LR}^\mathrm{E}v_\mathrm{d}`$, and $`m_{LR}^\nu =Y_{LR}^\nu v_\mathrm{u}`$ where $`v_\mathrm{d}=h_\mathrm{d}^0`$ and $`v_\mathrm{u}=h_\mathrm{u}^0`$, and the Lagrangian is of the form $`=\overline{\psi }_LY_{LR}h\psi _R+H.c.`$ The neutrino mass matrix $`m_{LL}^\nu `$ is given by the type I see-saw mechanism as
$$m_{LL}^\nu =m_{LR}^\nu M_{RR}^1m_{LR}^{\nu T},$$
(A.1)
in terms of the Dirac neutrino mass matrix $`m_{LR}^\nu `$ and the heavy Majorana mass matrix $`M_{RR}`$. In this convention the effective Majorana masses are given by the Lagrangian $`=\overline{\nu }_Lm_{LL}^\nu \nu ^c+H.c.`$ The change from flavour basis to mass eigenbasis can be performed with the unitary diagonalization matrices $`V_{E_\mathrm{L}},V_{E_\mathrm{R}}`$ and $`V_{\nu _\mathrm{L}}`$ by
$`V_{E_\mathrm{L}}m_{LR}^\mathrm{E}V_{E_\mathrm{R}}^{}=(\begin{array}{ccc}m_e& 0& 0\\ 0& m_\mu & 0\\ 0& 0& m_\tau \end{array}),V_{\nu _\mathrm{L}}m_{\mathrm{LL}}^\nu V_{\nu _\mathrm{L}}^T=(\begin{array}{ccc}m_1& 0& 0\\ 0& m_2& 0\\ 0& 0& m_3\end{array}).`$ (A.8)
The MNS matrix is then given by
$`U_{\mathrm{MNS}}=V_{e_\mathrm{L}}V_{\nu _\mathrm{L}}^{}.`$ (A.9)
We use the parameterization $`U_{\mathrm{MNS}}=U_{23}U_{13}U_{12}`$ with $`U_{23},U_{13},U_{12}`$ being defined as
$`U_{12}=\left(\begin{array}{ccc}c_{12}& s_{12}e^{i\delta _{12}}& 0\\ s_{12}e^{i\delta _{12}}& c_{12}& 0\\ 0& 0& 1\end{array}\right),`$ $`U_{13}=\left(\begin{array}{ccc}c_{13}& 0& s_{13}e^{i\delta _{13}}\\ 0& 1& 0\\ s_{13}e^{i\delta _{13}}& 0& c_{13}\end{array}\right),`$ (A.16)
$`U_{23}=\left(\begin{array}{ccc}1& 0& 0\\ 0& c_{23}& s_{23}e^{i\delta _{23}}\\ 0& s_{23}e^{i\delta _{23}}& c_{23}\end{array}\right)`$ (A.20)
where $`s_{ij}`$ and $`c_{ij}`$ stand for $`\mathrm{sin}(\theta _{ij})`$ and $`\mathrm{cos}(\theta _{ij})`$, respectively. $`\delta `$, the Dirac CP phase relevant for neutrino oscillations, is given by $`\delta =\delta _{13}\delta _{23}\delta _{12}`$.
The MNS matrix is thus constructed as a product of a unitary matrix from the charged lepton sector $`V^{E_L}`$ and a unitary matrix from the neutrino sector $`V_{}^{\nu _L}{}_{}{}^{}`$. Each of these unitary matrices may be parametrised as:
$$V^{}=P_2R_{23}R_{13}P_1R_{12}P_3$$
(A.21)
where $`R_{ij}`$ are a sequence of real rotations corresponding to the Euler angles $`\theta _{ij}`$, and $`P_i`$ are diagonal phase matrices. The Euler matrices are given by
$$R_{23}=\left(\begin{array}{ccc}1& 0& 0\\ 0& c_{23}& s_{23}\\ 0& s_{23}& c_{23}\end{array}\right)$$
(A.22)
$$R_{13}=\left(\begin{array}{ccc}c_{13}& 0& s_{13}\\ 0& 1& 0\\ s_{13}& 0& c_{13}\end{array}\right)$$
(A.23)
$$R_{12}=\left(\begin{array}{ccc}c_{12}& s_{12}& 0\\ s_{12}& c_{12}& 0\\ 0& 0& 1\end{array}\right)$$
(A.24)
where $`c_{ij}=\mathrm{cos}\theta _{ij}`$ and $`s_{ij}=\mathrm{sin}\theta _{ij}`$. The phase matrices are given by
$$P_1=\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\chi }& 0\\ 0& 0& 1\end{array}\right)$$
(A.25)
$$P_2=\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\varphi _2}& 0\\ 0& 0& e^{i\varphi _3}\end{array}\right)$$
(A.26)
$$P_3=\left(\begin{array}{ccc}e^{i\omega _1}& 0& 0\\ 0& e^{i\omega _2}& 0\\ 0& 0& e^{i\omega _3}\end{array}\right)$$
(A.27)
Thus we write
$$V_{}^{\nu _L}{}_{}{}^{}=P_2^{\nu _L}R_{23}^{\nu _L}R_{13}^{\nu _L}P_1^{\nu _L}R_{12}^{\nu _L}P_3^{\nu _L}$$
(A.28)
$$V_{}^{E_L}{}_{}{}^{}=P_2^{E_L}R_{23}^{E_L}R_{13}^{E_L}P_1^{E_L}R_{12}^{E_L}P_3^{E_L}$$
(A.29)
in terms of independent angles and phases for the left-handed neutrino and charged lepton sectors distinguished by the superscripts $`\nu _L`$ and $`E_L`$.
The MNS matrix can be expanded in terms of neutrino and charged lepton mixing angles and phases to leading order in the charged lepton mixing angles which are assumed small: <sup>5</sup><sup>5</sup>5Note that the sign of the last term in Eq.A.31 is reversed compared to the results quoted in . I am grateful to Stefan Antusch for correcting these results.
$`s_{23}e^{i\delta _{23}}`$ $``$ $`s_{23}^{\nu _L}e^{i\delta _{23}^{\nu _L}}\theta _{23}^{E_L}c_{23}^{\nu _L}e^{i\delta _{23}^{E_L}}`$ (A.30)
$`\theta _{13}e^{i\delta _{13}}`$ $``$ $`\theta _{13}^{\nu _L}e^{i\delta _{13}^{\nu _L}}\theta _{13}^{E_L}c_{23}^{\nu _L}e^{i\delta _{13}^{E_L}}\theta _{12}^{E_L}s_{23}^{\nu _L}e^{i(\delta _{23}^{\nu _L}\delta _{12}^{E_L})}`$ (A.31)
$`s_{12}e^{i\delta _{12}}`$ $``$ $`s_{12}^{\nu _L}e^{i\delta _{12}^{\nu _L}}+\theta _{13}^{E_L}c_{12}^{\nu _L}s_{23}^{\nu _L}e^{i(\delta _{23}^{\nu _L}\delta _{13}^{E_L})}\theta _{12}^{E_L}c_{23}^{\nu _L}c_{12}^{\nu _L}e^{i\delta _{12}^{E_L}}`$ (A.32)
where
$`\delta _{12}^{\nu _L}`$ $`=`$ $`\omega _1^{\nu _L}\omega _2^{\nu _L}`$ (A.33)
$`\delta _{13}^{\nu _L}`$ $`=`$ $`\omega _1^{\nu _L}\omega _3^{\nu _L}`$ (A.34)
$`\delta _{23}^{\nu _L}`$ $`=`$ $`\chi ^{\nu _L}+\omega _2^{\nu _L}\omega _3^{\nu _L}`$ (A.35)
$`\delta _{23}^{E_L}`$ $`=`$ $`\varphi _2^{E_L}+\varphi _3^{E_L}+\varphi _2^{\nu _L}\varphi _3^{\nu _L}+\chi ^{\nu _L}+\omega _2^{\nu _L}\omega _3^{\nu _L}`$ (A.36)
$`\delta _{13}^{E_L}`$ $`=`$ $`\varphi _3^{E_L}\varphi _3^{\nu _L}+\omega _1^{\nu _L}\omega _3^{\nu _L}`$ (A.37)
$`\delta _{12}^{E_L}`$ $`=`$ $`\chi ^{E_L}+\varphi _2^{E_L}\varphi _2^{\nu _L}\chi ^{\nu _L}+\omega _1^{\nu _L}\omega _2^{\nu _L}`$ (A.38)
In the quark sector an analagous procedure is followed. The Dirac mass matrices of the quarks are given by $`m_{LR}^\mathrm{D}=Y_{LR}^\mathrm{D}v_\mathrm{d}`$, and $`m_{LR}^\mathrm{U}=Y_{LR}^\mathrm{U}v_\mathrm{u}`$. The change from flavour basis to mass eigenbasis can be performed with the unitary diagonalization matrices $`V_{D_\mathrm{L}},V_{D_\mathrm{R}}`$ and $`V_{U_\mathrm{L}},V_{U_\mathrm{R}}`$ by
$`V_{D_\mathrm{L}}m_{LR}^\mathrm{D}V_{D_\mathrm{R}}^{}=(\begin{array}{ccc}m_d& 0& 0\\ 0& m_s& 0\\ 0& 0& m_b\end{array}),V_{U_\mathrm{L}}m_{LR}^\mathrm{U}V_{U_\mathrm{R}}^{}=(\begin{array}{ccc}m_u& 0& 0\\ 0& m_c& 0\\ 0& 0& m_t\end{array}),`$ (A.45)
The CKM matrix is then given by
$`U_{\mathrm{CKM}}=V_{U_\mathrm{L}}V_{D_\mathrm{L}}^{}.`$ (A.46)
We use the standard parameterization $`U_{\mathrm{CKM}}=R_{23}^{\mathrm{CKM}}U_{13}^{\mathrm{CKM}}R_{12}^{\mathrm{CKM}}`$ where we label the quark parameters as CKM to distinguish them from the (unlabelled) lepton mixing angles. If the CKM angles are given predominantly by the down mixing angles, then we may use the analagous results to those quoted above to obtain the corrections coming from the up sector. Thus for example the analagous relations to Eq.A.30-A.38 apply in the quark sector also, with the replacements $`\nu D`$ and $`EU`$. In the quark sector the phases $`\omega _i^{D_L}`$, $`\omega _i^{U_L}`$ are all undetermined and are used to remove phases from the MNS matrix. In particular $`\omega _i^{D_L}`$ may be used to set the phases $`\delta _{12}^{CKM}=\delta _{23}^{CKM}=0`$, with a single CKM phase remaining, $`\delta ^{\mathrm{CKM}}=\delta _{13}^{\mathrm{CKM}}`$.
## Appendix B Vacuum Alignment
In order to achieve the desired vacuum alignment in this model , we shall introduce the following additional superpotential terms:
$`W_{\mathrm{SB}}`$ $``$ $`A(\varphi _1^2\mathrm{\Lambda }_1^2)+B(\varphi _2^2\mathrm{\Lambda }_2^2)+C(\varphi _3^2\mathrm{\Lambda }_3^2)`$ (B.47)
$`+`$ $`D\varphi _1.\varphi _2+E\varphi _1.\varphi _3+F\varphi _2.\varphi _3`$ (B.48)
$`+`$ $`L\varphi _{12}.\stackrel{~}{\varphi }_{12}+M\varphi _{23}.\stackrel{~}{\varphi }_{23}`$ (B.49)
$`+`$ $`N\varphi _{123}.\varphi _{12}+O\varphi _{123}.\varphi _{23}`$ (B.50)
$`+`$ $`P((\varphi _{12}.\varphi _1)(\varphi _{12}.\varphi _2)\mathrm{\Lambda }_9^2)+Q((\stackrel{~}{\varphi }_{12}.\varphi _1)(\stackrel{~}{\varphi }_{12}.\varphi _2)\mathrm{\Lambda }_{10}^2)`$ (B.51)
$`+`$ $`R((\varphi _{23}.\varphi _2)(\varphi _{23}.\varphi _3)\mathrm{\Lambda }_{11}^2)+S((\stackrel{~}{\varphi }_{23}.\varphi _2)(\stackrel{~}{\varphi }_{23}.\varphi _3)\mathrm{\Lambda }_{12}^2)`$ (B.52)
$`+`$ $`T((\varphi _{123}.\varphi _1)(\varphi _{123}.\varphi _2)(\varphi _{123}.\varphi _3)\mathrm{\Lambda }_{13}^2)`$ (B.53)
where $`A,\mathrm{}T`$ are $`SO(3)`$ and Pati-Salam singlet superfields and $`\mathrm{\Lambda }_i`$ are independent heavy mass scales which we regard as arising from the VEVs of some $`SO(3)`$ singlet fields. Such VEVs could arise from some radiative symmetry breaking mechanism, for example . $`\mathrm{\Lambda }_i^2`$ can be taken to be real and positive by a suitable phase choice for the fields $`A,\mathrm{}T`$ . Note that in such an $`SO(3)`$ theory with real VEVs all the D-terms will be automatically zero, and the vacuum alignment is then achieved purely from the F-terms being minimised to zero, up to soft supersymmetry breaking perturbations. It is straightforward to deduce the required quantum numbers of these superfields under the symmetry group $`R\times Z_4^2\times Z_3^2\times U(1)`$ from the quantum number assignments of the flavons in Table 1, and the requirement that the superpotential terms given above are allowed.
The potential consists of F-terms of the form $`|F_X|^2`$, together with positive soft mass squareds for the flavon fields. Since $`\mathrm{\Lambda }_i^2`$ are real and positive this results in real VEVs as discussed in , which greatly simplifies the analysis, and crucially restricts the number of undetermined phases in the analysis, ultimately leading to a prediction for the neutrino oscillation phase. The purpose of the terms in Eq.B.47,B.48 is for the F-terms $`|F_X|^2`$, with $`X=A,\mathrm{}F`$, to be minimised by real three orthogonal VEVs for $`\varphi _{1,2,3}`$ of the form given in Eq.36. In particular the terms in Eq.B.47 drive the VEVs to be non-zero, and the the terms in Eq.B.48 together with the soft positive mass squareds then lead to real and orthogonal VEVs of the form given in Eq.36. The purpose of the remaining terms is to align the VEVs of the remaining fields relative to these basis vectors, in order to achieve the alignment for $`\varphi _{23},\varphi _{123}`$ as shown in Eq.36, as follows.
To achieve the alignment of $`\varphi _{23}`$ requires two fields $`\varphi _{23},\stackrel{~}{\varphi }_{23}`$. The purpose of the terms in Eq.B.52 is to drive non-zero VEVs for these two fields in the $`(2,3)`$ directions, and taken together with the positive soft mass squared terms, <sup>6</sup><sup>6</sup>6I am grateful to Graham Ross and Ivo de Medeiros-Varzielas for suggesting the use of soft mass terms to achieve this alignment. The use of soft mass terms for alignment is also discussed in . lead to a potential which is minimised when magnitudes of the VEVs along each of the directions is equal, for each of the two fields $`\varphi _{23},\stackrel{~}{\varphi }_{23}`$ separately. This is because for each of these flavons the potential takes the form $`V=m_{soft}^2(y^2+z^2)+b^2(yzM_2^2)^2`$, where all parameters are real and positive and $`y,z`$ are the components of the VEVs along the $`2,3`$ directions respectively. Such a potential is minimised by component VEVs of equal magnitude $`y^2=z^2`$. The term in Eq.B.49 proportional to $`M`$ then ensures that the two VEVs are orthogonal to each other, and without loss of generality this results in VEVs of the form:
$$\begin{array}{c}\stackrel{~}{\varphi }_{23}\left(\begin{array}{c}0\\ 1\\ 1\end{array}\right),\varphi _{23}\left(\begin{array}{c}0\\ 1\\ 1\end{array}\right).\hfill \end{array}$$
(B.54)
Following analagous arguments, the terms in Eq.B.51, together with positive soft mass terms and the term proportional to $`L`$ in Eq.B.49 leads, without loss of generality, to VEVs of the form:
$$\begin{array}{c}\stackrel{~}{\varphi }_{12}\left(\begin{array}{c}1\\ 1\\ 0\end{array}\right),\varphi _{12}\left(\begin{array}{c}1\\ 1\\ 0\end{array}\right).\hfill \end{array}$$
(B.55)
The alignment of $`\varphi _{123}`$ is achieved by the term in Eq.B.53 which drives the VEV, and ensures that all three components of the VEV are non-zero, and taken together the soft mass terms and F-terms in the potential which result from these terms imply that the component VEVs must have equal magnitude. The terms in Eq.B.50 then align these components to be orthogonal to both $`\varphi _{12}`$ and $`\varphi _{23}`$, resulting in the alignment assumed in Eq.36:
$$\begin{array}{c}\varphi _{123}\left(\begin{array}{c}1\\ 1\\ 1\end{array}\right).\hfill \end{array}$$
(B.56)
|
warning/0506/math0506256.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Let
$$\mathrm{\Gamma }_n=\left\{P=(p_1,p_2,\mathrm{},p_n)\right|p_i>0,\underset{i=1}{\overset{n}{}}p_i=1\},n2,$$
be the set of all complete finite discrete probability distributions. There are many information and divergence measures exists in the literature on information theory and statistics. Some of them are symmetric with respect to probability distributions, while others are not. Here we have divided these measure in these two categories. Through out the paper it is under stood that the probability distributions $`P,Q\mathrm{\Gamma }_n`$.
### 1.1 Non-Symmetric Measures
Here we shall give some non-symmetric measures of information. The most famous among them are $`\chi ^2`$divergence and Kullback-Leibler relative information. We understand by non symmetric measures are those that are not symmetric with respect to probability distributions $`P,Q\mathrm{\Gamma }_n`$. These measures as follows.
$``$ $`\chi ^2`$Divergence (Pearson )
$$\chi ^2(P||Q)=\underset{i=1}{\overset{n}{}}\frac{(p_iq_i)^2}{q_i}=\underset{i=1}{\overset{n}{}}\frac{p_i^2}{q_i}1$$
(1)
$``$ Relative Information (Kullback and Leibler )
$$K(P||Q)=\underset{i=1}{\overset{n}{}}p_i\mathrm{ln}(\frac{p_i}{q_i})$$
(2)
$``$ Relative J-Divergence (Dragomir et al. )
$$D(P||Q)=\underset{i=1}{\overset{n}{}}(p_iq_i)\mathrm{ln}\left(\frac{p_i+q_i}{2q_i}\right)$$
(3)
$``$ Relative Jensen-Shannon divergence (Sibson )
$$F(P||Q)=\underset{i=1}{\overset{n}{}}p_i\mathrm{ln}\left(\frac{2p_i}{p_i+q_i}\right)$$
(4)
$``$ Relative arithmetic-geometric divergence (Taneja )
$$G(P||Q)=\underset{i=1}{\overset{n}{}}\left(\frac{p_i+q_i}{2}\right)\mathrm{ln}\left(\frac{p_i+q_i}{2p_i}\right)$$
(5)
### 1.2 Symmetric Measures of Information
Here we shall give some symmetric measures of information. Some of them can be obtained from subsection 1.1. These measures as follows.
$``$ Hellinger Discrimination (Hellinger )
$$h(P||Q)=1B(P||Q)=\frac{1}{2}\underset{i=1}{\overset{n}{}}(\sqrt{p}_i\sqrt{q}_i)^2.$$
(6)
where
$$B(P||Q)=\underset{i=1}{\overset{n}{}}\sqrt{p_iq_i}$$
(7)
is the well-known Bhattacharyya distance.
$``$ Triangular Discrimination (Dacunha-Castelle )
$$\mathrm{\Delta }(P||Q)=2[1W(P||Q)]=\underset{i=1}{\overset{n}{}}\frac{(p_iq_i)^2}{p_i+q_i}.$$
(8)
where
$$W(P||Q)=\underset{i=1}{\overset{n}{}}\frac{2p_iq_i}{p_i+q_i}.$$
(9)
is the well-known harmonic mean divergence.
$``$ Symmetric Chi-square Divergence (Dragomir et al. )
$`\mathrm{\Psi }(P||Q)`$ $`=\chi ^2(P||Q)+\chi ^2(Q||P)`$
$`={\displaystyle \underset{i=1}{\overset{n}{}}}{\displaystyle \frac{(p_iq_i)^2(p_i+q_i)}{p_iq_i}}.`$ (10)
$``$ J-divergence (Jeffreys ; Kullback and Leibler )
$`J(P||Q)`$ $`=K(P||Q)+K(Q||P)`$
$`=D(P||Q)+D(Q||P)`$
$`={\displaystyle \underset{i=1}{\overset{n}{}}}(p_iq_i)\mathrm{ln}({\displaystyle \frac{p_i}{q_i}}).`$ (11)
$``$ Jensen-Shannon divergence (Sibson ; Burbea and Rao )
$`I(P||Q)`$ $`={\displaystyle \frac{1}{2}}[F(P||Q)+F(Q||P)].`$
$`={\displaystyle \frac{1}{2}}\left[{\displaystyle \underset{i=1}{\overset{n}{}}}p_i\mathrm{ln}\left({\displaystyle \frac{2p_i}{p_i+q_i}}\right)+{\displaystyle \underset{i=1}{\overset{n}{}}}q_i\mathrm{ln}\left({\displaystyle \frac{2q_i}{p_i+q_i}}\right)\right].`$ (12)
$``$ Arithmetic-Geometric divergence (Taneja )
$`T(P||Q)`$ $`={\displaystyle \frac{1}{2}}[G(P||Q)+G(Q||P)].`$
$`={\displaystyle \underset{i=1}{\overset{n}{}}}\left({\displaystyle \frac{p_i+q_i}{2}}\right)\mathrm{ln}\left({\displaystyle \frac{p_i+q_i}{2\sqrt{p_iq_i}}}\right).`$ (13)
After simplification, we can write
$$J(P||Q)=4[I(P||Q)+T(P||Q)]$$
and
$$D(Q||P)=\frac{1}{2}[F(P||Q)+G(P||Q)].$$
The measures $`I(P||Q)`$, $`J(P||Q),T(P||Q)`$, $`D(P||Q),F(P||Q)`$ and $`G(P||Q)`$ can be written in terms of $`K(P||Q)`$ as follows:
$`I(P||Q)`$ $`={\displaystyle \frac{1}{2}}[K(P||{\displaystyle \frac{P+Q}{2}})+K(Q||{\displaystyle \frac{P+Q}{2}})],`$
$`J(P||Q)`$ $`=K(P||Q)+K(Q||P),`$
$`T(P||Q)`$ $`={\displaystyle \frac{1}{2}}[K({\displaystyle \frac{P+Q}{2}}||P)+K({\displaystyle \frac{P+Q}{2}}||Q)],`$
$`D(P||Q)`$ $`={\displaystyle \frac{1}{2}}[K(Q||{\displaystyle \frac{P+Q}{2}})+K({\displaystyle \frac{P+Q}{2}}||Q)],`$
$`F(P||Q)`$ $`=K(P||{\displaystyle \frac{P+Q}{2}})`$
and
$`G(P||Q)`$ $`=K({\displaystyle \frac{P+Q}{2}}||P).`$
respectively.
The following parallelogram identity is also famous in the literature :
$$K(P||U)+K(Q||U)=K(P||\frac{P+Q}{2})+K(Q||\frac{P+Q}{2})+2K(\frac{P+Q}{2}||U),$$
for all $`P,Q,U\mathrm{\Gamma }_n`$
Some studies on information and divergence measures can be seen in Taneja , , . Also see on line book by Taneja .
From the symmetric measures we observe that the measure (7) is a part of measure (6) and the measure (9) is a part of measure (8). Thus we have the six measures (6), (8), (10)-(13) symmetric with respect to probability distributions.
The following inequalities are already known:
$$\frac{1}{2}h(P||Q)\frac{1}{4}\mathrm{\Delta }(P||Q)h(P||Q),$$
(14)
$$\frac{1}{4}\mathrm{\Delta }(P||Q)h(P||Q)\frac{1}{16}\mathrm{\Psi }(P||Q)$$
(15)
$$\mathrm{\Delta }(P||Q)\frac{1}{2}J(P||Q)\frac{1}{4}\mathrm{\Psi }(P||Q)$$
(16)
and
$$\frac{1}{4}\mathrm{\Delta }(P||Q)I(P||Q)\frac{\mathrm{log}2}{2}\mathrm{\Delta }(P||Q).$$
(17)
The inequalities (14) are due to LeCam and Dacunha-Castelle . The inequalities (15) are due to Taneja . The inequalities (16) are due to Dragomir et al. . Finally, the inequalities (17) are due to Tops$`รธ`$e .
Recently, Taneja proved the following inequalities:
$$\frac{1}{4}\mathrm{\Delta }(P||Q)I(P||Q)h(P||Q)\frac{1}{8}J(P||Q)T(P||Q)\frac{1}{16}\mathrm{\Psi }(P||Q).$$
(18)
In this paper, our aim is to relate the non-symmetric divergence measures with the symmetric measures given by (8), (11)-(13). In order to obtain these relationship we shall use the idea of Csiszรกr f-divergence and in some cases making restrictions on the probability distributions.
## 2 $`f`$Divergence and Information Measures
Given a convex function $`f:(0,\mathrm{})`$, the $`f`$divergence measure introduced by Csiszรกr is given by
$$C_f(P||Q)=\underset{i=1}{\overset{n}{}}q_if\left(\frac{p_i}{q_i}\right),$$
(19)
where $`P,Q\mathrm{\Gamma }_n`$.
The following theorem is well known in the literature (ref. Csiszรกr ).
###### Theorem 2.1.
Let the function $`f:(0,\mathrm{})`$ is differentiable convex and normalized, i.e., $`f(1)=0`$, then the Csiszรกr $`f`$divergence, $`C_f(P||Q)`$, given by (19) is nonnegative and convex in the pair of probability distribution $`(P,Q)\mathrm{\Gamma }_n\times \mathrm{\Gamma }_n`$.
The following theorem is due to Dragomir . It gives bounds on Csiszรกr f - divergence.
###### Theorem 2.2.
(Dragomir ). Let $`f:_+`$ be differentiable convex and normalized i.e., $`f(1)=0`$. Then
$$0C_f(P||Q)E_{C_f}(P||Q)$$
(20)
where
$$E_{C_f}(P||Q)=\underset{i=1}{\overset{n}{}}(p_iq_i)f^{}(\frac{p_i}{q_i}),$$
(21)
for all $`P,Q\mathrm{\Gamma }_n`$.
Let $`P,Q\mathrm{\Gamma }_n`$ be such that there exists $`r,R`$ with $`0<r\frac{p_i}{q_i}R<\mathrm{}`$, $`i\{1,2,\mathrm{},n\}`$, then
$$0C_f(P||Q)A_{C_f}(r,R),$$
(22)
where
$$A_{C_f}(r,R)=\frac{1}{4}(Rr)\left[f^{}(R)f^{}(r)\right].$$
(23)
Further, if we suppose that $`0<r1R<\mathrm{}`$, $`rR`$, then
$$0C_f(P||Q)B_{C_f}(r,R),$$
(24)
where
$$B_{C_f}(r,R)=\frac{(R1)f(r)+(1r)f(R)}{Rr}.$$
(25)
Moreover, the following inequalities hold:
$$E_{C_f}(P||Q)A_{C_f}(r,R),$$
(26)
$$B_{C_f}(r,R)A_{C_f}(r,R)$$
(27)
and
$$0B_{C_f}(r,R)C_f(P||Q)A_{C_f}(r,R).$$
(28)
The inequalities (26) and (28) can be seen in Dragomir , while the inequality (27) can be proved easily.
The following theorem is due to Taneja . It relates two $`f`$divergence measures.
###### Theorem 2.3.
Let $`f_1,f_2:I_+`$ be two differentiable convex functions which are normalized, i.e., $`f_1(1)=f_2(1)=0`$ and suppose that:
(i) $`f_1`$and $`f_2`$ are twice differentiable on $`(r,R)`$;
(ii) there exists the real constants $`m,M`$ such that $`m<M`$ and
$$m\frac{f_1^{\prime \prime }(x)}{f_2^{\prime \prime }(x)}M,f_2^{\prime \prime }(x)>0,x(r,R)$$
then we have
$$mC_{f_2}(P||Q)C_{f_1}(P||Q)MC_{f_2}(P||Q).$$
(29)
## 3 Bounds on Divergence Measures
Based on Theorems 2.1 and 2.2, we have the particular cases for the measures given in Section 1. These particular cases are given as examples, where the following the expression is frequently used:
$$L_1^1(a,b)=\{\begin{array}{cc}\frac{\mathrm{ln}b\mathrm{ln}a}{ba},\hfill & ab\hfill \\ a\hfill & a=b\hfill \end{array}$$
(30)
for all $`a>0`$, $`b>0`$.
###### Example 3.1.
(Relative J-Divergence). Let us consider
$$f_D(x)=(x1)\mathrm{ln}\left(\frac{x+1}{2}\right),x(0,\mathrm{})$$
(31)
in (19), then we have $`C_f(P||Q)=D(P||Q)`$.
Moreover,
$$f_D^{}(x)=\frac{x1}{x+1}+\mathrm{ln}\left(\frac{x+1}{2}\right)$$
(32)
and
$$f_D^{\prime \prime }(x)=\frac{x+3}{(x+1)^2}.$$
(33)
In view of (31), (32), Theorems 2.1 and 2.2, we have the following bounds on relative J-divergence:
$$0D(P||Q)E_D(P||Q)A_D(r,R)$$
(34)
and
$$0D(P||Q)B_D(r,R)A_D(r,R),$$
(35)
where
$`E_D(P||Q)`$ $`=D(P||Q)+\mathrm{\Delta }(P||Q),`$
$`A_D(r,R)`$ $`={\displaystyle \frac{1}{4}}(Rr)^2\left[{\displaystyle \frac{2}{(R+1)(r+1)}}+L_1^1(r+1,R+1)\right]`$
and
$`B_D(r,R)`$ $`=(R1)(1r)L_1^1(r+1,R+1).`$
###### Example 3.2.
(Relative Jensen-Shannon divergence). Let us consider
$$f_F(x)=x\mathrm{ln}\left(\frac{2x}{x+1}\right)\frac{x1}{2},x(0,\mathrm{})$$
(36)
in (19), then we have $`C_f(P||Q)=F(P||Q)`$.
Moreover,
$$f_F^{}(x)=\frac{1}{2}\frac{x1}{x+1}+\mathrm{ln}\left(\frac{2x}{x+1}\right)$$
(37)
and
$$f_{F_2}^{\prime \prime }(x)=\frac{1}{x(x+1)^2}.$$
(38)
In view of (36), (37), Theorems 2.1 and 2.2, we have the following bounds on relative Jensen-Shannon divergence:
$$0F(P||Q)E_F(P||Q)A_F(r,R)$$
(39)
and
$$0F(P||Q)B_F(r,R)A_F(r,R),$$
(40)
where
$`E_F(P||Q)`$ $`=D(Q||P){\displaystyle \frac{1}{2}}\mathrm{\Delta }(P||Q),`$
$`A_F(r,R)`$ $`={\displaystyle \frac{1}{4}}{\displaystyle \frac{(Rr)^2}{(R+1)(r+1)}}\left[L_1^1({\displaystyle \frac{r}{r+1}},{\displaystyle \frac{R}{R+1}})1\right]`$
and
$`B_{F_2}(r,R)`$ $`={\displaystyle \frac{1}{(Rr)}}\left[R\mathrm{ln}\left({\displaystyle \frac{2R}{R+1}}\right)r\mathrm{ln}\left({\displaystyle \frac{2r}{r+1}}\right)\right]`$
$`{\displaystyle \frac{rR}{(R+1)(r+1)}}L_1^1({\displaystyle \frac{r}{r+1}},{\displaystyle \frac{R}{R+1}}).`$
###### Example 3.3.
(Relative arithmetic-geometric divergence). Let us consider
$$f_G(x)=\frac{x+1}{2}\mathrm{ln}\left(\frac{x+1}{2x}\right)+\frac{x1}{2},x(0,\mathrm{})$$
(41)
in (19), then we have $`C_f(P||Q)=G(P||Q)`$.
Moreover,
$$f_G^{}(x)=\frac{1}{2}\left[\mathrm{ln}\left(\frac{x+1}{2x}\right)\frac{x1}{x}\right]$$
(42)
and
$$f_{G_2}^{\prime \prime }(x)=\frac{1}{2x^2(x+1)}.$$
(43)
In view of (41), (42), Theorems 2.1 and 2.2, we have the following bounds on relative arithmetic-geometric divergence:
$$0G(P||Q)E_G(P||Q)A_G(r,R)$$
(44)
and
$$0G(P||Q)B_G(r,R)A_G(r,R),$$
(45)
where
$`E_G(P||Q)`$ $`={\displaystyle \frac{1}{2}}[\chi ^2(Q||P)D(Q||P)].`$
$`A_G(r,R)`$ $`={\displaystyle \frac{(Rr)^2}{8rR}}\left[1L_1^1({\displaystyle \frac{r+1}{r}},{\displaystyle \frac{R+1}{R}})\right]`$
and
$`B_G(r,R)`$ $`={\displaystyle \frac{1}{2}}\mathrm{ln}\left[{\displaystyle \frac{(r+1)(R+1)}{4rR}}\right]{\displaystyle \frac{1Rr}{2rR}}L_1^1({\displaystyle \frac{r+1}{r}},{\displaystyle \frac{R+1}{R}})`$
###### Example 3.4.
(Triangular discrimination). Let us consider
$$f_\mathrm{\Delta }(x)=\frac{(x1)^2}{x+1},x(0,\mathrm{})$$
(46)
in (19), then we have $`C_f(P||Q)=\mathrm{\Delta }(P||Q)`$.
Moreover,
$$f_\mathrm{\Delta }^{}(x)=\frac{(x1)(x+3)}{(x+1)^2}$$
(47)
and
$$f_\mathrm{\Delta }^{\prime \prime }(x)=\frac{8}{(x+1)^3}.$$
(48)
In view of (46), (47), Theorems 2.1 and 2.2, we have the following bounds on triangular discrimination:
$$0\mathrm{\Delta }(P||Q)E_\mathrm{\Delta }(P||Q)A_\mathrm{\Delta }(r,R)$$
(49)
and
$$0\mathrm{\Delta }(P||Q)B_\mathrm{\Delta }(r,R)A_\mathrm{\Delta }(r,R),$$
(50)
where
$`E_\mathrm{\Delta }(P||Q)`$ $`={\displaystyle \underset{i=1}{\overset{n}{}}}\left({\displaystyle \frac{p_iq_i}{p_i+q_i}}\right)^2(p_i+3q_i),`$
$`A_\mathrm{\Delta }(r,R)`$ $`={\displaystyle \frac{(Rr)^2(R+r+2)}{(R+1)^2(r+1)^2}}`$
and
$`B_\mathrm{\Delta }(r,R)`$ $`={\displaystyle \frac{2(R1)(1r)}{(R+1)(1+r)}}.`$
###### Example 3.5.
(J-divergence). Let us consider
$$f_J(x)=(x1)\mathrm{ln}x,x(0,\mathrm{}),$$
(51)
in (19), then we have $`C_f(P||Q)=J(P||Q).`$
Moreover,
$$f_J^{}(x)=1x^1+\mathrm{ln}x,$$
(52)
and
$$f_J^{\prime \prime }(x)=\frac{x+1}{x^2}.$$
(53)
In view of (51), (52), Theorems 2.1 and 2.2, we have the following bounds on J-divergence:
$$0J(P||Q)E_J(P||Q)A_J(r,R)$$
(54)
and
$$0J(P||Q)B_J(r,R)A_J(r,R),$$
(55)
where
$`E_J(P||Q)`$ $`=J(P||Q)+\chi ^2(Q||P),`$
$`A_J(r,R)`$ $`={\displaystyle \frac{1}{4}}(Rr)^2\left[(rR)^1+L_1^1(r,R)\right]`$
and
$`B_J(r,R)`$ $`=(R1)(1r)L_1^1(r,R).`$
###### Example 3.6.
(Jensen-Shannon divergence). Let us consider
$$f_I(x)=\frac{x}{2}\mathrm{ln}x+\frac{x+1}{2}\mathrm{ln}\left(\frac{2}{x+1}\right),x(0,\mathrm{}),$$
(56)
in (19), then we have $`C_f(P||Q)=I(P||Q).`$
Moreover,
$$f_I^{}(x)=\frac{1}{2}\mathrm{ln}\left(\frac{2x}{x+1}\right),$$
(57)
and
$$f_I^{\prime \prime }(x)=\frac{1}{2x(x+1)}.$$
(58)
In view of (56), (57), Theorems 2.1 and 2.2, we have the following bounds on Jensen-Shannon divergence:
$$0I(P||Q)E_I(P||Q)A_I(r,R)$$
(59)
and
$$0I(P||Q)B_I(r,R)A_I(r,R),$$
(60)
where
$`E_I(P||Q)`$ $`={\displaystyle \frac{1}{2}}D(Q||P),`$
$`A_I(r,R)`$ $`={\displaystyle \frac{1}{8}}{\displaystyle \frac{(Rr)^2}{(R+1)(r+1)}}L_1^1({\displaystyle \frac{r}{r+1}},{\displaystyle \frac{R}{R+1}})`$
and
$`B_I(r,R)`$ $`={\displaystyle \frac{1}{2(Rr)}}[(R1)(r\mathrm{ln}r+(r+1)\mathrm{ln}\left({\displaystyle \frac{2}{r+1}}\right))`$ (61)
$`(1r)(R\mathrm{ln}R+(R+1)\mathrm{ln}\left({\displaystyle \frac{2}{R+1}}\right))].`$
###### Example 3.7.
(arithmetic-geometric divergence). Let us consider
$$f_T(x)=\left(\frac{x+1}{2}\right)\mathrm{ln}\left(\frac{x+1}{2\sqrt{x}}\right),x(0,\mathrm{}),$$
(62)
in (19), then we have $`C_f(P||Q)=T(P||Q).`$
Moreover,
$$f_T^{}(x)=\frac{1}{4}\left[1x^1+2\mathrm{ln}\left(\frac{x+1}{2\sqrt{x}}\right)\right],$$
(63)
and
$$f_T^{\prime \prime }(x)=\frac{1}{4}\left(\frac{1+x^2}{x^2+x^3}\right).$$
(64)
In view of (63), (64), Theorems 2.1 and 2.2, we have the following bounds on arithmetic-geometric divergence:
$$0T(P||Q)E_T(P||Q)A_T(r,R)$$
(65)
and
$$0T(P||Q)B_T(r,R)A_T(r,R),$$
(66)
where
$`E_T(P||Q)`$ $`={\displaystyle \frac{1}{4}}\chi ^2(Q||P)+{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{n}{}}}(p_iq_i)\mathrm{ln}\left({\displaystyle \frac{p_i+q_i}{2\sqrt{p_iq_i}}}\right),`$
$`A_T(r,R)`$ $`={\displaystyle \frac{1}{16}}(Rr)^2\left[(rR)^1+L_1^1(r+1,R+1)L_1^1(r,R)\right]`$
and
$`B_T(r,R)`$ $`={\displaystyle \frac{1}{2(Rr)}}[(R1)(r+1)\mathrm{ln}\left({\displaystyle \frac{r+1}{2\sqrt{r}}}\right)`$
$`+(1r)(R+1)\mathrm{ln}\left({\displaystyle \frac{R+1}{2\sqrt{R}}}\right)].`$
## 4 Relative J-Divergence and Inequalities
In this section we shall present bound on relative J-divergence in terms of symmetric measures (8), (11)-(13).
###### Proposition 4.1.
(Relative J-divergence and triangular discrimination). We have the following bounds:
$$\frac{(r+1)(r+3)}{8}\mathrm{\Delta }(P||Q)D(P||Q)\frac{(R+1)(R+3)}{8}\mathrm{\Delta }(P||Q),$$
(67)
###### Proof.
Let us consider
$$g_{D\mathrm{\Delta }}(x)=\frac{f_D^{\prime \prime }(x)}{f_\mathrm{\Delta }^{\prime \prime }(x)}=\frac{(x+1)(x+3)}{8},x(0,\mathrm{}),$$
(68)
where $`f_D^{\prime \prime }(x)`$ and $`f_\mathrm{\Delta }^{\prime \prime }(x)`$ are as given by (33) and (48) respectively.
From (68), we have
$$g_{D\mathrm{\Delta }}^{}(x)=\frac{2+x}{4}>0,x(0,\mathrm{}).$$
(69)
In view of (69), we conclude that
$$m=\underset{x[r,R]}{inf}g_{D\mathrm{\Delta }}(x)=\frac{(r+1)(r+3)}{8}$$
(70)
and
$$M=\underset{x[r,R]}{sup}g_{D\mathrm{\Delta }}(x)=\frac{(R+1)(R+3)}{8}.$$
(71)
Expressions (70) and (71) together with (29) give the required result. โ
###### Proposition 4.2.
(Relative J-divergence and J-divergence). We have the following bounds:
$$\frac{r^2(r+3)}{(r+1)^3}J(P||Q)D(P||Q)\frac{R^2(R+3)}{(R+1)^3}J(P||Q),$$
(72)
###### Proof.
Let us consider
$$g_{DJ}(x)=\frac{f_D^{\prime \prime }(x)}{f_J^{\prime \prime }(x)}=\frac{x^2(x+3)}{(x+1)^3},x(0,\mathrm{}),$$
(73)
where $`f_D^{\prime \prime }(x)`$ and $`f_J^{\prime \prime }(x)`$ are as given by (33) and (53) respectively.
From (73), we have
$$g_{DJ}^{}(x)=\frac{6x}{(x+1)^4}>0,x(0,\mathrm{}).$$
(74)
In view of (74), we conclude that
$$m=\underset{x[r,R]}{inf}g_{DJ}(x)=\frac{r^2(r+3)}{(r+1)^3}$$
(75)
and
$$M=\underset{x[r,R]}{sup}g_{DJ}(x)=\frac{r^2(r+3)}{(r+1)^3}.$$
(76)
Expressions (75) and (76) together with (29) give the required result. โ
###### Proposition 4.3.
(Relative J-divergence and Jensen-Shannon divergence). We have the following bounds:
$$\frac{2r(r+3)}{r+1}I(P||Q)D(P||Q)\frac{2R(R+3)}{R+1}I(P||Q),$$
(77)
###### Proof.
Let us consider
$$g_{DI}(x)=\frac{f_D^{\prime \prime }(x)}{f_I^{\prime \prime }(x)}=\frac{2x(x+3)}{x+1},x(0,\mathrm{}),$$
(78)
where $`f_D^{\prime \prime }(x)`$ and $`f_I^{\prime \prime }(x)`$ are as given by (33) and (58) respectively.
From (78), we have
$$g_{DI}^{}(x)=\frac{2(x^2+2x+3)}{(x+1)^2}>0,x(0,\mathrm{}).$$
(79)
In view of (79), we conclude that
$$m=\underset{x[r,R]}{inf}g_{DI}(x)=\frac{2r(r+3)}{r+1}$$
(80)
and
$$M=\underset{x[r,R]}{sup}g_{DI}(x)=\frac{2R(R+3)}{R+1}.$$
(81)
Expressions (80) and (81) together with (29) give the required result. โ
## 5 Relative Jensen-Shannon divergence and Inequalities
In this section we shall present bound on relative Jensen-Shannon divergence in terms of symmetric measures (8), (11)-(13).
###### Proposition 5.1.
(Relative Jensen-Shannon divergence and triangular discrimination). We have the following bounds:
$$\frac{R+1}{8R}\mathrm{\Delta }(P||Q)F(P||Q)\frac{r+1}{8r}\mathrm{\Delta }(P||Q),$$
(82)
###### Proof.
Let us consider
$$g_{F\mathrm{\Delta }}(x)=\frac{f_F^{\prime \prime }(x)}{f_\mathrm{\Delta }^{\prime \prime }(x)}=\frac{x+1}{8x},x(0,\mathrm{}),$$
(83)
where $`f_F^{\prime \prime }(x)`$ and $`f_\mathrm{\Delta }^{\prime \prime }(x)`$ are as given by (38) and (48) respectively.
From (83), we have
$$g_{F\mathrm{\Delta }}^{}(x)=\frac{1}{8x^2}<0,x(0,\mathrm{}).$$
(84)
In view of (84), we conclude that
$$m=\underset{x[r,R]}{inf}g_{F\mathrm{\Delta }}(x)=\frac{R+1}{8R}$$
(85)
and
$$M=\underset{x[r,R]}{sup}g_{F\mathrm{\Delta }}(x)=\frac{r+1}{8r}.$$
(86)
Expressions (85) and (86) together with (29) give the required result. โ
###### Proposition 5.2.
(Relative Jensen-Shannon divergence and J-divergence). We have the following bounds:
$$0F(P||Q)\frac{4}{27}J(P||Q),$$
(87)
###### Proof.
Let us consider
$$g_{FJ}(x)=\frac{f_F^{\prime \prime }(x)}{f_J^{\prime \prime }(x)}=\frac{x}{(x+1)^3},x(0,\mathrm{}),$$
(88)
where $`f_F^{\prime \prime }(x)`$ and $`f_J^{\prime \prime }(x)`$ are as given by (38) and (53) respectively.
From (88), we have
$$g_{FJ}^{}(x)=\frac{2x1}{(x+1)^4}\{\begin{array}{cc}0,\hfill & x\frac{1}{2}\hfill \\ 0,\hfill & x\frac{1}{2}\hfill \end{array},$$
(89)
In view of (89), we conclude that the function $`g_{FJ}(x)`$ is increasing in $`(0,\frac{1}{2})`$ and decreasing in $`(\frac{1}{2},\mathrm{})`$, and hence
$$M=\underset{x[r,R]}{sup}g_{FJ}(x)=g_{FJ}(\frac{1}{2})=\frac{4}{27}.$$
(90)
Now (90) together with (29) give the required result. โ
###### Proposition 5.3.
(Adjoint of relative Jensen-Shannon divergence and Jensen-Shannon divergence). We have the following bounds:
$$\frac{2}{R+1}I(P||Q)F(P||Q)\frac{2}{r+1}I(P||Q),$$
(91)
###### Proof.
Let us consider
$$g_{FI}(x)=\frac{f_F^{\prime \prime }(x)}{f_I^{\prime \prime }(x)}=\frac{2}{x+1},x(0,\mathrm{}),$$
(92)
where $`f_F^{\prime \prime }(x)`$ and $`f_I^{\prime \prime }(x)`$ are as given by (38) and (58) respectively.
From (92), we have
$$g_{FI}^{}(x)=\frac{2}{(x+1)^2}<0,x(0,\mathrm{}).$$
(93)
In view of (93), we conclude that
$$m=\underset{x[r,R]}{inf}g_{FI}(x)=\frac{2}{R+1}$$
(94)
and
$$M=\underset{x[r,R]}{sup}g_{FI}(x)=\frac{2}{r+1}.$$
(95)
Expressions (94) and (95) together with (29) give the required result. โ
###### Remark 5.1.
The inequalities (82) and (91) can also be as
$$r\zeta _t(P||Q)R,t=1and\mathrm{\hspace{0.17em}\hspace{0.17em}2},$$
where
$$\zeta _1(P||Q)=\frac{\mathrm{\Delta }(P||Q)}{8F(P||Q)\mathrm{\Delta }(P||Q)},$$
and
$$\zeta _3(P||Q)=\frac{2I(P||Q)F(P||Q)}{F(P||Q)}$$
respectively.
## 6 Relative arithmetic-geometric divergence and Inequalities
In this section we shall present bound on relative arithmetic-geometric divergence in terms of symmetric measures (8), (11)-(13).
###### Proposition 6.1.
(Relative arithmetic-geometric divergence and triangular discrimination). We have the following bounds:
$$\frac{(R+1)^2}{16R^2}\mathrm{\Delta }(P||Q)G(P||Q)\frac{(r+1)^2}{16r^2}\mathrm{\Delta }(P||Q),$$
(96)
###### Proof.
Let us consider
$$g_{G\mathrm{\Delta }}(x)=\frac{f_G^{\prime \prime }(x)}{f_\mathrm{\Delta }^{\prime \prime }(x)}=\frac{(x+1)^2}{16x^2},x(0,\mathrm{}),$$
(97)
where $`f_G^{\prime \prime }(x)`$ and $`f_\mathrm{\Delta }^{\prime \prime }(x)`$ are as given by (38) and (48) respectively.
From (97), we have
$$g_{G\mathrm{\Delta }}^{}(x)=\frac{x+1}{8x^3}<0,x(0,\mathrm{}).$$
(98)
In view of (98), we conclude that
$$m=\underset{x[r,R]}{inf}g_{G\mathrm{\Delta }}(x)=\frac{(R+1)^2}{16R^2}$$
(99)
and
$$M=\underset{x[r,R]}{sup}g_{G\mathrm{\Delta }}(x)=\frac{(r+1)^2}{16r^2}.$$
(100)
Expressions (99) and (100) together with (29) give the required result. โ
###### Proposition 6.2.
(Relative arithmetic-geometric divergence and J-divergence). We have the following bounds:
$$\frac{1}{2(R+1)^2}J(P||Q)G(P||Q)\frac{1}{2(r+1)^2}J(P||Q),$$
(101)
###### Proof.
Let us consider
$$g_{GJ}(x)=\frac{f_G^{\prime \prime }(x)}{f_J^{\prime \prime }(x)}=\frac{1}{2(x+1)^2},x(0,\mathrm{}),$$
(102)
where $`f_G^{\prime \prime }(x)`$ and $`f_J^{\prime \prime }(x)`$ are as given by (38) and (53) respectively.
From (102), we have
$$g_{GJ}^{}(x)=\frac{1}{(x+1)^3}<0,x(0,\mathrm{}).$$
(103)
In view of (98), we conclude that
$$m=\underset{x[r,R]}{inf}g_{GJ}(x)=\frac{1}{2(R+1)^2}$$
(104)
and
$$M=\underset{x[r,R]}{sup}g_{GJ}(x)=\frac{1}{2(r+1)^2}.$$
(105)
Expressions (104) and (105) together with (29) give the required result. โ
###### Proposition 6.3.
(Relative arithmetic-geometric divergence and Jensen-Shannon divergence). We have the following bounds:
$$\frac{1}{R}I(P||Q)G(P||Q)\frac{1}{r}I(P||Q),$$
(106)
###### Proof.
Let us consider
$$g_{GI}(x)=\frac{f_G^{\prime \prime }(x)}{f_I^{\prime \prime }(x)}=\frac{1}{x},x(0,\mathrm{}),$$
(107)
where $`f_G^{\prime \prime }(x)`$ and $`f_I^{\prime \prime }(x)`$ are as given by (38) and (58) respectively.
From (107), we have
$$g_{GI}^{}(x)=\frac{1}{x^2}<0,x(0,\mathrm{}).$$
(108)
In view of (108), we conclude that
$$m=\underset{x[r,R]}{inf}g_{GI}(x)=\frac{1}{R}$$
(109)
and
$$M=\underset{x[r,R]}{sup}g_{GI}(x)=\frac{1}{r}.$$
(110)
Expressions (109) and (110) together with (29) give the required result. โ
###### Proposition 6.4.
(Relative arithmetic-geometric divergence and arithmetic-geometric divergence). We have the following bounds:
$$\frac{2}{1+R^2}T(P||Q)G(P||Q)\frac{2}{1+r^2}T(P||Q),$$
(111)
###### Proof.
Let us consider
$$g_{GT}(x)=\frac{f_G^{\prime \prime }(x)}{f_T^{\prime \prime }(x)}=\frac{2}{1+x^2},x(0,\mathrm{}),$$
(112)
where $`f_G^{\prime \prime }(x)`$ and $`f_T^{\prime \prime }(x)`$ are as given by (38) and (64) respectively.
From (112), we have
$$g_{GT}^{}(x)=\frac{4x}{(1+x^2)^2}<0,x(0,\mathrm{}).$$
(113)
In view of (113), we conclude that
$$m=\underset{x[r,R]}{inf}g_{GT}(x)=\frac{2}{1+R^2}$$
(114)
and
$$M=\underset{x[r,R]}{sup}g_{GT}(x)=\frac{2}{1+r^2}.$$
(115)
Expressions (114) and (115) together with (29) give the required result. โ
###### Remark 6.1.
The inequalities (96), (101), (106) and (111) can also be as
$$r\xi _t(P||Q)R,t=1,2,3and\mathrm{\hspace{0.17em}\hspace{0.17em}4},$$
where
$$\xi _1(P||Q)=\frac{\sqrt{\mathrm{\Delta }(P||Q)}}{4\sqrt{G(P||Q)}\sqrt{\mathrm{\Delta }(P||Q)}},$$
$$\xi _2(P||Q)=\frac{\sqrt{J(P||Q)}\sqrt{2G(P||Q)}}{\sqrt{2G(P||Q)}},$$
$$\xi _2(P||Q)=\frac{I(P||Q)}{G(P||Q)},$$
and
$$\xi _4(P||Q)=\frac{\sqrt{2T(P||Q)G(P||Q)}}{\sqrt{G(P||Q)}}$$
respectively.
|
warning/0506/math0506183.html
|
ar5iv
|
text
|
# The triangle of operators, topologies, bornologies
## 1. Introduction
How can one describe a linear operator $`T`$ from a Banach space $`E`$ into a Banach space $`F`$? The usual way to describe $`T`$ is to state either the bornological property, via $`TU_E`$, or the topological property, via $`T^1U_F`$, of $`T`$, where $`U_E`$ (resp. $`U_F`$) is the closed unit ball of $`E`$ (resp. $`F`$). However, there are a lot of examples indicating that these two machineries are equivalent. For instance,
* $`T`$ is bounded (i.e., $`TU_E`$ is a bounded subset of $`F`$)
$``$ $`T`$ is continuous (i.e., $`T^1U_F`$ is a 0-neighborhood of $`E`$ in the norm topology);
* $`T`$ is of finite rank (i.e., $`TU_E\mathrm{conv}\{y_1,y_2,\mathrm{},y_n\}`$ for some $`y_1`$, $`y_2`$, $`\mathrm{}`$, $`y_n`$ in $`F`$)
$``$ $`T`$ is weak-norm continuous (i.e., $`T^1U_F`$ is a 0-neighborhood of $`E`$ in the weak topology); and
* $`T`$ is compact (i.e., $`TU_E`$ is totally bounded in $`F`$)
$``$ $`T`$ is continuous in the topology of uniform convergence on norm compact subsets of $`E^{}`$ (i.e., $`T^1U_FK^{}`$, the polar of a norm compact subset $`K`$ of the dual space $`E^{}`$ of $`E`$).
This is because the unit ball of a normed space simultaneously serves as a neighborhood of zero and a bounded set. It is, however, no longer true in the context of locally convex spaces (LCSโs, shortly). Mackey-Arensโ Theorem indicates that topologies (families of neighborhoods) and bornologies (families of bounded sets) are in dual pair (see e.g. ).
It is a long tradition of classifying special classes of locally convex spaces by families of continuous operators among them. A famous example is, of course, Grothendieckโs identification of the class of nuclear locally convex spaces. Other examples are those of Schwartz LCSโs, infraโSchwartz LCSโs and their โcoโspacesโ. After the great effort of Pietsch , it is now wellโknown that such suitable families of continuous operators are the soโcalled operator ideals.
There are many ways to utilize Grothendieckโs idea. For example, one can define a LCS $`X`$ to be nuclear ($`\mathrm{resp}.`$ Schwartz, infraโSchwartz) by asking that for each continuous seminorm $`p`$ on $`X`$, there is a continuous seminorm $`q`$ on $`X`$ with $`pq`$ such that the canonical map $`\stackrel{~}{Q}_{pq}`$ from $`\stackrel{~}{X}_q=\stackrel{~}{X/q^1(0)}`$ into $`\stackrel{~}{X}_p=\stackrel{~}{X/p^1(0)}`$ is nuclear ($`\mathrm{resp}.`$ precompact, weakly compact), where $`\stackrel{~}{}`$ denotes completion. It amounts to saying that the completion $`\stackrel{~}{X}`$ of $`X`$ is a topological projective limit $`\underset{}{\mathrm{lim}}\stackrel{~}{Q}_{pq}\stackrel{~}{X}_q`$ of Banach spaces of nuclear type ($`\mathrm{resp}.`$ precompact type, weakly compact type). The converse is also true, see Junek \[11, p. 139\]. We call such a LCS a Grothendieck space of nuclear ($`\mathrm{resp}.`$ precompact, weakly compact) type, or shortly a $`\mathrm{Groth}(๐)`$โspace ($`\mathrm{resp}.`$ $`\mathrm{Groth}(๐_p)`$โspace, $`\mathrm{Groth}(๐)`$โspace), where $`๐(\mathrm{resp}.๐_\mathrm{p},๐)`$ is the ideal of all nuclear ($`\mathrm{resp}.`$ precompact, weakly compact) operators between *Banach spaces*.
As a dual concept, a locally convex space $`X`$ is said to be a coโGrothendieck space of type $`๐`$, or shortly a coโ$`\mathrm{Groth}(๐)`$โspace, if for each infracomplete disk $`A`$ in $`X`$ there is an infracomplete disk $`B`$ in $`X`$ such that $`AB`$ and the canonical map $`J_{BA}`$ from $`X(A)=_{\lambda >0}\lambda A`$ into $`X(B)=_{\lambda >0}\lambda B`$ belongs to $`๐(X(A),X(B))`$. In other words, the convex bornological vector space $`X`$ equipped with the infracomplete bornology of $`X`$ is the bornological inductive limit $`\underset{}{\mathrm{lim}}J_{BA}X(A)`$ of Banach spaces of type $`๐`$. The converse is again true.
Another way to go is to define the ideal topology and the ideal bornology on each LCS associated to an operator ideal $`๐`$ *on LCSโs*. A continuous seminorm $`p`$ on a LCS $`X`$ is said to be an $`๐`$โcontinuous seminorm if the canonical map $`\stackrel{~}{Q}_p:X\stackrel{~}{X}_p`$ belongs to the injective hull $`๐^{\mathrm{inj}}`$ of $`๐`$. The topology on $`X`$ defined by the family of all such seminorms is called the $`๐`$โtopology of $`X`$. Similarly, an absolutely convex bounded set $`B`$ in $`X`$ is said to be $`๐`$โbounded if the canonical map $`J_B`$ from $`X(B)=_{\lambda >0}\lambda B`$ into $`X`$ belongs to the bornologically surjective hull $`๐^{\mathrm{bsur}}`$ of $`๐`$. The bornology on $`X`$ defined by the family of all such bounded sets is called the $`๐`$โbornology of $`X`$. A LCS $`X`$ is said to be $`๐`$โtopological (resp. $`๐`$โbornological) if the topology (resp. bornology) of $`X`$ coincides with the $`๐`$โtopology (resp. $`๐`$โbornology).
In we show that Grothendieck spaces are essentially a kind of $`๐`$โspaces. Thus these two different approaches coincide. In this paper, we will develop the duality theory of $`๐`$โtopological spaces and $`๐`$โbornological spaces. Basically, one may expect that a locally convex space $`X`$ is $`๐`$โtopological ($`\mathrm{resp}.๐`$โbornological) if and only if its strong dual $`X_\beta ^{}`$ is $`๐`$โbornological ($`\mathrm{resp}.๐`$โtopological). One can discover the same is true for Grothendieck spaces and coโGrothendieck spaces by observing the duality of topology and bornology and the duality of projective limits and inductive limits (see, e.g., ).
The following commutative diagram summaries our works.
The theory of operator ideal is founded by Pietsch and originated from the works of Grothendieck and Schatten . See also for more information. The idea of generating topologies and generating bornologies are due to Stephani and Franco and Piรฑeiro in the context of Banach spaces. The explicit construction (with all arrows shown in the diagram) of the (upper) triangle is given in , in which several applications to Banach space theory are demonstrated. When the underlying space is a fixed complex Hilbert space, West implements the triangle in the context of operator algebras and provides several applications with Conradie (see Section 2). In this paper, we shall complete the LCS version of the triangle. As an application, we shall show that in the study of LCSโs, the topological machinery of Randtke (via continuous seminorms) or the bornological machinery of Hogbe-Nlend (via convex bounded subsets) is as strong as that of the operator theoretical machinery of Grothendieck (via Banach space operators) (see e.g. ).
The author dedicates this paper to his late teacher, Professor Yau-Chuen Wong, who introduced the same concept of $`๐`$โtopology and $`๐`$-bornology through a great number of examples of special LCSโs as well as partially ordered locally convex spaces (see, ), although he did not employ the Pietschโs language (operator ideals) at his time. Together with , the current paper is a continuation of his ideal (see ).
## 2. Established examples in Hilbert spaces and Banach spaces
### 2.1. The triangle for Hilbert spaces
Let $`๐`$ be a von Neumann algebra of bounded linear operators on a Hilbert space $`H`$, and $`๐`$ an arbitrary non-zero two-sided ideal of $`๐`$.
* A locally convex topology $`๐ซ`$ of $`H`$ is called a generating topology if $`๐ซ`$ consists of norm open sets in $`H`$ such that all operators in $`๐`$ are $`๐ซ`$-to-$`๐ซ`$ continuous on $`H`$, i.e., the pre-images of $`๐ซ`$-open sets being $`๐ซ`$-open.
* A convex vector bornology $``$ of $`H`$ is called a generating bornology if $``$ consists of norm bounded subsets of $`H`$ such that all operators in $`๐`$ are $``$-to-$``$ bounded, i.e., sending $``$-bounded sets to $``$-bounded sets.
* The $`๐`$-*topology* $`๐ฏ(๐)`$ is the projective topology of $`H`$ induced by operators in $`๐`$, i.e., the weakest locally convex topology $`t`$ of $`H`$ such that operators in $`๐`$ are $`t`$-to-norm continuous.
* The $`๐`$-*bornology* $`(๐)`$ is the inductive bornology of $`H`$ induced by operators in $`๐`$, i.e., the smallest convex vector bornology $`b`$ of $`H`$ such that operators in $`๐`$ are norm-to-$`b`$ bounded.
* The polar of a subset $`A`$ in $`H`$ is
$$A^{}=\{xH:|a,x|1,aA\}.$$
Remark that the ideal $`๐`$ is
* *self-adjoint*, i.e., $`T๐`$ if and only if its Hilbert space adjoint map $`T^{}๐`$;
* injective, i.e., $`T๐`$ whenever $`ThSh,hH`$, for any $`S`$ in $`๐`$ and $`T`$ in $`๐`$; and
* surjective, i.e., $`T๐`$ whenever $`TU_HSU_H`$ for any $`S`$ in $`๐`$ and $`T`$ in $`๐`$.
###### Theorem 2.1 (West ).
1. 1. The $`๐`$-topology $`๐ฏ(๐)`$ is a generating topology.
2. The $`๐`$-bornology $`(๐)`$ is a generating bornology.
2. 1. The set $`๐ช(๐ซ)=๐(H_๐ซ,H)`$ of all $`๐ซ`$-to-norm continuous linear operators on $`H`$ is a two-sided ideals of $`๐`$.
2. The set $`๐ช()=B(H,H^{})`$ of all norm-to-$``$ bounded linear operators on $`H`$ is a two-sided ideals of $`๐`$.
3. 1. The polar $`๐ซ^{}=\{BH:B^{}\text{ is a }๐ซ\text{-neighborhood of }0\}`$ of a generating topology $`๐ซ`$ is a generating bornology.
2. The polar $`^{}=\{VH:V^{}\text{ is }\text{-bounded}\}`$ of a generating bornology $``$ is a generating topology.
4. The triangle of operators, topologies and bornologies is commutative:
1. $`๐ช(๐ฏ(๐))=๐`$, $`๐ช((๐))=๐`$.
2. $`๐ฏ(๐)^{}=(๐)`$, $`(๐)^{}=๐ฏ(๐)`$.
3. $`๐ฏ(๐ช(๐ซ))=๐ซ`$, $`(๐ช())=`$.
### 2.2. The triangle for Banach spaces
The Banach space version of the โtriangleโ is known to have many applications (cf. ). Let $`๐=\{๐(E,F):E,F\text{ are Banach spaces}\}`$ be an operator ideal on Banach spaces in the sense of Pietsch :
1. The components $`๐(E,F)`$ of $`๐`$ are non-zero subspaces of $`๐(E,F)`$.
2. $`RTS๐(E_0,F_0)`$ whenever $`R๐(F,F_0)`$, $`T๐(E,F)`$ and $`S๐(E_0,E)`$ for arbitrary Banach spaces $`E,E_0,F`$, and $`F_0`$.
An operator ideal $`๐`$ is said to be symmetric if $`T๐(E,F)`$ ensures its Banach space dual map $`T^{}๐(F^{},E^{})`$, and completely symmetric if $`T๐(E,F)T^{}๐(F^{},E^{})`$.
Suppose for each Banach space $`E`$, we have a locally convex topology $`๐ซ(E)`$ consisting of norm open subsets of $`E`$ and a convex vector bornology $`(E)`$ consisting of norm bounded subsets of $`E`$. We call $`๐ซ=\{๐ซ(E):E\text{ is a Banach space}\}`$ a generating topology and $`=\{(E):E\text{ is a Banach space}\}`$ a generating bornology on Banach spaces, if operators in $`๐(E,F)`$ are $`๐ซ(E)`$-to-$`๐ซ(F)`$ continuous and $`(E)`$-to-$`(F)`$ bounded for all Banach spaces $`E`$ and $`F`$, respectively.
The polar $`๐ซ^{}`$ of a generating topology $`๐ซ`$ consists of components
$$๐ซ^{}(E)=\{BE:B^{}\text{ is }๐ซ(E^{})\text{-bounded}\}.$$
Similarly, the polar $`^{}`$ of a generating bornology $``$ consists of components
$$^{}(E)=\{VE:V^{}\text{ is a }๐ซ(E^{})\text{-neighborhood of }0\}.$$
###### Theorem 2.2 (Stephani and Wong and Wong ).
1. 1. The family of projective topologies $`๐ฏ(๐)(E)`$ of Banach spaces $`E`$ induced by operators in $`๐(E,)`$ forms a generating topology.
2. The family of inductive bornologies $`(๐)(F)`$ of Banach spaces $`F`$ induced by operators in $`๐(,F)`$ forms a generating bornology.
2. 1. The family of sets $`๐ช(๐ซ)(E,F)=๐(E_๐ซ,F)`$ of all $`๐ซ(E)`$-to-norm continuous linear operators from $`E`$ into $`F`$ forms an injective operator ideal.
2. The family of sets $`๐ช()(E,F)=B(E,F_{})`$ of all norm-to-$`(F)`$ bounded linear operators from $`E`$ into $`F`$ forms a surjective operator ideal.
3. 1. The polar $`๐ซ^{}`$ of a generating topology $`๐ซ`$ is a generating bornology.
2. The polar $`^{}`$ of a generating bornology $``$ is a generating topology.
4. The triangle of operators, topologies and bornologies is almost commutative.
1. $`๐ช(๐ฏ(๐))=๐^{\mathrm{inj}}`$ and $`๐ช((๐))=๐^{\mathrm{sur}}`$.
2. $`(๐)^{}=๐ฏ(๐)`$ if $`๐`$ is symmetric, and $`๐ฏ(๐)^{}=(๐)`$ if $`๐`$ is completely symmetric.
3. $`๐ฏ(๐ช(๐ซ))=๐ซ`$, $`(๐ช())=`$.
Note that the ideals $`๐`$ of all bounded operators, $`๐`$ of all bounded operator of finite rank and $`๐`$ of all compact operators are all injective, surjective and completely symmetric. These explain the equivalence of topological and bornological approaches for these operators demonstrated at the very beginning of this paper. On the other hand, the ideal $`๐`$ of nuclear operators is neither injective, surjective or completely symmetric (cf. ).
## 3. Notations and Preliminaries
The classic reference to the theory of operator ideals is, of course, Pietsch . See also Jarchow and Junek . For the theory of locally convex spaces, together with Wong , Schaefer is our favorite. HogbeโNlend serves as our main source of the theory of bornology.
Throughout this paper, all vector spaces have the same underlying scalar field $`๐`$. $`๐`$ is either the field $``$ of real numbers or the field $``$ of complex numbers. Locally convex topologies are always Hausdorff, and convex vector bornologies are always separated, i.e., no nonzero subspace is bounded. Operators always refer to linear maps without any topological or bornological assumption. $`U_N`$ always denotes the closed unit ball of a normed space $`N`$.
A subset $`B`$ of a LCS $`X`$ is said to be a disk if $`B`$ is absolutely convex, i.e., $`\alpha B+\beta BB`$ whenever $`|\alpha |+|\beta |1`$. A disk $`B`$ is said to be a $`\sigma `$disk, or absolutely $`\sigma `$convex if $`\mathrm{\Sigma }_n\lambda _nb_n`$ converges in $`B`$ whenever $`_n|\lambda _n|1`$ and $`b_nB`$, $`n=1,2,\mathrm{}`$. A bounded disk $`B`$ is said to be infracomplete (or a Banach disk) if the normed space $`X(B)=_{\lambda >0}\lambda B`$ equipped with the gauge $`\gamma _B`$ of $`B`$ as its norm is complete, where $`\gamma _B(x)=inf\{\lambda >0:x\lambda B\}`$, for each $`x`$ in $`X(B)`$. Any continuous image of a $`\sigma `$โdisk or an infracomplete bounded disk is still a $`\sigma `$โdisk or an infracomplete bounded disk, respectively. A LCS $`X`$ is said to be infracomplete if the von Neumann bornology $`_{\mathrm{von}}(X)`$, i.e., the original bornology induced by the topology of $`X`$, has a basis consisting of infracomplete subsets of $`X`$, or equivalently, $`\sigma `$โdisked subsets of $`X`$. In other words, $`(X,_{\mathrm{von}}(X))`$ is a complete convex bornological vector space.
Let $`X,X^{}`$ be a dual pair and $`BX`$. The (absolute) polar $`B^{}`$ of $`B`$ in $`X^{}`$ is defined by
$$B^{}=\{xX^{}:|b,x|1,bB\}.$$
Whenever $`AX^{}`$, denote by $`A^{}`$ the polar of $`A`$ taken in $`X_{\beta \beta }^{\prime \prime }`$, namely,
$$A^{}=\{xX_{\beta \beta }^{\prime \prime }:|a,x|1,aA\},$$
where $`X_{\beta \beta }^{\prime \prime }`$ is the strong bidual of $`X`$, while $`A^{}`$ denotes the polar of $`A`$ taken in $`X`$ with respect to the dual pair $`X,X^{}`$.
###### Proposition 3.1 (See, e.g., Wong \[31, pp. 224 and 227\]).
Let $`X`$ and $`Y`$ be LCSโs and $`T(X,Y)`$. We have
1. $`T(X_\sigma ,Y_\sigma )`$, where $`X_\sigma `$, $`Y_\sigma `$ denote the LCSโs in their weak topologies.
2. $`T(X_\tau ,Y_\tau )`$, where $`X_\tau `$, $`Y_\tau `$ denote the LCSโs in their Mackey topologies.
3. $`T^{}(Y_\beta ^{},X_\beta ^{})`$, where $`T^{}`$ is the dual map of $`T`$ and $`X_\beta ^{}`$ (resp. $`Y_\beta ^{}`$) is the strong dual of $`X`$ (resp. $`Y`$).
4. $`(TA)^{}=(T^{})^1A^{}`$ for all nonempty subset $`A`$ of $`X`$.
5. $`(T^{}B)^{}=T^1B^{}`$ for all nonempty subset $`B`$ of $`Y^{}`$.
6. $`(T^1W)^{}=T^{}W^{}`$ for all neighborhoods $`W`$ of $`0`$ in its Mackey topology $`\tau (Y,Y^{})`$.
Let $`X`$ and $`Y`$ be LCSโs. $`J`$ in $`๐(X,Y)`$ is called a (topological) injection if $`J`$ is one-to-one and relatively open. $`Q`$ in $`๐(X,Y)`$ is called a (topological) surjection if $`Q`$ is open (and thus $`Q`$ induces the topology of $`Y`$). $`Q^1`$ in $`๐(X,Y)`$ is called a bornological surjection if $`Q^1`$ is onto and induces the bornology of $`Y`$ (i.e., for each bounded subset $`B`$ of $`Y`$ there is a bounded subset $`A`$ of $`X`$ such that $`Q^1A=B`$).
An operator ideal $`๐`$ on LCSโs is said to be
* *injective* if $`JT๐(X,Y_0)`$ infers $`T๐(X,Y)`$, whenever $`T๐(X,Y)`$ and $`J๐(Y,Y_0)`$ is an injection for some LCS $`Y_0`$;
* *surjective* if $`TQ๐(X_0,Y)`$ infers $`T๐(X,Y)`$, whenever $`T๐(X,Y)`$ and $`Q๐(X_0,X)`$ is a surjection for some LCS $`X_0`$; and
* *bornologically surjective* if $`TQ^1๐(X_0,Y)`$ infers $`T๐(X,Y)`$, whenever $`T๐(X,Y)`$ and $`Q^1๐(X_0,X)`$ is a bornological surjection for some LCS $`X_0`$.
The *injective hull* $`๐^{\mathrm{inj}}`$, the *surjective hull* $`๐^{\mathrm{sur}}`$, and the *bornologically surjective hull* $`๐^{\mathrm{bsur}}`$ of $`๐`$ is the intersection of all injective, surjective, and bornologically surjective operator ideals containing $`๐`$, respectively. Note that for operator ideals on Banach spaces, the notions of surjectivity and bornological surjectivity coincide.
Associate to each normed space $`N`$ the Banach space $`N^{\mathrm{inj}}=ล_{\mathrm{}}(U_N^{})`$ and the injection $`J_N`$ in $`๐(N,N^{\mathrm{inj}})`$ defined by $`J_N(x)=(<x,a>)_{aU_N^{}}`$. Similarly, we define $`N^{\mathrm{sur}}`$ to be the normed space $`L_1(U_N)=\{(\lambda _x)\mathrm{}_1(U_N):_{xU_N}\lambda _xx`$ converges in $`N\}`$ and $`Q_N:N^{\mathrm{sur}}N`$ to be the surjection defined by $`Q_N((\lambda _x)_{xU_N})=_{xU_N}\lambda _xx`$. In case $`E`$ is a Banach space, it is wellโknown that $`E^{\mathrm{inj}}`$ has the extension property and $`E^{\mathrm{sur}}`$ has the lifting property, cf. .
###### Proposition 3.2 ().
1. Let $`๐`$ be an operator ideal on Banach spaces.
$`๐^{\mathrm{inj}}(E,F)`$ $`=`$ $`\{R๐(E,F):J_FR๐(E,F^{\mathrm{inj}})\},`$
$`๐^{\mathrm{sur}}(E,F)`$ $`=`$ $`\{S๐(E,F):SQ_E๐(E^{\mathrm{sur}},F)\}.`$
2. Let $`๐`$ be an operator ideal on LCSโs. We can associate to each LCS $`Y`$ a LCS $`Y^{\mathrm{}}`$ and an injection $`J_Y^{\mathrm{}}`$ from $`Y`$ into $`Y^{\mathrm{}}`$, and to each LCS $`X`$ a LCS $`X^1`$ and a bornological surjection $`Q_X^1`$ from $`X^1`$ onto $`X`$ such that
$`๐^{\mathrm{inj}}(X,Y)`$ $`=`$ $`\{R๐(X,Y):J_Y^{\mathrm{}}R๐(X,Y^{\mathrm{}})\},`$
$`๐^{\mathrm{bsur}}(X,Y)`$ $`=`$ $`\{S๐(X,Y):SQ_X^1๐(X,Y^1)\}.`$
Moreover, we have
$$๐^{\mathrm{inj}\mathrm{bsur}}=๐^{\mathrm{bsur}\mathrm{inj}}.$$
In case $`N`$ is a normed space, $`R๐(X,N)`$ and $`S๐(N,Y)`$,
$`J_NR๐(X,N^{\mathrm{inj}})`$ $``$ $`J_N^{\mathrm{}}R๐(X,N^{\mathrm{}}),`$
$`SQ_N^1๐(N^1,Y)`$ $``$ $`SQ_N๐(N^{\mathrm{sur}},Y).`$
## 4. The construction and the commutativity of the triangle
Let $`๐`$ be a class of locally convex spaces. Let $`X,Y๐`$. We denote by $`๐^b(X,Y)`$, $`๐(X,Y)`$ and $`L^\times (X,Y)`$ the collection of all operators from $`X`$ into $`Y`$ which are bounded (i.e., sending a 0-neighborhood to a bounded set), continuous, and locally bounded (i.e., sending bounded sets to bounded sets), respectively.
Denote by $`\sigma (X,X^{})`$ the weak topology of $`X`$ with respect to its dual space $`X^{}`$, while $`๐ซ_{\mathrm{ori}}(X)`$ is the original topology of $`X`$. We employ the notion $`_{\mathrm{fin}}(Y)`$ for the finite dimensional bornology of $`Y`$ which has a basis consisting of all convex hulls of finite sets. On the other hand, $`_{\mathrm{von}}(Y)`$ is used for the von Neumann bornology of $`Y`$ which consists of all topologically bounded subsets of $`Y`$. Ordering of topologies and bornologies are induced by set-theoretical inclusion, as usual. Moreover, we write briefly $`X_๐ซ`$ for a vector space $`X`$ equipped with a locally convex topology $`๐ซ`$ and $`Y^{}`$ for a vector space $`Y`$ equipped with a convex vector bornology $``$.
We now give the details of the โtriangleโ.
###### Definition 4.1.
1. (โOperatorsโ) A family $`๐=\{๐(X,Y):X,Y๐\}`$ of algebras of operators associated to each pair of spaces $`X`$ and $`Y`$ in $`๐`$ is called an operator ideal if
$`๐(X,Y)`$ is a nonzero vector subspace of $`๐(X,Y)`$ for all $`X`$, $`Y`$ in $`๐`$; and
$`RTS๐(X_0,Y_0)`$ whenever $`R๐(Y,Y_0)`$, $`T๐(X,Y)`$ and $`S๐(X_0,X)`$ for any $`X_0`$, $`X`$, $`Y`$ and $`Y_0`$ in $`๐`$.
2. (โTopologiesโ) A family $`๐ซ=\{๐ซ(X):X๐\}`$ of locally convex topologies associated to each space $`X`$ in $`๐`$ is called a generating topology if
$`\sigma (X,X^{})๐ซ(X)๐ซ_{\mathrm{ori}}(X)`$ for all $`X`$ in $`๐`$; and
$`๐(X,Y)๐(X_๐ซ,Y_๐ซ)`$ for all $`X`$ and $`Y`$ in $`๐`$.
3. (โBornologiesโ) A family $`=\{(Y):Y๐\}`$ of convex vector bornologies associated to each space $`Y`$ in $`๐`$ is called a generating bornology if
$`_{\mathrm{fin}}(Y)(Y)_{\mathrm{von}}(Y)`$ for all $`Y`$ in $`๐`$; and
$`๐(X,Y)L^\times (X^{},Y^{})`$ for all $`X`$ and $`Y`$ in $`๐`$.
Classical examples of these notions are the ideals $`๐_p`$ of precompact operators and $`๐`$ of absolutely summing operators (see e.g. ), the generating systems $`๐ซ_{pc}`$ of precompact topologies (see e.g. ) and $`๐ซ_{pn}`$ of prenuclear topologies (see e.g. \[17, p. 90\]), and the generating systems $`_{pc}`$ of precompact bornologies and $`_{pn}`$ of prenuclear bornologies (see e.g. ), respectively. An interesting fact about these examples is that we can visualize the notions of โoperatorsโ, โtopologiesโ and โbornologiesโ as vertices of a triangle, and they can be transformed to each other by actions represented as linking edges of the triangle.
###### Definition 4.2.
Let $`๐`$ be an operator ideal, $`๐ซ`$ a generating topology and $``$ a generating bornology on $`๐`$.
1. (โOperatorsโ $``$ โTopologiesโ) For each $`X_0`$ in $`๐`$, the $`๐`$topology of $`X_0`$, denoted by $`๐ฏ(๐)(X_0)`$, is the projective topology of $`X_0`$ with respect to the family
$$\{T๐(X_0,Y):Y๐\}.$$
In other words, a seminorm $`p`$ of $`X_0`$ is $`๐ฏ(๐)(X_0)`$โcontinuous if and only if there is a $`T`$ in $`๐(X_0,Y)`$ for some $`Y`$ in $`๐`$ and a continuous seminorm $`q`$ of $`Y`$ such that
$$p(x)q(Tx),xX_0.$$
In this case, we call $`p`$ an $`๐`$seminorm of $`X_0`$.
2. (โOperatorsโ $``$ โBornologiesโ) For each $`Y_0`$ in $`๐`$, the $`๐`$bornology of $`Y_0`$, denoted by $`(๐)(Y_0)`$, is the inductive bornology of $`Y_0`$ with respect to the family
$$\{T๐(X,Y_0):X๐\}.$$
In other words, a subset $`B`$ of $`Y_0`$ is $`(๐)(Y_0)`$โbounded if and only if there is a $`T`$ in $`๐(X,Y_0)`$ for some $`X`$ in $`๐`$ and a topologically bounded subset $`A`$ of $`X`$ such that
$$BTA.$$
In this case, we call $`B`$ an $`๐`$bounded subset of $`Y_0`$.
3. (โTopologiesโ $``$ โOperatorsโ) For $`X`$, $`Y`$ in $`๐`$, let
$$๐ช(๐ซ)(X,Y)=๐(X_๐ซ,Y)$$
and
$$๐ช^b(๐ซ)(X,Y)=๐^b(X_๐ซ,Y)$$
be the vector space of all continuous operators from $`X`$ into $`Y`$ which is still continuous with respect to the $`๐ซ(X)`$โtopology, and which send a $`๐ซ(X)`$โneighborhood of zero to a bounded set, respectively.
4. (โBornologiesโ $``$ โOperatorsโ) For $`X`$, $`Y`$ in $`๐`$, let
$$๐ช()(X,Y)=๐(X,Y)L^\times (X,Y^{})$$
and
$$๐ช^b()(X,Y)=๐^b(X,Y^{})$$
be the vector space of all continuous operators from $`X`$ into $`Y`$ which send bounded sets to $`(Y)`$โbounded sets, and which send a neighborhood of zero to an $`(Y)`$โbounded set, respectively.
5. (โTopologiesโ $``$ โBornologiesโ) For $`X`$, $`Y`$ in $`๐`$, the $`๐ซ^{}(Y)`$bornology of $`Y`$ (resp. $`^{}(X)`$topology of $`X`$) is defined to be the bornology (resp. topology) polar to $`๐ซ(X)`$ (resp. $`(Y)`$). More precisely,
* a bounded subset $`A`$ of $`Y`$ is $`๐ซ^{}(Y)`$โbounded if and only if its polar $`A^{}`$ is a $`๐ซ(Y_\beta ^{})`$โneighborhood of zero; and
* a neighborhood $`V`$ of zero of $`X`$ is a $`^{}(X)`$โneighborhood of zero if and only if $`V^{}`$ is $`(X_\beta ^{})`$โbounded.
###### Theorem 4.3.
Let $`๐`$ be an operator ideal, $`๐ซ`$ a generating topology and $``$ a generating bornology on $`๐`$. We have
1. $`๐ฏ(๐)=\{๐ฏ(๐)(X):X๐\}`$ is a generating topology on $`๐`$.
2. $`(๐)=\{(๐)(Y):Y๐\}`$ is a generating bornology on $`๐`$.
3. $`๐ช(๐ซ)=\{๐ช(๐ซ)(X,Y):X,Y๐\}`$ is an operator ideal on $`๐`$.
4. $`๐ช^b(๐ซ)=\{๐ช^b(๐ซ)(X,Y):X,Y๐\}`$ is an operator ideal on $`๐`$.
5. $`๐ช()=\{๐ช()(X,Y):X,Y๐\}`$ is an operator ideal on $`๐`$.
6. $`๐ช^b()=\{๐ช^b()(X,Y):X,Y๐\}`$ is an operator ideal on $`๐`$.
7. $`๐ซ^{}=\{๐ซ^{}(Y):Y๐\}`$ is a generating bornology on $`๐`$.
8. $`^{}=\{^{}(Y):Y๐\}`$ is a generating topology on $`๐`$.
###### Proof.
(1)โ(6), together with the Banach space version of (7) and (8), are done in . For the locally convex space version of (7), we first note that (GB<sub>1</sub>) follows from (GT<sub>1</sub>) and the bipolar theorem. To check (GB<sub>2</sub>), let $`X`$ and $`Y`$ be LCSโs and $`T๐(X,Y)`$. Let $`B`$ be a $`๐ซ^{}(X)`$โbounded subset of $`X`$ and we want to see that $`TB`$ is $`๐ซ^{}(Y)`$โbounded in $`Y`$. Since $`B^{}`$ is a $`๐ซ(X_\beta ^{})`$โneighborhood of zero of the strong dual $`X_\beta ^{}`$ of $`X`$, $`(TB)^{}=(T^{})^1B^{}`$ is a $`๐ซ(Y_\beta ^{})`$โneighborhood of zero of $`Y_\beta ^{}`$ as a consequence of (GT<sub>2</sub>) and the fact that $`T^{}๐(Y_\beta ^{},X_\beta ^{})`$. Hence, $`TB`$ is $`๐ซ^{}(Y)`$โbounded in $`Y`$, as asserted.
Finally, for (8) we note that (GT<sub>1</sub>) is plain. For (GT<sub>2</sub>), let $`X`$ and $`Y`$ be LCSโs and $`T๐(X,Y)`$. Let $`V`$ be a $`^{}(Y)`$-neighborhood of zero of $`Y`$ and we want to see that $`T^1V`$ is an $`(X)`$โneighborhood of zero of $`X`$. Since $`V^{}`$ is $`(Y_\beta ^{})`$โbounded in $`Y_\beta ^{}`$, $`(T^1V)^{}=T^{}V^{}`$ is a $`(X_\beta ^{})`$โbounded subset of $`X_\beta ^{}`$, as asserted. โ
###### Remark 4.4.
A seemingly more general setting is to define for generating topologies $`๐ซ`$ and $`๐ซ_1`$, and generating bornologies $``$ and $`_1`$ the operator ideals with components $`๐ช(๐ซ/๐ซ_1)(X,Y)=๐(X_๐ซ,Y_{๐ซ_1})`$, $`๐ช(/_1)(X,Y)=L^\times (X^{},Y^_1)๐(X,Y)`$ and $`๐ช(๐ซ/)(X,Y)=๐(X_๐ซ,Y^{})`$. However, they will not give rise to new tools to us. In fact, we have $`๐ช(๐ซ/๐ซ_1)=๐ช(๐ซ_1)^1๐ช(๐ซ)`$ , $`๐ช(/_1)=๐ช()๐ช(_1)^1`$ , and $`๐ช(๐ซ/)=๐ช^b()๐ช(๐ซ)=๐ช()๐ช^b(๐ซ)`$. Readers are referred to Pietschโs classic for information regarding quotients and products of operator ideals.
Let $`p`$ be a continuous seminorm of a LCS $`X`$ and $`B`$ an absolutely convex bounded subset of a LCS $`Y`$. Denote by $`X_p`$ the normed space $`X/p^1(0)`$ equipped with norm $`x+p^1(0)=p(x)`$, and by $`Y(B)`$ the normed space $`_{\lambda >0}\lambda B`$ equipped with norm $`r_B(x)=inf\{\lambda >0:x\lambda B\}`$. Let $`\stackrel{~}{X}_p`$ be the completion of $`X_p`$. Define $`Q_p:XX_p`$, $`\stackrel{~}{Q}_p:X\stackrel{~}{X}_p`$ and $`J_B:Y(B)Y`$ to be the canonical maps.
###### Theorem 4.5 ().
Let $`๐`$ be an operator ideal on LCSโs. We have
1. A continuous seminorm $`p`$ of $`X`$ is an $`๐`$โseminorm if and only if $`Q_p๐^{\mathrm{inj}}(X,X_p)`$ if and only if $`\stackrel{~}{Q}_p๐^{\mathrm{inj}}(X,\stackrel{~}{X}_p)`$.
2. A bounded disk $`B`$ of $`Y`$ is an $`๐`$โbounded set if and only if $`J_B๐^{\mathrm{bsur}}(Y(B),Y)`$. Whenever $`๐`$ is surjective, we can replace $`๐^{\mathrm{bsur}}`$ by $`๐^{\mathrm{sur}}`$.
For operator ideals $`๐`$ on Banach spaces, Stephani achieved that $`๐ช(๐ฏ(๐))=๐^{\mathrm{inj}}`$ and $`๐ช((๐))=๐^{\mathrm{sur}}`$. However, we have two constructions $`๐ช`$ and $`๐ช^b`$ in the context of LCSโs. Unlike the Banach space version, they give rise to different ideals. For example, let $`_{pc}`$ be the generating system of precompact bornologies (i.e., the bornologies determined by totally bounded convex sets). Then $`๐_p=๐ช^b(_{pc})`$ is the ideal of precompact operators (i.e., those sending a neighborhood of zero to a totally bounded set) and $`๐_p^{loc}=๐ช(_{pc})`$ is the ideal of locally precompact operators (i.e., those sending bounded sets to totally bounded sets). Randtke indicated that $`๐_p(X,Y)=๐_p^{loc}(X,Y)`$ holds for all LCS $`Y`$ if and only if $`X`$ is a Schwartz space. On the other hand, it is straightforward to make the following observation.
###### Proposition 4.6.
For a generating topology $`๐ซ`$ and a generating bornology $``$ on LCSโs, $`๐ช(๐ซ)`$ and $`๐ช^b(๐ซ)`$ give rise to the same ideal topology, namely
$$๐ฏ(๐ช(๐ซ))=๐ฏ(๐ช^b(๐ซ))=๐ซ,$$
and $`๐ช()`$ and $`๐ช^b()`$ give rise to the same ideal bornology, namely
$$(๐ช())=(๐ช^b())=.$$
Moreover, $`๐ช(๐ซ)`$ and $`๐ช^b(๐ซ)`$ are injective, $`๐ช()`$ is bornologically surjective and $`๐ช^b()`$ is surjective.
###### Proposition 4.7.
Let $`๐`$ be an operator ideal on LCSโs. We have
1. $`๐ช^b(๐ฏ(๐))๐^{\mathrm{inj}}๐ช(๐ฏ(๐))`$.
2. $`๐ช^b((๐))๐^{\mathrm{bsur}}๐ช((๐))`$.
###### Proof.
Let $`T`$ be a (topologically) bounded linear operator from a LCS $`X`$ into a LCS $`Y`$, i.e., $`T๐^b(X,Y)`$. Then there is a continuous seminorm $`p`$ of $`X`$ and an absolutely convex bounded subset $`B`$ of $`Y`$ such that $`T`$ sends $`V_p=\{xX:p(x)1\}`$ into $`B`$. It is plain that $`T`$ has a decomposition
$$\begin{array}{ccc}X& \stackrel{T}{}& Y\\ Q_p& & J_B& & \\ X_p& \underset{T_0}{}& Y(B),\end{array}$$
where $`T_0๐(X_p,Y(B))`$ is the unique bounded operator induced by $`T`$.
If $`T๐ช^b(๐ฏ(๐))(X,Y)`$ then $`p`$ can be chosen to be an $`๐`$โseminorm of $`X`$. By Theorem 4.5, $`Q_p๐^{\mathrm{inj}}(X,X_p)`$ and hence $`T=J_BT_0Q_p๐^{\mathrm{inj}}(X,Y)`$. Similarly, if $`T๐ช^b())(X,Y)`$ then $`B`$ can be chosen to be an $`๐`$โbounded subset of $`Y`$. By Theorem 4.5 again, $`J_B๐^{\mathrm{bsur}}(Y(B),Y)`$ and hence $`T=J_BT_0Q_p๐^{\mathrm{bsur}}(X,Y)`$. In other words, $`๐ช^b(๐ฏ(๐))๐^{\mathrm{inj}}`$ and $`๐ช^b((๐))๐^{\mathrm{bsur}}`$. The other inclusions follows from the injectivity of $`๐ช(๐ฏ(๐))`$ and the bornological surjectivity of $`๐ช(๐ฏ(๐))`$. โ
###### Proposition 4.8.
Let $`๐ซ`$ be a generating topology on LCSโs. If the operator ideal $`๐=๐ช(๐ซ)`$ is symmetric (resp. $`๐=๐ช^b(๐ซ)`$ is symmetric) then
$$๐ซ^{}(Y)=(๐ช(๐ซ))(Y)(\text{resp. }๐ซ^{}(Y)=(๐ช^b(๐ซ))(Y))$$
for all infrabarrelled LCS $`Y`$.
###### Proof.
Let $`B`$ be a bounded disk in $`Y`$. Suppose firstly that $`B`$ is $`๐`$-bounded. Then there is a normed space $`N`$ such that $`TU_NB`$. Hence $`B^{}(T^{})^1U_N^{}`$. Now, the symmetry of $`๐`$ implies $`T^{}๐(Y_\beta ^{},N^{})`$. Thus, $`B^{}`$ is an $`๐`$-neighborhood of zero of $`Y_\beta ^{}`$. It follows from $`๐ฏ(๐)=๐ฏ(๐ช(๐ซ))=๐ฏ(๐ช^b(๐ซ))=๐ซ`$ that $`B^{}`$ is $`๐ซ(Y_\beta ^{})`$-neighborhood of zero of $`Y_\beta ^{}`$. Hence $`B`$ is $`๐ซ^{}(Y)`$-bounded.
Conversely, assume that $`B`$ is $`๐ซ^{}`$-bounded in $`Y`$. In other words, $`B^{}`$ is an $`๐`$-neighborhood of zero of $`Y_\beta ^{}`$. Therefore, there is a Banach space $`F`$ and a $`T`$ in $`๐(Y_\beta ^{},F)`$ such that $`B^{}T^1U_F`$. Hence the second polar $`B^{}`$ of $`B`$ in $`Y_{\beta \beta }^{\prime \prime }`$ is $`๐`$-bounded since $`B^{}T^{}U_F^{}`$ and $`T^{}๐(F^{},Y_{\beta \beta }^{\prime \prime })`$. Let $`K_Y`$ be the canonical embedding of $`Y`$ into $`Y_{\beta \beta }^{\prime \prime }`$. The infrabarrelledness of $`Y`$ ensures that $`K_Y`$ is a topological injection. As a result, the inclusion $`K_YBB^{}`$ establishes the existence of a $`k_B`$ in $`๐(Y(B),Y_{\beta \beta }^{\prime \prime }(B^{}))`$ such that $`J_B^{}k_B=K_XJ_B`$. Then $`J_B(๐^{\mathrm{bsur}})^{\mathrm{inj}}(Y(B),Y)`$ because $`J_B^{}๐^{\mathrm{bsur}}(Y_{\beta \beta }^{\prime \prime }(B^{}),Y_{\beta \beta }^{\prime \prime })`$ by Theorem 4.5. However, $`(๐^{\mathrm{bsur}})^{\mathrm{inj}}=(๐^{\mathrm{inj}})^{\mathrm{bsur}}=๐^{\mathrm{bsur}}`$ since $`๐=๐ช(๐ซ)`$ (or $`๐=๐ช^b(๐ซ)`$) is always injective. This implies that $`B`$ is $`๐`$-bounded, i.e., $`(๐ช(๐ซ))`$-bounded in $`Y`$, by Theorem 4.5 again. โ
###### Proposition 4.9.
Let $``$ be a generating bornology on LCSโs. If the operator ideal $`๐=๐ช()`$ is symmetric (resp. $`๐=๐ช^b()`$ is symmetric) then
$$^{}(X)=๐ฏ(๐ช())(X)(\text{resp. }^{}(X)=๐ฏ(๐ช^b())(X))$$
for all infrabarrelled LCS $`X`$.
###### Proof.
Let $`V`$ be a closed, absolutely convex neighborhood of zero of $`X`$. Suppose firstly that $`V`$ is an $`๐`$-neighborhood of $`X`$ then there is a normed space $`N`$ and an $`T`$ in $`๐(X,N)`$ such that $`T^1U_NV`$. Hence $`V^{}T^{}U_N^{}`$ and thus $`V^{}`$ is $``$-bounded in $`X_\beta ^{}`$ since $`T^{}๐(N^{},X_\beta ^{})`$. So $`V`$ is an $`^{}(X)`$-bounded subset of $`X`$.
Conversely, assume that $`V`$ is an $`^{}`$-neighborhood of zero of $`X`$. Then $`V^{}`$ is $`(X_\beta ^{})`$-bounded in the strong dual space $`X_\beta ^{}`$ of $`X`$. Hence there is a normed space $`N`$ and an $`T`$ in $`๐(N,X_\beta ^{})`$ such that $`TU_NV^{}`$. Consequently, $`V^{}(T^{})^1U_N^{}`$ and thus $`V^{}`$ is an $`๐`$-neighborhood of zero of $`X_{\beta \beta }^{\prime \prime }`$ as $`T^{}๐(X_{\beta \beta }^{\prime \prime },N^{})`$. Since $`X`$ is infrabarrelled, $`K_X`$ is continuous. By (GT<sub>2</sub>), $`V=K_X^1V^{}=V^{}X`$ is an $`๐`$-neighborhood of zero of $`X`$. โ
###### Definition 4.10.
A generating topology $`๐ซ`$ on $`LCS`$โs is said to have the subspace property if whenever $`Y`$ is a subspace of a $`LCSX`$, $`Y_๐ซ`$ is also a subspace of $`X_๐ซ`$, i.e., the $`๐ซ`$โtopology of $`Y`$ coincides with the subspace topology inherited from the $`๐ซ`$โtopology of $`X`$. See Jarchow for the Banach space version.
Let $`๐`$ be an operator ideal on LCSโs or Banach spaces. $`๐^{\mathrm{dual}}`$ denotes the operator ideal with components
$$๐^{\mathrm{dual}}(X,Y)=\{T(X,Y):T^{}๐(Y_\beta ^{},X_\beta ^{})\}.$$
###### Proposition 4.11.
Let $`๐ซ`$ be a generating topology on $`LCS`$โs and $`X`$ be an infrabarrelled $`LCS`$. Then
1. $`๐ช^b(๐ซ)^{\mathrm{dual}}(X,Y)=๐ช^b(๐ซ^{})(X,Y),LCSY`$.
2. $`๐ช^b(๐ซ^{})^{\mathrm{dual}}(X,Y)๐ช^b(๐ซ)(X,Y),LCSY`$.
If, in addition, $`๐ช(๐ซ)`$ is symmetric or $`๐ซ`$ has the subspace property then
1. $`๐ช^b(๐ซ^{})^{\mathrm{dual}}(X,Y)=๐ช^b(๐ซ)(X,Y),LCSY`$.
###### Proof.
(a) Let $`T๐ช^b(๐ซ)^{\mathrm{dual}}(X,Y)`$, i.e., $`T^{}๐ช^b(๐ซ)(Y_\beta ^{},X_\beta ^{})`$. Then there is a $`๐ซ(Y_\beta ^{})`$โneighborhood $`V`$ of $`0`$ in $`Y_\beta ^{}`$ such that $`T^{}V`$ is bounded in $`X_\beta ^{}`$. Hence $`U=(T^{}V)^{}=T^1V^{}`$ is a closed bornivorous barrel in $`X`$. Since $`X`$ is infrabarrelled, $`U`$ is a $`0`$โneighborhood in $`X`$. Now $`TUV^{}`$ ensures that $`T๐ช^b(๐ซ^{})(X,Y)`$. Conversely, if $`T๐ช^b(๐ซ^{})(X,Y)`$ then there is a $`0`$โneighborhood $`U`$ in $`X`$ such that $`A=TU`$ is $`๐ซ^{}(Y)`$โbounded in $`Y`$. Hence $`A^{}=(T^{})^1U^{}`$ is a $`๐ซ(Y_\beta ^{})`$โneighborhood of $`0`$ in $`Y_\beta ^{}`$. Now $`T^{}A^{}U^{}`$ implies that $`T^{}๐ช^b(๐ซ)(Y_\beta ^{},X_\beta ^{})`$.
(b) Let $`T๐ช^b(๐ซ^{})^{\mathrm{dual}}(X,Y)`$, i.e., $`T^{}๐ช^b(๐ซ^{})(Y_\beta ^{},X_\beta ^{})`$. Then there is a $`0`$โneighborhood $`V`$ in $`Y_\beta ^{}`$ such that $`T^{}V`$ is $`๐ซ^{}(X_\beta ^{})`$โbounded in $`X_\beta ^{}`$. Hence $`U=(T^{}V)^{}`$ is a $`๐ซ(X_{\beta \beta }^{\prime \prime })`$โneighborhood of $`0`$ in $`X_{\beta \beta }^{\prime \prime }`$. Let $`U_0=K_X^1U`$. By the functorial property $`(GT_2)`$ of $`๐ซ`$, $`U_0`$ is a $`๐ซ(X)`$โneighborhood of $`0`$ in $`X`$. It is easy to see that $`TU_0V^{}`$ and thus $`T๐ช^b(๐ซ)(X,Y)`$.
(b) Assume, in addition to those in (b), that $`๐=๐ช(๐ซ)`$ is symmetric. Let $`T๐ช^b(๐ซ)(X,Y)`$. We want to verify that $`T^{}๐ช^b(๐ซ^{})(Y_\beta ^{},X_\beta ^{})`$. By assumption, there is a $`๐ซ(X)`$โneighborhood $`U`$ of $`0`$ in $`X`$ such that $`A=TU`$ is bounded in $`Y`$. Now $`A^{}=(T^{})^1U^{}`$ suggests us to check if $`U^{}`$ is $`๐ซ^{}(X_\beta ^{})`$โbounded in $`X_\beta ^{}`$. Since $`๐ซ=๐ฏ(๐ช(๐ซ))=๐ฏ(๐)`$, there is a Banach space $`F`$ and an $`R`$ in $`๐(X,F)`$ such that $`UR^1U_F`$. Therefore, $`U^{}R^{}U_F^{}`$ and thus $`U^{}((R^{})^{})^1U_{F^{\prime \prime }}`$ where $`(R^{})^{}`$ is the double adjoint of $`R`$ from $`X_{\beta \beta }^{\prime \prime }`$ into $`F^{\prime \prime }`$. Since $`๐`$ is symmetric, $`(R^{})^{}๐(X_{\beta \beta }^{\prime \prime },F^{\prime \prime })`$ and thus $`U^{}`$ is a $`๐ซ(X_{\beta \beta }^{\prime \prime })`$โneighborhood of $`0`$ in $`X_{\beta \beta }^{\prime \prime }`$. Consequently, $`U^{}`$ is a $`๐ซ^{}(X_\beta ^{})`$โbounded subset of $`X_\beta ^{}`$, as asserted.
Finally, if the subspace property of $`๐ซ`$ is assumed instead of the symmetry of $`๐ช(๐ซ)`$ then the $`๐ซ(X)`$โneighborhood $`U`$ of $`0`$ in $`X`$ above is induced from a $`๐ซ(X_{\beta \beta }^{\prime \prime })`$โneighborhood $`V`$ of $`0`$ in $`X_{\beta \beta }^{\prime \prime }`$, i.e., $`K_XU=VK_XX`$ and thus $`U^{}=V^{}`$ is $`๐ซ^{}(X_\beta ^{})`$โbounded in $`X_\beta ^{}`$, where $`K_X`$ is the evaluation map from $`X`$ into $`X_{\beta \beta }^{\prime \prime }`$. โ
## 5. LCSโs defined by operators, topologies and bornologies
###### Theorem 5.1.
Let $`๐`$ be an operator ideal on LCSโs and $`X`$ be a LCS. The following are all equivalent.
1. $`X`$ is $`๐`$โtopological.
2. For each continuous seminorm $`p`$ on $`X`$, $`Q_p๐^{\mathrm{inj}}(X,X_p)`$, or equivalently, $`\stackrel{~}{Q}_p๐^{\mathrm{inj}}(X,\stackrel{~}{X}_p)`$.
3. $`^b(X,Y)๐^{\mathrm{inj}}(X,Y)`$ for every LCS $`Y`$.
4. $`(X,F)=๐^{\mathrm{inj}}(X,F)`$ for every normed (or Banach) space $`F`$.
5. $`id_X(X_๐,X)`$, where $`X_๐`$ is the LCS $`X`$ equipped with the $`๐`$โtopology.
###### Proof.
(1)$``$(2) is contained in Theorem 4.5. (1)$``$(5) and (2)$``$(3)$``$(4) are trivial. (4)$``$(1) is due to Theorem 4.5 again. โ
In the following, $`๐`$ denotes either the class of all LCSโs or the class of all Banach spaces. The next result is a generalization of a result of Jarchow \[9, Proposition 3\].
###### Theorem 5.2 ().
Let $`๐`$ be a surjective operator ideal on $`๐`$. If $`X`$, $`Y๐`$ and $`Y`$ is a (topological) quotient space of $`X`$ then the $`๐`$โtopology of $`Y`$ is the quotient topology induced by the $`๐`$โtopology of $`X`$. In particular, a quotient space of an $`๐`$โtopological space is again an $`๐`$โtopological space.
###### Theorem 5.3.
Let $`๐`$ be an operator ideal on LCSโs and $`Y`$ be a LCS. The following are all equivalent.
1. $`Y`$ is $`๐`$โbornological.
2. $`J_B๐^{\mathrm{bsur}}(Y(B),Y)`$ for each bounded disk $`B`$ in $`Y`$.
3. $`^b(X,Y)๐^{\mathrm{bsur}}(X,Y)`$ for every LCS $`X`$.
4. $`(N,Y)=๐^{\mathrm{bsur}}(N,Y)`$ for every normed space $`N`$.
5. $`id_YL^\times (Y,Y^๐)`$, where $`Y^๐`$ is the convex bornological vector space $`Y`$ equipped with the $`๐`$โbornology.
In case $`Y`$ is infracomplete they are all equivalent to
1. $`(E,Y)=๐^{\mathrm{bsur}}(E,Y)`$ for every Banach space $`E`$.
If $`๐`$ is surjective we can replace $`๐^{\mathrm{bsur}}`$ by $`๐`$ in all of the above statements.
###### Proof.
(1)$``$(5) is by definition. It is plain that (2)$``$(3)$``$(4)$``$(4). (4)$``$(1) (or (4)$`{}_{}{}^{}1`$ in case $`Y`$ is infracomplete) and (1)$``$(2) is due to Theorem 4.5. The last assertion is a consequence of Proposition 3.2. โ
Let $`๐`$ be either the class of all LCSโs or the class of all Banach spaces.
###### Theorem 5.4.
Let $`๐`$ be an injective operator ideal on $`๐`$ and $`X,Y๐`$. If $`Y`$ is a (topological) subspace of $`X`$ then the $`๐`$โbornology of $`Y`$ is the subspace bornology inherited from the $`๐`$โbornology of $`X`$. In particular, a subspace of an $`๐`$โbornological space is again an $`๐`$โbornological space.
###### Theorem 5.5.
Let $`๐`$ be an operator ideal on LCSโs. Let $`๐ซ=๐ฏ(๐)`$ be the ideal topology on LCSโs generated by $`๐`$. A LCS $`X`$ is $`๐`$โtopological if and only if
$$^b(X,Y)๐ช(๐ซ)(X,Y)=๐ช^b(๐ซ)(X,Y)$$
for each LCS $`Y`$.
###### Proof.
By Theorem 5.1, if $`X`$ is $`๐`$โtopological then $`๐ช^b(๐ซ)(X,Y)=^b(X,Y)`$ and $`๐ช(๐ซ)(X,Y)=(X,Y)`$. The equality follows. Conversely, assume the equality holds for every LCS $`Y`$. It suffices to show that $`(X,N)๐ช^b(๐ซ)(X,N)`$ for each normed space. Let $`N_๐ซ`$ be the LCS given by equipping $`N`$ with the $`๐ซ(N)`$โtopology. By the functorial property $`(GT_2)`$ of $`๐ซ`$, any $`T`$ in $`(X,N)`$ also belongs to $`(X_๐ซ,N_๐ซ)=๐ช(๐ซ)(X,N_๐ซ)`$. By the hypothesis, $`T๐ช^b(๐ซ)(X,N_๐ซ)`$. Since $`๐ซ(N)`$ is compatible with the dual pair $`(N,N^{})`$ by $`(GT_1)`$, we have $`T๐ช^b(๐ซ)(X,N)`$. It follows the desired assertion. โ
###### Remark 5.6.
If we let $`=(๐)`$ then a LCS $`Y`$ being $`๐`$โbornological implies
$$^b(X,Y)๐ช()(X,Y)=๐ช^b()(X,Y)$$
for each LCS $`X`$. We do not know if the converse is true.
Let $`๐`$ be an operator ideal on LCSโs. Denote by $`๐_๐น`$ the operator ideal defined on Banach spaces such that $`๐_๐น(E,F)=๐(E,F)`$ for every pair $`E`$ and $`F`$ of Banach spaces. Conversely, let $`๐`$ be an operator ideal on Banach spaces. There are many ways to extend $`๐`$ to an operator ideal $`๐_0`$ on LCSโs in the sense that $`(๐_0)_๐น=๐`$. In , Pietsch mentioned six different ways to extend $`๐`$ to an operator ideal on LCSโs. Among them, we are interested in
$`๐^{inf}`$ $`=`$ $`\{RS_0T๐(X,Y):T๐(X,X_0),S_0๐(X_0,Y_0),R๐(Y_0,Y)\},`$
$`๐^{\mathrm{rup}}`$ $`=`$ $`\{S๐(X,Y):B๐(Y,Y_0),A๐(X,X_0),S_0๐(X_0,Y_0)\text{ such that }BS=S_0A\},`$
$`๐^{\mathrm{lup}}`$ $`=`$ $`\{S๐(X,Y):B๐(X_0,X),A๐(Y_0,Y),S_0๐(X_0,Y_0)\text{ such that }SB=AS_0\},`$
$`๐^{sup}`$ $`=`$ $`\{S๐(X,Y):RST๐(X_0,Y_0),\text{ for all }T๐(X_0,X)\text{ and }R๐(Y,Y_0)\}.`$
Here, $`X,Y`$ run through all LCSโs and $`X_0,Y_0`$ run through all Banach spaces.
###### Definition 5.7 ().
Let $`๐`$ be an operator ideal on Banach spaces. We call a continuous seminorm $`p`$ on a LCS $`X`$ a $`\mathrm{Groth}(๐)`$seminorm if there is a continuous seminorm $`q`$ on $`X`$ such that $`pq`$ and $`\stackrel{~}{Q}_{pq}๐(\stackrel{~}{X}_q,\stackrel{~}{X}_p)`$. The $`\mathrm{Groth}(๐)`$topology on $`X`$ is defined to be the locally convex (Hausdorff) topology on $`X`$ which has a subbase determined by all $`\mathrm{Groth}(๐)`$โseminorms.
A LCS $`X`$ is a $`\mathrm{Groth}(๐)`$โspace if its topology coincides with the $`\mathrm{Groth}(๐)`$โtopology. It is equivalent to say that the identity map $`id_X๐^{\mathrm{rup}}(X,X)`$.
Let $`๐ซ`$ be a generating topology on Banach spaces. Define $`๐ซ^๐(X)`$ on each LCS $`X`$ to be the coarsest locally convex (Hausdorff) topology on $`X`$ among those $`๐ซ_0(X)`$ such that the inclusion
$$(X,F)(X_{๐ซ_0},F_๐ซ)$$
holds for every Banach space $`F`$. It is clear that for each LCS $`X`$, $`๐ซ_\sigma (X)๐ซ^๐(X)๐ซ_{\mathrm{ori}}(X)`$ and a continuous seminorm $`p`$ on $`X`$ is $`๐ซ^๐(X)`$โcontinuous if and only if there is a Banach space $`F`$, an $`S`$ in $`(X,F)`$ and a $`๐ซ(F)`$โcontinuous seminorm $`r`$ on $`F`$ such that $`p(x)r(Sx)`$ for all $`x`$ in $`X`$.
###### Lemma 5.8.
$`๐ซ^๐=\{๐ซ^๐(X):XLCS\}`$ is the minimal extension of $`๐ซ`$ to LCSโs.
###### Proof.
It is easy to see that $`๐ซ^๐`$ is a generating topology on LCSโs. Let $`E`$ be a Banach space. By definition of $`๐ซ^๐`$, $`๐ซ^๐(E)๐ซ(E)`$. On the other hand, $`id_E(E,E)(E_{๐ซ^๐},E_๐ซ)`$ implies $`๐ซ^๐(E)๐ซ(E)`$. So $`๐ซ^๐`$ is an extension of $`๐ซ`$ to LCSโs. The minimality of $`๐ซ^๐`$ is obvious. โ
###### Theorem 5.9.
Let $`๐`$ be an operator ideal on Banach spaces. The minimal extension $`๐ซ^๐`$ of $`๐ซ=๐ฏ(๐)`$ coincides with the $`\mathrm{Groth}(๐)`$โtopology.
###### Proof.
Without loss of generality, we can assume that $`๐`$ is injective since $`๐ฏ(๐)=๐ฏ(๐^{\mathrm{inj}})`$ by Theorem 4.5. Let $`p=rS`$ be a $`๐ซ^๐`$โcontinuous seminorm on a LCS $`X`$ where $`S(X,F)`$ and $`r`$ is a $`๐ซ`$โcontinuous seminorm on a Banach space $`F`$. Then we have $`\stackrel{~}{Q}_p(x)_{\stackrel{~}{X}_p}=\stackrel{~}{Q}_r(Sx)_{\stackrel{~}{F}_r}`$ for all $`x`$ in $`X`$. It follows that there is an isometry $`S_0`$ in $`(\stackrel{~}{X}_p,\stackrel{~}{F}_r)`$ such that $`S_0\stackrel{~}{Q}_p=\stackrel{~}{Q}_rS`$. Note that $`\stackrel{~}{Q}_r๐(F,\stackrel{~}{F}_r)`$. Define a continuous seminorm $`q`$ on $`X`$ by $`q(x)=\stackrel{~}{Q}_rSx`$. Now $`q(x)\stackrel{~}{Q}_rSx=p(x)`$ and we have an $`S_2`$ in $`(\stackrel{~}{X}_q,F)`$ induced by $`S`$. Since $`S_0`$ is an injection and $`S_0\stackrel{~}{Q}_{pq}=\stackrel{~}{Q}_rS_2๐(\stackrel{~}{X}_q,\stackrel{~}{F}_r)`$, $`\stackrel{~}{Q}_{pq}๐(\stackrel{~}{X}_q,\stackrel{~}{X}_p)`$, i.e., $`p`$ is a $`\mathrm{Groth}(๐)`$โseminorm.
Conversely, if $`p`$ is a $`\mathrm{Groth}(๐)`$โcontinuous seminorm on $`X`$ then there is a continuous seminorm $`q`$ on $`X`$ with $`pq`$ such that $`\stackrel{~}{Q}_{pq}๐(\stackrel{~}{X}_q,\stackrel{~}{X}_p)=((\stackrel{~}{X}_q)_๐ซ,\stackrel{~}{X}_p)`$ by Theorem 2.2(4a). In other words, the seminorm $`r`$ on $`\stackrel{~}{X}_q`$ defined by $`r(y)=\stackrel{~}{Q}_{pq}(y)_{\stackrel{~}{X}_p}`$, $`y\stackrel{~}{X}_q`$, is $`๐ซ`$โcontinuous. Note that $`\stackrel{~}{Q}_p=\stackrel{~}{Q}_{pq}\stackrel{~}{Q}_q`$ implies that $`p(x)=\stackrel{~}{Q}_{pq}\stackrel{~}{Q}_q(x)=r(\stackrel{~}{Q}_qx)`$. It simply says that $`p`$ is a $`๐ซ^๐`$โcontinuous seminorm. โ
###### Theorem 5.10.
Let $`๐`$ be an operator ideal on Banach spaces with $`๐ซ=๐ฏ(๐)`$. Then $`๐ช(๐ซ^๐)=(๐^{\mathrm{inj}})^{\mathrm{rup}}`$.
###### Proof.
Let $`X`$ and $`Y`$ be LCSโs. Assume $`T๐ช(๐ซ^๐)(X,Y)`$. Then for every Banach space $`F`$ and $`S`$ in $`(Y,F)`$, $`ST๐ช(๐ซ^๐)(X,F)=(X_{๐ซ^๐},F)`$. Hence there is a $`๐ซ^๐`$โcontinuous seminorm $`p`$ on $`X`$ such that $`STxp(x)`$. By Theorem 5.9, there is a continuous seminorm $`q`$ on $`X`$ such that $`pq`$ and $`\stackrel{~}{Q}_{pq}๐^{\mathrm{inj}}(\stackrel{~}{X}_q,\stackrel{~}{X}_p)`$. Let $`R๐(\stackrel{~}{X}_p,F)`$ is induced by the inequality $`STxp(x)`$. It is then not difficult to see that $`ST=R\stackrel{~}{Q}_{pq}\stackrel{~}{Q}_q`$, and thus $`T(๐^{\mathrm{inj}})^{\mathrm{rup}}(X,Y)`$.
Conversely, assume $`T(๐^{\mathrm{inj}})^{\mathrm{rup}}(X,Y)`$. Then for every continuous seminorm $`p`$ on $`Y`$ there exists a Banach space $`E`$, an $`R`$ in $`(X,E)`$ and an $`S`$ in $`๐^{\mathrm{inj}}(E,\stackrel{~}{Y}_p)`$ such that $`\stackrel{~}{Q}_pT=SR`$. Now $`S(E_๐ซ,\stackrel{~}{Y}_p)`$ and $`R(X,E)(X_{๐ซ^๐},E_๐ซ)`$ imply $`\stackrel{~}{Q}_pT=SR(X_{๐ซ^๐},\stackrel{~}{Y}_p)`$. Since it is true for every continuous seminorm $`p`$ on $`Y`$, $`T(X_{๐ซ^๐},Y)`$, i.e., $`T๐ช(๐ซ^๐)(X,Y)`$. โ
###### Definition 5.11 ().
Let $`๐`$ be an operator ideal on Banach spaces. A bounded $`\sigma `$โdisk $`A`$ in a LCS $`X`$ is said to be $`\mathrm{Groth}(๐)`$bounded in $`X`$ if there is a bounded $`\sigma `$โdisk $`B`$ in $`X`$ such that $`AB`$ and the canonical map $`J_{BA}๐(X(A),X(B))`$. Note that, in this case, both $`X(A)`$ and $`X(B)`$ are Banach spaces. The $`\mathrm{Groth}(๐)`$bornology on a LCS $`X`$ is defined to be the convex vector bornology on $`X`$ with a subbase consisting of $`\mathrm{Groth}(๐)`$โbounded $`\sigma `$โdisks in $`X`$.
A LCS is a co$`\mathrm{Groth}(๐)`$space, if all bounded $`\sigma `$โdisks in $`X`$ are $`\mathrm{Groth}(๐)`$โbounded. It is equivalent to say that $`id_X๐^{\mathrm{lup}}(X,X)`$.
Let $``$ be a generating bornology on Banach spaces. We define, for each LCS $`X`$, a convex vector bornology $`^๐(X)`$ on $`X`$ to be the smallest convex (separated) vector bornology among those $`_0`$ on $`X`$ such that
$$(E,X)L^\times (E^{},X^_0)$$
holds for every Banach space $`E`$. It is easy to see that $`_{\mathrm{fin}}(X)^๐(X)_{\mathrm{von}}(X)`$ and the family of subsets $`B`$ in $`X`$ in the form of $`B=TA`$ for some $`T(E,X)`$ and $``$โbounded set $`A`$ in a Banach space $`E`$ forms a basis of the bornology $`^๐(X)`$ for each LCS $`X`$.
###### Lemma 5.12.
Let $``$ be a generating bornology on Banach spaces. $`^๐`$ is the minimal extension of $``$ to LCSโs.
###### Proof.
Similar to Lemma 5.8. โ
###### Theorem 5.13.
Let $`๐`$ be an operator ideal on Banach spaces. The minimal extension $`^๐`$ of $`=(๐)`$ coincides with the $`\mathrm{Groth}(๐)`$โbornology.
###### Proof.
Without loss of generality, we can assume that $`๐`$ is surjective since $`(๐)=(๐^{\mathrm{sur}})`$ by Theorem 4.5. Let $`A`$ be a $`\mathrm{Groth}(๐)`$โbounded $`\sigma `$โdisk in a LCS $`X`$. By definition, there is a bounded $`\sigma `$โdisk $`B`$ in $`X`$ such that $`AB`$ and $`J_{BA}๐(X(A),X(B))=L^\times (X(A),X(B)^{})`$. In other words, $`C=J_{BA}U_{X(A)}`$ is $``$โbounded in $`X(B)`$. Now $`AJ_AU_{X(A)}=J_BJ_{BA}U_{X(A)}=J_BC`$ implies that $`A`$ is $`M^๐`$โbounded.
Conversely, if $`A=SB`$ is $`^๐`$โbounded in $`X`$ with some $`S`$ in $`(E,X)`$ and $``$โbounded $`\sigma `$โdisk $`B`$ in a Banach space $`E`$. Let $`C=\lambda SU_E`$ for some $`\lambda >0`$ such that $`\lambda U_EB`$. We have $`CA`$. Let $`S_0๐(E(B),X(A))`$ and $`S_2๐(E,X(C))`$ be induced by $`S`$. Since $`B`$ is $``$โbounded in $`E`$, $`J_B๐(E(B),E)`$ and $`J_{CA}S_0=S_2J_B๐(E(B),X(C))`$. Finally the surjectivity of $`S_0`$ ensures that $`J_{CA}๐(X(A),X(C))`$. โ
###### Theorem 5.14.
Let $`๐`$ be an operator ideal on Banach spaces with $`=(๐)`$. Then
$$๐ช(^๐)(X,Y)(๐^{\mathrm{sur}})^{\mathrm{lup}}(X,Y),LCS\text{โs }X,Y.$$
If $`X`$ is infracomplete (in particular, a Banach space) then we have
$$๐ช(^๐)(X,Y)=(๐^{\mathrm{sur}})^{\mathrm{lup}}(X,Y),LCSY.$$
###### Proof.
Similar to a previous theorem except that we shall use Theorem 5.13 instead of Theorem 5.9. The introduction of the infracompleteness is merely to give us a chance to utilize the extension condition. โ
We provide a new proof for the following result.
###### Theorem 5.15 ().
Let $`๐`$ be an operator ideal on Banach spaces. The Groth$`(๐^{\mathrm{inj}})`$โtopology coincides with the $`๐^{\mathrm{rup}}`$โtopology on every LCS, and the Groth$`(๐^{\mathrm{sur}})`$โbornology coincides with the $`๐^{\mathrm{lup}}`$โbornology on every infracomplete LCS. In particular, we have
1. A LCS $`X`$ is a Groth$`(๐^{\mathrm{inj}})`$โspace if and only if $`X`$ is an $`๐^{\mathrm{rup}}`$โtopological space.
2. An infracomplete LCS $`X`$ is a coโGroth$`(๐^{\mathrm{sur}})`$โspace if and only if $`X`$ is an $`๐^{\mathrm{lup}}`$โbornological space.
3. The $`๐`$โtopology (resp. $`๐`$โbornology) coincides with the Groth$`(๐^{\mathrm{inj}})`$โtopology (resp. Groth$`(๐^{\mathrm{sur}})`$โbornology) on Banach spaces.
###### Proof.
Let $`๐ซ=๐ฏ(๐)`$ and $`=(๐)`$ be the ideal topology and the ideal bornology on Banach spaces generated by $`๐`$, respectively. Let $`p`$ be a continuous seminorm on a LCS $`X`$. We observe the following equivalences:
1. $`p`$ is a $`\mathrm{Groth}(๐^{\mathrm{inj}})`$โcontinuous seminorm on $`X`$.
2. $`p`$ is an $`๐ช(๐ซ^๐)`$โcontinuous seminorm on $`X`$ by Theorem 5.9.
3. $`p`$ is an $`(๐^{\mathrm{inj}})^{\mathrm{rup}}`$โcontinuous seminorm on $`X`$ by Theorem 5.10.
4. $`\stackrel{~}{Q}_p[(๐^{\mathrm{inj}})^{\mathrm{rup}}]^{\mathrm{inj}}(X,\stackrel{~}{X}_p)`$ by Theorem 4.5.
5. $`\stackrel{~}{Q}_p(๐^{\mathrm{rup}})^{\mathrm{inj}}(X,\stackrel{~}{X}_p)`$ by \[28, Proposition 3.5\].
6. $`p`$ is an $`๐^{\mathrm{rup}}`$โcontinuous seminorm on $`X`$ by Theorem 4.5.
For the bornological case, assuming that $`X`$ is infracomplete, we have for each bounded $`\sigma `$โdisk $`A`$ in $`X`$:
1. $`A`$ is a coโ$`\mathrm{Groth}(๐^{\mathrm{sur}})`$โbounded set in $`X`$.
2. $`A`$ is an $`๐ช(^๐)`$โbounded set in $`X`$ by Theorem 5.13.
3. $`A`$ is an $`(๐^{\mathrm{sur}})^{\mathrm{lup}}`$โbounded set in $`X`$ by Theorems 4.5 and 5.14.
4. $`J_A[(๐^{\mathrm{sur}})^{\mathrm{lup}}]^{\mathrm{bsur}}(X(A),X)`$ by Theorem 4.5.
5. $`J_A(๐^{\mathrm{lup}})^{\mathrm{bsur}}(X(A),X)`$ by \[28, Proposition 3.5\].
6. $`A`$ is an $`๐^{\mathrm{lup}}`$โbounded set in $`X`$ by Theorem 4.5.
###### Proposition 5.16.
Let $``$ be a generating bornology on LCSโs and $`๐=๐ช^b()`$. The $`๐`$โtopology coincides with the Grothendieck topology generated by $`๐`$ on every LCS.
###### Proof.
It is easy to see that $`๐๐_๐น^{\mathrm{rup}}`$. The result follows from Theorems 5.9 and 5.15. โ
###### Proposition 5.17.
Let $``$ be a generating bornology on LCSโs and $`๐=๐ช()`$. Then the $`๐`$โtopology coincides with the $`(๐_๐น^{\mathrm{inj}})^{sup}`$โtopology on each infracomplete LCS.
###### Proof.
Since $`๐(๐_๐น^{\mathrm{inj}})^{sup}`$, $`๐ฏ(๐)`$ is always weaker than $`๐ฏ((๐_๐น^{\mathrm{inj}})^{sup})`$ on each LCS. By \[28, Corollary 3.2\], $`(๐_๐น^{\mathrm{inj}})^{sup}`$ is injective. Let $`p`$ be a $`๐ฏ((๐_๐น^{\mathrm{inj}})^{sup})`$โcontinuous seminorm on an infracomplete LCS $`X`$. Then $`\stackrel{~}{Q}_p(๐_๐น^{\mathrm{inj}})^{sup}(X,\stackrel{~}{X}_p)`$. Let $`B`$ be a bounded $`\sigma `$โdisk in $`X`$. Now $`J_{\stackrel{~}{X}_p}\stackrel{~}{Q}_pJ_B๐_๐น^{\mathrm{inj}}(X(B),\stackrel{~}{X}_p^{\mathrm{inj}})`$ implies $`\stackrel{~}{Q}_pJ_B๐_๐น^{\mathrm{inj}}(X(B),\stackrel{~}{X}_p)`$ and then again implies $`J_{\stackrel{~}{X}_p}\stackrel{~}{Q}_pJ_B๐_๐น(X(B),\stackrel{~}{X}_p^{\mathrm{inj}})`$ by Proposition 3.2. Consequently, $`J_{\stackrel{~}{X}_p}\stackrel{~}{Q}_p๐_๐น=๐ช(_๐น)`$. It turns out that $`\stackrel{~}{Q}_p๐^{\mathrm{inj}}`$, or equivalently, $`p`$ is an $`๐`$โcontinuous seminorm by Theorem 4.5. โ
###### Example 5.18.
Let $`X=๐^{(I)}`$ be the locally convex direct sum of card $`(I)`$ many $`๐`$โs where the index set $`I`$ is uncountable. $`X`$ is infracomplete. Let $`_{pc}`$ be the generating bornology of precompact sets ($`=`$ totally bounded sets). Then $`๐ช^b(_{pc})=๐_p`$, the ideal of all precompact operators and $`๐ช(_{pc})=๐_p^{\mathrm{loc}}`$, the ideal of all locally precompact operators, i.e., those sending bounded sets onto precompact sets. $`๐_p`$ is surjective but not bornologically surjective and $`๐_p^{\mathrm{loc}}`$ is bornologically surjective. Now $`id_X๐_p^{\mathrm{loc}}`$ implies $`X`$ is a $`๐_p^{\mathrm{loc}}`$โtopological space. On the other hand, $`X`$ is not a $`๐_p`$โtopological space (cf. \[6, p. 40\]). This serves as a counterโexample of $`๐^{sup}`$โtopology $`=๐^{inf}`$โtopology and $`๐^{sup}`$โtopological spaces $`=๐^{inf}`$โtopological spaces, although we always have $`๐^{\mathrm{rup}}`$โtopology $`=๐^{inf}`$โtopology and $`๐^{\mathrm{rup}}`$โtopological spaces $`=๐^{inf}`$โtopological spaces. By the way, $`X`$ is both $`๐_p`$โbornological and $`๐_p^{\mathrm{loc}}`$โ bornological, i.e., a coโSchwartz space but not a Schwartz space.
###### Proposition 5.19.
Let $`๐ซ`$ be a generating topology on LCSโs and $`๐=๐ช^b(๐ซ)`$. The $`๐`$โbornology coincides with the Grothendieck bornology generated by $`๐`$ on each infracomplete LCS.
###### Proof.
It follows from the easy fact $`๐๐_๐น^{\mathrm{lup}}`$ and Theorems 5.13 and 5.15. โ
###### Proposition 5.20.
Let $`๐ซ`$ be a generating topology on LCSโs and $`๐=๐ช(๐ซ)`$. Then the $`๐`$โbornology coincides with the $`(๐_๐น^{\mathrm{sur}})^{sup}`$โbornology on every infracomplete LCS.
###### Proof.
Similar to Proposition 5.17. โ
###### Example 5.21.
Let $`X=๐^I`$ be the product space of card$`(I)`$ many $`๐`$โs where the index set is uncountable. $`X`$ is infracomplete. Let $`๐ซ_{pc}`$ be the generating topology defined by the precompact seminorms, i.e., $`๐ซ_{pc}=๐ฏ(๐_p)`$, where $`๐_p`$ is the ideal of all precompact operators between LCSโs (see Wong ). Then $`๐ช^b(๐ซ_{pc})`$ is the ideal $`๐_p^b`$ of all quasiโSchwartz ($`=`$ precompactโbounded, cf. Rankte ) operators between LCSโs. $`๐ช(๐ซ_{pc})`$ is the ideal of those continuous operators between LCSโs which are still continuous when the domain space $`X`$ equipped with the (coarser) precompact topology $`๐ซ_{pc}(X)`$. $`X`$ is not a $`๐_p^b`$โbornological space since otherwise (by Theorem 5.3) we would have the canonical embedding from $`๐^{(I)}`$ into $`๐^I`$ being quasiโSchwartz and this is not the case as shown in \[11, p. 399\]. $`X`$ is, however, an $`๐ช(๐ซ_{pc})`$โbornological space since all bounded sets in $`X`$ are precompact. This serves as a counterโexample of $`๐^{sup}`$โbornology $`=๐^{inf}`$โbornology and $`๐^{sup}`$โ bornological spaces $`=๐^{inf}`$โbornological spaces, although we always have $`๐^{\mathrm{rup}}`$โbornology $`=๐^{inf}`$โ bornology on every infracomplete LCS. By the way, $`X`$ is both $`๐_p^b`$โtopological and $`๐ช(๐ซ_{pc})`$โtopological, i.e., a Schwartz space but not a coโSchwartz space.
###### Remark 5.22.
It may be interesting to study the $`๐^{sup}`$โtopology and the $`๐^{sup}`$โbornology for an operator ideal $`๐`$ on Banach spaces. Propositions 5.17 and 5.20 suggest the conjectures that $`๐ช(๐ฏ(๐))=(๐_๐น^{\mathrm{inj}})^{sup}`$ and $`๐ช((๐))=(๐_๐น^{\mathrm{sur}})^{sup}`$ where $`๐`$ is an operator ideal on LCSโs.
###### Theorem 5.23.
Let $`๐`$ be an operator ideal on LCSโs, and $`X`$ a LCS. Then
1. $`X`$ is $`๐^{\mathrm{dual}}`$โbornological $`X_\beta ^{}`$ is $`๐`$โtopological.
If, in addition, $`X`$ is infrabarrelled then
1. $`X`$ is $`๐^{\mathrm{dual}}`$โtopological $`X_\beta ^{}`$ is $`๐`$โbornological.
1. $`X_\beta ^{}`$ is $`๐^{\mathrm{dual}}`$โbornological $`X`$ is $`๐`$โtopological.
If, in addition to all above, $`๐`$ is also injective then
1. $`X_\beta ^{}`$ is $`๐^{\mathrm{dual}}`$โtopological $`X`$ is $`๐`$โbornological
###### Proof.
(a) Let $`V`$ be an absolutely convex, closed $`0`$โneighborhood in $`X_\beta ^{}`$. Then $`V^{}`$ is bounded and hence $`๐^{\mathrm{dual}}`$โbounded in $`X`$. So that there is a normed space $`N`$, a $`T`$ in $`๐^{\mathrm{dual}}(N,Y)`$ with $`TU_NV^{}`$. Consequently, $`(TU_N)^{}=(T^{})^1U_N^{}V^{}=V`$ and $`T^{}๐(X_\beta ^{},N^{})`$. It follows that $`V`$ is an $`๐`$โneighborhood of $`0`$ in $`X_\beta ^{}`$.
(b) Let $`B`$ be a bounded set in $`X_\beta ^{}`$. Then $`B^{}`$ is a closed bornivorous barrel in $`X`$, and hence a $`0`$โneighborhood, and consequently an $`๐^{\mathrm{dual}}`$โneighborhood of $`0`$ in $`X`$. Therefore there is a Banach space $`F`$, an $`T`$ in $`๐^{\mathrm{dual}}(X,F)`$ such that $`B^{}T^1(U_F)`$. It follows $`BB^{}T^{}U_F^{}`$. Since $`T^{}๐(F^{},X_\beta ^{})`$, $`B`$ is $`๐`$โbounded in $`X_\beta ^{}`$.
(c) Let $`V`$ be an absolutely convex, closed $`0`$โneighborhood in $`X`$. Then $`V^{}`$ is bounded and hence $`๐^{\mathrm{dual}}`$โbounded in $`X_\beta ^{}`$. So that there is a normed space $`N`$ and a $`T`$ in $`๐^{\mathrm{dual}}(N,X_\beta ^{})`$ such that $`TU_NV^{}`$ and thus $`(T^{})^1U_N^{}V^{}`$, where $`V^{}`$ is the polar of $`V^{}`$ in the strong bidual $`X_{\beta \beta }^{\prime \prime }`$ of $`X`$. It follows that $`V^{}`$ is an $`๐`$โneighborhood of $`0`$ in $`X_{\beta \beta }^{\prime \prime }`$. Since the evaluation map $`K_X:XX_{\beta \beta }^{\prime \prime }`$ is continuous, $`K_X^1(V^{})=V`$ is an $`๐`$โneighborhood of $`0`$ in $`X`$ by (GT<sub>2</sub>).
(d) Let $`B`$ be an absolutely convex bounded set in $`X`$. It suffices to check that $`J_B๐^{\mathrm{bsur}}(X(B),X)`$. Note that $`B^{}`$ is a neighborhood of $`0`$, and hence an $`๐^{\mathrm{dual}}`$โneighborhood of $`0`$ in $`X_\beta ^{}`$. Hence there exist a Banach space $`F`$ and a $`T`$ in $`๐^{\mathrm{dual}}(X_\beta ^{},F)`$ such that $`B^{}T^1(U_F)`$. Hence $`B^{}T^{}U_F^{}`$. Since $`T^{}๐(F^{},X_{\beta \beta }^{\prime \prime })`$, $`B^{}`$ is $`๐`$โbounded in $`X_{\beta \beta }^{\prime \prime }`$ and thus $`J_B^{}๐^{\mathrm{bsur}}(X_{\beta \beta }^{\prime \prime }(B^{}),X_{\beta \beta }^{\prime \prime })`$. Now $`K_XBB^{}`$ ensures that there is a $`K_B`$ in $`(X(B),X_{\beta \beta }^{\prime \prime }(B^{}))`$ such that $`J_B^{}K_B=K_XJ_B`$. Hence $`K_XJ_B๐^{\mathrm{bsur}}(X(B),X_{\beta \beta }^{\prime \prime })`$ and it follows
$$J_B(๐^{\mathrm{bsur}})^{\mathrm{inj}}(X(B),X)=(๐^{\mathrm{inj}})^{\mathrm{bsur}}(X(B),X)=๐^{\mathrm{bsur}}(X(B),X),$$
by Proposition 3.2. โ
###### Theorem 5.24.
Let $`๐`$ be a symmetric operator ideal (i.e., $`๐๐^{\mathrm{dual}}`$) on LCSโs, and $`X`$ an infrabarrelled LCS. Then
1. $`X`$ is $`๐`$โtopological $`X_\beta ^{}`$ is $`๐`$โbornological.
If, in addition, $`๐`$ is injective, then
1. $`X`$ is $`๐`$โbornological $`X_\beta ^{}`$ is $`๐`$โtopological.
###### Proof.
A consequence of Theorem 5.23. โ
###### Theorem 5.25.
Let $`๐ซ`$ be a generating topology on LCSโs and $`X`$ be an infrabarrelled LCS. Then
1. $`X`$ is $`๐ช^b(๐ซ^{})`$โtopological $`X_\beta ^{}`$ is $`๐ช^b(๐ซ)`$โbornological.
2. $`X`$ is $`๐ช^b(๐ซ^{})`$โbornological $`X_\beta ^{}`$ is $`๐ช^b(๐ซ)`$โtopological.
3. $`X_\beta ^{}`$ is $`๐ช^b(๐ซ^{})`$โtopological $`X`$ is $`๐ช^b(๐ซ)`$โbornological.
4. $`X_\beta ^{}`$ is $`๐ช^b(๐ซ^{})`$โbornological $`X`$ is $`๐ช^b(๐ซ)`$โtopological.
In case $`๐ช(๐ซ)`$ is symmetric or $`๐ซ`$ has the subspace property, all above implications become equivalences.
###### Proof.
We prove (c) only and all others are similar. Suppose $`X_\beta ^{}`$ is $`๐ช^b(๐ซ^{})`$โtopological. By Proposition 4.11(a), $`X_\beta ^{}`$ is also $`๐ช^b(๐ซ)^{\mathrm{dual}}`$โtopological. Note that $`๐ช^b(๐ซ)`$ is injective. Hence by Theorem 5.23(d), $`X`$ is $`๐ช^b(๐ซ)`$โbornological. In case $`๐ช(๐ซ)`$ is symmetric or $`๐ซ`$ has the subspace property, if $`X`$ is $`๐ช^b(๐ซ)`$โbornological, $`X`$ is also $`๐ช^b(๐ซ^{})^{\mathrm{dual}}`$โ bornological by Proposition 4.11(b). By Theorem 5.23(a), $`X_\beta ^{}`$ is $`๐ช^b(๐ซ^{})`$โtopological, as asserted. โ
###### Example 5.26.
Let $`๐ซ_\sigma `$ be the generating system of $`\sigma (X,X^{})`$โtopology on each LCS $`X`$. $`๐ซ_\sigma ^{}(X)`$ is thus the convex bornology $`(X)`$ consisting of those bounded subsets $`B`$ of $`X`$ whose polars $`B^{}`$ are $`\sigma (X_\beta ^{},X_{\beta \beta }^{\prime \prime })`$โneighborhoods of $`0`$ in $`X_\beta ^{}`$. Now both $`๐ช^b(๐ซ_\sigma )`$ and $`๐ช^b(๐ซ_\sigma ^{})`$ define the ideal $`๐`$ of continuous operators of finite rank. Moreover, $`๐ซ_\sigma `$ has the subspace property. Theorem 5.25 applies and says that for an infrabarrelled LCS $`X`$, we have
1. $`X`$ is $`๐`$โtopological if and only if $`X_\beta ^{}`$ is $`๐`$โbornological;
2. $`X`$ is $`๐`$โbornological if and only if $`X_\beta ^{}`$ is $`๐`$โtopological.
Unlike the case of Banach spaces, an $`๐`$โtopological or $`๐`$โbornological LCS need not be of finite dimension. For examples, the LCS $`๐^I`$ is $`๐`$โtopological and the LCS $`๐^{(I)}`$ is $`๐`$โbornological. This is because the weak topology $`\sigma (๐^I,๐^{(I)})`$ of $`๐^I`$ coincides with the product topology of $`๐^I`$ and every bounded set in $`๐^{(I)}`$ is of finite dimension. Here the index set $`I`$ is arbitrary.
## 6. Examples and Applications
This last section is devoted to examples and applications, showing the powerful techniques developed in the previous sections. Many other elegant applications of the theory of Grothendieck spaces and coโGrothendieck spaces can be found, for example, in , , , , . The concepts of $`๐`$โtopological spaces and $`๐`$โbornological spaces are also wellโdeveloped in the context in , , , , , and, in particular, .
### 6.1. Schwartz spaces and co-Schwartz spaces
###### Definition 6.1 (see, e.g., \[30, p. 14\]).
A continuous seminorm $`p`$ on a LCS $`X`$ is said to be precompact if there exists a $`(\lambda _n)`$ in $`c_0`$ and an equicontinuous sequence $`\{x_n^{}\}`$ in $`X^{}`$ such that
$$p(x)sup\{|\lambda _nx,x_n^{}|:n1\},xX.$$
Denote by $`๐ซ_{pc}(X)`$ the locally convex (Hausdorff) topology on $`X`$ defined by all precompact seminorms on $`X`$. It is easy to see that $`๐ซ_{pc}=\{๐ซ_{pc}(X):X`$ is a $`LCS\}`$ is a generating topology. It is a classical result (cf. or ) that $`p`$ is a precompact seminorm on a LCS $`X`$ if and only if the canonical map $`Q_p:XX_p`$ is precompact. Then, $`๐_p=๐ช(๐ซ_{pc})`$ is the ideal of all precompact operators, and $`๐_p^b=๐ช^b(๐ฏ(๐_p))`$ is the ideal of all quasiโSchwartz (i.e., precompactโbounded) operators between LCSโs.
###### Definition 6.2.
A LCS $`X`$ is said to be a Schwartz space if every continuous seminorm $`p`$ on $`X`$ is precompact.
We provide a new proof of the following classical result.
###### Theorem 6.3 (see \[30, pp. 17 and 26\]).
Let $`X`$ be a LCS. The following are all equivalent.
1. $`X`$ is a Schwartz space.
2. For each continuous seminorm $`p`$ on $`X`$ there is a continuous seminorm $`q`$ on $`X`$ such that $`pq`$ and the canonical map $`Q_{pq}`$ belongs to $`๐_p(X_q,X_p)`$.
3. $`Q_p๐_p(X,X_p)`$ for every continuous seminorm $`p`$ on $`X`$.
4. For any $`0`$โneighborhood $`U`$ in $`X`$ there exists a $`0`$โneighborhood $`V`$ in $`X`$ such that $`VU`$ and the canonical map from $`X^{}(U^{})`$ into $`X^{}(V^{})`$ is precompact.
5. $`(X,N)=๐_p(X,N)`$ for every normed (or Banach) space $`N`$.
6. $`๐_p^b(X,Y)=^b(X,Y)`$ for every LCS $`Y`$.
7. $`๐_p^b(X,Y)=๐_p(X,Y)`$ for every LCS $`Y`$.
8. $`(X,N)=๐_p^b(X,N)`$ for every normed (or Banach) space $`N`$.
9. $`X`$ is a $`๐_p`$โtopological space.
###### Proof.
(a)$``$(c)$``$(e)$``$(i) are due to Theorem 5.1 and the injectivity of $`๐_p`$. (a)$``$(b) because $`๐_p=๐ช^b(_{pc})`$ where $`_{pc}`$ is the generating bornology of precompact sets and Proposition 5.16 applies. (b)$``$(d) follows from the complete symmetry of the restriction $`(๐_p)_๐น`$ of $`๐_p`$ to Banach spaces, i.e., $`(๐_p)_๐น^{\mathrm{dual}}=(๐_p)_๐น`$. (a)$``$(f)$``$(h) are consequences of Theorem 5.1. (i)$``$(g) is contained in Proposition 4.7 and Theorem 5.5. Finally, for (g)$``$(h), denote by $`N_\sigma `$ the LCS $`(N,\sigma (N,N^{}))`$. For every $`T`$ in $`(X,N)`$, $`T(X,N_\sigma )=๐_p(X,N_\sigma )`$. Hence, by (g), $`T๐_p^b(X,N_\sigma )=๐_p^b(X,N)`$ since $`N`$ and $`N_\sigma `$ carry the same (von Neumann) bornology. โ
###### Definition 6.4.
A LCS $`Y`$ is said to be a coโSchwartz space if its strong dual $`Y_\beta ^{}`$ is a Schwartz space.
###### Theorem 6.5.
Let $`Y`$ be a LCS. Consider the following statements.
1. $`Y`$ is a coโSchwartz space.
2. For each bounded disk $`B`$ in $`Y`$ there is a bounded disk $`A`$ in $`Y`$ with $`BA`$ such that the canonical map $`J_{AB}`$ from $`Y(B)`$ into $`Y(A)`$ belongs to $`๐_p(Y(B),Y(A))`$.
3. $`J_B๐_p(Y(B),Y)`$ for each bounded disk $`B`$ in $`Y`$.
4. $`(N,Y)=๐_p(N,Y)`$ for every normed space $`N`$.
5. $`^b(X,Y)=๐_p(X,Y)`$ for every LCS $`X`$.
6. $`Y`$ is a $`๐_p`$โbornological space.
We have (a)$``$(b)$``$(c)$``$ (d)$``$(e)$``$(f).
###### Proof.
(a)$``$(b) follows from the equivalence (a)$``$(b) in the last theorem and the complete symmetry of $`(๐_p)_๐น`$. (c)$``$(d)$``$(e)$``$(f) are just examples of Theorem 5.3. (b)$``$(c) is trivial. Finally, the LCS $`๐^I`$, where the index set $`I`$ is uncountable, furnishes a counterโexample of the missing implication. โ
###### Proposition 6.6.
Let $`X`$ be an infrabarrelled LCS. Then $`X`$ is a Schwartz space ($`\mathrm{resp}.`$ coโSchwartz space) if and only if $`X_\beta ^{}`$ is a coโSchwartz space ($`\mathrm{resp}.`$ Schwartz space) if and only if $`X_{\beta \beta }^{\prime \prime }`$ is a Schwartz space ($`\mathrm{resp}.`$ coโSchwartz space).
###### Proof.
Repeat applying Theorems 6.3 and 6.5, and the complete symmetry of $`(๐_p)_๐น`$. โ
###### Remark 6.7.
Besides the ideals of precompact operators and quasiโSchwartz operators, one can also employ the ideal $`_{\text{im}}`$ of limit operators to define Schwartz spaces and coโSchwartz spaces. See for some other internal characterization of Schwartz spaces due to the introduction of $`_{\text{im}}`$.
Similar to Schwartz space we can relate infraโSchwartz spaces to the ideal $`๐`$ of weakly compact operators between LCSโs. Incidentally, readers should have no difficulty to figure out that $`๐_p`$โbornological spaces are, in fact, semiโMontel spaces and $`๐`$โbornological spaces are exactly semiโreflexive spaces. We leave these to the interested readers and refer them to for more information about the classical theory of these spaces.
### 6.2. Nuclear spaces and co-nuclear spaces
###### Definition 6.8.
A continuous seminorm $`p`$ on a LCS $`X`$ is called an absolutely summing seminorm ($`=`$ prenuclear seminorm in ) if there exists a $`\sigma (X^{},X)`$โclosed equicontinuous subset $`B`$ of $`X^{}`$ and a positive Radon measure $`\mu `$ on $`B`$ such that
$$p(x)_B|x,x^{}|๐\mu (x^{}),xX.$$
Let $`๐ซ_{\mathrm{as}}(X)`$ be the locally convex (Hausdorff) topology on $`X`$ generated by the family of all absolutely summing seminorms on $`X`$. It is easy to see that the system $`๐ซ_{\mathrm{as}}=\{๐ซ_{\mathrm{as}}(X):X`$ a $`LCS\}`$ is a generating topology. A continuous operator $`T`$ from a LCS $`X`$ into a LCS $`Y`$ is said to be absolutely summing if $`T๐ช(๐ซ_{\mathrm{as}})(X,Y)=(X_{๐ซ_{\mathrm{as}}},Y)`$. In case $`X`$ and $`Y`$ are Banach spaces, $`T`$ is absolutely summing if and only if $`T`$ sends every weakly summable series in $`X`$ to an absolutely summable series in $`Y`$. Denote by $`๐=๐ช(๐ซ_{\mathrm{as}})`$ the injective ideal of all absolutely summing operators between LCSโs, and by $`๐^b=๐ช^b(๐ซ_{\mathrm{as}})`$ the injective ideal of prenuclearโbounded operators .
A continuous operator $`T`$ from a LCS $`X`$ into a LCS $`Y`$ is said to be nuclear if there exist a $`(\lambda _n)`$ in $`ล_1`$, an equicontinuous sequence $`\{a_n\}`$ in $`X^{}`$ and a sequence $`\{y_n\}`$ contained in an infracomplete bounded disk $`B`$ in $`Y`$ such that $`T=\mathrm{\Sigma }_n\lambda _na_ny_n`$, i.e., $`Tx=\mathrm{\Sigma }_n\lambda _na_n(x)y_n`$ for each $`x`$ in $`X`$. Denote by $`๐`$ the ideal of all nuclear operators between LCSโs. Note that $`๐`$ is symmetric. It is more or less classical that $`๐=๐_๐น^{\mathrm{rup}}`$ \[30, p. 76\], $`๐=๐_๐น^{inf}`$ \[30, p. 144\] and $`๐_๐น^3๐_๐น๐_๐น`$ \[30, p. 145\] (in fact, we have $`๐_๐น^2๐_๐น,(๐_๐น^{\mathrm{inj}})^2๐_๐น`$, cf. ).
###### Definition 6.9.
A LCS $`X`$ is said to be nuclear if every continuous seminorm $`p`$ on $`X`$ is absolutely summing. A LCS $`Y`$ is said to be coโnuclear if its strong dual $`Y_\beta ^{}`$ is nuclear.
We provide a new proof of the following classical result.
###### Theorem 6.10 (see \[30, pp. 149 and 157\]).
Let $`X`$ be a LCS. The following are all equivalent.
1. $`X`$ is a nuclear space.
2. $`Q_p๐(X,X_p)`$ for every continuous seminorm $`p`$ on $`X`$.
3. For each continuous seminorm $`p`$ on $`X`$ there exists a continuous seminorm $`q`$ on $`X`$ with $`pq`$ such that the canonical map $`Q_{pq}๐(X_q,X_p)`$.
4. $`id_X๐(X,X)`$.
5. $`๐(X,Y)=(X,Y)`$ for every LCS $`Y`$.
6. $`๐(X,N)=(X,N)`$ for every normed space $`N`$.
7. $`\stackrel{~}{Q}_p๐(X,\stackrel{~}{X}_p)`$ for every continuous seminorm $`p`$ on $`X`$.
8. $`๐(X,F)=(X,F)`$ for every Banach space $`F`$.
9. For each continuous seminorm $`p`$ on $`X`$ there exists a continuous seminorm $`q`$ on $`X`$ with $`pq`$ such that the canonical map $`\stackrel{~}{Q}_{pq}๐(\stackrel{~}{X}_q,\stackrel{~}{X}_p)`$.
10. For each $`0`$โneighborhood $`V`$ in $`X`$ there is a $`0`$โneighborhood $`U`$ in $`X`$ with $`UV`$ such that the canonical map $`X^{}(V^{})X^{}(U^{})`$ is nuclear.
11. $`๐^b(X,Y)=^b(X,Y)`$ for every LCS $`Y`$.
12. $`^b(X,Y)๐(X,Y)`$ for every LCS $`Y`$.
13. $`๐_p(X,Y)๐^b(X,Y)`$ for every LCS $`Y`$.
14. $`^b(X,Y)๐(X,Y)=๐^b(X,Y)`$ for every LCS $`Y`$.
15. $`X`$ is a $`๐`$โtopological space.
16. $`X`$ is a $`๐`$โtopological space.
###### Proof.
(a) $``$ (b) $``$ (d) $``$ (e) $``$ (f) $``$ (l) $``$ (o) $``$ (p) are due to Theorem 5.1. Since $`๐=๐_๐น^{\mathrm{rup}}`$, we have (a) $``$ (c) by Theorem 5.15. (a) $``$ (n) $``$ (k) are due to Proposition 4.7 and Theorem 5.5. (k) $``$ (m) is obvious. To prove (m) $``$ (k) we employ the same trick as in Theorem 6.3. (c) $``$ (i) follows from the fact that $`๐^3๐๐`$. (i) $``$ (g) $``$ (h) are trivial. (h) $``$ (l) because $`๐๐`$ and $`๐`$ is injective. (i) $``$ (j) is ensured by the symmetry of $`๐_๐น`$.
Finally, we prove (j) $``$ (i). Let $`V_p=\{xX:p(x)1\}`$ be the $`0`$โneighborhood associated to a continuous seminorm $`p`$ on $`X`$. By (j), there is a continuous seminorm $`q`$ on $`X`$ such that $`V_qV_p`$ (i.e., $`pq`$) and $`\stackrel{~}{Q}_{pq}^{}:X^{}(V_p^{})X^{}(V_q^{})`$ is nuclear. By the symmetry of $`๐_๐น`$, $`\stackrel{~}{Q}_{pq}^{\prime \prime }:(X^{}(V_q^{}))^{}(X^{}(V_p^{}))^{}`$ is nuclear, too. Hence $`\stackrel{~}{Q}_{pq}^{\prime \prime }`$ is absolutely summing. Now $`(X^{}(V_q^{}))^{}`$ and $`(X^{}(V_p^{}))^{}`$ are isometrically isomorphic to $`X_q^{\prime \prime }`$ and $`X_p^{\prime \prime }`$, respectively. By the injectivity of $`๐`$, $`\stackrel{~}{Q}_{pq}`$ is absolutely summing. Repeating the same argument, we shall have continuous seminorms $`q_1`$ and $`q_2`$ on $`X`$ such that $`qq_1q_2`$ and $`\stackrel{~}{Q}_{qq_1}`$ and $`\stackrel{~}{Q}_{q_1q_2}`$ are both absolutely summing. Now $`pq_2`$ and $`\stackrel{~}{Q}_{pq_2}=\stackrel{~}{Q}_{pq}\stackrel{~}{Q}_{qq_1}\stackrel{~}{Q}_{q_1q_2}๐_๐น^3๐_๐น`$, and we are done. โ
###### Remark 6.11.
There are concepts of quasiโnuclearโseminorms, quasiโnuclear operators and quasiโnuclearโbounded operators, cf. . They can be used to define nuclear spaces like $`๐`$ and $`๐`$. However, they are simply, respectively, the $`๐`$โseminorms, $`๐^{\mathrm{inj}}`$โoperators and $`(๐ฏ(๐^{\mathrm{inj}}))^b`$โoperators. Using the same kind of argument in Theorem 6.10, one can easily prepare a longer list of equivalences. We leave this to the interested readers.
###### Theorem 6.12.
Let $`Y`$ be an infrabarrelled LCS. The following are all equivalent.
1. $`Y`$ is a coโnuclear space.
2. For each bounded disk $`B`$ in $`Y`$ there is a bounded disk $`A`$ in $`Y`$ with $`BA`$ such that the canonical map $`J_{AB}`$ from $`Y(B)`$ into $`Y(A)`$ is nuclear.
3. $`J_B๐(Y(B),Y)`$ for every bounded disk $`B`$ in $`Y`$.
4. $`(N,Y)=๐(N,Y)`$ for every normed space $`N`$.
5. $`^b(X,Y)๐(X,Y)`$ for every LCS $`X`$.
6. $`Y`$ is an $`๐`$โbornological space.
###### Proof.
Assume first that $`Y`$ is coโnuclear and $`B`$ is a bounded disk in $`Y`$. Then $`B^{}`$ is a $`0`$โneighborhood in $`Y_\beta ^{}`$. Hence there is a bounded disk $`A`$ in $`Y`$ with $`BA`$ such that the canonical map $`Y^{\prime \prime }(B^{})Y^{\prime \prime }(A^{})`$ is absolutely summing. Since $`๐`$ is injective, the canonical map $`J_{AB}`$ from $`Y(B)`$ into $`Y(A)`$ is also absolutely summing. Do this twice more and we shall get (a) $``$ (b) since $`๐_๐น^3๐_๐น`$. (b) $``$ (c), (d), (e) and each one of them $``$ (f) are straightforward. We consider (f) $``$ (a). Note that $`๐`$ is symmetric. Now Theorem 5.23(a) gives the desired conclusion. โ
###### Proposition 6.13.
Let $`X`$ be an infrabarrelled LCS. Then $`X`$ is a nuclear space ($`\mathrm{resp}.`$ coโnuclear space) if and only if $`X_\beta ^{}`$ is a coโnuclear space ($`\mathrm{resp}.`$ nuclear space) if and only if $`X_{\beta \beta }^{\prime \prime }`$ is a nuclear space ($`\mathrm{resp}.`$ coโnuclear space).
###### Proof.
In view of Theorem 5.25, it suffices to mention that the generating system $`๐ซ_{\mathrm{as}}`$ of absolutely summing topology has the subspace property. As a result, an infrabarrelled LCS $`X`$ is nuclear if and only if $`X_{\beta \beta }^{\prime \prime }`$ is nuclear. The other implications follow from this. โ
Since $`๐๐_p`$ we have the wellโknown
###### Proposition 6.14.
All nuclear ($`\mathrm{resp}.`$ coโnuclear) spaces are Schwartz ($`\mathrm{resp}.`$ coโSchwartz) spaces.
### 6.3. Permanence properties
We collect some results from about the permanence properties of Grothendieck spaces and coโGrothendieck spaces.
###### Theorem 6.15 (\[11, Junek\]).
Let $`๐`$ be an operator ideal on Banach spaces.
1. Any product of $`\mathrm{Groth}(๐)`$โspaces is a $`\mathrm{Groth}(๐)`$โspace.
2. Any locally convex direct sum of coโ$`\mathrm{Groth}(๐)`$โspaces is a coโ$`\mathrm{Groth}(๐)`$โspace.
3. If $`๐`$ is equivalent to some injective ideal then any subspace of a $`\mathrm{Groth}(๐)`$โspace is a $`\mathrm{Groth}(๐)`$โspace.
4. If $`๐`$ is equivalent to some surjective ideal then any quotient space of a coโ$`\mathrm{Groth}(๐)`$โspace is a coโ$`\mathrm{Groth}(๐)`$โspace.
5. If $`๐`$ is injective then any projective limit of $`\mathrm{Groth}(๐)`$โspaces is a $`\mathrm{Groth}(๐)`$โspace.
6. If $`๐`$ is injective then any subspace of a coโ$`\mathrm{Groth}(๐)`$โspace is a coโ$`\mathrm{Groth}(๐)`$โspace.
If one applies them together with other results in the earlier parts of this paper to Schwartz spaces, infraโSchwartz spaces, nuclear spaces, and their โcoโspacesโ, one can obtain a long list of permanence properties of these spaces, cf. or .
### 6.4. Other applications
Along the same line of reasoning in this paper one can develop similar applications of operator ideals to the theory of tensor products, partially ordered locally convex spaces and $`C^{}`$โalgebras.
It is of no doubt that the initial idea of operator ideals comes from tensor products. In \[13, p. 49\], Michor suggested a method to construct a tensor norm $`๐^{}`$ associated to each operator ideal $`๐`$ on Banach spaces (see for details about quasiโnormed ideal). See also and those famous works of A. Grothendieck and R. Schatten.
Let $`E`$ and $`F`$ be Banach spaces and $`๐`$ be an operator ideal on Banach spaces with ideal norm $`\alpha `$. Define $`_๐^{}`$ on $`EF`$ by
$$x_iy_i_๐^{}=sup\{|y_i,Tx_i|:T๐(E,F^{}),\alpha (T)1\}.$$
$`_๐^{}`$ turns out to be a reasonable cross norm. We denote by $`E_๐F`$ the $`๐`$tensor product of $`E`$ and $`F`$, that is, the completion of $`EF`$ under $`_๐^{}`$. Y. C. Wong \[30, p. 279\] showed that if $`๐`$ is the normed ideal $`(๐,P)`$ of all absolutely summing operators between Banach spaces, we would have $`(E_๐F)^{}(๐(E,F^{}),P)`$.
In general, let $`๐`$ be an operator ideal on LCSโs. We can define a tensor product topology associated to $`๐`$ by a family of $`๐`$โbilinear forms. A continuous bilinear form $`b`$ on $`X\times Y`$ is said to be an $`๐`$bilinear form, if there is a $`T`$ in $`๐(X,Y_\beta ^{})`$ such that $`b(x,y)=y,Tx`$. We write $`b=b_T`$ in this case. Detailed properties of $`b_T`$ can be found in . See also for other comments. If $`๐`$ is equipped with some locally convex topology (see ) then we can define similar seminorms like the one as $`_๐^{}`$. It might be interesting to investigate this kind of theory.
There is also an established theory of ideal topologies on partially ordered locally convex spaces. We give only one example here and refer interested readers to . Let $`(X,X_+,๐ฏ)`$ be a locally solid space. A continuous seminorm $`p`$ on $`X`$ is said to be a $`(PL)`$seminorm if there exists a positive $`f`$ in $`X^{}`$ such that $`p(x)sup\{g(x):fgf\}`$, $`xX`$. It turns out that a continuous seminorm $`p`$ on a locally solid space $`X`$ is a $`(PL)`$โseminorm if and only if $`Q_p`$ is a coneโabsolutely summing operator from $`X`$ onto $`X_p`$. Moreover, we have a list of characterizations of $`๐ฏ`$ to be the topology of uniform convergence on all order intervals as those appeared in Theorems 6.3 and 6.10 (see \[29, p. 136\]). We would like to mention that in the case of partially ordered locally convex spaces, or Banach lattices, the correct concept of operator ideals may be the soโcalled operator modules. For more information about operator modules, see Schwarz .
Finally, we finish this paper with a result of Jarchow . Let $`H`$ be a Hilbert space and $`A`$ be a $`C^{}`$โsubalgebra of $`B(H)`$.
###### Proposition 6.16 (\[10, Jarchow\]).
The $`๐`$โtopology of $`A`$, i.e., the ideal topology on $`A`$ generated by weakly compact operators, is the finest locally convex topology on $`A`$ which coincides with the strong (i.e., the double strong) operator topology on bounded subsets of $`A`$. The completion of $`A`$ under this topology is $`(A^{},\tau (A^{},A^{}))`$.
|
warning/0506/cond-mat0506736.html
|
ar5iv
|
text
|
# Topological transition in a two-dimensional model of liquid crystal
## 1 Introduction
The molecules of liquid crystals may often be described by long, neutral rigid rods which interact through electrostatic dipolar or higher order multi-polar interactions. This is at the origin of the natural introduction of Legendre polynomials for the description of the orientational transition between a disordered, isotropic high temperature phase and an ordered nematic phase at lower temperature . When such a material is cooled at even lower temperatures, other ordered phases may be encountered, the description of which requires more realistic potentials (see e.g. Ref. ).
Lattice models of nematic-isotropic transitions capture the essentials of the above description. The molecules are represented by $`n`$component unit vectors $`๐_๐ซ`$, hereafter called โspinsโ, located on the sites $`๐ซ`$ of a simple hyper-cubic lattice. The interaction between molecules is restricted to the nearest neighbours $`๐ซ,๐ซ^{}`$, so that the radial dependence is kept constant and the angular dependence enters through a $`k`$th order Legendre polynomial<sup>1</sup><sup>1</sup>1Even order Legendre polynomials guarantee the local $`Z_2`$ symmetry $`๐_๐ซ๐_๐ซ`$., $`P_k(\mathrm{cos}\alpha _{๐ซ,๐ซ^{}})`$, where $`\alpha _{๐ซ,๐ซ^{}}=\widehat{(๐_๐ซ,๐_๐ซ^{})}`$ is the angle between vectors $`๐_๐ซ`$ and $`๐_๐ซ^{}`$. The intensity of the interaction energy is measured through a parameter $`ฯต`$. The Hamiltonian of a lattice liquid crystal is thus given by
$$H=ฯต\underset{๐ซ,๐ซ^{}}{}P_k(\mathrm{cos}\alpha _{๐ซ,๐ซ^{}}),$$
(1)
and the relevant parameters for the investigation of the phase transition are the space dimension $`D`$, the โspinโ dimensionality or equivalently, its symmetry $`O(n)`$, and the โsymmetryโ $`k`$ of the interaction $`P_k`$.
Usual $`XY`$ and Heisenberg models correspond, within this description, to $`k=1`$ and $`n=2`$ and 3, respectively. In two dimensions ($`D=2`$), these models exhibit quite different behaviours. In the case of the $`XY`$ model ($`O(2)`$ abelian symmetry), a topological transition described by Berezinskiฤญ, and Kosterlitz and Thouless takes place, governed by the condensation of vortices. This is not prevented by the Mermin-Wagner-Hohenberg theorem which states that spin models with continuous symmetry cannot have any long range ordered phase. To leading order in the high-temperature expansion, one gets for the correlation function an exponential decay $`๐_{๐ซ_1}๐_{๐ซ_2}K^{|๐ซ_1๐ซ_2|}`$ ($`K=ฯต/k_BT`$), while at low temperatures in the harmonic approximation, the Hamiltonian becomes quadratic and leads to the Gaussian model which implies that $`๐_{๐ซ_1}๐_{๐ซ_2}|๐ซ_1๐ซ_2|^{1/2\pi K}`$, i.e. a temperature-dependent spin-spin critical exponent $`\eta _{XY}(T)=\frac{k_BT}{2\pi ฯต},T0`$. The low temperature (LT) phase of the $`XY`$ model is a quasi-long-range ordered phase (QLRO) with vanishing magnetisation $`M^2(T)=lim_{|๐ซ_1๐ซ_2|\mathrm{}}๐_{๐ซ_1}๐_{๐ซ_2}=0`$, or a critical phase. The Heisenberg model ($`O(3)`$, non-abelian symmetry) on the other hand has no transition at any finite temperature (asymptotic freedom) . This difference with $`XY`$ model is at first surprising, since the low temperature limit of the Heisenberg model is essentially described by a similar spin wave approximation (SWA): the longitudinal modes of the $`O(n)`$ model are frozen and only the transverse modes are activated, leading essentially to two Gaussian models. This apparent contradiction between the asymptotic freedom of the Heisenberg model and the topological transition at finite temperature for the $`XY`$ model finds its origins in the stability of topological defects in the latter case, while the โthird spin dimensionโ makes the vortices unstable at any temperature in the former model. Further, the question of the accessibility of the thermodynamic limit is worth studying. Berezinskiฤญ and Blank noticed long time ago that a really large, but finite $`XY`$ system always possesses a non-zero magnetisation . The size being limited, $`|๐ซ_1๐ซ_2|`$ becomes at most as large as the linear size $`L`$ and a finite order parameter follows, $`M_L(T)L^{\frac{1}{2}\eta (T)}`$. More precisely, $`M_L(T)=O(1)`$ as long as $`L\mathrm{e}^{2/\eta (T)}`$ a condition which can be fulfilled for any $`L`$ by considering small enough temperatures. With this result in mind, we expect for the model considered hereafter that a spin wave solution will be found at low enough temperature and thus we search for an algebraic decay of the spin-spin correlation function from an effectively non-zero order parameter, although there is at most QLRO at low temperatures.
Changing the value of $`k`$ in Eq. (1) modifies the symmetry of the spin-spin interaction and gives rise to new features<sup>2</sup><sup>2</sup>2In the same spirit, the case of symmetry-breaking magnetic fields $`h_k\mathrm{cos}k\theta `$ added to the $`XY`$ model and changing the phase diagram was investigated by Josรฉ et al .. When $`k`$ increases, one may indeed expect a qualitative change in the nature of the transition, like in the case of discrete spin symmetries (Potts model) . The value $`k=2`$ was intensively studied. It still corresponds to the $`XY`$ model for $`O(2)`$ spin symmetry, while it leads to the $`RP^2`$ or Lebwohl-Lasher model for $`3`$component spin vectors. The nature of the transition in this latter case is still under discussion , but a recent study reported new evidences, extremely convincing, in favour of a topological transition . The transition is driven by topologically stable point defects known as $`\frac{1}{2}`$ disclination points. Considering still larger values of $`k`$, there is a proof of asymptotic freedom in the large$`n`$ limit, for values of $`k`$ (in the interaction term $`(1+\mathrm{cos}\theta )^k`$) which do not exceed a critical $`k_c4.537\mathrm{}`$ . This is again discussed in a recent preprint from Caracciolo et al. . Above this value the transition becomes of first order, a result which does not violate Mermin-Wagner-Hohenberg theorem, since the correlation length is finite at the transition. For finite value of $`n`$, the question of the nature of the transition at high $`k`$ is still a challenging problem. In the context of orientational transitions in liquid crystals, Legendre polynomials rather than $`\mathrm{cos}^k\theta `$ interactions are introduced, and we are led to the Hamiltonian of Eq. (1). The 3-vector model with $`P_4`$ interactions was already considered in Refs. where convincing evidence for a first order transition was reported.
In this paper, we study the behaviour of an abelian spin model, namely $`O(2)`$ rotation group with $`P_4`$like spin interactions. The model will be referred to as $`P_4O(2)`$ for simplicity. We are mainly interested in the low temperature properties of the model where comparisons with analytic predictions are possible due to the simplicity of the SW approximation. The techniques used combine temperature analysis, FSS (Finite-Size Scaling) and conformal techniques (Finite-Shape Scaling - FShS - to plagiarize the famous acronym).
## 2 Definition of the model and of the observables
In Refs. , the following Hamiltonian $`H_{P_4O(3)}=ฯต_{๐ซ,๐ซ^{}}P_4(๐_๐ซ๐_๐ซ^{})`$ was considered, where $`๐_๐ซ=(\sigma _๐ซ^x,\sigma _๐ซ^y,\sigma _๐ซ^z)`$, $`|๐_๐ซ|=1`$ and $`P_4(x)=\frac{1}{8}(35x^430x^2+3)`$. For $`2`$component vectors, in the completely disordered phase $`\mathrm{cos}^2\theta =\frac{1}{2}`$ and $`\mathrm{cos}^4\theta =\frac{3}{8}`$. In order to keep the same symmetry in the interaction than in the $`P_4O(3)`$ model, but to normalise it between 0 and 1 in the limits of completely disordered and completely ordered phases respectively, we modify slightly the Hamiltonian to include pair interactions of the type $`Q_4(x)AP_4(x)+\mathrm{const}=\frac{8}{55}(35x^430x^2+\frac{15}{8})`$. The corresponding Hamiltonian is thus defined by
$$H_{P_4O(2)}=ฯต\underset{๐ซ,๐ซ^{}}{}Q_4(๐_๐ซ๐_๐ซ^{}),$$
(2)
with now $`๐_๐ซ=(\sigma _๐ซ^x,\sigma _๐ซ^y)`$, $`|๐_๐ซ|=1`$. A qualitative description of the transition is provided by the temperature behaviour of the energy density, the specific heat, the order parameters and the corresponding susceptibilities. The internal energy is defined from the thermal average of the Hamiltonian density,
$$u_{P_4O(2)}(T)=(DL^D)^1H_{P_4O(2)}$$
(3)
and the specific heat follows from fluctuation dissipation theorem,
$$L^DT^2C_v(T)=(H_{P_4O(2)})^2H_{P_4O(2)}^2.$$
(4)
Brackets denote the thermal average. The definition of the scalar order parameter (sometimes called nematisation) is deduced from the local second-rank order parameter tensor,
$$q^{\alpha \beta }(๐ซ)=\sigma _๐ซ^\alpha \sigma _๐ซ^\beta \frac{1}{2}\delta ^{\alpha \beta }.$$
(5)
After space average, the traceless tensor $`L^D_๐ซq^{\alpha \beta }(๐ซ)`$ admits two opposite eigenvalues $`\pm \frac{1}{2}\eta `$ corresponding to eigenvectors $`๐ง_+`$ and $`๐ง_{}`$. The order parameter density is defined after thermal averaging by
$$q_2(T)=\eta .$$
(6)
This quantity has the same physical content but is more stable numerically than a direct estimation of $`2(๐_๐ซ๐ง_+)^21`$. Another order parameter may be defined simply by inspection of the structure of the Hamiltonian,
$$q_4(T)=L^D\underset{๐ซ}{}Q_4(๐_๐ซ๐ง_+).$$
(7)
The associated susceptibilities are defined by the fluctuations of the order parameter densities, e.g.
$$\chi _{q_2}(T)=\frac{4L^D}{k_BT}(\eta ^2\eta ^2).$$
(8)
## 3 Thermal behaviour
In this section, we illustrate the behaviour of the various thermodynamic quantities as the temperature varies. It gives a first idea of the nature of the transition.
The simulations are performed using a standard Wolff algorithm suited to the expression of the nearest neighbour interaction . The spins are located on the vertices of a simple square lattice of size $`L^2`$ with periodic boundary conditions in the two directions. We use $`L=24`$, $`32`$, $`48`$, $`64`$ and $`128`$ with $`10^6`$ equilibrium steps (measured as the number of flipped Wolff clusters) and $`10^6`$ Monte Carlo steps (MCS) for the evaluation of thermal averages. The autocorrelation time (at $`k_BT/ฯต=0.2`$, $`L=16`$) is of order of $`30`$ MCS, hence the numbers of iterations that we used correspond roughly to $`3.10^4`$ independent measurements for the smallest size and is still safe at $`L=128`$.
We deliberately did not use histogram reweighting which would require a huge amount of simulation time to be reliable. We have adopted a different strategy here, producing more simulations with less iterations, but with a better control of the errors than at the ends of the histograms. The temperature dependence of thermodynamic quantities is plotted in Fig. 1 for different system sizes. The behaviour of the energy density clearly displays a difference between the regimes of low and high temperatures. This is the signature of a transition and this naive statement is corroborated by the behaviours of the other physical quantities, the specific heat $`C_v`$, the order parameters $`q_2`$ and $`q_4`$, and the corresponding susceptibilities. The specific heat close to the maximum does not seem to increase substantially with the system size. This might be the sign of an essential singularity <sup>3</sup><sup>3</sup>3Or at least the sign of a non diverging specific heat with non-positive exponent $`\alpha `$. around a temperature $`k_BT_c/ฯต0.700.75`$. From the behaviour of the order parameters, we may suspect a smooth transition, since there is no sharp jump. The susceptibilities display a non conventional behaviour at low temperature, increasing with the system size, which indicates a likely topological transition with a critical low temperature phase where the susceptibility diverges at any temperature (note the logarithmic scale for vertical axis).
The probability distributions of the energy density and the order parameter is also instructive. Both quantities are shown in Fig. 2. The distributions have a simple shape with single peaks, and this is still true for temperatures below the transition, a result which suggests a continuous transition. We note that due to the finite-size of the system, the order parameter is finite.
## 4 Finite-Size Scaling
In this section we investigate the properties of the low temperature phase using Finite-Size Scaling technique. In the critical low temperature phase of a model which displays a topological transition (the paradigmatic $`XY`$ model serves as a guide), the physical quantities behave like at criticality for a second-order phase transition, with power law behaviours of the system size. The difference is that in the critical phase, the critical exponents depend on the temperature and for any temperature below the transition one has e.g.
$$q_2(T)L^{\frac{1}{2}\eta _{q_2}(T)}$$
(9)
$$\chi _{q_2}(T)L^{2\eta _{q_2}(T)}.$$
(10)
Here $`\eta _{q_2}(T)`$ denotes the correlation function critical exponent, defined by
$$Q_2(\mathrm{cos}(\theta _{๐ซ_1}\theta _{๐ซ_2}))|๐ซ_1๐ซ_2|^{\eta _{q_2}(T)}.$$
(11)
In Fig. 3 the FSS behaviour of the order parameter densities and of the corresponding susceptibilities is shown on log-log scales vs the system size $`L`$ (we used $`L=16`$, 32, 48, 64, 128 and 256) for different values of $`T`$ below the expected transition. The linear behaviour on this scale indicates power laws, and the different slopes at different temperatures is the result of exponents which depend on the value of the temperature. Starting from a fully ordered system at $`T=0`$, at low temperature we can expect a small disorientation of the molecules and the spin-wave approximation becomes correct. For $`O(2)`$ model with nearest neighbour interactions described by arbitrary polynomial in $`\mathrm{cos}\alpha _{๐ซ,๐ซ^{}}`$, one is led to an effective harmonic Hamiltonian $`\frac{1}{2}ฯต_{๐ซ,๐ซ^{}}l(\theta _๐ซ\theta _๐ซ^{})^2`$. It yields power-law correlations,
$$\mathrm{cos}m(\theta _{๐ซ_1}\theta _{๐ซ_2})=\mathrm{e}^{{\scriptscriptstyle \frac{m^2}{2}}(\theta _{๐ซ_1}\theta _{๐ซ_2})^2}|๐ซ_1๐ซ_2|^{\eta _{ml}^{\mathrm{SW}}}$$
(12)
with a spin-wave decay exponent given by
$$\eta _{ml}^{\mathrm{SW}}=\frac{m^2}{l}\eta _{XY}(T)=\frac{m^2k_BT}{2l\pi ฯต}.$$
(13)
This expression, if confirmed numerically, will support the presence of a quasi-long-range ordered, scale-invariant phase at low temperatures. The exponent $`\eta _{q_2}(T)`$ ($`m=2`$) is accessible through $`q_2(T)`$ (or the corresponding susceptibility) via equations (9) and (10). We note that it is also accessible through the the FSS behaviour of $`q_4(T)`$, since the leading behaviour of $`Q_4(\theta _{๐ซ_1}\theta _{๐ซ_2})`$ is still governed by $`\mathrm{cos}2(\theta _{๐ซ_1}\theta _{๐ซ_2})`$. In Fig 3, one may extract numerical values of $`\eta `$ exponent. The values $`\eta _{q_2}(T)=0.0056`$, $`0.0117`$, $`0.0182`$, and $`0.0256`$ are obtained at $`k_BT/ฯต=0.1`$, $`0.2`$, $`0.3`$, and $`0.4`$, respectively. Using the small angle limit $`H_{P_4O(2)}=ฯต_{๐ซ,๐ซ^{}}Q_4(\mathrm{cos}\alpha _{๐ซ,๐ซ^{}})\frac{1}{2}ฯต_{๐ซ,๐ซ^{}}\frac{128}{11}(\theta _๐ซ\theta _๐ซ^{})^2,`$ we have $`l=\frac{128}{11}`$ which yields $`\eta _{2\frac{128}{11}}^{\mathrm{SW}}=\frac{11k_BT}{64\pi ฯต}`$. At $`k_BT/ฯต=0.1`$, $`0.2`$, $`0.3`$, and 0.4, we get $`\eta _{2\frac{128}{11}}^{\mathrm{SW}}(T)0.006`$, $`0.011`$, $`0.018`$, and $`0.022`$ which confirm the numerical values, since the SW exponent is always a lower bound for the exact exponent $`\eta _{q_2}(T)`$. The comparison is made visible in the figure where the SW values are plotted in full lines while the symbols represent the numerical data. As expected, the lower the temperature, the better the SW approximation.
## 5 Finite-Shape Scaling
Inspired by the results that we obtained for the Lebwohl-Lasher model or the XY model , we will now produce a complementary study using a rescaling of the density profiles. For that purpose, we assume the existence of a critical phase at low temperatures as suggested by the temperature dependence and FSS results. If the assumption is revealed incorrect, we will be led to some inconsistency.
The existence of a scale-invariant low temperature critical phase leads to conformally covariant density profiles or correlation functions at any temperature below the transition $`T_c`$. It is then advantageous to deduce the functional expression of the correlation functions or density profiles in a restricted geometry adapted to numerical simulations from a conformal mapping $`w(z)`$:
$$G(w_1,w_2)=|w^{}(z_1)|^{x_\sigma }|w^{}(z_2)|^{x_\sigma }G(z_1,z_2)$$
(14)
Here, $`w`$ labels the lattice sites in the transformed geometry (the one where the computations are really performed), $`z`$ is the corresponding point in the original one (usually the infinite plane where the two-point correlations take the standard power-law expression $`G(z_1,z_2)|z_1z_2|^{\eta _\sigma }`$), and $`x_\sigma =\frac{1}{2}\eta _\sigma `$ is the scaling dimension associated to the scaling field under consideration. The interest of such an approach lies in the full inclusion of the changes due to shape effects in the functional expression and we may copy the terminology FSS and call it Finite-Shape Scaling.
Rather than two-point correlation functions, it is even more convenient to work with density profiles $`m(w)`$ in a finite system with symmetry breaking fields along some surfaces in order to induce a non-vanishing local order parameter in the bulk. The density $`m(w)`$ will be $`q_2(๐ซ)=Q_2(๐_๐ซ๐ก_\mathrm{\Lambda })`$ or higher rank nematisation $`q_4(๐ซ)=Q_4(๐_๐ซ๐ก_\mathrm{\Lambda })`$. In the case of a square lattice $`\mathrm{\Lambda }`$ of size $`L\times L`$, with fixed boundary conditions along the four edges $`\mathrm{\Lambda }`$, one expects
$`m(w)`$ $``$ $`[\kappa (w)]^{\frac{1}{2}\eta _\sigma }`$
$`\kappa (w)`$ $`=`$ $`\mathrm{}\mathrm{m}\left[\mathrm{sn}{\displaystyle \frac{2\mathrm{K}w}{L}}\right]\times \left|\left(1\mathrm{sn}^2{\displaystyle \frac{2\mathrm{K}w}{L}}\right)\left(1k^2\mathrm{sn}^2{\displaystyle \frac{2\mathrm{K}w}{L}}\right)\right|^{\frac{1}{2}}`$ (15)
where $`w`$ stands for lattice site $`๐ซ`$. This expression easily follows from the expression of the order parameter profile decaying in the upper half-plane from a distant surface of spins constantly fixed in a given direction, $`m(z)y^{x_\sigma }`$, and from the conformal transformation of the upper half-plane $`z=x+iy`$, $`(0y<\mathrm{})`$, inside a square $`w=u+iv`$ of size $`L\times L`$, $`(L/2uL/2,0vL)`$, with open boundary conditions along the four edges, realized by a Schwarz-Christoffel transformation
$$w(z)=\frac{L}{2\mathrm{K}}\mathrm{F}(z,k),z=\mathrm{sn}\left(\frac{2\mathrm{K}w}{L}\right).$$
(16)
Here, F$`(z,k)`$ is the elliptic integral of the first kind, $`\mathrm{sn}(2\mathrm{K}w/L)`$ the Jacobian elliptic sine, K = K$`(k)=`$F$`(1,k)`$ the complete elliptic integral of the first kind, and the modulus $`k=0.171573`$ depends on the aspect ratio of $`\mathrm{\Lambda }`$ (here $`1`$).
The procedure is now to fit numerical data of the order parameter profile against expression (15). The result for the density profile of $`q_2(w)`$ is shown on a log-log scale in Fig. 4. At low temperatures, the resulting straight lines confirm the existence of a critical phase (a linear behaviour on this scale results from an algebraic decay in the original semi-infinite geometry). Above the deconfining transition (assuming that it is indeed the driving mechanism of the transition), the decay becomes faster, indicating a paramagnetic phase. The transition is approximately located at a temperature $`k_BT_c/ฯต0.700.75`$. The scenario is eventually completely consistent with a BKT transition. Furthermore the $`\eta `$ exponent again follows from the scaling of the density profile in Eq. (15) which provides an alternate determination of this quantity.
Another famous conformal mapping which has been applied to many two-dimensional critical systems is the logarithmic transformation $`w(z)=\frac{L}{2\pi }\mathrm{ln}z=\frac{L}{2\pi }\mathrm{ln}\rho +i\frac{L\phi }{2\pi }`$. It maps the infinite plane onto an infinitely long cylinder of perimeter $`L`$, and due to the one-dimensional character of this latter geometry, the correlation functions along the axis of the cylinder (let say in terms of the variable $`u=\frac{L}{2\pi }\mathrm{ln}\rho `$) decay exponentially at criticality, $`G(u_1,u_2)\mathrm{exp}[(u_2u_1)/\xi ]`$. The interesting result which makes this technique powerful is that the correlation length amplitude on the strip is universal and only determined by the corresponding $`\eta `$ exponent, $`\xi =\frac{L}{\pi \eta }`$. This relation is known as the gap-exponent relation in the context of quantum chains in $`1+1`$ dimensions. It was conjectured by several authors before Cardy proved it . Using MC simulations we cannot of course produce an infinitely long cylinder. It is however possible to perform simulations inside a rectangle $`L_1\times L`$ with $`L_1L`$ and periodic boundary conditions in both space directions (we get a very long torus). Due to the exponential decay of the correlation length, it is not necessary to explore really long distances $`u_2u_1`$ (typically $`u_2u_110L`$). A finite torus of long perimeter $`10^3L`$ thus only produces insignificant finite-size corrections to the gap-exponent relation. We performed the simulations at different temperatures in a system of size $`10\times 10000`$ and extracted the correlation function exponent from the linear behaviour
$$\mathrm{ln}Q_2[\mathrm{cos}(\theta _{u_2}\theta _{u_1}]=\mathrm{const}\frac{\pi \eta _{q_2}}{L}(u_2u_1).$$
(17)
This linear behaviour (in terms of the variable $`u_2u_1`$) is shown in the insert of Fig. 4 where for simplicity $`Q_2[\mathrm{cos}(\theta _{u_2}\theta _{u_1}]`$ is denoted $`G_2(u_2u_1)`$. It is impossible to apply this technique up to the transition temperature, since the strip system contains quite a large number of spins ($`10^5`$ while simulations in the square geometry are performed up to typically $`10^4`$ spins) and the autocorrelation time increases too fast.
## 6 Behaviour at the deconfining transition
Not only the low temperature behaviour is interesting. The value of the $`\eta `$ exponent at the BKT transition where some deconfining mechanism should lead to the proliferation of unbinded topological defects is also of primary interest. For that purpose, an accurate value of the transition temperature is needed. We performed a study of the crossing of $`U_4`$ Binder cumulant for very large statistics ($`30\times 10^6`$ MCS) and large system sizes (squares of $`L=64`$, 80, 96 and 128 with periodic boundary conditions). The results shown in Fig. 5 indicate a transition temperature of $`k_BT_{\mathrm{BKT}}/ฯต=0.7226`$.
Then this temperature is used to perform Finite-Shape Scaling using the technique already employed for the investigation of the low temperature phase, namely the algebraic decay of density profiles inside a square with fixed bounday conditions. These new simulations are really time-consuming, since the autocorrelation time increases in the low temperature phase when $`T`$ evolves towards the deconfining transition and a rather large number of Monte Carlo steps is needed to get a satisfying number of independent measurements. For sizes $`L=16`$, 32, 64, we used $`10^6`$ MCS for thermalization and $`30.10^6`$ for measurements, while โonlyโ $`20.10^6`$ for the largest size 128. For technical reasons, the exponential decay of two-point correlation functions along the torus cannot be applied at the BKT transition, since as already mentioned the system size being quite larger than in a square geometry, the number of MC iterations required is by far too large. In Fig. 5 we plot the โeffectiveโ exponent $`\eta _{\mathrm{eff}}(L)`$ measured at $`T_{\mathrm{BKT}}`$ for different system sizes as a function of the inverse size. An estimate of the thermodynamic limit value ($`L\mathrm{}`$) can be made using a polynomial fit (the results of quadratic and cubic fits are respectively $`0.118`$ and $`0.122`$). It is safer to keep the three largest sizes available, $`L=32`$, 64, and 128, for which a linear dependence of $`\eta _{\mathrm{eff}}(L)`$ with $`L^1`$ is observed. Taking into account the error bars, crossing the extreme straight lines leads to the following value for the correlation function exponent at the deconfining transition
$$\eta _{q_2}(T_{\mathrm{BKT}})=0.122\pm 0.007.$$
(18)
This value is essentially half the Kosterlitz value for the $`XY`$ model.
## 7 Conclusion
In Fig. 6 we plot as a function of the temperature the exponent $`\eta _{q_2}(T)`$ measured after conformal rescaling of the density profile and correlation functions at different system sizes and the FSS determination which follows from Fig. 3. Together with the exponent determined numerically, we report the result of the spin-wave approximation, shown in dotted line. The larger the size of the system, the better the agreement. Similar results (not shown here) are measured in the case of the higher-order nematisation, $`q_4(T)`$.
According to these results, the main outcome of the present work is the following:
\- $`P_4O(2)`$ model displays a BKT-like transition with QLRO in the LT phase where SWA nicely fits the nematisation temperature-dependent exponent $`\eta _{q_2}(T)`$ when $`T0`$.
The LT phase of the model is thus disordered. Nevertheless, in what we will call a physical limit, i.e. a finite but quite large size where the system contains a macroscopic number of spins, it is useful to consider that the system is partially ordered, as Fig. 2 seems to indicate.
These results find a partial interpretation through a naive comparison with clock model in 2D. Increasing the order of the interaction polynomial indeed increases the number of deep wells which stabilise the relative orientation of neighbouring spins. One is thus led to a system which is quite similar to a planar clock model with a finite number of states, unless the fact that here we keep a continuous spin symmetry which prevents from any โmagneticโ long-range order at finite temperature. The clock model is known to be in the Potts universality class when $`q=3`$, but at $`q4`$, it displays a QLRO phase before conventional ordering at lower temperatures . Combining these results with the requirements of Mermin-Wagner-Hohenberg theorem in the case of continuous spin symmetry gives a natural framework for the comprehension of our results for $`2`$component spin systems. Whatever the nearest-neighbour interaction (in $`P_1`$, $`P_2`$ or $`P_4`$) their behaviour seems to be always described by a BKT transition. The similar observation that a two-component nematic model renormalises in two dimensions towards the $`XY`$ model was already reported in Ref. . The transition is likely driven by a mechanism of condensation of defects, like in the $`XY`$ model, but due to the local $`Z_2`$ symmetry not only usual vortices carrying a charge $`\pm 1`$ are stable, but also disclination points carrying charges $`\pm 1/2`$ should be stable. The role of these defects might be studied in a similar way than in the recent work of Dutta and Roy , by the comparison of the the transition in the pure model and in a modified version where a chemical potential is artificially introduced in order to control the presence of defects.
## Acknowledgement
The work of A.I.F.S. is supported by a PCP cooperation programme (โFluides pรฉtroliersโ) between France and Venezuela. Thanks to the support from CINES in Montpellier for computational time. We benefited from instructive correspondence on $`O(n)`$ models with H. Kawamura, A. Pelissetto, D. Mouhanna and S. Korshunov who are gratefully acknowledged. B.B. is also indebted to Yu. Holovatch for stimulating discussions at the occasion of one of his visits in Nancy.
|
warning/0506/quant-ph0506226.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
A structure in which the dielectric constant varies periodically is called a photonic crystal. One of the most interesting properties of a photonic crystal is the existence of a photonic band gap . Radiation with a frequency that lies within the band gap cannot propagate in the photonic crystal structure. Photonic crystals are usually viewed as an optical analog of semiconductors that modify the properties of light similar to a microscopic atomic lattice that creates a semiconductor band-gap for electrons . Photonic band gap crystals offer unique ways to tailor light and the propagation of electromagnetic waves and have caused growing interest in recent years because it offers the possibility of controlling and manipulating light within a given frequency range through photonic band gap . Photonic band-gap materials have attracted much attention in recent years for theoretical and practical importance in fundamental science and application . The atom-photon interaction in photonic band gap materials has been found to exhibit many interesting new phenomena such as photon-atom bound states , spectral splitting , quantum interference dark line effect , phase control of spontaneous emission , transparency near band edge , and single-atom switching .
In a parallel development, considerable work was done recently on entanglement properties . The detection of entanglement is one of the fundamental problems in quantum information theory. From a theoretical point of view one can try to answer the question whether a given entirely known state is entangled or not, but despite a lot of progress in the last years , no general solution of this problem is known. In experiments, one aims at detecting entanglement without knowing the state completely. Bell inequalities and entanglement witnesses are the main tools to tackle this task. Interestingly, the concurrence of the ground state which is related to the entanglement of formation, has been shown to be strongly affected at the critical point . More precisely, in the one-dimension, it has been shown that the derivative of the concurrence with respect to the coupling constant diverges at the transition point, although the concurrence itself is not maximum. These pioneering results raise the question of the universality of these behaviors. Actually, the lack of exact solutions especially in higher dimensions implies a numerical treatment which often restrict the study to a small number of degrees of freedom.
Heisenbergโs uncertainty relations had tremendous impact in the field of quantum optics particularly in the context of the construction of coherent states and also for different physical systems as well as the reconstruction of quantum states. The minimization problem of finding the number-phase uncertainty state has been considered and minimum uncertainty state relations between number and phase uncertainty are presented . Many authors argued that , the Heisenberg inequality is too weak for practical purposes, which led them to the establishment of information theoretic uncertainty relations.
Our aim of the present paper is to consider the dynamics of a system of three-level atoms with dipole interaction in presence of the photonic band gap and study the concurrence and the entropic uncertainty relation for number and phase. With applying some approximations, one can deal with the quantization of the electromagnetic field modes of a homogeneous, but anisotropic medium which can then be made to form one of the sandwich layers in the slab structure under consideration involving two semi-infinite periodic photonic crystals. With the electromagnetic modes quantized, one can evaluate the entanglement degree, and explore its variations with the controllable parameters of the system. To reach our goal we have to find an exact analytic solution of the time dependent Schrรถdinger equation of the system. We show that a reasonable amount of entanglement can be achieved in a system of three-level atoms with dipole interaction in presence of the photonic band gap and essentially we establish deeper connections between entropic uncertainty relations and entanglement.
The organization of this paper is as follows: in section 2, we give an overview of effective medium approach and dispersions, followed by subsection 2.1 where we introduce our Hamiltonian model and give exact analytic solution for the Schrรถdinger equation in the frame of the dressed state formalism. In section 3, we employ the analytical results obtained in section 2 to investigate the properties of the entanglement degree due to the concurrence, and classify the behavior in several parameter regimes assuming that the electromagnetic field is in a coherent state in subsection 3.1. In section 4, we essentially establish deeper connections between entropic uncertainty relations and entanglement. Numerical results for the phase entropy are discussed in the subsection 4.1 for two different cases; one is the resonant and the other is the off-resonant case. The prospects for experimental observation of our predictions are analyzed in section 5. Finally, a summary of the main points of this work ends the paper and a few avenues for further investigations are indicated in section 6.
## 2 Effective medium approach
The effective-medium approach can be applied to situations in which all three regions of the structure possess frequency-dependent dielectric functions. In fact, the rapid pace of the technological progress in solid-state quantum computing gives one a hope that the specific prescriptions towards building robust qubits and their assemblies discussed in this work can be implemented in future devices. In this regard, very promising fields where the concept of nonlinear localized modes may find practical applications is the quantum computation of photonic band gap materials, periodic dielectric structures that produce many of the same phenomena for photons as the crystalline atomic potential does for electrons . Nonlinear photonic crystals (or photonic crystals with embedded nonlinear impurities) create an ideal environment for the generation and observation of nonlinear localized photonic modes. Much theoretical work has been done on the properties of finite one dimensional photonic band gap (PBG) crystals , including recent calculations of the thermal emissivity of such one-dimensional structures . The strong angular dependence of the gap effect with a one-dimensional structure has motivated successful experimental work with three-dimensional structures . In particular, the existence of such modes for the frequencies in the photonic band gaps has been predicted for $`2D`$ and $`3D`$ photonic crystals with Kerr nonlinearity. Nonlinear localized modes can also be excited at nonlinear interfaces with quadratic nonlinearity , or along dielectric waveguide structures possessing a nonlinear Kerr-type response .
The system that we consider here consists of a dielectric cavity occupying the region $`0<z<r`$ and the photonic crystals occupy the regions $`z>r`$ and $`z<0`$. For long wavelength fields and in the effective medium approach, the photonic crystal has the optical characteristic of a uniaxial medium . Moreover we shall specialize to uniaxial media, so that our system has only two principal axes with the $`z`$axis as the optical axis. In this case, the components of the dielectric tensor appropriate to the photonic crystals can be written as
$$ฯต=\left(\begin{array}{c}ฯต\\ 0\\ 0\end{array}\begin{array}{c}0\\ ฯต\\ 0\end{array}\begin{array}{c}0\\ 0\\ ฯต_z\end{array}\right),$$
(1)
where, $`ฯต=ฯต_0ฯต^{||},`$ $`ฯต_z=ฯต_0ฯต_z.`$ The dielectric tensor components for the two semi-infinite crystals can be written in the following forms $`ฯต_1^{||}=(\eta _1d_1+\eta _2d_2)/d_{12},ฯต_{z1}=\eta _1\eta _2d_{12}/(\eta _1d_2+\eta _2d_1),`$ and $`ฯต_2^{||}=(\eta _3d_3+\eta _4d_4)/d_{34},ฯต_{z2}=\eta _3\eta _4d_{34}/(\eta _3d_4+\eta _4d_3),`$ where the $`d_{ij}=d_i+d_j,`$ the subscripts 1 and 2 on $`ฯต_i^{||}`$ and $`ฯต_{zi}`$ refer to the first and second photonic crystal. The dielectric functions $`\eta _1`$ and $`\eta _2,`$ one or both of which may be frequency dependent. The photonic crystals are treated using the effective medium approach, which pertains to any layer structure formed by alternate periodic stacking of two types of layers of locally isotropic materials of thicknesses d<sub>1</sub> and $`d_2`$ (see figure 1).
In this paper we are concerned with the interface polaritons which are characterized by imaginary wave vectors normal to the interfaces such that the waves are decaying with distance from the interfaces at $`z=0`$ and $`z=r`$ into the outer regions and are hyperbolic in the slab . To see the salient features of the effective medium description we shall ignore retardation effects, which amounts to ignoring throughout ($`\omega /c`$) terms. In this case dispersion relation for the surface polaritons takes the form
$$k_sr=\mathrm{arctan}h\left(\frac{k_s}{ฯต_s}\times \frac{(k_1/ฯต_1^{||})+(k_2/ฯต_2^{||})}{(k_s/ฯต_s)+(k_1k_2/ฯต_1^{||}ฯต_2^{||})}\right).$$
(2)
The dispersion relations obtained from the Maxwell wave equation of this system lead to two distinct equations $`k_s^2=k_{||}^2\omega ^2ฯต_s/c^2,k_i^2=ฯต_i^{||}k_{||}^2/ฯต_{zi}\omega ^2ฯต_i^{||}/c^2,`$ where s refers to the slab cavity.
It is important to note that infinite and semi-infinite photonic crystals have the same band structure . The only difference is the existence of surface modes in the case of semi-infinite structure. The main feature of all 1D photonic crystals is that although forbidden gaps exist for most given values of the tangential component of the wave vector ($`k`$), there is not an absolute nor complete photonic band gap if all possible values of the tangential component of the wave vector are considered . Having determined the modes we can now quantize the fields associated with these modes using the usual quantization procedure the single-mode quantized field takes the form
$$E(\widehat{x},t)=E_0\widehat{a}(\widehat{k}_{||})\mathrm{exp}[i(\widehat{k}_{||}.\widehat{x}\omega t)]+H.C.,$$
(3)
where $`E_0`$ is the strength of the electric field, $`\widehat{k}_{||}`$ is the wave vector, $`\widehat{x}`$ is the position operator and $`\widehat{a}`$ the annihilation operator.
### 2.1 The model and methods of solution
Accurate potentials are of course required for a quantitatively correct prediction of the behavior and properties of real quantum systems. However, even qualitative conclusions drawn from simulations employing inaccurate or invalidated potentials can be problematic. The most appropriate form of the potential depends largely upon the properties of interest to the simulators. Now we consider the interaction of the abovementioned modes with a three-level atom in three different configurations, namely, $`V,`$ Lambda- and cascade-type. The transition in the 3-level atom is characterized by the dipole matrix element $`\lambda _{ij}.`$ The operator $`\widehat{S}_{ii}`$ describes the atomic population of level $`|i_A`$ with energy $`\omega _j,(j=a,b,c)`$ and the operator $`\widehat{S}_{ij},(ij)`$ describes the transition from level $`|i_A`$ to level $`|j_A`$.
The total Hamiltonian of this system is $`\widehat{H}=\widehat{H}_0+\widehat{H}_{int}`$. The $`3`$ eigenstates, $`|\xi _i`$ and corresponding eigenenergies, $`\alpha _i`$ are assumed to be known. The total wave-function may be expanded in terms of the known eigenstates, namely
$$|\mathrm{\Psi }(t)=A_1(t)|\xi _1+A_2(t)|\xi _2+A_3(t)|\xi _3.$$
(4)
With atomic units, using Schrรถdinger equation, we obtain the coupled equations for our three-level system, namely
$$i\frac{A_j(t)}{t}=r_jA_j(t)+\underset{k=1}{\overset{3}{}}H_{jk}A_k(t),$$
(5)
where $`\widehat{H}_0|\xi _i=r_i|\xi _i`$ and $`H_{jk}=\xi _j\left|\widehat{H}_{int}\right|\xi _k.`$ These equations are exact for any three-level atom. In the interaction picture, let us consider a three-level system described, in an appropriate rotating frame, by the Hamiltonian
$$\widehat{H}_{int}=\mathrm{\Delta }_1\widehat{S}_{11}+\mathrm{\Delta }_2\widehat{S}_{33}+\lambda _{21}\widehat{R}_1\widehat{S}_{21}+\lambda _{32}\widehat{R}_2\widehat{S}_{32}+\lambda _{21}^{}\widehat{R}_1^{}\widehat{S}_{12}+\lambda _{32}^{}\widehat{R}_2^{}\widehat{S}_{23}.$$
(6)
The atom-field couplings $`\lambda _{ij}`$ are given by $`\lambda _{ij}=Y\mu _{ij}.E,`$where $`E`$ is the quantized electric field given by equation (3) and $`\mu _{ij}`$ is the matrix dipole moment coupling between the state $`i`$ and $`j`$. The $`Y`$ factor accounts for local field effects and is given by $`Y=3ฯต_s(\omega )/(2ฯต_s(\omega )+1),`$ where $`ฯต_s(\omega )`$ is given in equation (1). It is easy to write $`\lambda _{ij}`$ in the following form
$$\lambda _{ij}=\frac{3ฯต_s(\omega )}{2ฯต_s(\omega )+1}.\frac{\left(\omega /\omega _T\right)^2\left(\omega _L/\omega _T\right)^2}{\left(\omega /\omega _T\right)^2\eta ^2},$$
(7)
where $`\eta ^2=[`$ $`2ฯต_s(\omega )(\omega _L/\omega _T)^2+1]/[2ฯต_s(\omega )+1]`$. The transitions between the three levels may occur in three different configurations depending upon the relationship between the energies $`E_1,E_2`$ and $`E_3`$ of levels $`1,2`$ and $`3`$. The possible configurations are (i) the $`V`$-type corresponding to $`E_2<E_1<E_3`$, (ii) the $`\mathrm{\Lambda }`$type or Raman configuration corresponding to $`E_1<E_3<E_2`$ and (iii) the $`\mathrm{\Xi }`$type or ladder-type corresponding $`E_1<E_2<E_3`$. Each of the two pairs of levels can be coupled by only one-mode or two-mode. The field operators in the abovementioned three types are (i) $`F_1=\widehat{a}^{},F_2=\widehat{b}`$ for $`V`$-type, (ii) $`F_1=\widehat{a},F_2=\widehat{b}^{}`$ for $`\mathrm{\Lambda }`$ -type and (iii) $`F_1=\widehat{a},F_2=\widehat{b}`$ for $`\mathrm{\Xi }`$-type with $`\widehat{a}=\widehat{b}`$ if both pairs of levels are coupled by the same mode.
In order to solve equations (5), we assume that
$$G(t)=A(t)+xB(t)+yC(t),$$
(8)
which means that
$$i\frac{dG(t)}{dt}=\left(r_1+v_1^{}y\right)\left\{A(t)+\frac{r_2x+v_2^{}y}{r_1+v_1^{}y}B(t)+\frac{v_2x+r_3y}{r_1+v_1^{}y}C(t)\right\},$$
(9)
where $`v_1`$ and $`v_2`$ are given using equations (5) and (6). We seek $`G(t)`$ such that $`i\stackrel{.}{G}(t)=zG(t)`$. This hold if
$$y=\frac{v_2x+r_3y}{r_1+v_1^{}y},x=\frac{r_2x+v_2^{}y}{r_1+v_1^{}y},z=r_1+v_1^{}y.$$
After some algebra this leads to a cubic equation which has three eigenvalues $`x_i(y_i)`$ which determine the $`z_i`$. There are also three corresponding eigenfunctions $`G_j(t)=G_j(0)\mathrm{exp}(iz_jt)`$, where
$$G_j(t)=M_{j1}A(t)+M_{j2}B(t)+M_{j3}C(t),$$
(10)
where
$$M_{ji}=\left(\begin{array}{c}1\\ 1\\ 1\end{array}\begin{array}{c}x_1\\ x_2\\ x_3\end{array}\begin{array}{c}y_1\\ y_2\\ y_3\end{array}\right).$$
(11)
Now, we express the unperturbed state amplitude $`A(t),B(t)`$ and $`C(t)`$ in terms of the dressed state amplitude $`R_j`$
$$F_i(t)=\underset{j=1}{\overset{3}{}}M_{ij}^1G_j(t)=\underset{j=1}{\overset{3}{}}M_{ij}^1G_j(0)\mathrm{exp}(iz_jt),$$
(12)
$`F_{1,2,3}(t)=A,B,C.`$ Using the above equations, we can write
$`A(t)`$ $`=`$ $`{\displaystyle \frac{1}{D}}\left[(x_2y_3y_2x_3)e^{iz_1t}+(x_3y_1y_3x_1)e^{iz_2t}+(x_1y_2y_1x_2)e^{iz_3t}\right],`$
$`B(t)`$ $`=`$ $`{\displaystyle \frac{1}{D}}\left[(y_2y_3)e^{iz_1t}+(y_3y_1)e^{iz_2t}+(y_1y_2)e^{iz_3t}\right],`$ (13)
$`C(t)`$ $`=`$ $`{\displaystyle \frac{1}{D}}\left[(x_2x_3)e^{iz_1t}(x_3x_1)e^{iz_2t}(x_1x_2)e^{iz_3t}\right],`$
where $`D=det(M)=x_1y_2+x_2y_3+x_3y_1x_1y_3x_2y_1x_3y_2.`$ We have thus completely determined the dynamics of a three-level system in the presence of photonic crystal.
The picture in this case is of the three-level system in the presence of photonic band gap and the detuning, rather than the usual picture of the three-level Jaynes-Cummings model (JCM) system. The important point to note here is that, using the above analytic approach, any three-level Hamiltonian is likewise exactly solvable, with precisely similar eigenvectors and eigenvalues that are obtained directly using equations (4) and (6). In Ref. an analytic approach is proposed for three-level systems, based on the Riccati nonlinear differential equation. However, the solution obtained is valid only in certain situations. On the other hand, our analytic approach removed the restriction that considered in the previous work and this solution is valid for any three-level system.
Next, we discuss a frequently encountered phenomena of particular interest in which we define the entanglement measure of the present system.
## 3 Concurrence
Quantum entanglement has recently been attracted much attention as a potential resource for communication and information processing . Entanglement is usually arise from quantum correlations between separated subsystems which can not be created by local actions on each subsystem. The concept of concurrence originates from the seminal work of Hill and Wootters where the exact expression of the entanglement of formation of a system of two qubits was derived. They showed that the entanglement of formation, an entropic entanglement monotone, is a convex monotonic increasing function of the concurrence.
It has been shown that the concurrence of a mixed two-qubit state, $`C(\rho _{AB})`$, can be expressed in terms of the minimum average pure-state concurrence, $`C\left(|\psi _{AB}\right)`$, where the minimum is taken over all possible ensemble decompositions of $`\rho _{AB}.`$ So that, the concurrence is defined of a mixed state $`\rho `$ for $`2\times 2`$ quantum systems, in the following form
$$C(\rho )=\mathrm{max}\left(\sigma _1\sigma _2\sigma _3\sigma _4\right),$$
(14)
where the $`\sigma _i`$ are the square roots of the eigenvalues of the product matrix $`Q`$, the singular values (by convention sorted in descending fashion), all of which are non-negative real quantities
$$Q=\sqrt{\rho }^T\sigma _y\sigma _y\sqrt{\rho },$$
(15)
$`\sigma _y`$ is the well-known Pauli matrix, and $`\sqrt{\rho }`$ is any matrix satisfying $`\sqrt{\rho }=\sqrt{\rho }^{}.`$ The importance of this measure follows from the direct connection between concurrence and entanglement of formation $`E_f`$
$$E_f\left(\rho \right)=\mu _+\mathrm{ln}\mu _+\mu _{}\mathrm{ln}\mu _{},$$
(16)
where
$$\mu _\pm =\frac{1}{2}\left(1\pm \sqrt{1C(\rho )^2}\right).$$
(17)
One can prove that $`\rho `$ is separable if and only if the concurrence is zero.
Let us now turn our attention to the definition of the concurrence of a pure state on a $`(N\times K)`$dimensional Hilbert space $`\mathrm{}=\mathrm{}_N\mathrm{}_K.`$ The flip operator $`F`$ acting on an arbitrary Hermitian operator $`A`$ on $`\mathrm{}`$ can be written as
$$F(A):=A+(trA)\mathrm{\Pi }(tr_NA)\mathrm{\Pi }_K\mathrm{\Pi }_N(tr_KA),$$
(18)
where $`tr_N`$ and $`tr_k`$ the partail traces over $`\mathrm{}_N`$ and $`\mathrm{}_K,`$ respectively. We denote by $`\mathrm{\Pi }_N`$ and $`\mathrm{\Pi }_K`$ the identity on $`\mathrm{}_N`$ and $`\mathrm{}_K,`$ respectively. The expectation value $`\psi \left|F(\rho _\psi )\right|\psi ,`$ where $`\rho _\psi =|\psi \psi |`$, is non-negative for all pure states and equals zero if and only if $`|\psi `$ is a product state. This allows to define the concurrence of any arbitrary bipartite pure state as
$`C\left(|\psi \right)`$ $`=`$ $`\sqrt{\psi \left|F(\rho _\psi )\right|\psi }`$ (19)
$`=`$ $`\sqrt{2\left(\psi |\psi ^2tr(\rho _N^2)\right)},`$
where $`\rho _N=tr_K\left(\rho _\psi \right)`$ is the reduced density operator of dimension N. For a normalized state, $`\psi |\psi =1,`$ it interpolates monotonously between zero for product states and $`\sqrt{\frac{2(N1)}{N}}`$ for maximally entangled states.
To investigate the concurrence for the system under consideration, we have to evaluate the reduce atomic density matrix $`\rho __A=tr__F\rho (t),`$ which can be written as
$$\rho __A=\underset{i=1,2,3}{}\rho _{_{ii}}|ii\left|+\underset{i,j=1,2,3,ij}{}\rho _{_{ij}}\right|ij|,$$
(20)
where $`\rho _{ij}(t)=i|\rho __A(t)|j,i,j=1,2`$ and $`3.`$ Using equations (19) and (20), we can write the concurrence in the following form
$$C\left(|\psi \right)=\sqrt{2\underset{i,j=1,2,3,ij}{}\left(\rho _{_{ii}}\rho _{_{jj}}\rho _{_{ij}}\rho _{_{ji}}\right)}.$$
(21)
Although the concurrence and therefore the results we obtain are not restricted to the standard one-mode three-level system, we will use that language throughout most of the paper.
Having specified the various photonic crystal and field amplitude parameters, we will present in the following subsection the results of our numerical analysis of the concurrence.
### 3.1 Numerical results
For applications in real systems, we consider the dipole emitters with frequencies in the reststrahl band of GaAs. In this subsection we will discuss the time dependence of the concurrence, which considered as an entanglement measure. We will consider the commonly used state as initial condition for the cavity field: the coherent state, which may be applicable in different situations. As might be expected, the behavior of the three-level system changes dramatically depending on the initial field state. Throughout this subsection the quantity to be examined is the concurrence $`C\left(|\psi \right).`$
In figure 3, we present the oscillatory behavior of the concurrence $`C\left(|\psi \right)`$ against the scaled time $`\lambda _1t`$ and the mean photon number $`\overline{n}`$ for different values of the detuning parameter, where $`\mathrm{\Delta }=0`$ for Fig. 3a and $`\mathrm{\Delta }=5\lambda _1`$ for Fig 3b. We consider a specific system in which the cavity is taken as GaAs with $`ฯต_0=10.89,`$ $`\eta =1.085,`$ $`\omega /\omega _T=2,`$ $`\omega _0/\omega _T=1,`$ $`\mathrm{}\omega _L=36.29`$meV, $`\mathrm{}\omega _T=33.25`$meV. The photonic crystals parameters are given by the arbitrary set, $`d_1=500`$ร
, $`d_2=300`$ร
, $`ฯต_1=9,`$ $`ฯต_2=1.3,`$ $`d_3=500`$ร
, $`d_4=400`$ร
, $`ฯต_3=10,`$ $`ฯต_4=1.5`$ and $`L=1.5d.`$ The general behavior due to the coherent state of the field does not contain any surprises it is quite broad, corresponding to the standard quantum limit. The value of concurrence at the first maximum is 1, which is quit remarkable, see figure 3a. After the time goes on, we see that the maximum value of the concurrence decreases with small amplitude of the oscillations. As the mean photon number increased, the number of oscillations decreased.
The effect of the parameter which describes the mismatch between the atomic frequency and the mean frequency of the cavity mode has been considered in figure 3b. We set the other parameters as the same as in figure 3a, and $`\mathrm{\Delta }=5\lambda _1`$. As $`\mathrm{\Delta }`$ is increased the behavior of the three-level system becomes increasingly erratic. Shorter revival times cause successive revivals to overlap and interfere so that the time evolution appears irregular. The detuning parameter at which irregularity emerges is closely tied to the mean-photon number: the higher the mean-photon number, the smaller the detuning needed to produce irregular behavior. Larger detuning also results in decreased revival amplitude due to the larger number of frequencies in the sum, which causes the rephasing to be less complete. However, a signature of the revivals persists as a return to the bare Rabi frequency even at mean-photon number high enough that the behavior looks random and the revival amplitude is essentially washed out. From our further calculations (which are not presented here), we point out that as we increase the value of the detuning one can see that the revival time is also prolonged, however the period of fluctuations is decreasing. Detuning affects the revival time by elongating it and the maximum value of the entanglement degree becomes smaller and smaller. Similar to the case of a two-level atom, detuning shifted the atomic occupation probability around which it oscillates upward meaning that the energy is stored in the atomic system.
Now we will turn our attention to the effect on the concurrence of the mode frequency $`\omega /\omega _T`$. In particular, we consider $`ฯต_0=10.89,`$ $`\eta =1.085,`$ $`\omega _0/\omega _T=1`$ and for different values of the scaled time, where $`\lambda _1t=\pi /2`$ for Fig. 4a and $`\lambda _1t=3\pi /2`$ for Fig 4b. Our particular observation is the maximum entanglement occurs near the band edges, which corresponds to $`\omega =1.085\omega _T`$. Near the band edges the wave vector parallel to the interface reaches its maximum value, and this corresponds to the first two relatively small peaks around the point $`1.085`$. In the gap region or the reststrahl region of the $`GaAs`$ system no electromagnetic fields can propagate and coupling is therefore suppressed. The extra peaks around the point $`1.085`$ are attributed to local field effects and can be understood from looking at equation (7) where $`\lambda _{ij}`$ has a pole at $`\eta =\omega /\omega _T.`$ One has to bear in mind that the above calculation did not take into explicit account the spatial dependence of the coupling parameters. Therefore, a more careful calculation would have to take into account the nonstationary property of the present system. The above model calculations suggest that physical parameters such as mode frequency, mode-atom coupling and cavity dielectric have important effects on the entanglement. One can see that the oscillations collapse after few Rabi periods and after an interval of time in which the concurrence is constant, the oscillations reappear again. This revival then collapses and a new revival begins.
This behavior highlights once again the role of the functional form of the modified Rabi frequencies in controlling the time evolution of the concurrence. Rabi frequencies which obtained in the present model are similar to that obtained from the standard three-level model but involving a frequency-dependent dielectric function. An important point to keep in mind when comparing the results presented here with results from the usual three-level system in the absence of the photonic band gap is that: they are give a different feature relative to the entanglement. This raises an interesting question: can one use the present system in building quantum logic gates? Calculations and detailed discussion of this issue will be presented in a forthcoming paper.
## 4 Phase entropy
One of the most striking features of quantum mechanics is the property that certain observable cannot simultaneously be assigned arbitrarily precise values. This property does not compromise claims of completeness for the theory, since it may consistently be asserted that such observable cannot simultaneously be measured to an arbitrary accuracy . The Shannon entropies associated with the photon number distribution $`P_m`$ and phase probability distribution $`P(\theta ,t),`$
$`P_m`$ $`=`$ $`m|\rho (t)|m,`$
$`P(\theta ,t)`$ $`=`$ $`\theta |\rho (t)|\theta ,`$ (22)
where $`|m`$ is the Fock state and $`|\theta `$ is the phase state, are given respectively by
$`R_N`$ $`=`$ $`{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}P_m\mathrm{ln}P_m,`$
$`R_\psi `$ $`=`$ $`{\displaystyle \underset{2\pi }{}}\left(P(\theta ,t)\mathrm{ln}P(\theta ,t)\right)๐\theta .`$ (23)
The entropic uncertainty relations for the number and phase distribution determine the lower bound on the sum of the Shannon entropies $`R_N`$ and $`R_\psi :`$
$$R_N+R_\psi \mathrm{ln}(2\pi ).$$
(24)
This equality is satisfied by a Fock state for which $`R_N=0`$ and $`R_\psi =\mathrm{ln}(2\pi ).`$ Other physical states give an entropic sum greater than $`\mathrm{ln}(2\pi ).`$ Specifically, for a coherent state we find that the sum is $`\mathrm{ln}(e\pi )`$ for the mean photon number greater than one, i.e.
$$R_N+R_\psi \mathrm{ln}(e\pi ).$$
(25)
The lower bound for the position-momentum entropic uncertainty relation is also given by right-hand side of this equation i.e $`\mathrm{ln}(e\pi ).`$
The single-mode of the Pegg-Barnett phase formalism which of interest in the field of quantum optics can be constructed from the single-mode phases to take the form
$$P(\theta ,t)=\underset{s\mathrm{}}{lim}\left(\frac{s+1}{2\pi }\right)\theta _m|\rho (t)|\theta _m,$$
(26)
$`|\theta _m`$ is a phase state of the mode,
$$|\theta _m=\frac{1}{\sqrt{(s+1)}}\underset{n=0}{\overset{s}{}}e^{in\theta _m}|n,$$
(27)
where $`\theta _m=\theta _{}+\frac{2\pi m}{s+1},`$ and $`m=0,1,\mathrm{}s,`$ and $`\theta _{}`$ arbitrary. Equation (26) defines a particular basis set of $`(s+1)`$ mutually orthogonal phase states.
Using the standard procedure , the phase probability distribution, the expectation value and the variance of the Hermitian phase operator may be obtained for the field. Since the coherent field at $`t=0`$ belongs to a class of partial phase states, we have chosen the reference phase $`\theta _0`$ as $`\theta _0=\beta \frac{\pi s}{s+1},`$ and introduced the new phase labels $`\zeta =m\frac{1}{2}s`$ where $`m=0,1,2,\mathrm{},s.`$ Then as $`s`$ tends to infinity the summation may be transformed into an integral after replacing $`\frac{2\pi \zeta }{s+1}`$ by $`\theta ,`$ and $`\frac{2\pi }{s+1}`$ by $`d\theta .`$ This leads to continuous phase probability distribution, where
$$P(\theta ,t)=\frac{1}{2\pi }\left(1+2\underset{n>m}{\overset{\mathrm{}}{}}\left\{A_{n,m}(t)\mathrm{cos}[\theta (nm)]+B_{n,m}(t)\mathrm{sin}[\theta (nm)]\right\}\right),$$
(28)
where $`A_{n,m}(t)`$ and $`B_{n,m}(t)`$ are given by
$`A_{n,m}(t)`$ $`=`$ $`Re\left\{A_n(t)A_m^{}(t)+B_n(t)B_m^{}(t)+C_n(t)C_m^{}(t)\right\},`$
$`B_{n,m}(t)`$ $`=`$ $`Im\left\{A_n(t)A_m^{}(t)+B_n(t)B_m^{}(t)+C_n(t)C_m^{}(t)\right\}.`$ (29)
The phase probability distribution is normalized according to $`_\pi ^\pi P(\theta ,t)๐\theta =1.`$
### 4.1 Numerical results
In what follows we shall display some general arguments based on the equality sign in the Heisenberg uncertainty relations that to demonstrate the phase entropy of a general three-level system in the presence of photonic band gab when the initial state of the field is assumed to be in a coherent state.
In figure 5a, we have plotted the phase probability distribution $`P(\theta ,t)`$ as a function of the scaled time $`\lambda _1t`$ and $`\theta `$ taking into consideration the presence of the photonic band gap. For example at time $`\lambda _1t=0`$ we realize that the phase distribution $`P(\theta ,t)`$ starts with a single-peaked structure at $`\theta =0`$ corresponding to the initial coherent state. Then as the time develops the peak splits into two peaks moving into two opposite directions.
However the amplitudes of the split peaks fluctuate in time giving a top like shape until the two peaks reach the values $`\theta =\pm \pi `$ at middle of the revival time but in this range the amplitudes of the peaks do not show any fluctuations. The picture changes greatly as time develops further (say $`\lambda _1t>40`$) where we find that the two-peak profile breaks up into multi peak with reduction of the amplitudes of these peaks. Thus the phase distribution shows diffusion as well as bifurcation. Different features are visible when we consider the off-resonant case and the behavior of the phase probability distribution is changed dramatically (see figure 5b). In this case we observe that there is a diffusion of the peaks at earlier time.
In figure 6, we consider the behavior of the phase probability distribution against the mode frequency $`\omega /\omega _T`$ and $`\theta `$ for the same parameters as in figure 5, while in this figure, we keep the the scaled time $`\lambda _1t`$ fixed, where, $`\lambda _1t=\pi /2`$ for figure 6a and $`\lambda _1t=3\pi /2`$ for figure 6b. One may clearly see that the phase probability distribution is discontinuous near the band edges. This corresponds to the zero value at the point $`1.085`$. We can prove, in an analogous manner to the equation (7), that the $`\omega /\omega _T=1.085`$ is a pole of the atom-field coupling can not avoided. It is interesting to see that, the phase probability distribution does not depend on the mode frequency for a fixed value of $`\theta ,`$ except at the point $`1.085`$. As the time increased, the only difference is that, the phase probability distribution peak splits into two peaks moving into two opposite directions, keeping the symmetry around the point $`\theta =0.`$
In figure 7, we plot the number entropy $`R_N`$ and the phase entropy $`R_\psi `$ as functions of the scaled time $`\lambda _1t`$. The initial state of the field is considered as a coherent state. We specifically present the results for the same values of figure 5. It should be noted that at a special choice of the mean-photon number parameter, the situation becomes interesting, where the Rabi frequency has a minimum value at $`\overline{n}`$. In this case we find that the general behavior of the entropies $`R_N`$ and $`R_\psi `$ and with an initially coherent field exhibit irregular structures instead of the regular structure resembling those manifested by the number or vacuum states cases.
Here it is interesting to note that the periodic oscillations are observed for a short period of the interaction time only. When we consider smaller mean-photon number, the regularity behavior of the oscillations in the entropies $`R_N`$ and $`R_\psi `$ are still obvious (see figure 8) where we have considered the initial mean photon number $`\overline{n}=10`$. However, the number of oscillations is increased. Also it is interesting to point out that at the revival time optimal phase entropy is attained in all the cases which means that the atom has achieved an almost pure state, this has been observed all through our figures. The number and phase entropic uncertainties for a weak coherent state follow those of underlying states superposition.
In figure 9 we plot the entropies $`R_N`$ and $`R_\psi `$ against the mode-frequency $`\omega `$ in units of $`\omega _T`$ for different values of the scaled time. Now, where the atom-field coupling is proportional to $`\lambda _{ij}`$ this explains the origin of the second peak in this figure. It is interesting to note the dependence of these entropies on the mode-frequency, with different values of the scaled time. We wonder, as a possible generalization of this concept, whether there exist another family of similar oscillations if we consider two-qubit system. In such a case, the properties of these systems would probably be of interest, in order to bring further insight and knowledge about entanglement and quantum logic gates for multi-partite systems. As the scaled time increased, a characteristic feature of the entropies $`R_N`$ and $`R_\psi `$ is quit interesting, where more oscillations exist, also, only around the resistable region for the number entropy but less number of oscillations exist for the phase entropy (see figure 10).
Far from the resistable region the entropies behavior observed here does not depend on the mode frequency and the intensity of the initial field mode.
Given a $`3D`$ photonic crystal with a complete gap, one has the possibility of introducing a defect in the structure which will create a localized state in the gap. If this is a point-like defect then the photon mode will be completely localized about a point. In figure 10, we show the zero point associated with the defect created by removing a small amount of dielectric from one of the vertical dielectric columns of the crystal structure. The resulting defect mode has a state near mid gap. One feature that should be highlighted in this context is the appearance of a frequency gap between the pair of interface dispersion. This gap is present only when the two photonic crystal regions are different, and disappears when they are identical.
## 5 Experimental prospects
The perfect semiconductor crystal is quite elegant and beautiful, but it becomes ever more useful when it is doped. Likewise, the perfect photonic crystal can become of even greater value when a defect is introduced . The point to make about photonic crystals is that they are very empty structures, consisting of about 78 empty space. But in a sense they are much emptier than that. They are emptier and quieter than even the vacuum, since they contain not even zero-point fluctuations within the forbidden frequency band. Our model system consist of a three-level atom located inside a photonic band gap material. There are several ways of placing such an atom inside a photonic crystal. From a material standpoint, it is possible to dope an existing photonic band gap material using ion beam implantation methods. For instance, it has recently been shown that $`Er^{3+}`$ ions implanted into bulk silicon exhibit sharp free-atom-like spectra . Intense temperature-dependent photoluminescent (PL) at $`1.54\mu m`$ is observed in the system at low temperatures (when the host material is crystalline, Er-related PL is quenched at temperatures above $`80K`$ so that it cannot be detected at room temperatures). This wavelength is particularly significant because it corresponds to the minimum absorption of silica fibre-based optical communication system. Because the PL at $`1.54\mu m`$ is due to the spin-orbit split $`{}_{}{}^{4}I_{13/2}^{}^4I_{15/2}`$ of $`4f`$ electrons in the $`Er^{3+}`$ ions which are shielded by outer $`5s^25p^6`$ shells, the influence of the host lattice on the luminescence wavelength is weak. (The key to the success of erbium is that the upper level of the amplifying transition $`{}_{}{}^{4}I_{13/2}^{}`$ is separated by a large energy gap from the next-lowest level $`{}_{}{}^{4}I_{15/2}^{}`$ so that its lifetime is very long and mostly radiative. In spite of the screening of the atomic transition by the outer shells, it is likely that thermal phonons in the silicon host would cause significant dephasing of the quantum degrees of freedom within the erbium 4f shell. Consequently, such a system must be cooled to liquid helium temperatures. Such experiments appear to be nearly within the reach of current technology. Although it has not yet been demonstrated, the system consisting of a multi-level system coupled to a multi-mode appears to be another potential candidate for achieving new features. Such systems are potentially interesting for their ability to process information in a novel way and might find application in models of quantum logic gates. Therefore, atoms or trapped ions + cavities in a presence of photonic band gap represent, in our opinion, a very promising system for quantum information processing.
## 6 Conclusion
In this communication the quantum electrodynamic properties of a three-level atom embedded in a photonic band gap material were investigated. We have focused on the application of the effective-medium theory to the present problem in a nanoscale dielectric cavity QED situation. The effective-medium approach can in fact be applied to situations in which all three regions of the structure possess frequency-dependent dielectric functions. Specifically, the combined effects of coherent control by an external driving field and photon localization facilitated by a photonic band gap on entanglement from a three-level atom embedded in a photonic band gap material were examined. Exact solutions of the wave function in the Schrรถdinger picture have been obtained within rotating wave approximation. In particular, we have chosen to focus on three-level system coupled to a single mode. Observation of the three-level system may offer some insight into the quantum nature of the resonator, just as atoms provide a sensitive probe for the nonclassical nature of electromagnetic fields. The observation of revivals, which are a strictly nonclassical phenomenon, would give evidence for the quantum nature of the quantum system.
The results point to a number of interesting features, which arise from the variation of the adjustable parameters of the system, namely, the mode-frequency, dipole vector orientation, dipole position within the slab, the slab width, and the photonic crystal parameters: layer widths and dielectric functions. Our investigations for the entanglement, collapse-revival phenomena, and phase and number entropic uncertainty relations in the presence of the photonic band gap as compared with the usual three-level model are summarized as follows:-
i) The concurrence behavior is reflect the pattern of collapse and revival which is qualitatively similar to that of the usual three-level model but with reduced amplitude. In case of a smaller mean photon number and for initially excited atom the usual pattern in the three-level model of collapse and revival changes to rapid fluctuations of interference patterns for all time considered. In this way, our concurrence function contains all the information necessary to identify the entanglement of a given state. Nevertheless, it depends on the particular choice of the mode-frequency.
ii) The phase entropy can be used to measure entanglement of the system presented here with explicitly atom-field coupling in the presence of photonic band gap. We would like to point out that the phase Shannon entropic considered for the presented model has not been treated in this manner before.
iii) The photonic band gap introduces sudden changes in the concurrence and phase entropy due to the variation of these quantities with mode frequency. This feature attributed to the fact that in the photonic band gap region electromagnetic modes are not allowed to propagate into the dielectric slab and hence no interaction can take place in this region. Theory predicts analytically this behavior for a GaAs system at $`\omega =\eta \omega _T`$.
Finally, we emphasize the fact that without any conditions it was possible to obtain exact analytic solution which reproduce the most important features of the three-level atom interacting with a cavity one- or two-mode in the presence of photonic band gap. A similar set of equations have been derived in for a three-level system using some approximations, based on the Riccati nonlinear differential equation. In contrast, the method used here gives exact analytic solutions without any conditions.
Acknowledgment
I acknowledge the hospitality and financial support from the Center for Computational and Theoretical Sciences, Kulliyyah of Science, IIUM, Malaysia where the final version of the paper was prepared. Also, helpful discussions with Prof. A.-S. F. Obada and Prof. M. R. B. Wahiddin are gratefully acknowledged.
|
warning/0506/cs0506036.html
|
ar5iv
|
text
|
# Non prefix-free codes for constrained sequences
## I Introduction
Variable length codes are usually considered to have to satisfy the Kraft inequality (), as it was proved that this is a necessary condition for unique decodability in . Unique decodability is always defined as a characteristic of the set of code words only, and no reference to the type of source is considered. Thus, imposing unique decodability we are requiring any sequence of symbols being distinguishable. This fact is clearly perfectly acceptable if the source can produce any sequence of symbol with non-null probability as, for example, in the memoryless source case. But what happens if the source is not memoryless and, more precisely, not all sequences are allowed? Consider for example a source with three symbols $`A`$, $`B`$ and $`C`$, and suppose that symbol $`A`$ can never be followed by $`B`$. This fact is a characteristic of the source that we can consider known for the encoding problem, as we actually do in the classic Shannon paradigm where no bits are assigned to null-probability events. Thus, the decoder is usually supposed to know that an event is not possible, at least from the fact that that no code word is assigned to that event. Thus, suppose we use the codewords $`{}_{}{}^{}0_{}^{}`$, $`{}_{}{}^{}1_{}^{}`$ and $`{}_{}{}^{}01_{}^{}`$ for our source symbols $`A`$, $`B`$ and $`C`$ respectively. It is easy to see that any sequence of symbols that can be generated by the source is uniquely specified by concatenating the codewords of every single symbol. This is because with this code one may only confuse sequences of symbols with sequences that our source cannot generate. The objective of this paper is to propose the notion of unique decodability of a set of codewords relatively to a specific information source. We show simple examples of Markov sources that can be efficiently encoded by non-prefix free codes. Our examples also show that there are sources for which a coding technique exists such that the expected number of bits of the code for the first $`n`$ symbols of the source is strictly smaller than the entropy of those first $`n`$ symbols (for every $`n`$). This fact is often erroneously considered impossible, and this error arises from considering the Kraft inequality a necessary condition for any type of sources.
## II A curious Markov chain example
Suppose, so as to analyze more throughly the example of the introduction, we have a source generating symbols $`X_1,X_2,X_3,\mathrm{}`$ extracted from the set $`๐ณ=\{A,B,C\}`$ following the Markov chain graphically shown in figure 1.
In formulae, if we call $`๐_i`$ the probability distribution row vector on $`๐ณ`$ at step $`n`$, we have $`๐_{i+1}=๐_i๐`$, where $`๐`$ is the transition probability matrix
$$๐=\left[\begin{array}{ccc}1/2& 0& 1/2\\ 1/4& 1/2& 1/4\\ 1/4& 1/2& 1/4\end{array}\right]$$
(1)
If we suppose the initial state has uniform probability $`๐_1=[1/3,1/3,1/3]`$, it is easy to verify that the process is stationary, i.e $`๐_i=๐_1`$ for every $`i`$. Thus, the entropy $`H(X_1,X_2,\mathrm{},X_k)`$ of the first $`k`$ symbols can be easily computed. We have
$$\begin{array}{c}H(X_1,X_2,\mathrm{},X_k)=\hfill \\ \hfill H(X_1)+H(X_2|X_1)+\mathrm{}+H(X_k|X_{k1})=\\ \hfill H(X_1)+H(X_2|X_1)(k1)=\mathrm{log}(3)+\frac{4}{3}(k1)\end{array}$$
(2)
Let us consider now the codeword assignment $`A0`$, $`B1`$, $`C01`$. This code is clearly not prefix-free but, as explained in the introduction, when used for this source no ambiguity can arise. If $`l(X)`$ is the length of the code associated to $`X`$, we can compute the expected number of bits in coding the first $`k`$ symbols as
$$\begin{array}{c}E[l(X_1X_2X_3\mathrm{}X_k)]=E\left[\underset{i=1}{\overset{k}{}}l(X_i)\right]=\hfill \\ \hfill \underset{i=1}{\overset{k}{}}E[l(X_i)]=k\left(\frac{1}{3}+\frac{1}{3}+\frac{2}{3}\right)=\frac{4}{3}k\end{array}$$
(3)
It is easy to verify that this value is strictly smaller than the entropy $`H(X_1,X_2,\mathrm{},X_k)`$ above computed. What happens with our coding procedure? And what happens when the number of symbols goes to infinity? Looking carefully at our example, we note that our coding strategy uses an expected number of $`4/3`$ bits for coding the first symbol, while its entropy is $`\mathrm{log}3`$. For the following symbols, in turn, the entropies $`H(X_2|X_1)`$, $`H(X_3|X_2)\mathrm{}`$ equal $`4/3`$ bits, and thus they have exactly the same value as the number of bits used by our code. So, we can say that our code only gains in the first symbol and not substantially (as it is obvious from AEP for ergodic sources). But this fact is somehow interesting; our code assigns to the first symbol a number of bits smaller than its entropy, using the memory properties of the source, without affecting unique decodability. Thus, if we consider the case when only a finite number of symbols are given the problem of finding the optimal coding strategy arises. Furthermore, we should consider that in the more general case, when higher order constraint are eventually present, the problem becomes much more intriguing.
## III Kraft inequality for constrained sequences
McMillan showed in that a necessary condition for the unique decodability of a set of $`n`$ codewords is that their lengths $`l_1,l_2,\mathrm{},l_n`$ satisfy the Kraft inequality
$$\underset{i=1}{\overset{n}{}}2^{l_i}1$$
(4)
Karush () gave a simple proof of this fact by considering that for every $`k>0`$ the following inequality must be satisfied
$$\left(\underset{i}{}2^{l_i}\right)^kkl_{\text{max}}$$
(5)
The term on the left hand side of (5) can be written as the sum of weights of codes of possible sequences of $`k`$ symbols. For example, a sequence starting with $`x_1,x_3,x_2,\mathrm{}`$ gives a term $`2^{l_1}2^{l_3}2^{l_2}\mathrm{}`$ in the expantion of the left hand side of (5). In order to have only one sequence assigned to every code the above inequality is necessary. But if we only want to distinguish between the possible sequences generated by a constrained source, we may rewrite the condition in a less stringent form. Let us consider once more as an example the source of fig. 1 with $`l_1`$, $`l_2`$ and $`l_3`$ lengths assigned respectively to $`A`$, $`B`$ and $`C`$. Thus, terms on the expansion of left hand side of (5) that contains $`\mathrm{}2^{l_1}2^{l_2}\mathrm{}`$ should not be considered as $`B`$ cannot follow $`A`$ in a source sequence. Let us consider the matrix
$$๐=\left[\begin{array}{ccc}2^{l_1}& 0& 2^{l_1}\\ 2^{l_2}& 2^{l_2}& 2^{l_2}\\ 2^{l_3}& 2^{l_3}& 2^{l_3}\end{array}\right];$$
(6)
it is possible to verify that the really necessary correspondent of eq. (5) for our source should be written, for $`k>0`$, as
$$\left[\begin{array}{ccc}1& 1& 1\end{array}\right]๐^{k1}\left[\begin{array}{c}2^{l_1}\\ 2^{l_2}\\ 2^{l_3}\end{array}\right]kl_{\text{max}}$$
(7)
It is possible to show that a necessary condition for this inequality to be satisfied for every $`k`$ is that the matrix $`๐`$ has spectral radius<sup>1</sup><sup>1</sup>1The spectral radius of a matrix is defined as the greatest modulus of its eigenvalues. at most equal to 1. We state and prove this fact in the general case.
###### Theorem 1
Let $`๐`$ be an irreducible $`n\times n`$ stochastic matrix and $`๐ฅ=[l_1,l_2,\mathrm{},l_n]`$ a vector of $`n`$ integers. Let $`๐`$ be the $`n\times n`$ matrix such that
$$๐_{ij}=\{\begin{array}{cc}0\hfill & \text{if }P_{ij}=0\hfill \\ 2^{l_i}\hfill & \text{if }P_{ij}>0\hfill \end{array}$$
(8)
Then, a necessary condition for the codeword lengths $`l_1`$, $`l_2`$, โฆ, $`l_n`$ to be lengths of a uniquely decodable code for a Markov source with transition probability matrix $`๐`$ is that $`\rho (๐)1`$, where $`\rho (๐)`$ is the spectral radius of $`๐`$.
###### Proof:
We follow Karushโs proof of McMillan theorem. Suppose without loss of generality that the set of our source symbols is $`๐ณ=\{1,2,\mathrm{},n\}`$, and call $`๐ณ^{(k)}`$ the set of all sequences of $`k`$ symbols that can be produced by the source. Let us set $`๐=2^๐ฅ=[2^{l_1},2^{l_2},\mathrm{},2^{l_n}]`$ and define, for $`k>0`$,
$$๐_k^T=๐^{k1}๐^T.$$
(9)
Then it is easy to see by induction that the $`i`$-th component of $`๐_k`$ is written as
$$๐_k^i=\underset{x_1,x_2,\mathrm{},x_k}{}2^{l_{x_1}l_{x_2}\mathrm{}l_{x_k}}$$
(10)
where the sum runs over all elements $`(x_1,x_2,\mathrm{},x_k)`$ of $`๐ณ^{(k)}`$ with $`x_1=i`$. So, if we call $`\mathrm{๐}_n`$ the row vector composed of $`n`$ 1โs, we have
$$\mathrm{๐}_n๐^{k1}๐^T=\underset{x_1,x_2,\mathrm{},x_k}{}2^{l_{x_1}l_{x_2}\mathrm{}l_{x_k}}$$
(11)
where the sum now runs over all elements of $`๐ณ^{(k)}`$. Thus, reindexing the sum with respect to the total length $`l=l_{x_1}+l_{x_2}+\mathrm{}+l_{x_k}`$ and calling $`c_l`$ the number of sequences of $`๐ณ^{(k)}`$ to which correspond a code of length $`l`$, we have
$$\mathrm{๐}_n๐^{k1}๐^T=\underset{l=kl_{\text{min}}}{\overset{kl_{\text{max}}}{}}c_l2^l$$
(12)
where $`l_{\text{min}}`$ and $`l_{\text{max}}`$ are respectively the maximum and the minimum of the values $`l_i,i=1,2,\mathrm{},n`$. Since the code is uniquely decodable, all $`c_l`$ sequences of length $`l`$ must be different and so they are at most $`2^l`$. This implies that, for every $`k>0`$, we must have
$$\mathrm{๐}_n๐^{k1}๐^T\underset{l=kl_{\text{min}}}{\overset{kl_{\text{max}}}{}}2^l2^l=k(l_{\text{max}}l_{\text{min}}+1)$$
(13)
Now, note that the irreducible matrix $`๐`$ is also nonnegative. Thus, for the Perron-Frobenius theorem (see for details), its spectral radius $`\rho (๐)`$ is also an eigenvalue<sup>2</sup><sup>2</sup>2Note that in general the spectral radius is not an eigenvalue as it is defined as the maximum of $`|\lambda |`$ over all eigenvalues $`\lambda `$., with algebraic multiplicity 1 and with positive associated eigenvector. As the vectors $`\mathrm{๐}_n`$ and $`๐`$ are both positive, this implies that the term on the left hand side of eq. (13) asymptotically grows like $`\rho (๐)^{k1}`$. On the contrary, the right hand side term only grows linearly with $`k`$; so, taking the limit as $`k\mathrm{}`$ in eq. (13) we conclude that $`\rho (๐)1`$. โ
We note that if the $`๐`$ matrix has all strictly positive entries, the matrix $`๐`$ reduces to have all equal columns, and its spectral radius is exactly $`2^{l_i}`$. Thus, for non-constrained sequences, we obtain the classic Kraft inequality. Furthermore, as the spectral radius of a nonnegative positive matrix increases if any of the elements increases, we note that the situation $`\rho (๐)=1`$ is a strong condition on $`๐`$ and $`๐ฅ`$. In the sense that if for a given matrix $`๐`$ there is a decodable code with codeword lengths $`l_i,i=1,\mathrm{},n`$ such that $`\rho (๐)=1`$, then there is no decodable code with lengths $`l_i^{}`$ if $`l_i^{}<l_i`$ for some $`i`$. Also, it is not possible to remove constraints from the Markov chain while keeping unique decodability property.
The most important remark, however, concerns the non sufficiency of the stated condition. In fact, while the classic Kraft inequality is a necessary and sufficient condition for the existence of a uniquely decodable code for an unconstrained sequence, the found inequality $`\rho (๐)1`$ is unfortunately only necessary, and not sufficient. We discuss this point in the next section, where we propose an extension of the Sardinas Patterson test for testing the unique decodability of a code for a constrained sequence.
## IV Non sufficiency and Sardinas Patterson test
In the preceding sections we have shown that the classic Kraft inequality is not, in general, a necessary condition for the unique decodability of a constrained sequence, and we have found a necessary condition under this hypothesis. Unfortunately, the found condition is not sufficient and trivial examples show this fact. We note that the only parameter determining the matrix $`๐`$ are the length vector $`๐ฅ`$ and the graph associated to the Markov chain, i.e. the state pairs with positive transition probability. Thus, we only consider the transition graphs of the sources without taking into account the value of the transition probabilities. Consider for example a source with three symbols $`A,B`$ and $`C`$ with transition graph as shown in fig. 2(a). It is easy to see that if $`๐ฅ=[1,1,1]`$ then $`\rho (๐)=1`$; anyway, it is clearly impossible to decode the sequences of the source if we assign only one bit to every symbol. In general, we may consider that it is not possible to have a decodable code with more than $`2^i`$ codewords of length $`i`$, because otherwise even the initial state cannot be recovered. Anyway, still imposing this additional condition does not suffice. Take for example a code with $`๐ฅ=[1,1,2]`$ for a source with transition graph as shown in fig. 2(b); we have $`\rho (๐)<1`$, only two codewords of 1 bit and one codeword of 2 bits, but still a decodable code with those lenghts does not exist ($`A0`$ imposes $`B1`$, and consequently $`C11`$, but so $`BCB`$ and $`CC`$ have the same code).
The above examples show that the question of finding a sufficient condition for the unique decodability of codes for constrained sequences appears to be more complicated than with unconstrained sequences. A positive fact is that it is possible to extend the Sardinas Patterson (SP) test () to the case of our interest. Given a set of codewords, the SP test allows to establish in a finite number of steps if the code is uniquely decodable for unconstrained sequences. Here we modify the classic algorithm for the case of constrained ones. The generalization is straightforward so that we do not give here a formal proof, as it would merely be a rewriting of that for the classic SP test, for which we refer the reader to \[6, th. 2.2.1\].
Suppose our source symbol set is $`๐ณ=\{1,2,\mathrm{},n\}`$ and let us call $`W=\{W_i\}_{i=1,\mathrm{},n}`$ the set of associated codewords. For $`i=1,2,\mathrm{},n`$ we call $`F_i=\{W_j|P_{ij}>0\}`$ the subset of $`W`$ containing all codewords that can follow $`W_i`$ in a source sequence. We construct a sequence of sets $`S_1,S_2,\mathrm{}`$ in the following way. To form $`S_1`$ we consider all pairs of codewords of $`W`$; if a codeword $`W_i`$ is a prefix of another codeword $`W_j`$, i.e. $`W_j=W_iA`$ we put the suffix $`A`$ into $`S_1`$. In order to consider only the possible sequences, we have to remember the codewords that have generated every suffix; thus, let us say that we mark the obtained suffix $`A`$ with two labels, thus we indicate it with $`{}_{i}{}^{}A_{j}^{}`$. We do this for every $`i`$ and $`j`$. Then, for $`n>1`$, $`S_n`$ is constructed by comparing elements of $`S_{n1}`$ and elements of $`W`$; for a generic element $`{}_{l}{}^{}B_{m}^{}`$ of $`S_{n1}`$ we consider the subset $`F_l`$ of $`W`$:
* If a codeword $`W_kF_l`$ is equal to $`{}_{l}{}^{}B_{m}^{}`$ the algorithm stops and the code is not decodable,
* if $`{}_{l}{}^{}B_{m}^{}`$ is prefix of a codeword $`W_r={}_{l}{}^{}B_{m}^{}C`$ we put the labelled $`{}_{m}{}^{}C_{r}^{}`$ suffix into $`S_n`$,
* if if instead a codeword $`W_s`$ is prefix of $`{}_{l}{}^{}B_{m}^{}=W_sD`$, we place the labelled suffix $`{}_{s}{}^{}D_{m}^{}`$ into $`S_n`$.
The code is uniquely decodable if and only if item a) is never reached. Note that the algorithm can be stopped after a finite number of steps; there are in fact only a finite number of possible different sets $`S_i`$ and so the sequence $`S_i,i=1,2,\mathrm{}`$ is either finite or periodic. We note that the code is *finite delay* uniquely decodable if the sequence $`S_i`$ is finite and *infinite delay* uniquely decodable if the sequence is periodic.
As an example of SP test for constrained sequences we consider the transition graphs shown in fig. 3. For both cases we use codewords 0, 1, 01 and 10 for $`A,B,C`$ and $`D`$ respectively. For the graph of fig. 3(a) we obtain $`S_1=\{_A1_C,{}_{B}{}^{}0_{D}^{}\}`$, $`S_2=\mathrm{}`$. Thus the code is finite delay uniquely decodable and we can indeed verify that we need to wait at most two bits for decoding a symbol. For the graph of fig. 3(b), instead, we have $`S_1=\{_A1_C,{}_{B}{}^{}0_{D}^{}\}`$, $`S_2=\{_C0_D,{}_{D}{}^{}1_{C}^{}\}`$, $`S_3=S_2`$ and then $`S_i=S_2`$ for every other $`i>3`$. So, the code is still uniquely decodable but with infinite delay; in fact it is not possible to distinguish the sequences $`ADDD\mathrm{}`$ and $`CCC\mathrm{}`$ until they are finished, so that the delay may be as long as we want.
## V Entropy rate and average lengths
In the second section we have seen an example of Markov chain that is efficiently encoded with a non prefix-free code. Here we note that the same thing happens with the sources of fig 3 with the indicated codewords if the transition probability matrix associated are
$$๐=\left[\begin{array}{cccc}1/2& 0& 1/2& 0\\ 0& 1/2& 0& 1/2\\ 1/4& 1/4& 1/4& 1/4\\ 1/4& 1/4& 1/4& 1/4\end{array}\right]$$
(14)
and
$$๐=\left[\begin{array}{cccc}1/2& 0& 1/4& 1/4\\ 0& 1/2& 1/4& 1/4\\ 0& 1/2& 1/4& 1/4\\ 1/2& 0& 1/4& 1/4\end{array}\right]$$
(15)
respectively, and the initial state is uniformly distributed. For this sources the entropy of the first $`k`$ symbols is $`(3k+1)/2`$ while our code uses on average $`3k/2`$ bits. Thus we cannot say that, even for stationary ergodic processes<sup>3</sup><sup>3</sup>3Note that in this paper we have considered only stationary processes; for nonstationary processes there may be still more surprising results., the minimum average length of the code for the first $`k`$ symbols is greater than or equal to their entropy. The only thing we can say is about the entropy rate.
Given a Markov source $`X_1X_2X_3\mathrm{}`$ with transition matrix $`๐`$, let $`\mu =[\mu _1,\mu _2,\mathrm{},\mu _n]`$ be the stationary distribution, and $``$ the entropy rate. Using the asymptotic equipartition property for ergodic sources (Shannon-McMillan theorem, ), we deduce that for every uniquely decodable code we must use at lest $``$ bits per symbol on average and thus $`๐ฅ\mu ^T`$. Note that this does not imply that $`E[l(X_1)]H(X_1)`$, nor, in general, that for some fixed $`n`$ we have $`E[l(X_1,X_2,\mathrm{},X_k)]H(X_1,X_2,\mathrm{},X_k)`$ (see \[8, Th. 5.4.2\]), and our examples precisely show that in fact this is not true. The point is that for sources with memory we can use codes that do not respect the classic Kraft inequality and thus the inequality $`E[l(X)]H(X)`$ cannot be proved. Thus, we should be careful in formulating the converse theorem for variable length codes.
Anyway, the examples shown in this and in preceding sections leave many open questions. We have shown examples of transition graphs with associated codeword lengths such that the obtained code is uniquely decodable. For every one of these graphs we have shown that there is a transition probability matrix $`๐`$ such that the entropy rate of the source exactly equals the average length per symbol of our code. We should ask if this is only a coincidence or if it is the rule. Furthermore, we may ask what happens in terms of efficiency with our code for a generic matrix $`๐`$ on a given graph; is it possible to find an optimal codeword assignment as we do with Huffman codes in the classic framework? Again, what happens if we consider blocks of more than one symbols? We clearly obtain a different transition graph and the efficiency cannot be lower; but is there a gain in some cases or not?
## VI Some considerations
It is interesting to consider the proposed coding approach from the point of view of the encoding complexity. Consider for example the stationary Markov chain with transition matrix of eq. (14) and uniform initial distribution. Note that it is possible to encode the first $`k`$ symbols of the source using exactly as many bits as the entropy of those symbols using Huffman codes. Two bits are used for the first symbol, which correspond to its entropy. Then, at every step we use a different Huffman code, depending on the preceding step, and use on average $`H(X_{k+1}|X_k)==3/2`$ bits. Note that this technique actually fits with the conditional entropy idea in the sense that it really encodes each symbol given the preceding one. This implies that the encoder must trace the state of the source and choose the code for the new symbol. On the contrary, the non prefix-free codeword assignment indicated in fig. 3 allows a very simple encoding phase, as there is a fixed mapping from symbols to code bits, with the same (slightly better) compression performance. The point is that we are making a different use of the decoder knowledge about possible transitions. Note that, even for the Huffman code, we are supposing that the decoder exactly knows what transitions are possible and what are not. The difference is that with the non prefix-free code we are making the decoder more active. This relates the presented idea to other developed coding paradigms. We should note in fact, that in practice the proposed approach was already used in other contexts. One of the oldest examples may be that of modulo-PCM codes () for numerical sequences; here only the modulo-4 value of every sample is encoded, leaving to the decoder the task of understanding the original value using its knowledge on the memory of the source. In that case the used code is even a singular code<sup>4</sup><sup>4</sup>4In our examples here we have considered only non singular codes with singular extension. The non singularity of the code is required in our setting where we want to be able to decode a sequence composed of one single symbol. but under certain hypothesis this does not affect the decodability. Similar ideas are then used in the recently reemerged theme of distributed source coding (see for example and references therein). Let us consider for a moment the problem of noiseless separated coding of dependent sequences. The well known Slepian-Wolf theorem () says that two correlated memoryless sources, $`X`$ and $`Y`$, can be separately lossless coded at rates $`R(X)`$ and $`R(Y)`$ respectively when jointly decoded, if $`R(X)H(X|Y)`$, $`R(Y)H(Y|X)`$ and $`R(X)+R(Y)H(H,Y)`$. Cover extended the result to the case of general ergodic sources in . Roughly speaking, the used encoding process at every encoder consists on considering large blocks of symbols; the set of all such blocks is split into disjoint bins and only the index of the bin that contains the extracted block is encoded. At the decoder the original block for both sequences $`X`$ and $`Y`$ is recovered by extracting from the pair of specified bins the only pair of jointly typical blocks. It is interesting to note that this encoding technique actually uses singular codes in order to achieve compression, leaving to the decoder the task of disambiguating, based on the joint statistic of the two sources.
The same idea is now being used, in order to shift the complexity from the encoder to the decoder, in what may be called โsource coding based on distributed source coding principlesโ. Special attention in this field is being payed to the case of video coding (see for example ). In this contest the memory of the video source is not exploited in the encoding phase but in the decoding one. Again, roughly speaking, in the encoding phase no motion compensation and prediction is applied and singular codes are used for the data compression task. In the decoding phase, on the contrary, the memory of the source is exploited in order to remove ambiguity by using motion compensation. It is interesting to see that practical architectural specification for this video coding techniques have been presented only in recent years, even if the idea of such an approach to video coding was already patented by Witsenhausen and Wyner in late โ70s (). Moreover, it is also interesting to note that there is not much difference in this approach with respect to the idea behind the modulo-PCM coding above mentioned.
As a final comment on the general problem of finding variable length non prefix-free codes for a given constrained sequence, we note that there are many connections with the area of coding for constrained channels (see for example ). The relation between the transition graphs of our sources and the possible codeword assignments may find interesting counterparts (and maybe answers) in that field, where graph theory has already been successfully applied.
|
warning/0506/hep-th0506143.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
T-duality invariance, the result of interchanging small radius with large radius, $`R`$ $``$ $`\alpha ^{}/R`$, is a spontaneously broken symmetry in String/M Theory: in other words, a T-duality transformation on an embedding target space coordinate will, in general, map a given background of String/M theory to a different background of the same theory. To be specific, the circle-compactified $`E_8`$$`\times `$$`E_8`$ heterotic oriented closed string theory is mapped under a T-duality to the circle-compactified $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic oriented closed string theory . The circle-compactified type IIA oriented closed string theory is mapped to the circle-compactified type IIB oriented closed string theory . And, finally, the unoriented type IB open and closed string theory is mapped under a T-duality transformation to the, rather unusual, type I unoriented open and closed string theory: the integer-moded open string momentum modes associated with the compact coordinate are mapped to the integer-moded closed string winding modes in the T-dual type I theory .
The results in this paper are a beautiful illustration of the significance of the Euclidean T-duality transformations linking the six different weakly coupled string theory limits of the String/M Duality web at finite temperature.<sup>2</sup><sup>2</sup>2The idea of pursuing the significance of Euclidean T-duality transformations on the different string theory limits of the String/M Duality Web came to me after reading Polchinskiโs discussion of thermal duality in the closed bosonic string theory given in his textbook , an approach I first considered in . The observation that the one-loop string free energy has a first principles derivation from the Polyakov path integral, therefore expressible as an integral over the fundamental domain of the modular group, is apparent in Polchinskiโs analysis of the closed bosonic string ensemble in ; I simply applied that observation to the six supersymmetric string theories. For the heterotic string, I discovered that some of the necessary inputs for the one-loop vacuum amplitude already existed in the literature , except that their relationship to finite temperature string theory had not been noted. My presentation of the type I and type I strings in was guided by the well-established isomorphism to the heterotic strings , motivated also by the implications for the physics of the low energy finite temperature gauge theory limit . Finally, the impossibility of a type II canonical ensemble became apparent to me in the follow-up work , where I also discovered the significance of the temperature dependent Wilson line in both heterotic and type I ensembles, the connection to conjectures for strong-weak coupling duality without supersymmetry , and the $`T^2`$ growth of the string free energy at high temperatures. The realization that the conventional lore of the Hagedorn transition is flawed came to me in , and was explored further in subsequent works . Thus, the presentation that follows in this paper has been determined by pedagogy, rather than the chronology of my own understanding. It is clear that a Wick rotation on $`X^0`$ maps the noncompact $`SO(9,1)`$ Lorentz invariant background of a given supersymmetric string theory to the corresponding $`SO(10)`$ invariant background, with an embedding time coordinate of Euclidean signature. The Wick-rotated $`SO(10)`$ invariant background arises naturally in any formulation of equilibrium string statistical mechanics in the canonical ensemble, the statistical ensemble characterized by fixed temperature and fixed spatial volume $`(\beta ,V)`$. The Polyakov path integral over connected world-surfaces is formulated in a target spacetime of fixed spacetime volume. Thus, the one-loop vacuum functional in the $`SO(10)`$ invariant background computes precisely the sum over connected one-loop vacuum graphs in the target space $`R^9`$ in the finite temperature vacuum at temperature $`T`$ . The appearance of a tachyonic mode in the string thermal spectrum is an indication that the worldsheet conformal field theory is no longer at a fixed point of the 2d Renormalization Group (RG): the tachyon indicates a relevant flow of the 2d RG. The question of significance is then as follows: does the relevant flow terminate in a new infrared fixed point? If so, the new fixed point determines the true thermal string vacuum. An equilibrium statistical mechanics of strings requires that this fixed point belong to a fixed line parameterized by inverse temperature $`\beta `$ : the precise analog under Wick rotation of the line of fixed points parameterized by the radius of a compact spatial coordinate in the $`SO(9,1)`$ vacuum.
Target spacetime supersymmetry, and its spontaneous breaking in the thermal vacuum along the line of fixed points parameterized by $`\beta `$, introduces new features into this discussion. We must require compatibility with the expected properties of the low energy field theory limit where the contribution from massive string modes has been suppressed, namely, those of a 10D finite temperature supersymmetric gauge theory. We must also require consistency with string theoretic symmetries of both worldsheet, and target space, origin in the Wick rotated $`SO(10)`$ invariant background. In particular, Euclidean T-duality transformations must link the thermal vacua of the six different supersymmetric string theories in pairs: heterotic $`E_8`$$`\times `$$`E_8`$ and $`\mathrm{Spin}(32)/\mathrm{Z}_2`$, type IIA and type IIB, and type IB and type I. Remarkably, we will find as a direct consequence of the T-duality transformations, that the tachyonic thermal instabilities arising in all previous attempts to formulate an equilibrium supersymmetric string statistical mechanics in the canonical ensemble for the heterotic and type I strings are simply absent. For the type II superstrings, in the absence of Dbranes or fluxes, we will find that there is no canonical ensemble: the thermal tachyon free background also has target space supersymmetry, as a consequence of modular invariance.
Having clarified that the canonical ensemble of heterotic and type I strings is well-defined at all temperatures, including the temperature regime far above the string mass scale, $`\alpha ^{1/2}`$, we will establish several new results in this paper. First, we show that in either ensemble the growth of the free energy, $`F(\beta )`$, at high temperatures far beyond the string scale is only as fast as in a 2d quantum field theory. Thus, as was conjectured with only limited intuition as far back as 1988 by Atick and Witten for closed string theories , there is a dramatic reduction in the growth of the free energy at high temperatures. More recently , Polchinski has shown that the $`T^2`$ growth in the free energy at high temperatures is a direct consequence of the thermal self-duality of the vacuum functional in the closed bosonic string theory. We will show in this paper that Polchinskiโs observation extends to both the heterotic closed, and type I unoriented open and closed string ensembles. In either case, the $`T^2`$ growth in the string free energy at high temperatures follows as a direct consequence of the Euclidean T-duality transformations that link the thermal ground states of the supersymmetric string theories in pairs. Furthermore, we will find that the full high temperature expansion in powers of $`\beta `$ for the string free energy can be obtained explicitly upon term-by-term evaluation of the one-loop modular integrals in the string mass level expansion.
Our starting point in this paper is the generating functional of connected one-loop vacuum string graphs, $`W(\beta )`$ $``$ $`\mathrm{ln}`$ $`\mathrm{Z}(\beta )`$, where $`\mathrm{Z}(\beta )`$ is the canonical partition function. $`W(\beta )`$ is derived from first principles in the Polyakov path integral formalism following .The free energy, $`F(\beta )`$, and vacuum energy density, $`\rho (\beta )`$, can be directly inferred from $`W(\beta )`$. Let us recall the basic thermodynamic identities of the canonical ensemble :
$$F=W/\beta =V\rho ,P=\left(\frac{F}{V}\right)_T,U=T^2\left(\frac{W}{T}\right)_V,S=\left(\frac{F}{T}\right)_V,C_V=T\left(\frac{S}{T}\right)_V.$$
(1)
Note that $`W(\beta )`$ is an intensive thermodynamic variable without explicit dependence on the spatial volume. $`F`$ is the Helmholtz free energy of the ensemble of strings, $`U`$ is the internal energy, and $`\rho `$ is the finite temperature effective potential, or vacuum energy density, at finite temperature. $`S`$ and $`C_V`$ are, respectively, the entropy and specific heat of the canonical ensemble. The pressure of the string ensemble simply equals the negative of the vacuum energy density, as is true for a cosmological constant, just as in an ideal fluid with negative pressure . The enthalpy, $`H`$$`=`$$`U`$$`+`$$`PV`$, the Helmholtz free energy, $`F`$$`=`$$`U`$$`TS`$, and the Gibbs function, also known as the Gibbs free energy, is $`G`$$`=`$$`U`$$``$$`TS`$$`+`$$`PV`$. As a result of these relations, all of the thermodynamic potentials of the string ensemble have been give a simple, first-principles, formulation in terms of the Polyakov path integral. Notice, in particular, that since $`P`$$`=`$$`\rho `$, the one-loop contribution to the Gibbs free energy of the string ensemble vanishes identically!
We should address the expected Jeans instability of a gravitating statistical ensemble , and the basic definition of the thermodynamic limit in the presence of gravity. Consider an ensemble of total mass, $`M`$, and Schwarschild radius, $`R_S`$, with one-loop vacuum energy density, $`\rho `$ $`=`$ $`F(\beta )/R^{D1}`$. Recall that the Newtonian gravitational coupling, $`G_N`$$``$$`g_s^2`$, where $`g_s`$ is the closed string coupling. Since $`M`$ $``$ $`\rho R_S^{D1}`$, we have the relation:
$$R_S>>\left(\frac{1}{g_s^2\rho (\beta )}\right)^{1/2}.$$
(2)
In other words, strictly speaking, it is only possible to take the infinite volume limit $`V`$ $``$ $`\mathrm{}`$, $`\alpha ^{}`$ $``$ $`0`$, when the energy density per unit volume of the gravitating ensemble happens to be zero, or in the limit of weak string coupling .
## 2 The Impossibility of a Type II Canonical Ensemble
We will begin by explaining why there can be no equilibrium type II superstring ensemble in the absence of Dbranes or background fluxes, either of which introduces Yang-Mills gauge fields in the low energy field theory limit of the finite temperature superstring theory.
As explained at the outset in this paper, an equilibrium ensemble of type II strings requires that we identify a tachyon-free nonsupersymmetric background at all temperatures starting from zero. Let us examine some consequences of finding such a solution. Our interest is in a one-parameter family of nonsupersymmetric backgrounds of the type II superstring with Euclidean target spacetime $`R^9`$$`\times `$$`S^1`$; a line of fixed points of the worldsheet RG parametrized by the single parameter, $`T`$, the inverse of the circumference of the $`S^1`$, whose physical interpretation is temperature. Continuity of the vacuum functional as a function of the parameter $`T`$ requires that, at least for small values of $`T`$, $`W(T)`$ take the form of an integral over the modular group of genus one Riemann surfaces with a modular invariant integrand. In addition, the leading terms in the closed string mass level expansion for small $`T`$ must have a self-consistent interpretation in terms of the field theoretic modes of a 10D supergravity field theory at finite temperature. Thus, the target space supersymmetry of the zero temperature vacuum must be spontaneously broken at all temperatures different from zero. And the existence of a stable gravitating ensemble in thermodynamic equilibrium requires the absence of tachyonic thermal modes in the string mass spectrum. Finally, we must check that the vacuum energy density satistfies the criterion for the absence of the classical Jeans instability in a gravitating ensemble . Thus, we can succinctly state the infrared consistency conditions required of an equilibrium ensemble of type II strings at finite temperatures:
* the absence of thermal tachyons in the mass level expansion.
* the spontaneous breaking of target space supersymmetry at low temperatures, $`T`$ $`>`$ $`0`$.
* the demonstration of a $`T^{10}`$ growth of the free energy at low temperatures, when we isolate the leading contribution to the string mass level expansion from the low energy, field theoretic supergravity modes alone.
We will find that the first two conditions are incompatible: modular invariance of the one-loop type II vacuum amplitude turns out to be extremely restrictive. The spontaneous breaking of supersymmetry in the 10D type II superstring vacuum at temperatures different from zero necessarily implies the existence of a whole slew of low temperature tachyonic momentum modes. This invalidates the possibility of a type II string canonical ensemble, even at low temperatures far below the string scale.
In order to understand the stringent limitations on the expression for the one-loop vacuum amplitude in the finite temperature vacuum imposed by modular invariance, recall that the type II string mass level expansion results from a generic combination of the four, holomorphic and anti-holomorphic, Jacobi theta functions, weighted by, a priori, undetermined phases. In addition, the zero temperature mass formulae in each spin structure sector will now acquire additional contributions from thermal winding and thermal momentum modes. Let us begin by examining the shifts in the ground state energy of the NS-NS vacuum of the type II superstring at finite temperature. Recall that this state is tachyonic in the zero temperature Hilbert space, eliminated from the mass level expansion for either supersymmetric type II superstring theory by a judicious choice of phases in the vacuum amplitude. It will be easy to verify that there is no corresponding choice of modular invariant phases that can eliminate the tachyonic thermal modes of the NS-NS sector, while spontaneously breaking supersymmetry at all temperatures different from zero.
Suppressing the oscillator contributions, the ground state energy in the NS-NS sector for physical states satisfying the level matching constraint takes the form:
$$(\mathrm{mass})_L^2=(\mathrm{mass})_R^2=\frac{4}{\alpha ^{}}\left[\frac{1}{2}+\frac{1}{2}\left(\frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}+\frac{\beta ^2w^2}{4\pi ^2\alpha ^{}}\right)\right].$$
(3)
Notice that the pure momentum and pure winding states, $`(n,0)`$ and $`(0,w)`$, are potential tachyons that enter into the general expression for the level expansion. In the absence of oscillator excitations, each pure momentum mode, $`(\pm n,0)`$, is tachyonic upto some critical temperature, $`T_n^2`$ $`=`$ $`1/2n^2\pi ^2\alpha ^{}`$, beyond which it turns marginal (massless). Conversely, each pure winding mode $`(0,\pm w)`$, turns tachyonic beyond some critical temperature, $`T_w^2`$ $`=`$ $`w^2/8\pi ^2\alpha ^{}`$. Finally, recall that under a thermal duality transformation, the type IIA superstring is mapped to the type IIB superstring:
$$\beta _{\mathrm{IIA}}\beta _{\mathrm{IIB}}=4\pi ^2\alpha ^{}/\beta _{\mathrm{IIA}},(n,w)_{\mathrm{IIA}}(n^{}=w,w^{}=n)_{\mathrm{IIB}},$$
(4)
thus interchanging the identification of momentum and winding modes in the type IIA and type IIB thermal specta. In other words, in the absence of a nontrivial Ramond-Ramond sector, the result for the IIA and IIB one-loop vacuum functionals always coincides. In either case, upon compactification on the circle of radius $`\beta /2\pi `$, the zero mode spectrum takes the form :
$$p_L=\frac{2\pi n}{\beta }+\frac{w\beta }{2\pi \alpha ^{}},p_R=\frac{2\pi n}{\beta }\frac{w\beta }{2\pi \alpha ^{}},$$
(5)
so that the contribution to the path integral from the $`(n,w)`$th sector is:
$$\mathrm{exp}\left[\pi \tau _2\left(\frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}+\frac{w^2\beta ^2}{4\pi ^2\alpha ^{}}\right)+2\pi inw\tau _1\right].$$
(6)
Thus, upon compactifying either type II superstring on $`R^9`$$`\times `$$`S^1`$, the one-loop vacuum amplitude takes the form:
$`W_{\mathrm{II}}(\beta )`$ $`=\beta L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{}}{\displaystyle \frac{d^2\tau }{4\tau _2^2}}(\tau _2)^{8/2}[\eta (\tau )\overline{\eta }(\overline{\tau })]^{8/2}`$ (9)
$`\times {\displaystyle \frac{1}{4}}\left[({\displaystyle \frac{\mathrm{\Theta }_{00}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_{01}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_{10}}{\eta }})^4\pm ({\displaystyle \frac{\mathrm{\Theta }_{11}}{\eta }})^4\right]\left[({\displaystyle \frac{\overline{\mathrm{\Theta }}_{00}}{\overline{\eta }}})^4({\displaystyle \frac{\overline{\mathrm{\Theta }}_{01}}{\overline{\eta }}})^4({\displaystyle \frac{\overline{\mathrm{\Theta }}_{10}}{\overline{\eta }}})^4\pm ({\displaystyle \frac{\overline{\mathrm{\Theta }}_{11}}{\overline{\eta }}})^4\right]`$
$`\times {\displaystyle \underset{n,w=\mathrm{}}{\overset{\mathrm{}}{}}}\mathrm{exp}[\pi \tau _2({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}+{\displaystyle \frac{w^2\beta ^2}{4\pi ^2\alpha ^{}}})+2\pi inw\tau _1],`$
Notice that the familiar choice of phase that projects out the tachyonic NS-NS vacuum at zero temperature, namely, the relative minus sign between (00) and (01) spin structure sectors, also suffices to eliminate all of the tachyonic momentum and winding modes from contributing to the string mass level expansion. This is the precise type II analog of the thermal tachyon spectrum of the closed bosonic string ensemble , except that they have been rendered unphysical due to the choice of phase. In addition, we find that modular invariance has simultaneously forced the particular combination of worldsheet fermionic spin structures displayed in the expression above. Thus, target spacetime supersymmetry is unbroken! In other words, there is no viable type II string canonical ensemble. We need to introduce a new feature into the one-loop string vacuum amplitude that can loosen up the stringent constraints on the combination of fermionic spin structures arising from modular invariance. This new feature turns out to be the introduction of Yang-Mills gauge fields and, consequently, the possibility of spontaneous supersymmetry breaking via a temperature dependent Wilson line .
## 3 Free Energy of the Heterotic Canonical Ensemble
Unlike the type II superstrings where the constraints from modular invariance proved far too restrictive to permit a tachyon-free thermal ground state, the heterotic string theory has a Yang-Mills sector. This introduces the possibility of a Wilson line gauge background, loosening the constraints from modular invariance while enabling a tachyon-free thermal vacuum and, consequently, a self-consistent formulation of the equilibrium canonical ensemble. Consider the ten-dimensional supersymmetric $`E_8`$$`\times `$$`E_8`$ theory at zero temperature. The $`\alpha ^{}`$$``$$`0`$ low energy field theory limit is 10D $`N`$$`=`$$`1`$ supergravity coupled to $`E_8`$$`\times `$$`E_8`$ Yang-Mills gauge fields. What happens to the supersymmetric ground state of this theory at finite temperature? We wish to derive an analogue of the zero temperature vacuum functional which describes a stable finite temperature ground state: thermal tachyons should be absent, target space supersymmetry must be spontaneously broken at finite temperature, and the free energy must grow as $`T^{10}`$ at low temperatures when only massless field theory modes are excited. Most importantly, since we wish to preserve the finiteness and perturbative renormalizability of the zero temperature ground state at finite temperature, it is important to preserve the invariance of the one-loop vacuum functional under the modular group of the torus. Finally, we must require self-consistency with the thermal duality transformations: the $`E_8`$$`\times `$$`E_8`$ and $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string theories are related by the Euclidean timelike T-duality transformation:
$$\beta _{\mathrm{E}_8\times \mathrm{E}_8}\beta _{\mathrm{SO}(32)}=4\pi ^2\alpha ^{}/\beta _{\mathrm{E}_8\times \mathrm{E}_8},(n,w)_{\mathrm{E}_8\times \mathrm{E}_8}(n^{}=w,w^{}=n)_{\mathrm{SO}(32)},$$
(10)
thus interchanging the identification of momentum and winding modes in the $`E_8`$$`\times `$$`E_8`$ and $`SO(32)`$ thermal spectra. Thus, the finite temperature vacuum functional is required to interpolate between the following two spacetime supersymmetric limits: in the $`\beta `$$``$$`\mathrm{}`$ limit we recover the vacuum functional of the supersymmetric $`E_8`$$`\times `$$`E_8`$ heterotic string, while in the $`\beta `$$`=`$$`0`$ limit we must recover, instead, the vacuum functional of the supersymmetric $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string.
It is helpful to begin by examining the resulting shifts in the ground state energy of the (NS$`+`$,NS$`+`$) vacuum due to the winding and momentum modes in the mass spectrum obtained upon compactification on the circle of radius $`\beta /2\pi `$. Suppressing the oscillator contributions, the ground state energy in this sector for physical states satisfying the level matching constraint takes the form:
$$(\mathrm{mass})_L^2=(\mathrm{mass})_R^2=\frac{4}{\alpha ^{}}\left[1+\frac{1}{2}๐ค_L^2+\frac{1}{2}\left(\frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}+\frac{\beta ^2w^2}{4\pi ^2\alpha ^{}}\right)\right]=\frac{4}{\alpha ^{}}\left[\frac{1}{2}+\frac{1}{2}\left(\frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}+\frac{\beta ^2w^2}{4\pi ^2\alpha ^{}}\right)\right],$$
(11)
where $`๐ค_L`$ belongs in the lattice $`E_8`$$`\times `$$`E_8^{}`$, and either $`E_8`$ lattice is spanned by the vectors:
$`(n_1,\mathrm{},n_8),(n_1+\frac{1}{2},\mathrm{},n_8+\frac{1}{2}),\mathrm{where}\mathrm{all}\mathrm{n}_\mathrm{i}\mathrm{Z},\mathrm{and}{\displaystyle \underset{\mathrm{i}=1}{\overset{8}{}}}\mathrm{n}_\mathrm{i}2\mathrm{Z}.`$ (12)
The corresponding result for the $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ lattice takes the form:
$`(n_1,\mathrm{},n_{16}),(n_1+\frac{1}{2},\mathrm{},n_{16}+\frac{1}{2}),\mathrm{where}\mathrm{all}\mathrm{n}_\mathrm{i}\mathrm{Z},\mathrm{and}{\displaystyle \underset{\mathrm{i}=1}{\overset{16}{}}}\mathrm{n}_\mathrm{i}2\mathrm{Z}.`$ (13)
As for the type II superstrings, the pure momentum and pure winding states, $`(n,0)`$ and $`(0,w)`$, are potential tachyons that enter into the general expression for the level expansion. In the absence of oscillator excitations, each pure momentum mode, $`(\pm n,0)`$, is tachyonic upto some critical temperature, $`T_n^2`$ $`=`$ $`1/2n^2\pi ^2\alpha ^{}`$, beyond which it turns marginal (massless). Conversely, each pure winding mode $`(0,\pm w)`$, turns tachyonic beyond some critical temperature, $`T_w^2`$ $`=`$ $`w^2/8\pi ^2\alpha ^{}`$. Notice that retaining the relative minus sign between the contributions from (00) and (01) sectors of the right-moving superconformal field theory in the supersymmetric vacuum functional, as in the previous section, would eliminate both $`O((q\overline{q})^{1/2})`$ contributions to the level expansion, in addition to all of the $`\beta `$ dependent tachyonic thermal modes. Thus, for example, all of the potential tachyons in the (NS$`+`$,NS$`+`$) sector of the circle compactified supersymmetric ground state are eliminated from the level expansion for physical states in one fell swoop. But can this mechanism work in the absence of spacetime supersymmetry? We will need to identify a different chiral modular invariant, one that spontaneously breaks spacetime supersymmetry at finite temperature. The key lies in the introduction of a temperature dependent Wilson line background .
### 3.1 Axial Gauge and the Euclidean Timelike Wilson Line
The generating functional of connected one-loop vacuum string graphs in the stable finite temperature vacuum will be given by an expression of the form:
$$W_{\mathrm{het}}(\beta )=\beta L^9(4\pi ^2\alpha ^{})^5_{}\frac{d^2\tau }{4\tau _2^2}\tau _2^4Z_{\mathrm{het}}(\beta ),$$
(14)
where the function $`Z_{\mathrm{het}}(\beta )`$ denotes the desired level expansion for the thermal mass spectrum of the heterotic ensemble at generic temperatures that can satisfy the interpolations to the zero temperature limit required by the thermal duality transformations. In particular, we wish to identify a suitable interpolating expression for $`W_{\mathrm{het}}(\beta )`$ which satisfies the infrared consistency conditions at generic values of $`\beta `$, matching smoothly with the known vacuum functional of the supersymmetric $`E_8`$$`\times `$$`E_8`$ string theory at zero temperature ($`\beta `$$`=`$$`\mathrm{}`$):
$`W(\beta )|_{\mathrm{T}=0}=`$ $`\beta L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{}}\left\{{\displaystyle \frac{d^2\tau }{4\tau _2^2}}(\tau _2)^4[\eta (\tau )\overline{\eta }(\overline{\tau })]^8\right\}`$ (17)
$`\times \left[\{({\displaystyle \frac{\mathrm{\Theta }_{00}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_{01}}{\eta }})^4\}\{({\displaystyle \frac{\mathrm{\Theta }_{10}}{\eta }})^4\pm ({\displaystyle \frac{\mathrm{\Theta }_{11}}{\eta }})^4\}\right]`$
$`\times {\displaystyle \frac{1}{4}}\left[\left({\displaystyle \frac{\mathrm{\Theta }_{00}}{\eta }}\right)^8+\left({\displaystyle \frac{\mathrm{\Theta }_{01}}{\eta }}\right)^8+\left({\displaystyle \frac{\mathrm{\Theta }_{10}}{\eta }}\right)^8+\left({\displaystyle \frac{\mathrm{\Theta }_{11}}{\eta }}\right)^8\right]^2,`$
while also recovering the vacuum functional of the T-dual supersymmetric $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string in the presence of the T-dual Wilson line background:
$`W(\beta )|_{\mathrm{T}^{}=0}=`$ $`\beta L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{}}\left\{{\displaystyle \frac{d^2\tau }{4\tau _2^2}}(\tau _2)^4[\eta (\tau )\overline{\eta }(\overline{\tau })]^8\right\}`$ (20)
$`\times \left[\{({\displaystyle \frac{\mathrm{\Theta }_{00}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_{01}}{\eta }})^4\}\{({\displaystyle \frac{\mathrm{\Theta }_{10}}{\eta }})^4\pm ({\displaystyle \frac{\mathrm{\Theta }_{11}}{\eta }})^4\}\right]`$
$`\times {\displaystyle \frac{1}{4}}\left[\left({\displaystyle \frac{\mathrm{\Theta }_{00}}{\eta }}\right)^{16}+\left({\displaystyle \frac{\mathrm{\Theta }_{01}}{\eta }}\right)^{16}+\left({\displaystyle \frac{\mathrm{\Theta }_{10}}{\eta }}\right)^{16}+\left({\displaystyle \frac{\mathrm{\Theta }_{11}}{\eta }}\right)^{16}\right].`$
It turns out that the desired expression for $`\mathrm{Z}_{\mathrm{het}}(\beta )`$, and the pair of T-dual Wilson line backgrounds necessary for this interpolation, can be inferred using extant results in the heterotic string literature.
The argument proceeds as follows. The modular invariant possibilities for the sum over spin structures in the 10d heterotic string have been classified, both by free fermion and by orbifold techniques , and there is a unique nonsupersymmetric and tachyon-free ground state with gauge symmetry $`SO(16)`$$`\times `$$`SO(16)`$. In the fermionic formulation, the GSO projection differs from that of the $`E_8`$$`\times `$$`E_8`$ heterotic string as follows:
$$e^{\pi \stackrel{~}{F}+\alpha +\alpha ^{}}=e^{\pi iF+\alpha ^{}+\stackrel{~}{\alpha }}=e^{\pi i\stackrel{~}{F^{}}+\stackrel{~}{\alpha }+\alpha }=+1,$$
(21)
where $`\alpha `$, $`\alpha ^{}`$, and $`\stackrel{~}{\alpha }`$, equal, $`+1`$, or $`0`$, respectively, in the Ramond, or Neveu-Schwarz, sector for each of the three 8-fermion blocks: right-moving, left-moving gauge, and left-moving gauge . The โall-positiveโ choice of worldsheet chiralities in every sector of the supersymmetric $`E_8`$$`\times `$$`E_8`$ theory are therefore altered to the more complicated, and nonsupersymmetric, projection:
$`(\mathrm{NS}+,\mathrm{NS}+,\mathrm{NS}+),(\mathrm{NS},\mathrm{NS},\mathrm{R}+),(\mathrm{NS},\mathrm{R}+,\mathrm{NS}),(\mathrm{R}+,\mathrm{NS},\mathrm{NS}),`$ (22)
$`(\mathrm{NS}+,\mathrm{R},\mathrm{R}),(\mathrm{R},\mathrm{NS}+,\mathrm{R}),(\mathrm{R},\mathrm{R},\mathrm{NS}+),(\mathrm{R}+,\mathrm{R}+,\mathrm{R}+).`$ (23)
This choice of twisted boundary conditions introduces discrete torsion in the, alternative, orbifold description of this theory , raising the ground state energy in the sectors that previously provided the massless chiral fermions of the supersymmetric heterotic string. Most illuminating, perhaps, is to understand this theory as the result of taking the noncompact limit of either circle-compactified $`E_8`$$`\times `$$`E_8`$ or $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string theories, upon introduction of a Wilson line background that spontaneously breaks both spacetime supersymmetry, as well as breaking the nonabelian gauge symmetry to $`SO(16)`$$`\times `$$`SO(16)`$. This interpretation is clarified by invoking the interpolating vacuum functional approach .
Begin with the one-loop vacuum functional of the circle compactified nonsupersymmetric and tachyon free $`SO(16)`$$`\times `$$`SO(16)`$ heterotic string theory , where the $`S^1`$ is a circle of radius $`\beta /2\pi `$:
$`W_{\mathrm{SO}(16)\times \mathrm{SO}(16)}(\beta )=`$ $`\beta L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{}}\left\{{\displaystyle \frac{d^2\tau }{4\tau _2^2}}(\tau _2)^4[\eta (\tau )\overline{\eta }(\overline{\tau })]^8\right\}`$ (26)
$`\times {\displaystyle \frac{1}{4}}\left[({\displaystyle \frac{\mathrm{\Theta }_2}{\eta }})^8({\displaystyle \frac{\mathrm{\Theta }_4}{\eta }})^8({\displaystyle \frac{\overline{\mathrm{\Theta }_3}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_2}{\eta }})^8({\displaystyle \frac{\mathrm{\Theta }_3}{\eta }})^8({\displaystyle \frac{\overline{\mathrm{\Theta }_4}}{\eta }})^4({\displaystyle \frac{\mathrm{\Theta }_3}{\eta }})^8({\displaystyle \frac{\mathrm{\Theta }_4}{\eta }})^8({\displaystyle \frac{\overline{\mathrm{\Theta }_2}}{\eta }})^4\right]`$
$`\times {\displaystyle \underset{n,w=\mathrm{}}{\overset{\mathrm{}}{}}}\mathrm{exp}[\pi \tau _2({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}+{\displaystyle \frac{w^2\beta ^2}{4\pi ^2\alpha ^{}}})+2\pi inw\tau _1].`$
It is easy to verify by inspection that the integrand in this expression is invariant under the one-loop modular group. The expression also satisfies some of the infrared consistency conditions for a viable thermal ground state outlined in section 2: there are no thermal tachyons. However, the noncompact $`\beta `$ $``$ $`0`$ limit of this expression does not lead to a spontaneous restoration of target spacetime supersymmetry since we recover, instead, the nonsupersymmetric $`O(16)`$$`\times `$$`O(16)`$ 10D modular invariant. We must incorporated a temperature dependent Wilson line background in order to interpolate between 10D supersymmetric and 9D nonsupersymmetric ground states.
The spontaneous restoration of spacetime supersymmetry in the noncompact limit within the larger family of one-parameter continuously connected 9D $`SO(16)`$$`\times `$$`SO(16)`$ heterotic string vacua with nontrivial Wilson line gauge backgrounds was discovered in independent numerical investigations by Itoyama and Taylor and Ginsparg and Vafa . The plots in the latter paper indicate clearly the distinction between taking the noncompact limit of the expression in Eq. (26): where the vacuum energy density goes thru a minimum at the self-dual point without touching zero, and taking the noncompact limit in the presence of a Wilson line background. It is in the latter case that the target spacetime supersymmetry is spontaneously restored at $`T`$ $`=`$ $`0`$. Additional massless gauge bosons appear as we approach the $`T`$ $`=`$ $`0`$ limit, such that we recover the full $`SO(32)`$ gauge symmetry . Recall that the gauge lattice vector $`๐ค_L`$ in the equivalent worldsheet bosonic description of $`W_{\mathrm{het}}(\beta )`$ will belong in a $`(1,17)`$-component self-dual lattice of Lorentzian signature. Generic points in the lattice can be reached by Lorentz boosts that preserve the Lorentzian self-duality property, a pre-requisite for modular invariance, and each describes a heterotic string vacuum that can be reached by continuous interpolation from the theory above. The precise form of a generic vector in the self-dual lattice $`\mathrm{\Gamma }^{(1,17)}`$ is given by :
$$(p_R^0,p_L^0,๐ค_L)(p_R^0,p_L^0,๐ค_L^{})=(p_R^0๐ค๐^0\frac{w^0\beta }{4\pi }๐_0๐^0,p_L^0๐ค๐^0\frac{w^0\beta }{4\pi }๐_0๐^0,๐ค_L+w^0\left(\frac{\beta }{2\pi }\right)๐_0),$$
(27)
where:
$$p_L^0=\frac{2\pi n}{\beta }+\frac{w\beta }{2\pi \alpha ^{}},p_R^0=\frac{2\pi n}{\beta }\frac{w\beta }{2\pi \alpha ^{}},n,w\mathrm{Z}^+.$$
(28)
In particular, it is illuminating to consider the two lattice boosts that, respectively, interpolate between the nonsupersymmetric and tachyon-free 9D $`SO(16)`$$`\times `$$`SO(16)`$ vacuum functional and the pair of 10D $`E_8`$$`\times `$$`E_8`$, and $`SO(32)`$, supersymmetric string limits. The result for the required lattice-boosts is due to Ginsparg . Starting with the supersymmetric 10D $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ string, compactify on a circle of radius $`\beta /2\pi `$ with Wilson line gauge background: $`\beta ๐^0`$ $`=`$ $`2\pi (1^8;0^8)`$. The closed form expression for the interpolating one-loop vacuum functional describing the exactly marginal flow to the 10D $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string vacuum can be found in the paper of Itoyama and Taylor <sup>3</sup><sup>3</sup>3The vacuum functional of interest to us has been called the โTwist IIโ model by the authors of , as is clear from the plot in Figure 1 of their paper. The T-duality transform of the expression in Eq. (13) of Ref. \[\] gives the desired expression for $`\mathrm{Z}_{\mathrm{het}}(\beta )`$; a typo in the last line of Eq. (13) has been corrected. The flow to the 10D $`E_8`$$`\times `$$`E_8`$ vacuum, instead, would be described by the interpolation with the T-dual Wilson line background, as in .:
$`Z_{\mathrm{het}}(\beta )=`$ $`{\displaystyle \frac{1}{8}}[\eta (\tau )\overline{\eta }(\overline{\tau })]^8[\eta (\tau )]^{16}[\overline{\eta }(\overline{\tau })]^4\times \{(_0+_{1/2})[\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_4^8\mathrm{\Theta }_2^8\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8]`$ (35)
$`+\left(๐ช_0๐ช_{1/2}\right)\left[\frac{1}{2}\overline{\mathrm{\Theta }}_3^4\left(\mathrm{\Theta }_2^{16}+\mathrm{\Theta }_3^{16}+\mathrm{\Theta }_4^{16}\right)\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\right]`$
$`+\left(_0_{1/2}\right)\left[\frac{1}{2}\overline{\mathrm{\Theta }}_2^4\left(\mathrm{\Theta }_2^{16}+\mathrm{\Theta }_3^{16}+\mathrm{\Theta }_4^{16}\right)\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_4^8\mathrm{\Theta }_2^8+\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\right]`$
$`+(๐ช_0+๐ช_{1/2})[\frac{1}{2}\overline{\mathrm{\Theta }}_4^4(\mathrm{\Theta }_2^{16}+\mathrm{\Theta }_3^{16}+\mathrm{\Theta }_4^{16})+\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_2^8\mathrm{\Theta }_4^8]\}.`$
As explained above, we use the T-dual of the standard basis for thermal mode summations introduced in , in terms of which the one-loop modular transformations are especially easy to verify:
$$_0:n\mathrm{Z},w\mathrm{even}._{1/2}:n\mathrm{Z}+1/2,w\mathrm{even}.๐ช_0:n\mathrm{Z},w\mathrm{odd}.๐ช_{1/2}:n\mathrm{Z}+1/2,w\mathrm{odd}.$$
(37)
A $`\tau `$$``$$`1/\tau `$ transformation leaves $`(_0`$$`+`$$`_{1/2})`$ and $`(๐ช_0`$$``$$`๐ช_{1/2})`$ invariant, while interchanging $`(_0`$$``$$`_{1/2})`$ and $`(๐ช_0`$$`+`$$`๐ช_{1/2})`$. Under $`\tau `$$``$$`\tau `$$`+`$$`1`$, the first three summations are invariant, while $`๐ช_{1/2}`$ maps to the negative of itself. It is easy to verify the invariance of this expression under a $`\tau `$ $``$ $`1/\tau `$ transformation. Under a $`\tau `$ $``$ $`\tau +1`$ transformation, the first and third lines within curly brackets transform with a minus sign, while the second and fourth are interchanged with a minus sign. The overall minus sign is accounted for by the transformation of the eta functions. As before, $`w`$ and $`n`$ denote thermal windings, and thermal momenta, respectively. Notice the appearance of both half-integer and integer Matsubara frequencies along this flow. Notice, also, that because of the inclusion of a Wilson line background, the interpolating functional describing the flow to the T-dual $`E_8`$$`\times `$$`E_8`$ vacuum cannot be inferred directly from the expression above, because the T-duality acts nontrivially on the gauge field, in addition to interchanging momenta and windings. We must start afresh with the 10D $`E_8`$$`\times `$$`E_8`$ vacuum functional, compactifying on a circle of radius $`\beta `$ with temperature dependent Wilson line background as before. The interpolation takes the form:
$`Z_{\mathrm{het}}(\beta )=`$ $`{\displaystyle \frac{1}{8}}[\eta (\tau )\overline{\eta }(\overline{\tau })]^8[\eta (\tau )]^{16}[\overline{\eta }(\overline{\tau })]^4\times \{(_0+_{1/2})[\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_4^8\mathrm{\Theta }_2^8\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8]`$ (44)
$`+\left(๐ช_0๐ช_{1/2}\right)\left[\frac{1}{2}\overline{\mathrm{\Theta }}_3^4\left(\mathrm{\Theta }_2^8+\mathrm{\Theta }_3^8+\mathrm{\Theta }_4^8\right)^2\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\right]`$
$`+\left(_0_{1/2}\right)\left[\frac{1}{2}\overline{\mathrm{\Theta }}_2^4\left(\mathrm{\Theta }_2^8+\mathrm{\Theta }_3^8+\mathrm{\Theta }_4^8\right)^2\overline{\mathrm{\Theta }}_4^4\mathrm{\Theta }_4^8\mathrm{\Theta }_2^8+\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_2^8\mathrm{\Theta }_3^8\right]`$
$`+(๐ช_0+๐ช_{1/2})[\frac{1}{2}\overline{\mathrm{\Theta }}_4^4(\mathrm{\Theta }_2^8+\mathrm{\Theta }_3^8+\mathrm{\Theta }_4^8)^2+\overline{\mathrm{\Theta }}_3^4\mathrm{\Theta }_3^8\mathrm{\Theta }_4^8\overline{\mathrm{\Theta }}_2^4\mathrm{\Theta }_2^8\mathrm{\Theta }_4^8]\}.`$
It should be noted that in the presence of the timelike Wilson line backgrounds, the Yang-Mills gauge group is no longer a full $`SO(32)`$, nor a full $`E_8`$$`\times `$$`E_8`$, except in their respective noncompact limits: $`\beta `$ $``$ $`0`$. At generic values of $`\beta `$, we have the gauge group $`SO(16)`$$`\times `$$`SO(16)`$$`\times `$$`U(1)`$. Enhanced gauge symmetries of rank $`17`$ will appear as we vary $`(\beta ,A_0(\beta ))`$, subject to the Lorentzian (17,1) self-duality conditions that ensure modular invariance. This is the usual phenomenon of Wilson line gauge symmetry breaking, a hallmark of heterotic string phenomenology. From the perspective of the low energy Yang-Mills gauge theory at finite temperature, the usual axial gauge quantization, $`๐_0`$$`=`$$`0`$, has been replaced by a modified axial gauge quantization, which also spontaneously breaks the nonabelian gauge symmetry: $`\beta ๐_0`$$`=`$$`2\pi (1^8;0^8)`$ . We emphasize that the necessity for a Wilson line background in the finite temperature quantization arose as a consequence of our insistence on preserving the finiteness, and perturbative renormalizability, of the zero temperature ground state at finite temperatures.
Let us derive from our expression for the string free energy the growth with temperature in the low energy field theory limit, namely, large $`\beta `$. It will suffice to carry out this check in a single sector of the thermal spectrum, for instance, $`๐ช_0`$, with unit winding, but summing over all integer Matsubara modes, $`n`$ $``$ $`\mathrm{Z}`$. We emphasize that modular invariance is not at issue here, since our interest is only in a field theoretic check of the scaling behavior as a function of $`T`$ in the leading term of the string level expansion. Expanding the Jacobi theta functions in powers of $`q\overline{q}`$ $`=`$ $`e^{4\pi \tau _2}`$, and extracting the coefficients of the $`O(1)`$ term in the level expansion of the zero thermal winding number sector of $`๐ช_0`$, gives the following leading contribution from massless target spacetime bosons:
$`F(\beta )|_{(\mathrm{w}=1;\mathrm{n}\mathrm{Z})}=`$ $`22^8L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle _{\sqrt{1\tau _1^2}}^{\mathrm{}}}{\displaystyle \frac{d\tau _2}{4\tau _2^2}}(\tau _2)^4`$ (47)
$`\times {\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\mathrm{exp}[\pi \tau _2({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}+{\displaystyle \frac{\beta ^2}{4\pi ^2\alpha ^{}}})+2\pi in\tau _1]`$
$`=`$ $`2^9L^9T^{10}{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\left(1+{\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^4}}\right)^5{\displaystyle _{1/2}^{1/2}}๐\tau _1\mathrm{exp}\left[2\pi in\tau _1\right]\mathrm{\Gamma }(5,\sqrt{1\tau _1^2}).`$ (48)
The leading coefficient of the product of eta functions is $`(\overline{q}^{1/2}q)`$ in heterotic string theory, and we isolate the compensating powes in the $`q`$ expansion. Notice that all of the massless target space fermions of the zero temperature vacuum have acquired a tree-level mass that is linear in temperature, another self-consistency check for the low energy finite temperature gauge theory. The degeneracy at the massless level accounts for both the 64 bosonic states in the spin 2 gravity multiplet, as well as the 8.496 bosonic states in the spin 1 Yang-Mills multiplet. As an aside, as mentioned before in our discussion of the type II string level expansion in section 2, each degree of freedom in the spin 1 multiplet contributes with half as much weight as those in the spin 2 multiplet. The reason is that the level expansion counts target space degeneracies in terms of equivalent 2d free bosonic oscillator modes. The leading $`T^{10}`$ dependence in the free energy arises from the Matsubara spectrum; the subleading temperature dependence comes from the background of stringy winding modes.
### 3.2 Thermal Duality and High Temperature Scaling Behavior
The $`T^2`$ high temperature scaling behavior of the free energy can be inferred more elegantly by application of the T-duality transformation linking the thermal ground states of the two heterotic string theories. That argument proceeds as follows. In the closed bosonic string theory, the generating functional for connected one-loop vacuum string graphs is invariant under the thermal self-duality transformation: $`W(T)`$ $`=`$ $`W(T_c^2/T)`$, at the string scale, $`T_c`$ $`=`$ $`1/2\pi \alpha ^{1/2}`$. It was pointed out by Polchinski that we can infer the following thermal duality relation which holds for both the Helmholtz free energy, $`F(T)`$ $`=`$ $`TW(T)`$, and the vacuum energy density, $`\rho (T)`$ $`=`$ $`TW(T)/V`$ of the closed bosonic string ensemble:
$$F(T)=\frac{T^2}{T_C^2}F(\frac{T_C^2}{T}),\rho (T)=\frac{T^2}{T_C^2}\rho (\frac{T_C^2}{T}).$$
(49)
In the case of the heterotic string, we will show that the thermal duality relation instead relates, respectively, the free energies of the $`E_8`$$`\times `$$`E_8`$ and $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ theories. Since we deal with a supersymmetric string theory, it is convenient to restrict ourselves to the contributions to the vacuum energy density from target space bosonic degrees of freedom alone:
$$F(T)_{E_8\times E_8}=\frac{T^2}{T_C^2}F(\frac{T_C^2}{T})_{\mathrm{Spin}(32)/\mathrm{Z}_2},\rho (T)_{E_8\times E_8}=\frac{T^2}{T_C^2}\rho (\frac{T_C^2}{T})_{\mathrm{Spin}(32)/\mathrm{Z}_2}.$$
(50)
Consider the high temperature limit of this expression:
$$\underset{T\mathrm{}}{lim}\rho (T)_{E_8\times E_8}=\underset{T\mathrm{}}{lim}\frac{T^2}{T_C^2}\rho (\frac{T_C^2}{T})_{\mathrm{Spin}(32)/\mathrm{Z}_2}=\underset{(T_C^2/T)0}{lim}\frac{T^2}{T_C^2}\rho (\frac{T_C^2}{T})_{\mathrm{Spin}(32)/\mathrm{Z}_2}=\frac{T^2}{T_C^2}\rho (0)_{\mathrm{Spin}(32)/\mathrm{Z}_2},$$
(51)
where $`\rho (0)`$ is the contribution to the cosmological constant, or vacuum energy density, at zero temperature from target space bosonic degrees of freedom alone. Note that it is finite. Thus, at high temperatures, the contribution to the free energy of either heterotic ensemble from target space bosonic degrees of freedom alone grows only as fast as $`T^2`$. In other words, the growth in the number of target spacetime bosonic degrees of freedom at high temperature in the heterotic string ensemble is only as fast as in a two-dimensional field theory. This is significantly slower than the $`T^{10}`$ growth of the high temperature degrees of freedom expected in the ten-dimensional low energy field theory. Notice that the thermal duality transformation interchanges the thermal winding and thermal momentum modes of the corresponding heterotic ensembles. This is the reason for the simple $`T^2`$ scaling behavior found in this duality relation.
Thus, at high temperatures far above the string scale, the leading behavior is a significantly slower $`T^2`$ growth in the closed superstring free energy. Such a $`T^2`$ growth was first conjectured by Atick and Witten in 1989 . Subsequently, it was shown to hold for the free energy of the closed bosonic string ensemble by Polchinski as a direct consequence of its thermal self-duality property.
### 3.3 High Temperature Expansion of Free Energy
As a final bit of closed string methodology, we will now show how the modular integrals appearing in the result for the one-loop heterotic string vacuum amplitude can be carried out in closed form by the procedure of term-by-term integration. The result will be a power series expansion for the one-loop free energy of the canonical ensemble expressed as an exact function of temperature.
We begin with a Poisson resummation on the infinite sum over winding modes, in order to put our expression for $`F(T)_{\mathrm{het}}`$ in a form suitable for a high temperature expansion valid in the regime $`\beta `$ $``$ $`0`$. Denoting the coefficients in the level expansion by $`b_m^{(\mathrm{het})}`$, the result takes the form:
$`F_{\mathrm{het}}=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle _{\sqrt{1\tau _1^2}}^{\mathrm{}}}{\displaystyle \frac{d\tau _2}{4\tau _2^2}}\tau _2^4{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{w=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}`$ (53)
$`\times b_m^{(\mathrm{het})}\mathrm{exp}\left[4\pi m\tau _2\right]\mathrm{exp}\left[\pi \tau _2\left({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}+{\displaystyle \frac{w^2\beta ^2}{4\pi ^2\alpha ^{}}}\right)+2\pi inw\tau _1\right]`$
$`=`$ $`L^9(4\pi ^2\alpha ^{})^5\beta ^1{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle _{\sqrt{1\tau _1^2}}^{\mathrm{}}}{\displaystyle \frac{d\tau _2}{4\tau _2^2}}\tau _2^4\left({\displaystyle \frac{\tau _2}{4\pi ^2\alpha ^{}}}\right)^{1/2}{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}`$ (55)
$`\times b_m^{(\mathrm{het})}\mathrm{exp}\left[4\pi m\tau _2\right]\mathrm{exp}\left[\pi \tau _2\left({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}\right)\pi (pn\tau _1)^2{\displaystyle \frac{4\pi ^2\alpha ^{}}{\tau _2\beta ^2}}\right].`$
Notice that the three integer-valued infinite summations in the expression for $`F_{\mathrm{het}}(\beta )`$ given here account for, respectively, closed string mass level, thermal windings, and thermal momenta. It is convenient to split the range of integration for the modular variable $`\tau _2`$ as follows:
$$_{\sqrt{1\tau _1^2}}^{\mathrm{}}๐\tau _2_0^{\mathrm{}}๐\tau _2_0^{\sqrt{1\tau _1^2}}๐\tau _2,$$
(56)
where in the second term we must take care to express the integrand in a form that is manifestly bounded within the specified domain of integration. We will find that the former can be recognized as the Bessel function, $`K_\nu (y)`$, with a power series representation in the argument $`y`$:
$`{\displaystyle _0^{\mathrm{}}}๐xx^{\nu 1}e^{\frac{\delta }{x}\gamma x}=2\left({\displaystyle \frac{\delta }{\gamma }}\right)^{\nu /2}K_\nu \left(2\sqrt{\delta \gamma }\right),\mathrm{Re}\delta >0,\mathrm{Re}\gamma >0`$ (57)
$`\mathrm{where}\delta {\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}(pn\tau _1)^2,\gamma 4\pi m+4\pi ^3\alpha ^{}n^2/\beta ^2,\mathrm{and}\nu 11/2.`$ (58)
The latter can be expressed in terms of the standard Whittaker functions, $`W_{\frac{\nu +1}{2},\frac{\nu }{2}}(y)`$, with integral representation:
$`{\displaystyle _0^u}๐xx^{\nu 1}e^{\delta /x}=\delta ^{(\nu 1)/2}u^{(1+\nu )/2}e^{\delta /2u}W_{\frac{\nu +1}{2},\frac{\nu }{2}}\left(\delta /u\right)`$ (59)
$`\mathrm{where}\delta {\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}(pn\tau _1)^2,u\sqrt{1\tau _1^2},\mathrm{and}\nu {\displaystyle \frac{11}{2}}+k,k\mathrm{Z}^+,`$ (60)
following substitution of the Taylor expansion of the function $`\mathrm{exp}[\gamma \tau _2]`$ in the integrand. Note that, with this substitution, we can verify that the integrand is always bounded in the region $`|\tau _2|<1`$, term-by-term in the Taylor expansion. Substituting in the expression for the free energy given in Eq. (55) we obtain the result:
$`F_{\mathrm{het}}=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle _{\sqrt{1\tau _1^2}}^{\mathrm{}}}{\displaystyle \frac{d\tau _2}{4\tau _2^2}}\tau _2^4{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{w=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}`$ (62)
$`\times b_m^{(\mathrm{het})}\mathrm{exp}\left[4\pi m\tau _2\right]\mathrm{exp}\left[\pi \tau _2\left({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}+{\displaystyle \frac{w^2\beta ^2}{4\pi ^2\alpha ^{}}}\right)+2\pi inw\tau _1\right]`$
$`=`$ $`L^9(4\pi ^2\alpha ^{})^5\beta ^1{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle _{\sqrt{1\tau _1^2}}^{\mathrm{}}}{\displaystyle \frac{d\tau _2}{4\tau _2^2}}\tau _2^4\left({\displaystyle \frac{\tau _2}{4\pi ^2\alpha ^{}}}\right)^{1/2}{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}`$ (64)
$`\times b_m^{\mathrm{het}}\mathrm{exp}\left[4\pi m\tau _2\right]\mathrm{exp}\left[\pi \tau _2\left({\displaystyle \frac{4\pi ^2\alpha ^{}n^2}{\beta ^2}}\right)\pi (pn\tau _1)^2{\displaystyle \frac{4\pi ^2\alpha ^{}}{\tau _2\beta ^2}}\right]`$
$`=`$ $`{\displaystyle \frac{1}{4}}L^9(4\pi ^2\alpha ^{})^5\beta ^1{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}b_m^{(\mathrm{het})}(pn\tau _1)^{11/2}`$ (69)
$`\times \left[n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right]^{11/4}K_{11/2}\left({\displaystyle \frac{8\pi ^2\alpha ^{}}{\beta ^2}}(pn\tau _1)\left(n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right)^{1/2}\right)`$
$`{\displaystyle \frac{1}{4}}L^9(4\pi ^2\alpha ^{})^5\beta ^1{\displaystyle _{1/2}^{1/2}}๐\tau _1{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{b_m^{(\mathrm{het})}}{k!}}`$
$`\times \left[{\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}\left(n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right)\right]^k\left[{\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}(pn\tau _1)^2\right]^{(k13/2)/2}(1\tau _1^2)^{(k9/2)/4}`$
$`\times \mathrm{exp}\left[{\displaystyle \frac{2\pi ^3\alpha ^{}}{\beta ^2}}{\displaystyle \frac{(pn\tau _1)^2}{\sqrt{1\tau _1^2}}}\right]W_{(k9/2)/2,(k11/2)/2}\left({\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}{\displaystyle \frac{(pn\tau _1)^2}{\sqrt{1\tau _1^2}}}\right).`$
Expressing, in turn, the Bessel, Whittaker, and exponential functions as convergent power series expansions in their respective arguments, enables all of the $`\tau _1`$ dependence in the integrand to be extracted in the form of an algebraic power series. The term-by-term integration over $`\tau _1`$ can then be carried out explicitly. Note the useful relations $`W_{\lambda ,\mu }(y)`$$`=`$$`W_{\lambda ,\mu }(y)`$, $`K_\nu (y)`$$`=`$$`K_\nu (y)`$. We will make use of the following power series formulae:
$`K_{11/2}(y)=\sqrt{{\displaystyle \frac{\pi }{2y}}}e^y{\displaystyle \underset{r=0}{\overset{5}{}}}{\displaystyle \frac{(5+r)!}{r!(5r)!}}(2y)^r,\mathrm{and}y{\displaystyle \frac{8\pi ^2\alpha ^{}}{\beta ^2}}(pn\tau _1)\left(n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right)^{1/2}.`$ (71)
Notice that, term-by-term, the $`\tau _1`$ dependence in this series is remarkably simple. Also, notice that the second subscript, $`\mu `$, in the Whittaker function is such that $`2\mu `$ is half-integer. We can use a functional relation, and the power series representation:
$`W_{\lambda ,\mu }(z)={\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}\mu \lambda )}}M_{\lambda ,\mu }(z)+{\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}+\mu \lambda )}}M_{\lambda ,\mu }(z)`$ (72)
$`M_{\mu +{\scriptscriptstyle \frac{1}{2}},\mu }(z)=\left({\displaystyle \frac{1}{2\mu }}\right)z^{{\scriptscriptstyle \frac{1}{2}}\mu }e^{{\scriptscriptstyle \frac{1}{2}}z}`$ (73)
$`\mathrm{where}z{\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}{\displaystyle \frac{(pn\tau _1)}{\sqrt{1\tau _1^2}}},\lambda =(k9/2)/2,\mu (k11/2)/2.`$ (74)
Prior to $`\tau _1`$ integration, it is convenient to include the Taylor expansion of the exponential, $`e^z`$, as well as the powers of binomials, $`(pn\tau _1)`$, $`(1\tau _1^2)^{1/2}`$, accompanying the Whittaker function in the expression in Eq. (LABEL:eq:typeIr), defining the coefficients, $`w_{(r,s)}^{(\pm )}(\beta ,\alpha ^{})`$, and numerical subscripts, $`\sigma _\pm (\lambda ,\mu )`$, as follows:
$`(pn\tau _1)^{(k13)/2}(1\tau _1^2)^{(k9/2)/4}e^{{\scriptscriptstyle \frac{1}{2}}z}W_{(k9/2)/2,(k11/2)/2}(z)`$ (75)
$`{\displaystyle \underset{r=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{s=0}{\overset{\mathrm{}}{}}}\tau _1^{r+s+\sigma _\pm }\left[{\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}\mu \lambda )}}w_{(r,s)}^{(+)}+{\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}+\mu \lambda )}}w_{(r,s)}^{()}\right].`$ (76)
A similar expansion holds for the term proportional to $`K_{11/2}(y)`$, with the coefficient $`\kappa (\alpha ^{},\beta )`$ defined by the relation, $`y`$ $``$ $`(pn\tau _1)\kappa `$. The result for $`F_{\mathrm{het}}`$ consequently takes the form:
$`F_{\mathrm{het}}={\displaystyle \frac{1}{4}}L^9(4\pi ^2\alpha ^{})^5{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}b_m^{(\mathrm{het})}\left[n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right]^{11/4}`$ (77)
$`\times {\displaystyle _{1/2}^{1/2}}d\tau _1(pn\tau 1)^{11/2}\left[\sqrt{{\displaystyle \frac{\pi }{2\kappa (pn\tau _1)}}}e^{\kappa (pn\tau _1)}{\displaystyle \underset{r=0}{\overset{5}{}}}{\displaystyle \frac{(5+r)!}{r!(5r)!}}(2\kappa (pn\tau _1))^r\right]`$ (78)
$`{\displaystyle \frac{1}{4}}L^9\beta ^1(4\pi ^2\alpha ^{})^5{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{p=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{b_m^{(\mathrm{het})}}{k!}}\left[n^2+{\displaystyle \frac{m\beta ^2}{\pi ^2\alpha ^{}}}\right]^k\left[{\displaystyle \frac{4\pi ^3\alpha ^{}}{\beta ^2}}\right]^{(3k13/2)/2}`$ (79)
$`\times {\displaystyle _{1/2}^{1/2}}d\tau _1\left\{{\displaystyle \underset{r=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{s=0}{\overset{\mathrm{}}{}}}\tau _1^{r+s+\sigma _\pm }[{\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}\mu \lambda )}}w_{(r,s)}^{(+)}+{\displaystyle \frac{\mathrm{\Gamma }(2\mu )}{\mathrm{\Gamma }(\frac{1}{2}+\mu \lambda )}}w_{(r,s)}^{()}]\right\}.`$ (80)
With these substitutions, all of the $`\tau _1`$ integrals appearing in our expression for $`F_{\mathrm{het}}`$ can be evaluated explicitly. Recall that $`m`$ and $`n`$ denote heterotic string mass level, and thermal momentum number, respectively. The infinite summation over $`p`$ resulted from the Poisson resummation of thermal winding numbers, necessary to cast the expression for the free energy in a form suitable for a high temperature expansion in $`\beta `$. Thus, following the one-loop modular integrations, we obtain an exact high temperature expansion for the one-loop string free energy as a function of $`\beta `$, and the degeneracies, $`b_m^{(\mathrm{het})}`$, in the heterotic string mass level expansion.
Finally, it should be noted that, using the power series representation in $`\beta `$ given above, one could demonstrate the analyticity of infinitely many thermodynamic potentials for the heterotic ensemble. Namely, potentials defined by taking arbitrarily large number of temperature derivatives in the vicinity of the self-dual temperature. This is clear evidence that the thermal duality transition linking the high temperature behavior of one heterotic ensemble to the low temperature behavior of the other ensemble lies within the Kosterlitz-Thouless universality class .
## 4 Free Energy of the Unoriented Type IB Ensemble
The free energy of the unoriented open and closed type IB string ensemble at one-loop order in the string coupling receives contributions from surfaces of four different worldsheet topologies : torus, annulus, Mobius strip, and Klein bottle. Given our analysis of the heterotic and type II superstring ensembles in previous sections, we expect that the Yang-Mills fields will play an essential role in enabling discussion of an equilibrium canonical ensemble. As explained in section 2, pure type II closed string theories have no Yang-Mills gauge fields, but upon coupling to Dbranes, the full unoriented open and closed string theory indeed contains timelike Wilson line backgrounds. This is the case we shall consider in this section.
The torus contribution sums over closed oriented worldsheets, and the result therefore follows from the analysis of the type IIB superstring described in section 2. In particular, there are no tachyon-free thermal ground states that break target space supersymmetry and, consequently, no equilibrium canonical ensemble. The functional, $`W_{\mathrm{tor}}(\beta )`$, therefore vanishes identically. Thus, at one-loop order in string perturbation theory, the oriented closed string sector will not contribute to the free energy of the type IB thermal vacuum.
Thus, our focus must shift to the open and unoriented string sectors of the type IB thermal spectrum. An equilibrium description of type IB string statistical mechanics in the canonical ensemble requires a tachyon-free thermal spectrum in the full temperature range, in addition to the absence of massless Ramond-Ramond sector tadpoles. The vacuum functional at one-loop order in the presence of $`N`$ $`=`$ $`2^5`$ D9branes can be written in the general form:
$`W_{\mathrm{IB}}(\beta )=`$ $`\beta L^9(4\pi ^2\alpha ^{})^5{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^5\mathrm{Z}_{\mathrm{IB}}(\beta ),`$ (81)
where the function $`\mathrm{Z}_{\mathrm{IB}}(\beta )`$ denotes the level expansion of the type IB unoriented thermal ensemble, with contributions from annulus, Mobius strip, and Klein bottle.
Notice that there is no constraint analogous to modular invariance on the open string mass spectrum. In addition, there are no thermal winding modes. Consider the contribution to the path integral from the $`n`$th Matsubara mode:
$$e^{4\pi ^3\alpha ^{}n^2t/\beta ^2},n,n+\frac{1}{2}\mathrm{Z}.$$
(82)
Focussing on the $`t`$$``$$`\mathrm{}`$ IR limit of the string amplitude at first, at low temperatures $`\beta `$ $``$ $`\mathrm{}`$, we see that the field theoretic modes with low values of $`n`$ dominate. But in the opposite $`t`$ $``$ $`0`$ limit, we must include arbitrarily high values of $`n`$ in the Matsubara spectrum. What about the high temperature limit, with $`\beta `$ $``$ $`0`$? Even in the $`t`$ $``$ $`\mathrm{}`$ limit, modes with arbitrarily high $`n`$ are now important. Thus, a thermal duality transformation to the Euclidean T-dual type I ensemble would illuminate the high temperature behavior of the ensemble. As in the case of the heterotic string ensemble, the key will lie in introducing a temperature dependent Wilson line gauge background in order to achieve the spontaneous breaking of supersymmetry without the introduction of thermal tachyons. We must simultaneously require that $`\mathrm{Z}_{(\mathrm{IB})}(\beta )`$ preserve the Ramond-Ramond sector massless tadpole cancellations necessary for infrared finiteness and perturbative renormalizability of the finite temperature vacuum.
Recall that the Wilson line introduced in the $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ heterotic string breaks the nonabelian gauge group to $`SO(16)`$$`\times `$$`SO(16)`$, while spontaneously breaking supersymmetry. Let us compactify the type IB SO(32) string on a circle of radius $`\beta /2\pi `$, turning on the temperature dependent Wilson line :
$$\beta A_0=(1^8,0^8),\theta _i=2\pi ,i=1,\mathrm{},8;\theta _i=0,i=9,\mathrm{},16.$$
(83)
This is the SO(16)$`\times `$SO(16) type IB vacuum, and the Wilson line wrapping the $`X^0`$ coordinate has spontaneously broken supersymmetry in addition to a partial breaking of the nonabelian gauge symmetry in the supersymmetric SO(32) vacuum. Thus, we wish to find the finite temperature vacuum functional that interpolates between the tachyon-free and nonsupersymmetric type IB SO(16)$`\times `$SO(16) string at finite $`\beta `$, and the supersymmetric type IB $`SO(32)`$ string in the limit where the Wilson loop is infinite radius.
The Euclidean T-duality transformation, $`\beta `$ $``$ $`\beta ^{}`$$`=`$$`4\pi ^2\alpha ^{}/\beta `$, maps the configuration of D9branes coincident with an $`O9`$ plane to a configuration of a pair of $`O8`$ planes, each coincident with 16 D8branes, at the two endpoints of an interval of length $`\beta ^{}`$. The thermal modes of the type IB ensemble correspond to a Matsubara-like frequency spectrum with timelike momentum: $`p_n`$$`=`$$`2n\pi /\beta `$, where $`n`$, $`n+\frac{1}{2}`$ $``$ $`\mathrm{Z}`$; for the T-dual type I ensemble this role is played by the infinite summations over thermal winding modes. Note that the D8branes are each paired with their images, so that the gauged background has a total of $`8^2`$ oriented stretched strings of length $`\beta ^{}`$. Supersymmetry is spontaneously broken due to the presence of the stretched strings, which contribute with positive thermal tension to the vacuum energy of the finite temperature vacuum:
$$๐ฏ(\beta )=\beta ^2/4\pi ^2\alpha ^{},$$
(84)
contributing an overall shift in the masses of the thermal tachyons in the NS sector. The GSO projections on worldsheet fermions eliminates all tachyons from the physical state spectrum, as in the supersymmetric vacuum.
The cancellation of both RR-RR, and NS-NS, sector massless tadpoles, and the absence of open string tachyons, in a nonsupersymmetric and tachyon-free 9D SO(16)$`\times `$SO(16) type IIB orientifold, was shown by Blum and Dienes in . As a consequence, both the cylinder and the Klein bottle amplitude vanish identically in the 9D vacuum, and the result for the vacuum energy density can be written entirely in terms of the contribution from the Mobius strip:
$`\mathrm{Z}_{\mathrm{Mob}}^{\mathrm{NS}\mathrm{NS}}(\beta )=`$ $`\left[_0+_{1/2}\right]\left({\displaystyle \frac{\mathrm{\Theta }_{01}(it;0)\mathrm{\Theta }_{10}(it;0)}{\eta (it)\mathrm{\Theta }_{00}(it)}}\right)^4`$ (85)
$`\mathrm{Z}_{\mathrm{Mob}}^{\mathrm{R}\mathrm{R}}(\beta )=`$ $`\left[_0_{1/2}\right]\left({\displaystyle \frac{\mathrm{\Theta }_{01}(it;0)\mathrm{\Theta }_{10}(it;0)}{\eta (it)\mathrm{\Theta }_{00}(it)}}\right)^4.`$ (86)
The authors of Ref. have computed the individual contributions from the Mobius strip and torus amplitudes numerically down to values of the radius at which the divergence in the torus amplitude sets in, as a consequence of the closed string (high temperature) thermal tachyon. The asymptotic approach to the supersymmetric SO(32) vacuum at low temperatures is evident in their plot. As explained earlier, we believe it is more meaningful to interpret these results as evidence that the worldsheet RG flow is towards the IR stable supersymmetric vacuum in the closed string sector: the one-loop vacuum amplitude computes the tree level mass spectrum, and at this order in string perturbation theory the supergravity and Yang-Mills sectors have not as yet โcommunicatedโ. Thus, supersymmetry remains unbroken in the gravitational sector, and the nonvanishing one-loop vacuum energy density of the type I model comes wholly from the Mobius strip.
We believe the correct physical interpretation of the Blum-Dienes result is as follows. Unlike the heterotic case, in the type I strings, there are thermal interpolations between the 10D supersymmetric SO(32) heterotic vacuum and either the 9D nonsupersymmetric and nontachyonic type IB, or the T-dual type I, SO(16)$`\times `$SO(16) string vacua. The former has only thermal momentum modes, and the latter only thermal winding modes. The result for the free energy of the type I ensemble takes the form:
$`F_\mathrm{I}^{}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^5[\eta (it)]^8\left[{\displaystyle \frac{\mathrm{\Theta }_{01}(it;0)\mathrm{\Theta }_{10}(it;0)}{\eta (it)\mathrm{\Theta }_{00}(it)}}\right]^4`$ (88)
$`\times {\displaystyle \underset{w=\mathrm{}}{\overset{\mathrm{}}{}}}\{e^{2\pi \beta ^2w^2/4\pi ^2\alpha ^{}}+e^{2\pi \beta ^2(w+{\scriptscriptstyle \frac{1}{2}})^2]/4\pi ^2\alpha ^{}}\},`$
or the Euclidean T-dual transformed result for the type IB ensemble:
$`F_{\mathrm{IB}}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^5[\eta (it)]^8\left[{\displaystyle \frac{\mathrm{\Theta }_{01}(it;0)\mathrm{\Theta }_{10}(it;0)}{\eta (it)\mathrm{\Theta }_{00}(it)}}\right]^4`$ (90)
$`\times {\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\{e^{8\pi ^3\alpha ^{}n^2/\beta ^2}+e^{8\pi ^3\alpha ^{}(n+{\scriptscriptstyle \frac{1}{2}})^2]/\beta ^2}\},`$
Thus, we have found evidence for a stable equilibrium ensemble of type I unoriented open and closed strings in the limit of weak string coupling.
### 4.1 The Low Energy Limit of $`F_{\mathrm{IB}}(T)`$
At low temperatures far below the string mass scale, $`\beta `$$`>>`$$`\alpha ^{1/2}`$, we expect not to excite any thermal modes beyond the lowest-lying field theoretic modes in the open string spectrum. A useful check of self-consistency with the low energy finite temperature field theory limit is to verify the expected $`T^{10}`$ growth of the free energy. For simplicity, we will restrict ourselves to the sector of the thermal spectrum with half-integer moding of Matsubara frequencies. An analogous expression holds for the sector with integer moding.
We begin with the one-loop contribution to the type IB string vacuum functional given in Eq. (90), including both integer and half-integer Matsubara frequencies. Expand the integrand of the modular integral in powers of $`q`$$`=`$$`e^{2\pi t}`$, thus isolating the $`t`$$``$$`\mathrm{}`$ asymptotics of the modular integral which is dominated by the massless modes in the open string spectrum. We have:
$`F_{\mathrm{IB}}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^5[\eta (it)]^8\left[{\displaystyle \frac{\mathrm{\Theta }_{01}(it;0)\mathrm{\Theta }_{10}(it;0)}{\eta (it)\mathrm{\Theta }_{00}(it)}}\right]^4`$ (92)
$`\times {\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}[e^{8\pi ^3\alpha ^{}n^2/\beta ^2}+e^{8\pi ^3\alpha ^{}(n+{\scriptscriptstyle \frac{1}{2}})^2/\beta ^2}].`$
Notice that at sufficiently low temperatures, an infinity of Matsubara modes can contribute to the $`\beta `$ $``$ $`\mathrm{}`$ limit of this expression since the integral on $`t`$ ranges over \[$`0`$, $`\mathrm{}`$\]. Thus, we must include the full sum over the full Matsubara spectrum when computing the leading low energy field theory limit of the string free energy.
We recognize in our expression the integral representation of an Euler gamma function with negative argument, $`\mathrm{\Gamma }(9/2)`$, which may be defined by either the product formula for gamma functions, or by analytic continuation in the argument of the gamma function. Thus, performing the explicit integration over $`t`$ for the leading term in the level expansion on $`m`$, and restricting to the summation over just the half-integer moded Matsubara spectrum for illustration, gives the result:
$`F_{\mathrm{short}}=`$ $`L^9(4\pi ^2\alpha ^{})^52{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^52^4e^{8\alpha ^{}\pi ^3(n+{\scriptscriptstyle \frac{1}{2}})^2t/\beta ^2}`$ (93)
$`=`$ $`L^9(4\pi ^2\alpha ^{})^52^4\left[\zeta (10,\frac{1}{2})\right]\left[(2\pi )^5\mathrm{\Gamma }\left(5\right)(4\pi ^2\alpha ^{})^5\beta ^{10}\right]`$ (94)
$``$ $`\left[L^9(4\pi ^2\alpha ^{})^5\right](\beta _0/\beta )^{10},`$ (95)
where $`L^9`$ denotes the nine-dimensional spatial volume. $`\mathrm{\Gamma }(n)`$ would have to be defined by analytic continuation, invoking a complex integral representation. The factors within the second pair of square brackets in Eq. (95) result from the integral over the world-sheet modulus, $`t`$. The modular integration is followed by the infinite summations on even and odd integers, giving the result within the first pair of square brackets. The summations have been expressed in terms of the Riemann zeta function $`\zeta (z,q)`$:
$$\underset{n=0}{\overset{\mathrm{}}{}}(n+\frac{1}{2})^z\zeta (z,\frac{1}{2}),\zeta (n,\frac{1}{2})=\frac{1}{(n+1)(n+2)}B_{n+2}^{}(x)|_{x={\scriptscriptstyle \frac{1}{2}}},$$
(96)
and $`B_n(x)`$ is a Bernoulli polynomial, $`B_n^{}(x)`$$`=`$$`nB_{n1}(x)`$. The parameter, $`\beta _0`$, has the dimensions of an inverse temperature, and it characterizes the asymptotic limit of the free energy for the low energy supersymmetric gauge theory. Recall that the string vacuum functional is dimensionless. Thus, we have demonstrated that the one-loop free energy, $`F(\beta )`$$`=`$$`W_{\mathrm{IB}}(\beta )/\beta `$, grows as $`T^{10}`$ in the asymptotic low temperature regime. This is precisely the behavior expected of a ten-dimensional, finite temperature supersymmetric gauge theory. Explicitly, we have:
$$\beta _0=[2^4(B_{11}(\frac{1}{2})/11)\mathrm{\Gamma }(5)]^{1/10}(4\pi ^2\alpha ^{})^{1/2}.$$
(97)
### 4.2 High Temperature Expansion
Conversely, when we include the contribution from all of the massive string modes, we find a $`T^2`$ growth in the string free energy at temperatures far above the string scale. Set $`t`$$``$$`1/t`$ in the expression for the level expansion, and expand in powers of $`e^{\pi /t}`$. The ultraviolet asymptotics of the mass level expansion dominates the high temperature regime, giving the following leading contribution to $`F_{\mathrm{IB}}(\beta )`$:
$`F_{\mathrm{IB}}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}{\displaystyle \frac{t^1}{\eta (i/t)^8}}`$ (98)
$`\times 2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\left[{\displaystyle \frac{\mathrm{\Theta }_{01}(i/t;0)\mathrm{\Theta }_{10}(i/t;0)}{\eta (i/t)\mathrm{\Theta }_{00}(i/t)}}\right]^4e^{8\alpha ^{}\pi ^3(n+{\scriptscriptstyle \frac{1}{2}})^2t/\beta ^2}`$ (99)
$`=`$ $`L^9(4\pi ^2\alpha ^{})^5{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle _0^{\mathrm{}}}๐tt^2\left[2^4+O(e^{\pi /t})\right]e^{8\alpha ^{}\pi ^3(n+{\scriptscriptstyle \frac{1}{2}})^2t/\beta ^2}`$ (100)
$`=`$ $`L^9(4\pi ^2\alpha ^{})^5\left[2^4\zeta (2,\frac{1}{2})\right]\left[\mathrm{\Gamma }(1)(2\pi )(4\pi ^2\alpha ^{})\beta ^2\right]`$ (101)
$``$ $`L^9(4\pi ^2\alpha ^{})^5(\beta _1/\beta )^2.`$ (102)
Thus, we find that the free energy at one-loop order in every perturbative string ensemble, whether open or closed, bosonic or supersymmetric, grows as $`T^2`$ at temperatures far above the string scale, a conjecture originally made by Atick and Witten for closed string theories alone, and later shown to be a consequence of thermal self-duality in the closed bosonic string theory by Polchinski .
Including all of the corrections to the ultraviolet asymptotic limit of the open string mass spectrum, we find that the term-by-term integration over the modulus, $`t`$, can be expressed in terms of the Bessel function $`K_1(x)`$. The free energy takes the form:
$`F_{\mathrm{IB}}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^52{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{dt}{2t}}t^1\left[2^4+{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}b_m^{(\mathrm{IB})}e^{2\pi m/t}\right]e^{8\alpha ^{}\pi ^3(n+{\scriptscriptstyle \frac{1}{2}})^2t/\beta ^2}`$ (103)
$`=`$ $`L^9(4\pi ^2\alpha ^{})^5\left[(\beta _1/\beta )^2+{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle _0^{\mathrm{}}}๐tt^2{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}b_m^{(\mathrm{IB})}e^{2\pi m/t}e^{8\alpha ^{}\pi ^3(n+{\scriptscriptstyle \frac{1}{2}})^2t/\beta ^2}\right]`$ (104)
where the $`b_m^{(\mathrm{IB})}`$ are the coefficients in the mass level expansion for the type IB thermal mass spectrum, obtained by expanding in powers of $`q`$$`=`$$`e^{2\pi /t}`$:
$$๐ต_{\mathrm{IB}}(i/t)[\eta (i/t)]^8\left[\frac{\mathrm{\Theta }_{01}(i/t;0)\mathrm{\Theta }_{10}(i/t;0)}{\eta (i/t)\mathrm{\Theta }_{00}(i/t)}\right]^4\underset{m=0}{\overset{\mathrm{}}{}}b_m^{(\mathrm{IB})}e^{2\pi m/t}.$$
(106)
Performing the integration over $`t`$ results in the expression:
$`F_{\mathrm{IB}}(\beta )=`$ $`L^9(4\pi ^2\alpha ^{})^5(\beta _1/\beta )^2`$ (108)
$`L^9(4\pi ^2\alpha ^{})^5{\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}b_m^{(\mathrm{IB})}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}[2\pi (4\pi ^2\alpha ^{})(n+\frac{1}{2})^2/\beta ^2)][2({\displaystyle \frac{x}{2}})^1K_1(x)]^1,`$
where $`\beta ^2x^2`$$`=`$$`64\pi ^4\alpha ^{}m(n+\frac{1}{2})^2`$ identifies the argument of the Bessel function. We emphasize that this result for the type IB free energy is valid at all temperatures, and includes all of the massive string modes, as well as the full half-integer moded Matsubara frequency spectrum. A similar expression can be derived for the integer moded frequency spectrum. The factor in square brackets in the first equality arises from the modular integration. The infinite summation over thermal mode numbers therefore yields $`\zeta (1,\frac{1}{2})`$. The large $`x`$, or high temperature, asymptotics of the Bessel function indicates clearly that all of the corrections to the leading $`T^2`$ growth of the string free energy are exponentially suppressed. The asymptotic expression for the one-loop free energy therefore takes the form:
$$F_{\mathrm{IB}}(T)/L^9(4\pi ^2\alpha ^{})^5=\beta _1^2T^2\underset{m=1}{\overset{\mathrm{}}{}}\underset{n=1}{\overset{\mathrm{}}{}}A_mTe^{\beta _2(m,n)T}.$$
(110)
where the leading $`T^2`$ growth is found to be corrected by a negligible $`Te^T`$ dependence, at high temperatures. Recall that $`\beta _1`$ takes the form:
$`\beta _1^2=`$ $`\left[(2\pi )2^4\mathrm{\Gamma }(1)B_3(\frac{1}{2})/3\right]\beta _C^2,`$ (111)
where we have substituted for $`\zeta (2,\frac{1}{2})`$. As before, $`\mathrm{\Gamma }(n)`$ has to be defined by analytic continuation. Notice the minus sign in the coefficient, reversing the sign of the free energy, and clarifying that the dominant UV contribution is from target space fermionic modes.
It is helpful to check the corresponding scaling behavior as a function of temperature for the first few thermodynamic potentials. The internal energy of the canonical ensemble, $`U`$ $`=`$ $`\left(W/\beta \right)_V`$, displays the same scaling behavior as the free energy. The entropy is given by the expression:
$$S=\beta ^2\left(\frac{F}{\beta }\right)_V=\beta ^2\left[\beta ^2W(\beta )\beta ^1\left(\frac{W}{\beta }\right)_V\right],$$
(112)
and we infer that it scales at high temperatures as $`\beta ^1`$. Finally, since the specific heat at constant volume is given by:
$$C_V=\beta \left(\frac{S}{\beta }\right)_V,$$
(113)
we infer that it also scales as $`\beta ^1`$ at high temperatures. The corresponding scaling behavior at low temperatures, in agreement with the expected result for the low energy gauge theory limit can be extracted, as in the previous section, by considering instead the $`t`$$``$$`\mathrm{}`$ asymptotics of these expressions.
## 5 Conclusions
We have shown that all of the perturbatively renormalizable and anomaly-free supersymmetric string theories with Yang-Mills gauge fields: heterotic, type I, and type I admit stable and tachyon-free finite temperature ground states in which we can formulate an equilibrium statistical mechanics of strings in the weak coupling limit. Preceeding attempts to formulate an equilibrium statistical mechanics with a tachyon-free supersymmetric string canonical ensemble have failed for a variety of reasons; a detailed account appears in section 2 of hep-th/0409301v1. These works do not correctly incorporate the Euclidean T-duality transformations linking the thermal vacua of the supersymmetric string theories in pairs. Neither do they meet the infrared consistency conditions we have required in our analysis, matching self-consistently with the low energy finite temperature field theoretic physics. In addition, string theoretic consistency conditions, such as modular invariance, have often been violated. The results in this paper demonstrate irrefutably that the widespread misconception that the canonical ensemble of superstrings necessarily breaks down beyond a limiting (Hagedorn) temperature is simply wrong. Finally, in the case of the pure type IIA and type IIB thermal ground states, we show that it is not possible to eliminate low temperature tachyons from the thermal spectrum while meeting the constraints from modular invariance. This fact also precludes interpretation of thermal winding mode tachyons as indication of a string scale โHagedornโ phase transition in the type II theories: the canonical ensemble cannot be defined in the full temperature range, starting at $`T`$$`=`$$`0`$.
For each of the six superstring ensembles, we have shown that the growth of the free energy at high temperatures far above the string scale is only as fast as that in a 2d quantum field theory: $`F(T)`$ $``$ $`T^2`$. This behavior exemplifies the equivalence of perturbatively renormalizable superstring theories to the 2d gauge theory of diffeomorphism and Weyl invariances. We have simultaneously verified that the low energy field theoretic limit of our expressions for the string free energy in each case recovers the expected $`T^{10}`$ growth characteristic of a 10D quantum field theory, at low temperatures where only the lowest-lying massless field theory modes contribute. These results follow directly from an explicit evaluation of the one-loop modular integrals. The $`T^2`$ growth of the high temperature string free energy was originally conjectured by Atick and Witten in . We have shown that the $`T^2`$ scaling can also be inferred from the thermal duality relations linking the thermal ground states of the superstring ensembles in pairs. This is in precise analogy with the $`T^2`$ scaling of the free energy of the closed bosonic string as a consequence of its thermal self-duality, as shown by Polchinski in . In fact, we can infer the existence of a duality phase transition in the Kosterlitz-Thouless universality class mapping the finite temperature ground state of the $`E_8`$$`\times `$$`E_8`$ heterotic string to its $`\mathrm{Spin}(32)/\mathrm{Z}_2`$ Euclidean T-dual, analogous to the self-duality phase transition observed in the closed bosonic string ensemble . The Kosterlitz-Thouless universality class is characterized by an infinite hierarchy of thermodynamic potentials displaying analyticity at the critical temperature, in this case, $`T_C`$ $`=`$ $`1/2\pi \alpha ^{1/2}`$ .
We will close with mention of an important insight that applies more broadly to the development of a fundamental, and nonperturbative, formulation for String/M Theory. We remind the reader that perturbative string theory as formulated in the worldsheet formalism is inherently background dependent: the โheat-bathโ representing the embedding target space of fixed spatial volume and fixed inverse temperature is forced upon us, together with any external background fields characterizing the target spacetime geometry. Thus, we are ordinarily restricted to the canonical ensemble of statistical mechanics. We should caution the reader that while an immense, and largely conjectural, literature exists on proposals for microcanonical ensembles of weakly-coupled strings , the conceptual basis of these treatments is full of holes. Some of the pitfalls have been described in . One could argue that, strictly speaking, the microcanonical ensemble is what is called for when discussing quantum cosmology, or the statistical mechanics of the Universe : the Universe is, by definition, an isolated closed system, and it is meaningless to invoke the canonical ensemble of the โfundamentalโ degrees of freedom. However, there remain many simpler questions in both early Universe cosmology, and in finite temperature gauge theories, that are approachable within the framework of the canonical ensemble. Limited use of the microcanonical ensemble under certain assumptions is also possible, but the constraints imposed by the Jeans instability, and by thermal back reaction, need to be kept in mind.
Acknowledgements: I would like to thank Joe Polchinski for urging me to be precise in applying the thermal (Euclidean T-duality) transformations linking the thermal ground states of the six superstring theories. Although this comment came in May 2001, as clarified in Footnote 2, the idea of pursuing the significance of thermal duality also comes from his previous work . I would like to acknowledge Hassan Firouzjahi for correcting a misleading typo in presentations of the type II NS-NS tachyon spectrum prior to Aug 2004. I also thank an anonymous reader, as well as Keith Dienes, Mike Lennek, and Lubos Motl, for pointing out an error in my presentations of modular invariance in the type II analysis prior to June 2005. Both corrections reinforce my conclusion in of the impossibility of a type II canonical ensemble in the absence of a Yang-Mills gauge sector. This research has been supported in part by the National Science Foundation, the Aspen Center for Physics, and the Kavli Institute for Theoretical Physics.
|
warning/0506/math0506128.html
|
ar5iv
|
text
|
# The Andrews-Stanley partition function and Al-Salam-Chihara polynomials
## 1 Introduction
For any integer partition $`\lambda `$, denote by $`\lambda ^{}`$ its conjugate and $`\mathrm{}(\lambda )`$ the number of its parts. Let $`๐ช(\lambda )`$ denote the number of odd parts of $`\lambda `$ and $`|\lambda |`$ the sum of its parts. R. Stanley () has shown that if $`t(n)`$ denotes the number of partitions $`\lambda `$ of $`n`$ for which $`๐ช(\lambda )๐ช(\lambda ^{})`$ ($`\mathrm{mod}4`$), then
$$t(n)=\frac{1}{2}\left(p(n)+f(n)\right),$$
where $`p(n)`$ is the total number of partitions of $`n`$, and $`f(n)`$ is defined by
$$\underset{n=0}{\overset{\mathrm{}}{}}f(n)q^n=\underset{i1}{}\frac{(1+q^{2i1})}{(1q^{4i})(1+q^{4i2})}.$$
Motivated by Stanleyโs problem, G.E. Andrews assigned the weight $`z^{๐ช(\lambda )}y^{๐ช(\lambda ^{})}q^{|\lambda |}`$ to each partition $`\lambda `$ and computed the corresponding generating function of all partitions with parts each less than or equal to $`N`$ (see Corollary 4.4). The following more general weight first appeared in Stanleyโs paper . Let $`a`$, $`b`$, $`c`$ and $`d`$ be commuting indeterminates. For each partition $`\lambda `$, define the *Andrews-Stanley partition functions* $`\omega (\lambda )`$ by
$$\omega (\lambda )=a^{_{i1}\lambda _{2i1}/2}b^{_{i1}\lambda _{2i1}/2}c^{_{i1}\lambda _{2i}/2}d^{_{i1}\lambda _{2i}/2},$$
(1.1)
where $`x`$ (resp. $`x`$) stands for the smallest (resp. largest) integer greater (resp. less) than or equal to $`x`$ for a given real number $`x`$. Actually it is more convenient to define the above weight through the Ferrers diagram of $`\lambda `$: one fills the $`i`$th row of the Ferrers diagram alternatively by $`a`$ and $`b`$ (resp. $`c`$ and $`d`$) if $`i`$ is *odd* (resp. *even*), the weight $`w(\lambda )`$ is then equal to the product of all the entries in the diagram. For example, if $`\lambda =(5,4,4,1)`$ then $`\omega (\lambda )`$ is the product of the entries in the following diagram for $`\lambda `$.
$`a`$ $`b`$ $`a`$ $`b`$ $`a`$ $`c`$ $`d`$ $`c`$ $`d`$ $`a`$ $`b`$ $`a`$ $`b`$ $`c`$
In C. Boulet has obtained results for the generating functions of all ordinary partitions and all strict partitions with respect to the weight (1.1) (see Corollary 3.6 and Corollary 4.5). On the other hand, A. Sills has given a combinatorial proof of Andrewsโ result, which has been further generalized by A. Yee by restricting the sum over partitions with parts each $`N`$ and length $`M`$.
In this paper we shall generalize Bouletโs results by summing the weight function $`\omega (\lambda )z^{\mathrm{}(\lambda )}`$ over all the ordinary (resp. strict) partitions with parts each $`N`$. It turns out that the corresponding generating functions are related to the basic hypergeometric series, namely the Al-Salam-Chihara polynomials and the associated Al-Salam-Chihara polynomials (see Theorem 3.4 and Theorem 4.3).
This paper can be regarded as a succession of , in which one of the authors gave a Pfaffian formula for the weighted sum $`{\displaystyle \omega (\lambda )s_\lambda (x)}`$ of the Schur functions $`s_\lambda (x)`$, where the sum runs over all ordinary partitions $`\lambda `$, and settled an open problem by Richard Stanley. Though it is not possible to specialize the Schur functions to $`z^{\mathrm{}(\lambda )}`$, we show in this paper that this approach still works, i.e., we can evaluate the weighted sum $`\omega (\lambda )z^{\mathrm{}(\lambda )}`$ by using Pfaffians and minor summation formulas as tools (, ), but, as an after thought, we also provide alternative combinatorial proofs.
In the last section we show the weighted sum $`{\displaystyle \omega (\mu )z^{\mathrm{}(\mu )}P_\mu (x)}`$ of Schurโs $`P`$-functions $`P_\mu (x)`$ (when $`z=2`$, this equals the weighted sum $`\omega (\mu )Q_\mu (x)`$ of Schurโs $`Q`$-functions $`Q_\mu (x)`$) can be expressed by a Pfaffian where $`\mu `$ runs over all strict partitions (with parts each $`N`$).
## 2 Preliminaries
A $`q`$-shifted factorial is defined by
$$(a;q)_0=1,(a;q)_n=(1a)(1aq)\mathrm{}(1aq^{n1}),n=1,2,\mathrm{}.$$
We also define $`(a;q)_{\mathrm{}}=_{k=0}^{\mathrm{}}(1aq^k)`$. Since products of $`q`$-shifted factorials occur very often, to simplify them we shall use the compact notations
$`(a_1,\mathrm{},a_m;q)_n=(a_1;q)_n\mathrm{}(a_m;q)_n,`$
$`(a_1,\mathrm{},a_m;q)_{\mathrm{}}=(a_1;q)_{\mathrm{}}\mathrm{}(a_m;q)_{\mathrm{}}.`$
We define an $`{}_{r+1}{}^{}\varphi _{r}^{}`$ basic hypergeometric series by
$`{}_{r+1}{}^{}\varphi _{r}^{}({\displaystyle \genfrac{}{}{0pt}{}{a_1,a_2,\mathrm{},a_{r+1}}{b_1,\mathrm{},b_r}};q,z)={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(a_1,a_2,\mathrm{},a_{r+1};q)_n}{(q,b_1,\mathrm{},b_r;q)_n}}z^n.`$
The Al-Salam-Chihara polynomial $`Q_n(x)=Q_n(x;\alpha ,\beta |q)`$ is, by definition (cf. \[11, p.80\]),
$`Q_n(x;\alpha ,\beta |q)`$ $`={\displaystyle \frac{(\alpha \beta ;q)_n}{\alpha ^n}}{}_{3}{}^{}\varphi _{2}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^n,\alpha u,\alpha u^1}{\alpha \beta ,0}};q,q),`$
$`=(\alpha u;q)_nu^n{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^n,\beta u^1}{\alpha ^1q^{n+1}u^1}};q,\alpha ^1qu),`$
$`=(\beta u^1;q)_nu^n{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^n,\alpha u}{\beta ^1q^{n+1}u}};q,\beta ^1qu^1),`$
where $`x=\frac{u+u^1}{2}`$. This is a specialization of the Askey-Wilson polynomials (see ), and satisfies the three-term recurrence relation
$$2xQ_n(x)=Q_{n+1}(x)+(\alpha +\beta )q^nQ_n(x)+(1q^n)(1\alpha \beta q^{n1})Q_{n1}(x),$$
(2.1)
with $`Q_1(x)=0`$, $`Q_0(x)=1`$.
We also consider a more general recurrence relation:
$$2x\stackrel{~}{Q}_n(x)=\stackrel{~}{Q}_{n+1}(x)+(\alpha +\beta )tq^n\stackrel{~}{Q}_n(x)+(1tq^n)(1t\alpha \beta q^{n1})\stackrel{~}{Q}_{n1}(x),$$
(2.2)
which we call the associated Al-Salam-Chihara recurrence relation. Put
$`\stackrel{~}{Q}_n^{(1)}(x)=u^n(t\alpha u;q)_n{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{t^1q^n,\beta u^1}{t^1\alpha ^1q^{n+1}u^1}};q,\alpha ^1qu),`$ (2.3)
$`\stackrel{~}{Q}_n^{(2)}(x)=u^n{\displaystyle \frac{(tq;q)_n(t\alpha \beta ;q)_n}{(t\beta uq;q)_n}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{tq^{n+1},\alpha ^1qu}{t\beta q^{n+1}u}};q,\alpha u),`$ (2.4)
where $`x=\frac{u+u^1}{2}`$. In , Ismail and Rahman have presented two linearly independent solutions of the associated Askey-Wilson recurrence equation (see also ). By specializing the parameters, we conclude that $`\stackrel{~}{Q}_n^{(1)}(x)`$ and $`\stackrel{~}{Q}_n^{(2)}(x)`$ are two linearly independent solutions of the associated Al-Salam-Chihara equation 2.2 (see \[5, p.203\]). Here, we use this fact and omit the proof. The series 2.3 and 2.4 are convergent if we assume $`|u|<1`$ and $`|q|<|\alpha |<1`$ (see \[5, p.204\]).
Let
$$W_n=\stackrel{~}{Q}_n^{(1)}(x)\stackrel{~}{Q}_{n1}^{(2)}(x)\stackrel{~}{Q}_{n1}^{(1)}(x)\stackrel{~}{Q}_n^{(2)}(x)$$
(2.5)
denote the Casorati determinant of the equation 2.2. Since $`\stackrel{~}{Q}_n^{(1)}(x)`$ and $`\stackrel{~}{Q}_n^{(2)}(x)`$ both satisfy the recurrence equation 2.2, it is easy to see that $`W_n`$ satisfies the recurrence equation
$$W_{n+1}=(1tq^n)(1t\alpha \beta q^{n1})W_n.$$
Using this equation recursively, we obtain
$$W_{n+1}=(tq,t\alpha \beta ;q)_nW_1,$$
which implies
$$W_1=\frac{lim_n\mathrm{}W_{n+1}}{(tq,t\alpha \beta ;q)_{\mathrm{}}}.$$
Using 2.3 and 2.4, we obtain
$`\underset{n\mathrm{}}{lim}W_{n+1}={\displaystyle \frac{u^1(t\alpha u,tq,t\alpha \beta ,\beta u;q)_{\mathrm{}}}{(t\beta uq,\alpha u;q)_{\mathrm{}}}}`$
(for the detail, see ). Thus we conclude that
$`W_1={\displaystyle \frac{u^1(t\alpha u,\beta u;q)_{\mathrm{}}}{(\alpha u,t\beta uq;q)_{\mathrm{}}}}.`$ (2.6)
In the following sections we need to find a polynomial solution of the recurrence equation 2.2 which satisfies a given initial condition, say $`\stackrel{~}{Q}_0(x)=\stackrel{~}{Q}_0`$ and $`\stackrel{~}{Q}_1(x)=\stackrel{~}{Q}_1`$. Since $`\stackrel{~}{Q}_n^{(1)}(x)`$ and $`\stackrel{~}{Q}_n^{(2)}(x)`$ are linearly independent solutions of 2.2, this $`\stackrel{~}{Q}_n(x)`$ can be written as a linear combination of these functions, say
$$\stackrel{~}{Q}_n(x)=C_1\stackrel{~}{Q}_n^{(1)}(x)+C_2\stackrel{~}{Q}_n^{(2)}(x).$$
If we substitute the initial condition $`\stackrel{~}{Q}_0(x)=\stackrel{~}{Q}_0`$ and $`\stackrel{~}{Q}_1(x)=\stackrel{~}{Q}_1`$ into this equation and solve the linear equation, then we obtain
$`C_1={\displaystyle \frac{1}{W_1}}\left\{\stackrel{~}{Q}_1\stackrel{~}{Q}_0^{(2)}(x)\stackrel{~}{Q}_0\stackrel{~}{Q}_1^{(2)}(x)\right\},`$
$`C_2={\displaystyle \frac{1}{W_1}}\left\{\stackrel{~}{Q}_0\stackrel{~}{Q}_1^{(1)}(x)\stackrel{~}{Q}_1\stackrel{~}{Q}_0^{(1)}(x)\right\}.`$
By 2.6, we obtain
$`\stackrel{~}{Q}_n(x)`$ $`={\displaystyle \frac{u(\alpha u,t\beta uq;q)_{\mathrm{}}}{(t\alpha u,\beta u;q)_{\mathrm{}}}}[\{\stackrel{~}{Q}_1\stackrel{~}{Q}_0^{(2)}(x)\stackrel{~}{Q}_0\stackrel{~}{Q}_1^{(2)}(x)\}\stackrel{~}{Q}_n^{(1)}(x)`$
$`+\{\stackrel{~}{Q}_0\stackrel{~}{Q}_1^{(1)}(x)\stackrel{~}{Q}_1\stackrel{~}{Q}_0^{(1)}(x)\}\stackrel{~}{Q}_n^{(2)}(x)]`$ (2.7)
with
$`\stackrel{~}{Q}_0^{(1)}(x)`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{t^1,\beta u^1}{t^1\alpha ^1u^1q}};q,\alpha ^1uq),`$
$`\stackrel{~}{Q}_1^{(1)}(x)`$ $`=u^1(1\alpha tu){}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{t^1q^1,\beta u^1}{t^1\alpha ^1u^1}};q,\alpha ^1uq),`$
$`\stackrel{~}{Q}_0^{(2)}(x)`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{tq,\alpha ^1uq}{t\beta uq}};q,\alpha u),`$
$`\stackrel{~}{Q}_1^{(2)}(x)`$ $`={\displaystyle \frac{u(1tq)(1t\alpha \beta )}{(1t\beta uq)}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{tq^2,\alpha ^1uq}{t\beta uq^2}};q,\alpha u).`$
Since
$`\underset{n\mathrm{}}{lim}u^n\stackrel{~}{Q}_n^{(1)}(x)`$ $`={\displaystyle \frac{(t\alpha u,\beta u;q)_{\mathrm{}}}{(u^2;q)_{\mathrm{}}}},`$
$`\underset{n\mathrm{}}{lim}u^n\stackrel{~}{Q}_n^{(2)}(x)`$ $`=0,`$
if we take the limit $`\underset{n\mathrm{}}{lim}u^n\stackrel{~}{Q}_n(x)`$, then we have
$$\underset{n\mathrm{}}{lim}u^n\stackrel{~}{Q}_n(x)=\frac{u(t\beta uq,\alpha u;q)_{\mathrm{}}}{(u^2;q)_{\mathrm{}}}\left\{\stackrel{~}{Q}_1\stackrel{~}{Q}_0^{(2)}(x)\stackrel{~}{Q}_0\stackrel{~}{Q}_1^{(2)}(x)\right\}.$$
(2.8)
In the later half of this section, we briefly recall our tools, i.e. partitions and Pfaffians. We follow the notation in concerning partitions and the symmetric functions. For more information about the general theory of determinants and Pfaffains, the reader can consult , and since, in this paper, we sometimes omit the details and give sketches of proofs.
Let $`n`$ be a non-negative integer and assume we are given a $`2n`$ by $`2n`$ skew-symmetric matrix $`A=(a_{ij})_{1i,j2n}`$, (i.e. $`a_{ji}=a_{ij}`$), whose entries $`a_{ij}`$ are in a commutative ring. The Pfaffian of $`A`$ is, by definition,
$$\mathrm{Pf}(A)=ฯต(\sigma _1,\sigma _2,\mathrm{},\sigma _{2n1},\sigma _{2n})a_{\sigma _1\sigma _2}\mathrm{}a_{\sigma _{2n1}\sigma _{2n}}.$$
where the summation is over all partitions $`\{\{\sigma _1,\sigma _2\}_<,\mathrm{},\{\sigma _{2n1},\sigma _{2n}\}_<\}`$ of $`[2n]`$ into $`2`$-elements blocks, and where $`ฯต(\sigma _1,\sigma _2,\mathrm{},\sigma _{2n1},\sigma _{2n})`$ denotes the sign of the permutation
$$\left(\begin{array}{cccc}1& 2& \mathrm{}& 2n\\ \sigma _1& \sigma _2& \mathrm{}& \sigma _{2n}\end{array}\right).$$
We call a partition $`\sigma =\{\{\sigma _1,\sigma _2\}_<,\mathrm{},\{\sigma _{2n1},\sigma _{2n}\}_<\}`$ of $`[2n]`$ into $`2`$-elements blocks a perfect matching or $`1`$-factor of $`[2n]`$, and let $`_n`$ denote the set of all perfect matchings of $`[2n]`$. We represent a perfect matching $`\sigma `$ graphically by embedding the points $`i[2n]`$ along the $`x`$-axis in the coordinate plane and representing each block $`\{\sigma _{2i1},\sigma _{2i}\}_<`$ by the curve connecting $`\sigma _{2i1}`$ to $`\sigma _{2i}`$ in the upper half plane. For instance, the graphical representation of $`\sigma =\{\{1,4\},\{2,5\},\{3,6\}\}`$ is the Figure 1 bellow.
If we write $`\mathrm{wt}(\sigma )=ฯต(\sigma )_{i=1}^na_{\sigma _{2i1}\sigma _{2i}}`$ for each perfect matching $`\sigma `$, then we can restate our definition as
$$\mathrm{Pf}(A)=\underset{\sigma _n}{}\mathrm{wt}(\sigma ).$$
(2.9)
A skew-symmetric matrix $`A=(a_{ij})_{1i,jn}`$ is uniquely determined by its upper triangular entries $`(a_{ij})_{1i<jn}`$. So we sometimes define a skew-symmetric matrix by describing its upper triangular entries.
Let $`O_{m,n}`$ denote the $`m\times n`$ zero matrix and let $`E_n`$ denote the identity matrix $`(\delta _{ij})_{1i<jn}`$ of size $`n`$. Here $`\delta _{ij}`$ denotes the Kronecker delta. We use the abbreviation $`O_n`$ for $`O_{n,n}`$.
For any finite set $`S`$ and any nonnegative integer $`r`$, let $`\left(\genfrac{}{}{0pt}{}{S}{r}\right)`$ denote the set of all $`r`$-element subsets of $`S`$. For example, $`\left(\genfrac{}{}{0pt}{}{[n]}{r}\right)`$ stands for the set of all multi-indices $`\{i_1,\mathrm{},i_r\}`$ such that $`1i_1<\mathrm{}<i_rn`$. Let $`m`$, $`n`$ and $`r`$ be integers such that $`rm,n`$ and let $`T`$ be an $`m`$ by $`n`$ matrix. For any index sets $`I=\{i_1,\mathrm{},i_r\}\left(\genfrac{}{}{0pt}{}{[m]}{r}\right)`$ and $`J=\{j_1,\mathrm{},j_r\}\left(\genfrac{}{}{0pt}{}{[n]}{r}\right)`$, let $`\mathrm{\Delta }_J^I(A)`$ denote the submatrix obtained by selecting the rows indexed by $`I`$ and the columns indexed by $`J`$. If $`r=m`$ and $`I=[m]`$, we simply write $`\mathrm{\Delta }_J(A)`$ for $`\mathrm{\Delta }_J^{[m]}(A)`$. Similarly, if $`r=n`$ and $`J=[n]`$, we write $`\mathrm{\Delta }^I(A)`$ for $`\mathrm{\Delta }_{[n]}^I(A)`$. It is essential that the weight $`\omega (\lambda )`$ can be expressed by a Pfaffain, which is a fact proved in :
###### Theorem 2.1.
Let $`n`$ be a non-negative integer. Let $`\lambda =(\lambda _1,\mathrm{},\lambda _{2n})`$ be a partition such that $`\mathrm{}(\lambda )2n`$, and put $`l=(l_1,\mathrm{},l_{2n})=\lambda +\delta _{2n}`$. Define a skew-symmetric matrix $`A=(\alpha _{ij})_{i,j0}`$ by
$$\alpha _{ij}=a^{(j1)/2}b^{(j1)/2}c^{i/2}d^{i/2}$$
for $`i<j`$. Then we have
$$\mathrm{Pf}\left[\mathrm{\Delta }_{I(\lambda )}^{I(\lambda )}\left(A\right)\right]_{1i,j2n}=(abcd)^{\left(\genfrac{}{}{0pt}{}{n}{2}\right)}\omega (\lambda ),$$
where $`I(\lambda )=\{l_{2n},\mathrm{},l_1\}`$.
A variation of this theorem for strict partitions is as follows.
###### Theorem 2.2.
Let $`n`$ be a nonnegative integer. Let $`\mu =(\mu _1,\mathrm{},\mu _n)`$ be a strict partition such that $`\mu _1>\mathrm{}>\mu _n0`$. Let $`K(\mu )=\{\mu _n,\mathrm{},\mu _1\}`$. Define a skew-symmetric matrix $`B=(\beta _{ij})_{i,j1}`$ by
$$\beta _{ij}=\{\begin{array}{cc}1,\hfill & \text{ if }i=1\text{ and }j=0\text{,}\hfill \\ a^{j/2}b^{j/2}z,\hfill & \text{ if }i=1\text{ and }j1\text{,}\hfill \\ a^{j/2}b^{j/2}z\hfill & \text{ if }i=0\text{,}\hfill \\ a^{j/2}b^{j/2}c^{i/2}d^{i/2}z^2,\hfill & \text{ if }i>0\text{,}\hfill \end{array}$$
(2.10)
for $`1i<j`$.
1. If $`n`$ is even, then we have
$$\mathrm{Pf}\left[\mathrm{\Delta }_{K(\mu )}^{K(\mu )}\left(B\right)\right]=\omega (\mu )z^{\mathrm{}(\mu )}.$$
(2.11)
2. If $`n`$ is odd, then we have
$$\mathrm{Pf}\left[\mathrm{\Delta }_{\{1\}K(\mu )}^{\{1\}K(\mu )}\left(B\right)\right]=\omega (\mu )z^{\mathrm{}(\mu )}.\mathrm{}$$
(2.12)
These theorems are easy consequences of the following Lemma which has been proved in \[8, Section 4, Lemma 7\].
###### Lemma 2.3.
Let $`x_i`$ and $`y_j`$ be indeterminates, and let $`n`$ is a non-negative integer. Then
$$\mathrm{Pf}[x_iy_j]_{1i<j2n}=\underset{i=1}{\overset{n}{}}x_{2i1}\underset{i=1}{\overset{n}{}}y_{2i}.\mathrm{}$$
## 3 Strict Partitions
A partition $`\mu `$ is strict if all its parts are distinct. One represents the associated shifted diagram of $`\mu `$ as a diagram in which the $`i`$th row from the top has been shifted to the right by $`i`$ places so that the first column becomes a diagonal. A strict partition can be written uniquely in the form $`\mu =(\mu _1,\mathrm{},\mu _{2n})`$ where $`n`$ is an non-negative integer and $`\mu _1>\mu _2>\mathrm{}>\mu _{2n}0`$. The length $`\mathrm{}(\mu )`$ is, by definition, the number of nonzero parts of $`\mu `$. We define the weight function $`\omega (\mu )`$ exactly the same as in 1.1. For example, if $`\mu =(8,5,3)`$, then $`\mathrm{}(\mu )=3`$, $`\omega (\mu )=a^6b^5c^3d^2`$ and its shifted diagram is as follows.
Let
$$\mathrm{\Psi }_N=\mathrm{\Psi }_N(a,b,c,d;z)=\omega (\mu )z^{\mathrm{}(\mu )},$$
(3.1)
where the sum is over all strict partitions $`\mu `$ such that each part of $`\mu `$ is less than or equal to $`N`$. For example, we have
$`\mathrm{\Psi }_0`$ $`=1,`$
$`\mathrm{\Psi }_1`$ $`=1+az,`$
$`\mathrm{\Psi }_2`$ $`=1+a(1+b)z+abcz^2,`$
$`\mathrm{\Psi }_3`$ $`=1+a(1+b+ab)z+abc(1+a+ad)z^2+a^3bcdz^3.`$
In fact, the only strict partition such that $`\mathrm{}(\mu )=0`$ is $`\mathrm{}`$, the strict partitions $`\mu `$ such that $`\mathrm{}(\mu )=1`$ and $`\mu _13`$ are the following three:
a
a
b
a
b
a
,
a
a
b
a
b
a
\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt\\}}\ ,
the strict partitions $`\mu `$ such that $`\mathrm{}(\mu )=2`$ and $`\mu _13`$ are the following three:
a
b
c
a
b
a
c
a
b
a
c
d
,
a
b
c
a
b
a
c
a
b
a
c
d
\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$d$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\ ,
and the strict partition $`\mu `$ such that $`\mathrm{}(\mu )=3`$ and $`\mu _13`$ is the following one:
a
b
a
c
d
a
.
a
b
a
c
d
a
\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$d$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\hskip 12.0pt&\hskip 12.0pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\ .
The sum of the weights of these strict partitions is equal to $`\mathrm{\Psi }_3`$. In this section we always assume $`|a|,|b|,|c|,|d|<1`$. One of the main results of this section is that the even terms and the odd terms of $`\mathrm{\Psi }_N`$ respectively satisfy the associated Al-Salam-Chihara recurrence relation:
###### Theorem 3.1.
Set $`q=abcd`$. Let $`\mathrm{\Psi }_N=\mathrm{\Psi }_N(a,b,c,d;z)`$ be as in 3.1 and put $`X_N=\mathrm{\Psi }_{2N}`$ and $`Y_N=\mathrm{\Psi }_{2N+1}`$. Then $`X_N`$ and $`Y_N`$ satisfy
$`X_{N+1}`$ $`=\left\{1+ab+a(1+bc)z^2q^N\right\}X_N`$
$`ab(1z^2q^N)(1acz^2q^{N1})X_{N1},`$ (3.2)
$`Y_{N+1}`$ $`=\left\{1+ab+abc(1+ad)z^2q^N\right\}Y_N`$
$`ab(1z^2q^N)(1acz^2q^N)Y_{N1},`$ (3.3)
where $`X_0=1`$, $`Y_0=1+az`$, $`X_1=1+a(1+b)z+abcz^2`$ and
$$Y_1=1+a(1+b+ab)z+abc(1+a+ad)z^2+a^3bcdz^3.$$
Especially, if we put $`X_N^{}=(ab)^{\frac{N}{2}}X_N`$ and $`Y_N^{}=(ab)^{\frac{N}{2}}Y_N`$, then $`X_N^{}`$ and $`Y_N^{}`$ satisfy
$`\left\{(ab)^{\frac{1}{2}}+(ab)^{\frac{1}{2}}\right\}X_N^{}`$ $`=X_{N+1}^{}a^{\frac{1}{2}}b^{\frac{1}{2}}(1+bc)z^2q^NX_N^{}`$
$`+(1z^2q^N)(1acz^2q^{N1})X_{N1}^{},`$ (3.4)
$`\left\{(ab)^{\frac{1}{2}}+(ab)^{\frac{1}{2}}\right\}Y_N^{}`$ $`=Y_{N+1}^{}a^{\frac{1}{2}}b^{\frac{1}{2}}c(1+ad)z^2q^NY_N^{}`$
$`+(1z^2q^N)(1a^2bc^2dz^2q^{N1})Y_{N1}^{},`$ (3.5)
where $`X_0^{}=1`$, $`Y_0^{}=1+az`$, $`X_1^{}=(ab)^{\frac{1}{2}}+a^{\frac{1}{2}}b^{\frac{1}{2}}(1+b)z+(ab)^{\frac{1}{2}}cz^2`$ and
$$Y_1^{}=(ab)^{\frac{1}{2}}+a^{\frac{1}{2}}b^{\frac{1}{2}}(1+b+ab)z+a^{\frac{1}{2}}b^{\frac{1}{2}}c(1+a+ad)z^2+a^{\frac{5}{2}}b^{\frac{1}{2}}cdz^3.$$
Thus 3.4 agrees with the associated Al-Salam-Chihara recurrence relation 2.2 where $`u=a^{\frac{1}{2}}b^{\frac{1}{2}}`$, $`\alpha =a^{\frac{1}{2}}b^{\frac{1}{2}}c`$, $`\beta =a^{\frac{1}{2}}b^{\frac{1}{2}}`$ and $`t=z^2`$, and 3.5 also agrees with 2.2 where $`u=a^{\frac{1}{2}}b^{\frac{1}{2}}`$, $`\alpha =a^{\frac{1}{2}}b^{\frac{1}{2}}c`$, $`\beta =a^{\frac{3}{2}}b^{\frac{1}{2}}cd`$ and $`t=z^2`$. One concludes that, when $`|a|,|b|,|c|,|d|<1`$, the solutions of 3.2 and 3.3 are expressed by the linear combinations of 2.3 and 2.4 as follows.
###### Theorem 3.2.
Assume $`|a|,|b|,|c|,|d|<1`$ and set $`q=abcd`$. Let $`\mathrm{\Psi }_N=\mathrm{\Psi }_N(a,b,c,d;z)`$ be as in 3.1.
1. Put $`X_N=\mathrm{\Psi }_{2N}`$. Then we have
$`X_N`$ $`={\displaystyle \frac{(az^2q,abc;q)_{\mathrm{}}}{(a,abcz^2;q)_{\mathrm{}}}}`$
$`\times \{(s_0^XX_1s_1^XX_0)(abcz^2;q)_N{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^Nz^2,b^1}{(abc)^1q^{N+1}z^2}};q,c^1q)`$
$`+(r_1^XX_0r_0^XX_1)(ab)^N{\displaystyle \frac{(qz^2,acz^2;q)_N}{(aqz^2;q)_N}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^{N+1}z^2,c^1q}{aq^{N+1}z^2}};q,abc)\},`$ (3.6)
where
$`r_0^X`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2,b^1}{(abc)^1z^2q}};q,c^1q),`$
$`s_0^X`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2q,c^1q}{az^2q}};q,abc),`$
$`r_1^X`$ $`=(1+abcz^2){}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2q^1,b^1}{(abc)^1z^2}};q,c^1q),`$
$`s_1^X`$ $`={\displaystyle \frac{ab(1z^2q)(1acz^2)}{1+az^2q}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2q^2,c^1q}{az^2q^2}};q,abc).`$
2. Put $`Y_N=\mathrm{\Psi }_{2N+1}`$. Then we have
$`Y_N`$ $`={\displaystyle \frac{(aq^2z^2,abc;q)_{\mathrm{}}}{(aq,abcz^2;q)_{\mathrm{}}}}`$
$`\times \{(s_0^YY_1s_1^YY_0)(abcz^2;q)_N{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^Nz^2,acd}{(abc)^1q^{N+1}z^2}};q,c^1q)`$
$`+(r_1^YY_0r_0^YY_1)(ab)^N{\displaystyle \frac{(qz^2,acqz^2;q)_N}{(aq^2z^2;q)_N}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^{N+1}z^2,c^1q}{aq^{N+2}z^2}};q,abc)\},`$ (3.7)
where
$`r_0^Y`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2,acd}{(abc)^1qz^2}};q,c^1q),`$
$`r_1^Y`$ $`=(1+abcz^2){}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^1z^2,acd}{(abc)^1z^2}};q,c^1q),`$
$`s_0^Y`$ $`={}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2q,c^1q}{aq^2z^2}};q,abc),`$
$`s_1^Y`$ $`={\displaystyle \frac{ab(1z^2q)(1acqz^2)}{1+aq^2z^2}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{z^2q^2,c^1q}{aq^3z^2}};q,abc).`$
If we take the limit $`N\mathrm{}`$ in 3.6 and 3.7, then by using 2.8, we obtain the following generalization of Bouletโs result (see Corollary 3.6).
###### Corollary 3.3.
Assume $`|a|,|b|,|c|,|d|<1`$ and set $`q=abcd`$. Let $`s_i^X`$, $`s_i^Y`$, $`X_i`$, $`Y_i`$ ($`i=0,1`$) be as in the above theorem. Then we have
$`{\displaystyle \underset{\mu }{}}\omega (\mu )z^{\mathrm{}(\mu )}`$ $`={\displaystyle \frac{(abc,az^2q;q)_{\mathrm{}}}{(ab;q)_{\mathrm{}}}}(s_0^XX_1s_1^XX_0)`$
$`={\displaystyle \frac{(abc,az^2q^2;q)_{\mathrm{}}}{(ab;q)_{\mathrm{}}}}(s_0^YY_1s_1^YY_0),`$ (3.8)
where the sum runs over all strict partitions and the first terms are as follows:
$`1+{\displaystyle \frac{a(1+b)}{1ab}}z+{\displaystyle \frac{abc(1+a+ad+abd)}{(1ab)(1q)}}z^2+{\displaystyle \frac{a^2q(1+b)(1+bc+abc+bq)}{(1ab)(1q)(1abq)}}z^3+O(z^4).`$
On the other hand, by plugging $`z=1`$ into 3.6 and 3.7, we conclude that the solutions of the recurrence relations 3.4 and 3.5 with the above initial condition are exactly the Al-Salam-Chihara polynomials, which give two finite versions of Bouletโs result.
###### Corollary 3.4.
Put $`u=\sqrt{ab}`$, $`x=\frac{u+u^1}{2}`$ and $`q=abcd`$. Let $`\mathrm{\Psi }_N(a,b,c,d;z)`$ be as in 3.1.
1. The polynomial $`\mathrm{\Psi }_{2N}(a,b,c,d;1)`$ is given by
$`\mathrm{\Psi }_{2N}(a,b,c,d;1)`$ $`=(ab)^{\frac{N}{2}}Q_N(x;a^{\frac{1}{2}}b^{\frac{1}{2}}c,a^{\frac{1}{2}}b^{\frac{1}{2}}|q),`$
$`=(a;q)_N{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^N,c}{a^1q^{N+1}}};q,bq).`$ (3.9)
2. The polynomial $`\mathrm{\Psi }_{2N+1}(a,b,c,d;1)`$ is given by
$`\mathrm{\Psi }_{2N+1}(a,b,c,d;1)`$ $`=(1+a)(ab)^{\frac{N}{2}}Q_N(x;a^{\frac{1}{2}}b^{\frac{1}{2}}c,a^{\frac{3}{2}}b^{\frac{1}{2}}cd|q)`$
$`=(a;q)_{N+1}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^N,c}{a^1q^N}};q,b).`$ (3.10)
Substituting $`a=zyq`$, $`b=z^1yq`$, $`c=zy^1q`$ and $`d=z^1y^1q`$ into Theorem 3.4(see ), then we immediately obtain the strict version of Andrewsโ result (see Corollary 4.4).
###### Corollary 3.5.
$$\underset{\genfrac{}{}{0pt}{}{\mu \text{ strict partitions}}{\mu _12N}}{}z^{๐ช(\mu )}y^{๐ช(\mu ^{})}q^{|\mu |}=\underset{j=0}{\overset{N}{}}\left[\genfrac{}{}{0pt}{}{N}{j}\right]_{q^4}(zyq;q^4)_j(zy^1q;q^4)_{Nj}(yq)^{2N2j},$$
(3.11)
and
$$\underset{\genfrac{}{}{0pt}{}{\mu \text{ strict partitions}}{\mu _12N+1}}{}z^{๐ช(\mu )}y^{๐ช(\mu ^{})}q^{|\mu |}=\underset{j=0}{\overset{N}{}}\left[\genfrac{}{}{0pt}{}{N}{j}\right]_{q^4}(zyq;q^4)_{j+1}(zy^1q;q^4)_{Nj}(yq)^{2N2j},$$
(3.12)
where
$$\left[\genfrac{}{}{0pt}{}{N}{j}\right]_q=\{\begin{array}{cc}\frac{(1q^N)(1q^{N1})\mathrm{}(1q^{Nj+1})}{(1q^j)(1q^{j1})\mathrm{}(1q)},\hfill & \text{ for }0jN\text{,}\hfill \\ 0,\hfill & \text{ if }j<0\text{ and }j>N\text{.}\hfill \end{array}$$
Letting $`N\mathrm{}`$ in Theorem 3.4 or setting $`z=1`$ in 3.8, we obtain the following result of Boulet (cf. \[2, Corollary 2\]).
###### Corollary 3.6.
(Boulet) Let $`q=abcd`$, then
$`{\displaystyle \underset{\mu }{}}\omega (\mu )={\displaystyle \frac{(a;q)_{\mathrm{}}(abc;q)_{\mathrm{}}}{(ab;q)_{\mathrm{}}}},`$ (3.13)
where the sum runs over all strict partitions.
To prove Theorem 3.1, we need several steps. Our strategy is as follows: write the weight $`\omega (\mu )z^{\mathrm{}(\mu )}`$ as a Pfaffain (Theorem 2.2) and apply the minor summation formula (Lemma 3.7) to make the sum of the weights into a single Pfaffian (Theorem 3.8). Then we make use of the Pfaffian to derive a recurrence relation (Proposition 3.9). We also give another proof of the recurrence relation by a combinatorial argument (Remark 3.10).
Let $`J_n`$ denote the square matrix of size $`n`$ whose $`(i,j)`$th entry is $`\delta _{i,n+1j}`$. We simply write $`J`$ for $`J_n`$ when there is no fear of confusion on the size $`n`$. We need the following result on a sum of Pfaffians \[18, Theorem of Section 4\].
###### Lemma 3.7.
Let $`n`$ be a positive integer. Let $`A=(a_{ij})_{1i,jn}`$ and $`B=(b_{ij})_{1i,jn}`$ be skew symmetric matrices of size $`n`$. Then
$`{\displaystyle \underset{t=0}{\overset{n/2}{}}}z^t{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2t}\right)}{}}\gamma ^{|I|}\mathrm{Pf}\left(\mathrm{\Delta }_I^I(A)\right)\mathrm{Pf}\left(\mathrm{\Delta }_I^I(B)\right)=\mathrm{Pf}\left[\begin{array}{cc}J_n{}_{}{}^{t}AJ_n& J_n\\ J_n& C\end{array}\right],`$ (3.14)
where $`|I|=_{iI}i`$ and $`C=(C_{ij})_{1i,jn}`$ is given by $`C_{ij}=\gamma ^{i+j}b_{ij}z`$.
This lemma is a special case of Lemma 5.4, so a proof will be given later.
Let $`S_n`$ denote the $`n\times n`$ skew-symmetric matrix whose $`(i,j)`$th entry is $`1`$ for $`0i<jn`$. As a corollary of Lemma 3.7, we obtain the following expression of the sum of the weight $`\omega (\mu )`$ by a single Pfaffian.
###### Theorem 3.8.
Let $`N`$ be a nonnegative integer.
$`\mathrm{\Psi }_N(a,b,c,d;z)=\mathrm{Pf}\left[\begin{array}{cc}S_{N+1}& J_{N+1}\\ J_{N+1}& B\end{array}\right],`$ (3.15)
where $`B=(\beta _{ij})_{0i<jN}`$ is the $`(N+1)\times (N+1)`$ skew-symmetric matrix whose $`(i,j)`$th entry $`\beta _{ij}`$ is defined in 2.10.
* By Theorem 2.2, we have
$$\mathrm{\Psi }_N(a,b,c,d;z)=\underset{t=0}{\overset{(N+1)/2}{}}\underset{\genfrac{}{}{0pt}{}{\mu =(\mu _1,\mathrm{},\mu _{2t})}{N\mu _1>\mathrm{}>\mu _{2t}0}}{}\mathrm{Pf}\left(\mathrm{\Delta }_{K(\mu )}^{K(\mu )}\left(B\right)\right).$$
If we take $`S=(1)_{0i<jN}`$, then we have $`\mathrm{Pf}\left(\mathrm{\Delta }_I^I(S)\right)=1`$ for any subset $`I[0,N]`$ of even cardinality. (For detailed arguments on sub-pfaffains, see ). Thus 3.15 follows from Lemma 3.7. $`\mathrm{}`$
For example, if $`N=3`$, then the Pfaffian in the right-hand side of 3.15 looks
$$\mathrm{Pf}\left[\begin{array}{cccccccc}0& 1& 1& 1& 0& 0& 0& 1\\ 1& 0& 1& 1& 0& 0& 1& 0\\ 1& 1& 0& 1& 0& 1& 0& 0\\ 1& 1& 1& 0& 1& 0& 0& 0\\ & & & & & & & \\ 0& 0& 0& 1& 0& az& abz& a^2bz\\ 0& 0& 1& 0& az& 0& abcz^2& a^2bcz^2\\ 0& 1& 0& 0& abz& abcz^2& 0& a^2bcdz^2\\ 1& 0& 0& 0& a^2bz& a^2bcz^2& a^2bcdz^2& 0\end{array}\right],$$
and this is equal to $`1+a(1+b+ab)z+abc(1+a+ad)z^2+a^3bcdz^3`$.
By performing elementary transformations on rows and columns of the matrix, we obtain the following recurrence relation:
###### Proposition 3.9.
Let $`\mathrm{\Psi }_N=\mathrm{\Psi }_N(a,b,c,d;z)`$ be as above. Then we have
$`\mathrm{\Psi }_{2N}=(1+b)\mathrm{\Psi }_{2N1}+(a^Nb^Nc^Nd^{N1}z^2b)\mathrm{\Psi }_{2N2},`$ (3.16)
$`\mathrm{\Psi }_{2N+1}=(1+a)\mathrm{\Psi }_{2N}+(a^{N+1}b^Nc^Nd^Nz^2a)\mathrm{\Psi }_{2N1},`$ (3.17)
for any positive integer $`N`$.
* Let $`A`$ denote the $`2(N+1)\times 2(N+1)`$ skew symmetric matrix $`\left[\begin{array}{cc}S_{N+1}& J_{N+1}\\ J_{N+1}& B\end{array}\right]`$ as on the right-hand side of 3.15. Here we assume row/column indices start at 0. So, for example, the row indices for the upper $`(N+1)`$ rows are $`i`$, $`i=0,\mathrm{},N`$, and the row indices for the lower $`(N+1)`$ rows are $`i+N+1`$, $`i=0,\mathrm{},N`$. Now, subtract $`a`$ times $`(j+N)`$th column from $`(j+N+1)`$th column if $`j`$ is odd, or subtract $`b`$ times $`(j+N)`$th column from $`(j+N+1)`$th column if $`j`$ is even, for $`j=N,N1,\mathrm{},1`$. To make our matrix skew-symmetric, subtract $`a`$ times $`(i+N)`$th row from $`(i+N+1)`$th row if $`i`$ is odd, or subtract $`b`$ times $`(i+N)`$th row from $`(i+N+1)`$th row if $`i`$ is even, for $`i=N,N1,\mathrm{},1`$. Next subtract $`(i+1)`$th row from $`i`$th row for $`i=0,1,\mathrm{},N1`$, then we also subtract $`(j+1)`$th column from $`j`$th column for $`j=0,1,\mathrm{},N1`$. By these transformations, we obtain a skew symmetric matrix $`A^{}=\left[\begin{array}{cc}P& Q\\ {}_{}{}^{t}Q& R\end{array}\right]`$, where $`P=(\delta _{i+1,j})_{0i<jN}`$, $`Q=(q_{ij})_{0i<jN}`$ and $`R=(r_{ij})_{0i<jN}`$ are given by
$`q_{ij}=\{\begin{array}{cc}1\hfill & \text{ if }i+j=N1\text{,}\hfill \\ 1\hfill & \text{ if }i=N\text{ and }j=0\text{,}\hfill \\ 1+a^{\chi (j\text{ is odd})}b^{\chi (j\text{ is even})}\hfill & \text{ if }i+j=N\text{ and }j1\text{,}\hfill \\ a^{\chi (j\text{ is odd})}b^{\chi (j\text{ is even})}\hfill & \text{ if }i+j=N+1\text{,}\hfill \\ 0\hfill & \text{ otherwise,}\hfill \end{array}`$
$`r_{ij}=\{\begin{array}{cc}az\delta _{1,j}\hfill & \text{ if }i=0\text{,}\hfill \\ a^{(i+1)/2}b^{(i+1)/2}c^{i/2}d^{i/2}z^2\delta _{i+1,j}\hfill & \text{ if }i>0\text{.}\hfill \end{array}`$
Here $`\chi (A)`$ stands for $`1`$ if the statement $`A`$ is true and $`0`$ otherwise. For example, if $`N=3`$, then $`A^{}`$ looks as follows:
$$\left[\begin{array}{cccccccc}0& 1& 0& 0& 0& 0& 1& 1+a\\ 1& 0& 1& 0& 0& 1& 1+b& a\\ 0& 1& 0& 1& 1& 1+a& b& 0\\ 0& 0& 1& 0& 1& a& 0& 0\\ & & & & & & & \\ 0& 0& 1& 1& 0& az& 0& 0\\ 0& 1& 1a& a& az& 0& abcz^2& 0\\ 1& 1b& b& 0& 0& abcz^2& 0& a^2bcdz^2\\ 1a& a& 0& 0& 0& 0& a^2bcdz^2& 0\end{array}\right].$$
By expanding $`\mathrm{Pf}(A^{})`$ along the first row/column, we obtain the desired formula. $`\mathrm{}`$
###### Remark 3.10.
Proposition 3.9 can be also proved by a combinatorial argument as follows.
* By definition, the generating function for strict partitions $`\mu =(\mu _1,\mu _2,\mathrm{})`$ such that $`\mu _1=2N`$ and $`\mu _22N2`$ is equal to
$$b(\mathrm{\Psi }_{2N1}\mathrm{\Psi }_{2N2}).$$
That for strict partitions such that $`\mu _1=2N`$ and $`\mu _2=2N1`$ is equal to
$$a^Nb^Nc^Nd^{N1}z^2\mathrm{\Psi }_{2N2}.$$
Finally the generating function of strict partitions such that $`\mu _12N1`$ is equal to $`\mathrm{\Psi }_{2N1}`$. Summing up we get 3.16. The same argument works to prove 3.17. $`\mathrm{}`$
Note that one can immediately derive Theorem 3.1 from Proposition 3.9 by substitution. Thus, if one use 2.7, then he immediately derive Theorem 3.2 by a simple computation.
* Let $`u=\sqrt{ab}`$, $`t=z^2`$ and $`q=abcd`$. By 3.4, $`X_N^{}`$ satisfies the associated Al-Salam-Chihara recurrence relation 2.2 with $`\alpha =a^{\frac{1}{2}}b^{\frac{1}{2}}c`$ and $`\beta =a^{\frac{1}{2}}b^{\frac{1}{2}}`$. Note that $`|u|<1`$ and $`|q|<|\alpha |<1`$ hold. Thus, by 2.7, we conclude that $`X_N`$ is given by 3.6. A similar argument shows that $`Y_N^{}`$ satisfies 2.2 with $`\alpha =a^{\frac{3}{2}}b^{\frac{1}{2}}c`$ and $`\beta =a^{\frac{1}{2}}b^{\frac{1}{2}}cd`$, which implies $`Y_N`$ is given by 3.7. $`\mathrm{}`$
* First, if we set $`z=1`$ in 3.6, we obtain that $`s_0^XX_1s_1^XX_0=1+a`$ and $`r_1^XX_0r_0^XX_1=0`$ which immediately imply
$$X_N=(abc;q)_N{}_{2}{}^{}\varphi _{1}^{}(\genfrac{}{}{0pt}{}{q^N,b^1}{(abc)^1q^{N+1}};q,c^1q).$$
In fact, it is easy to see that, when $`z=1`$, $`r_0^X=0`$ and $`r_1^X=1+abc+a(1+b)`$, which immediately implies $`r_1^XX_0r_0^XX_1=0`$. A similar computation shows that, if we put $`z=1`$, we have
$`s_0^X`$ $`={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(c^1q;q)_n}{(aq;q)_n}}(abc)^n,`$
$`s_1^X`$ $`=ab(1ac){\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(1q^{n+1})(c^1q;q)_n}{(aq;q)_{n+1}}}(abc)^n.`$
If we use these equalities, then we obtain
$`s_0^XX_1s_1^XX_0`$ $`=(1+a){\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(c^1q;q)_n}{(aq;q)_{n+1}}}(abc)^n\{a+abc+a(1+b)q^{n+1}\}`$
$`=(1+a)\left\{{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(c^1q;q)_n}{(aq;q)_n}}(abc)^n{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(c^1q;q)_{n+1}}{(aq;q)_{n+1}}}(abc)^{n+1}\right\}.`$
Thus the right-hand side equals $`1`$, and this proves 3.9. By a similar argument we can derive 3.10 from 3.7. The details are left to the reader. $`\mathrm{}`$
* We first claim that
$$\mathrm{\Psi }_{2N}(a,b,c,d;1)=\underset{k=0}{\overset{N}{}}\left[\genfrac{}{}{0pt}{}{N}{k}\right]_q(a;q)_k(c;q)_{Nk}(ab)^{Nk}.$$
(3.18)
Then 3.11 is an easy consequence of 3.18 by substituting $`azyq`$, $`bz^1yq`$, $`czy^1q`$ and $`dz^1y^1q`$. In fact, using $`(q^N;q)_k=\frac{(q;q)_N}{(q;q)_{Nk}}(1)^kq^{\left(\genfrac{}{}{0pt}{}{k}{2}\right)Nk}`$, we have
$`{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^N,c}{a^1q^{N+1}}};q,bq)={\displaystyle \underset{k=0}{\overset{N}{}}}\left[{\displaystyle \genfrac{}{}{0pt}{}{N}{k}}\right]_q{\displaystyle \frac{(c;q)_{Nk}}{(a^1q^{N+1};q)_{Nk}}}q^{\left(\genfrac{}{}{0pt}{}{Nk}{2}\right)N(Nk)}(bq)^{Nk}.`$
Substitute $`(a^1q^{N+1};q)_{Nk}=\frac{(a;q)_N}{(a;q)_k}a^{N+k}q^{\left(\genfrac{}{}{0pt}{}{N}{2}\right)+\left(\genfrac{}{}{0pt}{}{k}{2}\right)}`$ into this identity to show that the right-hand side equals
$`{\displaystyle \underset{k=0}{\overset{N}{}}}\left[{\displaystyle \genfrac{}{}{0pt}{}{N}{k}}\right]_q{\displaystyle \frac{(a;q)_k(c;q)_{Nk}}{(a;q)_N}}(ab)^{Nk}.`$
Finally, use 3.9 to obtain 3.18. The proof of 3.12 reduces to
$$\mathrm{\Psi }_{2N+1}(a,b,c,d;1)=\underset{k=0}{\overset{N}{}}\left[\genfrac{}{}{0pt}{}{N}{k}\right]_q(a;q)_{k+1}(c;q)_{Nk}(ab)^{Nk},$$
(3.19)
which is derived from 3.10 similarly. $`\mathrm{}`$
* By replacing $`k`$ by $`Nk`$ and letting $`N`$ to $`+\mathrm{}`$, we get
$`\underset{N\mathrm{}}{lim}\mathrm{\Psi }_{2N}(a,b,c,d;1)=(a;q)_{\mathrm{}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(c;q)_k}{(q;q)_k}}(ab)^k={\displaystyle \frac{(a;q)_{\mathrm{}}(abc;q)_{\mathrm{}}}{(ab;q)_{\mathrm{}}}}`$
where the last equality follows from the $`q`$-binomial formula (see ). Similarly we can derive the limit from 3.19.
Note that we can also derive 3.13 from 3.8 by the same argument as in the proof of Theorem 3.4. $`\mathrm{}`$
## 4 Ordinary Partitions
First we present a generalization of Andrewsโ result in . Let us consider
$$\mathrm{\Phi }_N=\mathrm{\Phi }_N(a,b,c,d;z)=\underset{\genfrac{}{}{0pt}{}{\lambda }{\lambda _1N}}{}\omega (\lambda )z^{\mathrm{}(\lambda )},$$
(4.1)
where the sum runs over all partitions $`\lambda `$ such that each part of $`\lambda `$ is less than or equal to $`N`$. For example, the first few terms can be computed directly as follows:
$`\mathrm{\Phi }_0=1,`$
$`\mathrm{\Phi }_1={\displaystyle \frac{1+az}{1acz^2}},`$
$`\mathrm{\Phi }_2={\displaystyle \frac{1+a(1+b)z+abcz^2}{(1acz^2)(1qz^2)}},`$
$`\mathrm{\Phi }_3={\displaystyle \frac{1+a(1+b+ab)z+abc(1+a+ad)z^2+a^3bcdz^3}{(1z^2ac)(1z^2q)(1z^2acq)}},`$
where $`q=abcd`$ as before. If one compares these with the first few terms of $`\mathrm{\Psi }_N`$, one can easily guess the following theorem holds:
###### Theorem 4.1.
For non-negative integer $`N`$, let $`\mathrm{\Phi }_N=\mathrm{\Phi }_N(a,b,c,d;z)`$ be as in 4.1 and $`q=abcd`$. Then we have
$`\mathrm{\Phi }_N(a,b,c,d;z)={\displaystyle \frac{\mathrm{\Psi }_N(a,b,c,d;z)}{(z^2q;q)_{N/2}(z^2ac;q)_{N/2}}},`$ (4.2)
where $`\mathrm{\Psi }_N=\mathrm{\Psi }_N(a,b,c,d;z)`$ is the generating function defined in 3.1. Note that $`\mathrm{\Psi }_N`$ is explicitly given in terms of basic hypergeometric functions in Theorem 3.2.
In fact, the main purpose of this section is to prove this theorem. Here we give two proofs, i.e. an algebraic proof (see Proposition 4.6 and Proposition 4.7) and a bijective proof (see Remark 4.8). Before we proceed to the proofs of this theorem we state the corollaries immediately obtained from this theorem and the results in Section 3. First of all, as an immediate corollary of Theorem 4.1 and Corollary 3.3, we obtain the following generalization of Bouletโs result (Corollary 4.5).
###### Corollary 4.2.
Assume $`|a|,|b|,|c|,|d|<1`$ and set $`q=abcd`$. Let $`s_i^X`$, $`s_i^Y`$, $`X_i`$, $`Y_i`$ ($`i=0,1`$) be as in Theorem 3.2. Then we have
$`{\displaystyle \underset{\lambda }{}}\omega (\lambda )z^{|\mu |}`$ $`={\displaystyle \frac{(abc,az^2q;q)_{\mathrm{}}}{(ab,acz^2,z^2q;q)_{\mathrm{}}}}(s_0^XX_1s_1^XX_0)`$
$`={\displaystyle \frac{(abc,a^2bcdz^2q;q)_{\mathrm{}}}{(ab,acz^2,z^2q;q)_{\mathrm{}}}}(s_0^YY_1s_1^YY_0),`$ (4.3)
where the sum runs over all partitions $`\lambda `$.
Theorem 4.1 and Theorem 3.4 also give the following corollary:
###### Corollary 4.3.
Put $`x=\frac{(ab)^{\frac{1}{2}}+(ab)^{\frac{1}{2}}}{2}`$ and $`q=abcd`$. Let $`\mathrm{\Phi }_N=\mathrm{\Phi }_N(a,b,c,d;z)`$ be as in 4.1.
1. The generating function $`\mathrm{\Phi }_{2N}(a,b,c,d;1)`$ is given by
$`\mathrm{\Phi }_{2N}(a,b,c,d;1)`$ $`={\displaystyle \frac{(ab)^{\frac{N}{2}}Q_N(x;a^{\frac{1}{2}}b^{\frac{1}{2}}c,a^{\frac{1}{2}}b^{\frac{1}{2}}|q)}{(q;q)_N(ac;q)_N}}`$
$`={\displaystyle \frac{(a;q)_N}{(q;q)_N(ac;q)_N}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^N,c}{a^1q^{N+1}}};q,bq).`$ (4.4)
2. The generating function $`\mathrm{\Phi }_{2N}(a,b,c,d;1)`$ is given by
$`\mathrm{\Phi }_{2N+1}(a,b,c,d;1)`$ $`={\displaystyle \frac{(1+a)(ab)^{\frac{N}{2}}Q_N(x;a^{\frac{1}{2}}b^{\frac{1}{2}}c,a^{\frac{3}{2}}b^{\frac{1}{2}}cd|q)}{(q;q)_N(ac;q)_{N+1}}}`$
$`={\displaystyle \frac{(a;q)_{N+1}}{(q;q)_N(ac;q)_{N+1}}}{}_{2}{}^{}\varphi _{1}^{}({\displaystyle \genfrac{}{}{0pt}{}{q^N,c}{a^1q^N}};q,b).`$ (4.5)
Let $`S_N(n,r,s)`$ denote the number of partitions $`\pi `$ of $`n`$ where each part of $`\pi `$ is $`N`$, $`๐ช(\pi )=r`$, $`๐ช(\pi ^{})=s`$. As before we immediately deduce the following result of Andrews (cf. \[1, Theorem 1\]) from Corollary 4.3.
###### Corollary 4.4.
(Andrews)
$$\underset{n,r,s0}{}S_{2N}(n,r,s)q^nz^ry^s=\frac{_{j=0}^N\left[\genfrac{}{}{0pt}{}{N}{j}\right]_{q^4}(zyq;q^4)_j(zy^1q;q^4)_{Nj}(yq)^{2N2j}}{(q^4;q^4)_N(z^2q^4;q^4)_N},$$
(4.6)
and
$$\underset{n,r,s0}{}S_{2N+1}(n,r,s)q^nz^ry^s=\frac{_{j=0}^N\left[\genfrac{}{}{0pt}{}{N}{j}\right]_{q^4}(zyq;q^4)_{j+1}(zy^1q;q^4)_{Nj}(yq)^{2N2j}}{(q^4;q^4)_N(z^2q^4;q^4)_{N+1}}.$$
(4.7)
Similarly, as in the strict case, we obtain immediately Bouletโs corresponding result for ordinary partitions (cf. \[2, Theorem 1\]).
###### Corollary 4.5.
(Boulet) Let $`q=abcd`$, then
$`{\displaystyle \underset{\lambda }{}}\omega (\lambda )={\displaystyle \frac{(a;q)_{\mathrm{}}(abc;q)_{\mathrm{}}}{(q;q)_{\mathrm{}}(ab;q)_{\mathrm{}}(ac;q)_{\mathrm{}}}},`$ (4.8)
where the sum runs over all partitions.
In order to prove Theorem 4.1 we first derive a recurrence formula for $`\mathrm{\Phi }_N(a,b,c,d;z)`$.
###### Proposition 4.6.
Let $`\mathrm{\Phi }_N=\mathrm{\Phi }_N(a,b,c,d;z)`$ be as before and $`q=abcd`$. Then the following recurrences hold for any positive integer $`N`$.
$`(1z^2q^N)\mathrm{\Phi }_{2N}=(1+b)\mathrm{\Phi }_{2N1}b\mathrm{\Phi }_{2N2},`$ (4.9)
$`(1z^2acq^N)\mathrm{\Phi }_{2N+1}=(1+a)\mathrm{\Phi }_{2N}a\mathrm{\Phi }_{2N1}.`$ (4.10)
* It suffices to prove that
$`\mathrm{\Phi }_{2N}=\mathrm{\Phi }_{2N1}+b(\mathrm{\Phi }_{2N1}\mathrm{\Phi }_{2N2})+z^2q^N\mathrm{\Phi }_{2N},`$ (4.11)
$`\mathrm{\Phi }_{2N+1}=\mathrm{\Phi }_{2N}+a(\mathrm{\Phi }_{2N}\mathrm{\Phi }_{2N1})+z^2acq^N\mathrm{\Phi }_{2N+1}.`$ (4.12)
Let $`_N`$ denote the set of partitions $`\lambda `$ such that $`\lambda _1N`$. The generating function of $`_N`$ with weight $`\omega (\lambda )z^{\mathrm{}(\lambda )}`$ is $`\mathrm{\Phi }_N=\mathrm{\Phi }_N(a,b,c,d;z)`$. We divide $`_N`$ into three disjoint subsets:
$$_N=_{N1}_N๐ฉ_N$$
where $`_N`$ denote the set of partitions $`\lambda `$ such that $`\lambda _1=N`$ and $`\lambda _2<N`$, and $`๐ฉ_N`$ denote the set of partitions $`\lambda `$ such that $`\lambda _1=\lambda _2=N`$. When $`N=2r`$ is even, it is easy to see that the generating function of $`_{2r}`$ equals $`b(\mathrm{\Phi }_{2r1}\mathrm{\Phi }_{2r2})`$, and the generating function of $`๐ฉ_{2r}`$ equals $`z^2q^r\mathrm{\Phi }_{2r}`$. This proves 4.11. When $`N=2r+1`$ is odd, the same division proves 4.12. $`\mathrm{}`$
By simple computation, one can derive the following identities from 4.9 and 4.10.
###### Proposition 4.7.
If we put
$$\mathrm{\Phi }_N(a,b,c,d;z)=\frac{F_N(a,b,c,d;z)}{(z^2q;q)_{N/2}(z^2ac;q)_{N/2}},$$
(4.13)
then,
$`F_{2N}=(1+b)F_{2N1}b(1z^2acq^{N1})F_{2N2},`$ (4.14)
$`F_{2N+1}=(1+a)F_{2N}a(1z^2q^N)F_{2N1}.`$ (4.15)
hold for any positive integer $`N`$.
* Substitute 4.13 into 4.9 and 4.10, and compute directly to obtain 4.14 and 4.15. $`\mathrm{}`$
* From 4.14 and 4.15, one easily sees that $`F_{2N}(a,b,c,d;z)`$ and $`F_{2N+1}(a,b,c,d;z)`$ satisfy exactly the same recurrence in Theorem 3.1. Further, from the above example, we see
$`F_0=1,`$
$`F_1=1+az,`$
$`F_2=1+a(1+b)z+abcz^2,`$
$`F_3=1+a(1+b+ab)z+abc(1+a+ad)z^2+a^3bcdz^3,`$
$`F_4=1+a(1+b)(1+ab)z+abc(1+a+ab+ad+abd+abcd)z^2`$
$`+a^3bcd(1+b)(1+bc)z^3+a^3b^3c^3dz^4.`$
Thus the first few terms of $`F_N(a,b,c,d;z)`$ agree with those of $`\mathrm{\Psi }_N(a,b,c,d;z)`$. We immediately conclude that $`F_N(a,b,c,d;z)=\mathrm{\Psi }_N(a,b,c,d;z)`$ for all $`N`$. $`\mathrm{}`$
###### Remark 4.8.
Here we also give another proof of Theorem 4.1 by a bijection, which has already been used by Boulet in the infinite case.
* Let $`๐ซ_N`$ (resp. $`๐_N`$) denote the set of partitions (resp. strict partitions) whose parts are less than or equal to $`N`$ and let $`_N`$ denote the set of partitions whose parts appear an even number of times and are less than or equal to $`N`$. We shall establish a bijection $`g:๐ซ_N๐_N\times _N`$ with $`g(\lambda )=(\mu ,\nu )`$ defined as follows. Suppose $`\lambda `$ has $`k`$ parts equal to $`i`$. If $`k`$ is even then $`\nu `$ has $`k`$ parts equal to $`i`$, and if $`k`$ is odd then $`\nu `$ has $`k1`$ parts equal to $`i`$. The parts of $`\lambda `$ which were not removed to form $`\nu `$, at most one of each cardinality, give $`\mu `$. It is clear that under this bijection, $`\omega (\lambda )=\omega (\mu )\omega (\nu )`$. It is easy to see that the generating function of $`_N`$ is equal to
$$\underset{j=1}{\overset{\frac{N}{2}}{}}\frac{1}{1z^2q^j}\times \underset{j=0}{\overset{\frac{N1}{2}}{}}\frac{1}{1z^2acq^j},$$
where $`q=abcd`$. As $`\frac{N1}{2}=\frac{N}{2}1`$, we obtain (4.13). $`\mathrm{}`$
At the end of this section we state another enumeration of the ordinary partitions, which is not directly related to Andrewsโ result, but obtained as an application of the minor summation formula of Pfaffians. Let
$$\mathrm{\Phi }_{N,M}=\mathrm{\Phi }_{N,M}(a,b,c,d)=\underset{\genfrac{}{}{0pt}{}{\lambda }{\lambda _1N,\mathrm{}(\lambda )M}}{}\omega (\lambda ),$$
where the sum runs over all partitions $`\lambda `$ such that $`\lambda `$ has at most $`M`$ parts and each part of $`\lambda `$ is less than or equal to $`N`$.
Again we use Lemma 3.7 and Theorem 2.1 to obtain the following theorem.
###### Theorem 4.9.
Let $`N`$ be a positive integer and set $`q=abcd`$. Then we have
$`{\displaystyle \underset{t=0}{\overset{N/2}{}}}\mathrm{\Phi }_{N2t,2t}(a,b,c,d)z^tq^{\left(\genfrac{}{}{0pt}{}{t}{2}\right)}=\mathrm{Pf}\left[\begin{array}{cc}S& J\\ J& C\end{array}\right],`$ (4.16)
where $`S=(1)_{0i<jN1}`$ and $`C=(a^{(j1)/2}b^{(j1)/2}c^{i/2}d^{i/2}z)_{0i<jN1}`$.
* As before, we take $`A=(1)_{0i<jN1}`$ and
$$B=(a^{(j1)/2}b^{(j1)/2}c^{i/2}d^{i/2})_{0i<jN1},$$
in Lemma 3.7, then 4.16 follows from Lemma 2.1. $`\mathrm{}`$
For example, if $`N=4`$, then the right-hand side of 4.16 becomes
$$\mathrm{Pf}\left[\begin{array}{cccccccc}0& 1& 1& 1& 0& 0& 0& 1\\ 1& 0& 1& 1& 0& 0& 1& 0\\ 1& 1& 0& 1& 0& 1& 0& 0\\ 1& 1& 1& 0& 1& 0& 0& 0\\ & & & & & & & \\ 0& 0& 0& 1& 0& z& az& abz\\ 0& 0& 1& 0& z& 0& acz& abcz\\ 0& 1& 0& 0& az& acz& 0& abcdz\\ 1& 0& 0& 0& abz& abcz& abcdz& 0\end{array}\right].$$
Let $`\stackrel{~}{\mathrm{\Phi }}_N=\stackrel{~}{\mathrm{\Phi }}_N(a,b,c,d;z)=\mathrm{Pf}\left[\begin{array}{cc}S& J\\ J& C\end{array}\right]`$ denote the right-hand side of 4.16. For example, we have $`\stackrel{~}{\mathrm{\Phi }}_1=1`$, $`\stackrel{~}{\mathrm{\Phi }}_2=1+z`$, $`\stackrel{~}{\mathrm{\Phi }}_3=1+(1+a+ac)z`$ and $`\stackrel{~}{\mathrm{\Phi }}_4=1+(1+a+ab+ac+abc+abcd)z+abcdz^2`$. Note that the partitions $`\lambda `$ such that $`\mathrm{}(\lambda )2`$ and $`\lambda _12`$ are the following six:
a
a
b
a
c
a
b
c
a
b
c
d
.
a
a
b
a
c
a
b
c
a
b
c
d
\emptyset\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\qquad\vbox{\vskip 3.0pt plus 1.0pt minus 1.0pt\offinterlineskip\halign{&\smvsquare{#}\cr\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$a$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$b$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$c$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt&\vbox{\hbox{\vrule width=0.5pt\vbox to12.0pt{\hrule height=0.5pt\vss\hbox to12.0pt{\hss{\smcellfont$d$}\hss}
\vss\hrule height=0.5pt}
\vrule width=0.5pt}
\kern-0.5pt}\kern-0.5pt\\}}\ .
The sum of their weights is equal to $`[z]\stackrel{~}{\mathrm{\Phi }}_4=1+a+ab+ac+abc+abcd`$.
The same argument as in the proof of Proposition 3.9 can be used to prove the following proposition.
###### Proposition 4.10.
Let $`\stackrel{~}{\mathrm{\Phi }}_N=\stackrel{~}{\mathrm{\Phi }}_N(a,b,c,d;z)`$ be as above. Then we have
$`\stackrel{~}{\mathrm{\Phi }}_{2N}=(1+b)\stackrel{~}{\mathrm{\Phi }}_{2N1}+(a^{N1}b^{N1}c^{N1}d^{N1}zb)\stackrel{~}{\mathrm{\Phi }}_{2N2},`$ (4.17)
$`\stackrel{~}{\mathrm{\Phi }}_{2N+1}=(1+a)\stackrel{~}{\mathrm{\Phi }}_{2N}+(a^Nb^{N1}c^Nd^{N1}za)\stackrel{~}{\mathrm{\Phi }}_{2N1},`$ (4.18)
for any positive integer $`N`$.
* Perform the same elementary transformations of rows and columns on $`\left[\begin{array}{cc}S& J\\ J& C\end{array}\right]`$ as we did in the proof of Proposition 3.9, and expand it along the last row/column. The details are left to the reader. $`\mathrm{}`$
###### Remark 4.11.
The recurrence equations 4.17 and 4.18 also can be proved combinatorially.
* Consider the generating function of partitions:
$$\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _12j+12t}}}{}w(\lambda )=\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _12j2t}}}{}w(\lambda )+\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _1=2j+12t}}}{}w(\lambda ).$$
(4.19)
Splitting the partitions $`\lambda `$ in the second sum of the right side into two subsets: $`\lambda _2<\lambda _1`$, and $`\lambda _2=\lambda _1`$. Now
$$\underset{\genfrac{}{}{0pt}{}{\lambda :\lambda _1>\lambda _2}{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _1=2j+12t}}}{}w(\lambda )=a\left(\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _12j2t}}}{}w(\lambda )\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _12j12t}}}{}w(\lambda )\right),$$
(4.20)
and
$$\underset{\genfrac{}{}{0pt}{}{\lambda :\lambda _1=\lambda _2}{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t}{\lambda _1=2j+12t}}}{}w(\lambda )=acq^{jt}\underset{\genfrac{}{}{0pt}{}{\lambda }{\genfrac{}{}{0pt}{}{\mathrm{}(\lambda )2t2}{\lambda _12j+12t}}}{}w(\lambda ).$$
(4.21)
Plugging (4.20) and (4.21) into (4.19) and then multiplying by $`z^tq^{\left(\genfrac{}{}{0pt}{}{t}{2}\right)}`$ and summing over $`t`$ we get (4.18). Similarly we can prove (4.17). $`\mathrm{}`$
###### Proposition 4.12.
Set $`U_N=\stackrel{~}{\mathrm{\Phi }}_{2N}`$ and $`V_N=\stackrel{~}{\mathrm{\Phi }}_{2N+1}`$, then, for $`N1`$,
$`U_{N+1}`$ $`=\left\{1+ab+ac(1+bd)q^{N1}z\right\}U_Na(bzq^{N1})(1czq^{N1})U_{N1},`$ (4.22)
$`V_{N+1}`$ $`=\left\{1+ab+(1+ac)zq^N\right\}V_Na(bzq^N)(1czq^{N1})V_{N1},`$ (4.23)
where $`U_0=1`$, $`V_0=1`$, $`U_1=1+z`$, $`V_1=1+(1+a+ac)z`$.
Thus $`U_N`$ and $`V_N`$ are also expressed by the solutions of the associated Al-Salam-Chihara polynomials.
## 5 A weighted sum of Schurโs $`P`$-functions
We use the notation $`X=X_n=(x_1,\mathrm{},x_n)`$ for the finite set of variables $`x_1`$, $`\mathrm{}`$, $`x_n`$. The aim of this section is to give some Pfaffian and determinantal formulas for the weighted sum $`\omega (\mu )z^{\mathrm{}(\mu )}P_\mu (x)`$ where $`P_\mu (x)`$ is Schurโs $`P`$-function.
Let $`A_n`$ denote the skew-symmetric matrix
$`\left({\displaystyle \frac{x_ix_j}{x_i+x_j}}\right)_{1i,jn}`$
and for each strict partition $`\mu =(\mu _1,\mathrm{},\mu _l)`$ of length $`ln`$, let $`\mathrm{\Gamma }_\mu `$ denote the $`n\times l`$ matrix $`\left(x_j^{\mu _i}\right)`$. Let
$$A_\mu (x_1,\mathrm{},x_n)=\left(\begin{array}{cc}A_n& \mathrm{\Gamma }_\mu J_l\\ J_l{}_{}{}^{t}\mathrm{\Gamma }_{\mu }^{}& O_l\end{array}\right)$$
which is a skew-symmetric matrix of $`(n+l)`$ rows and columns. Define $`\mathrm{Pf}_\mu (x_1,\mathrm{},x_n)`$ to be $`\mathrm{Pf}A_\mu (x_1,\mathrm{},x_n)`$ if $`n+l`$ is even, and to be $`\mathrm{Pf}A_\mu (x_1,\mathrm{},x_n,0)`$ if $`n+l`$ is odd. By \[14, Ex.13, p.267\], Schurโs $`P`$-function $`P_\mu (x_1,\mathrm{},x_n)`$ is defined to be
$$\frac{\mathrm{Pf}_\mu (x_1,\mathrm{},x_n)}{\mathrm{Pf}_{\mathrm{}}(x_1,\mathrm{},x_n)},$$
where it is well-known that $`\mathrm{Pf}_{\mathrm{}}(x_1,\mathrm{},x_n)=_{1i<jn}\frac{x_ix_j}{x_i+x_j}.`$ Meanwhile, by \[14, (8.7), p.253\], Schurโs $`Q`$-function $`Q_\mu (x_1,\mathrm{},x_n)`$ is defined to be $`2^{\mathrm{}(\lambda )}P_\mu (x_1,\mathrm{},x_n)`$.
In this section, we consider a weighted sum of Schurโs $`P`$-functions and $`Q`$-functions, i.e.,
$`\xi _N(a,b,c,d;X_n)={\displaystyle \underset{\genfrac{}{}{0pt}{}{\mu }{\mu _1N}}{}}\omega (\mu )P_\mu (x_1,\mathrm{},x_n),`$
$`\eta _N(a,b,c,d;X_n)={\displaystyle \underset{\genfrac{}{}{0pt}{}{\mu }{\mu _1N}}{}}\omega (\mu )Q_\mu (x_1,\mathrm{},x_n),`$
where the sums run over all strict partitions $`\mu `$ such that each part of $`\mu `$ is less than or equal to $`N`$. More generally, we can unify these problems to finding the following sum:
$$\zeta _N(a,b,c,d;z;X_n)=\underset{\genfrac{}{}{0pt}{}{\mu }{\mu _1N}}{}\omega (\mu )z^{\mathrm{}(\mu )}P_\mu (x_1,\mathrm{},x_n),$$
(5.1)
where the sum runs over all strict partitions $`\mu `$ such that each part of $`\mu `$ is less than or equal to $`N`$. One of the main results of this section is that $`\zeta _N(a,b,c,d;z;X_n)`$ can be expressed by a Pfaffian (see Corollary 5.6). Further, let us put
$`\zeta (a,b,c,d;z;X_n)=\underset{N\mathrm{}}{lim}\zeta _N(a,b,c,d;z;X_n)={\displaystyle \underset{\mu }{}}\omega (\mu )z^{\mathrm{}(\mu )}P_\mu (X_n),`$ (5.2)
where the sum runs over all strict partitions $`\mu `$. We also write
$`\xi (a,b,c,d;X_n)=\zeta (a,b,c,d;1;X_n)={\displaystyle \underset{\mu }{}}\omega (\mu )P_\mu (X_n),`$
where the sum runs over all strict partitions $`\mu `$. Then we have the following theorem:
###### Theorem 5.1.
Let $`n`$ be a positive integer. Then
$`\zeta (a,b,c,d;z;X_n)=\{\begin{array}{cc}\mathrm{Pf}\left(\gamma _{ij}\right)_{1i<jn}/\mathrm{Pf}_{\mathrm{}}(X_n)\hfill & \text{ if }n\text{ is even,}\hfill \\ \mathrm{Pf}\left(\gamma _{ij}\right)_{0i<jn}/\mathrm{Pf}_{\mathrm{}}(X_n)\hfill & \text{ if }n\text{ is odd,}\hfill \end{array}`$ (5.3)
where
$`\gamma _{ij}`$ $`={\displaystyle \frac{x_ix_j}{x_i+x_j}}+u_{ij}z+v_{ij}z^2`$ (5.4)
with
$`u_{ij}`$ $`={\displaystyle \frac{adet\left(\begin{array}{cc}x_i+bx_i^2& 1abx_i^2\\ x_j+bx_j^2& 1abx_j^2\end{array}\right)}{(1abx_i^2)(1abx_j^2)}},`$ (5.5)
$`v_{ij}`$ $`={\displaystyle \frac{abcx_ix_jdet\left(\begin{array}{cc}x_i+ax_i^2& 1a(b+d)x_i^2abdx_i^3\\ x_j+ax_j^2& 1a(b+d)x_j^2abdx_j^3\end{array}\right)}{(1abx_i^2)(1abx_j^2)(1abcdx_i^2x_j^2)}},`$ (5.6)
if $`1i,jn`$, and
$`\gamma _{0j}=1+{\displaystyle \frac{ax_j(1+bx_j)}{1abx_j^2}}z`$ (5.7)
if $`1jn`$.
Especially, when $`z=1`$, we have
$`\xi (a,b,c,d;X_n)=\{\begin{array}{cc}\mathrm{Pf}\left(\stackrel{~}{\gamma }_{ij}\right)_{1i<jn}/\mathrm{Pf}_{\mathrm{}}(X_n)\hfill & \text{ if }n\text{ is even,}\hfill \\ \mathrm{Pf}\left(\stackrel{~}{\gamma }_{ij}\right)_{0i<jn}/\mathrm{Pf}_{\mathrm{}}(X_n)\hfill & \text{ if }n\text{ is odd,}\hfill \end{array}`$ (5.8)
where
$`\stackrel{~}{\gamma }_{ij}=\{\begin{array}{cc}\frac{1+ax_j}{1abx_j^2}\hfill & \text{ if }i=0\text{,}\hfill \\ \frac{x_ix_j}{x_i+x_j}+\stackrel{~}{v}_{ij}\hfill & \text{ if }1i<jn\text{,}\hfill \end{array}with`$ (5.9)
$$\stackrel{~}{v}_{ij}=\frac{adet\left(\begin{array}{cc}x_i+bx_i^2& 1b(a+c)x_i^2abcx_i^3\\ x_j+bx_j^2& 1b(a+c)x_j^2abcx_j^3\end{array}\right)}{(1abx_i^2)(1abx_j^2)(1abcdx_i^2x_j^2)}.$$
(5.10)
We can generalize this result in the following theorem (Theorem 5.2) using the generalized Vandermonde determinant used in . Let $`n`$ be an non-negative integer, and let $`X=(x_1,\mathrm{},x_{2n})`$, $`Y=(y_1,\mathrm{},y_{2n})`$, $`A=(a_1,\mathrm{},a_{2n})`$ and $`B=(b_1,\mathrm{},b_{2n})`$ be $`2n`$-tuples of variables. Let $`V^n(X,Y,A)`$ denote the $`2n\times n`$ matrix whose $`(i,j)`$th entry is $`a_ix_i^{nj}y_i^{j1}`$ for $`1i2n`$, $`1jn`$, and let $`U^n(X,Y;A,B)`$ denote the $`2n\times 2n`$ matrix $`\left(\begin{array}{cc}V^n(X,Y,A)& V^n(X,Y,B)\end{array}\right).`$ For instance if $`n=2`$ then $`U^2(X,Y;A,B)`$ is
$$\left(\begin{array}{cccc}a_1x_1& a_1y_1& b_1x_1& b_1y_1\\ a_2x_2& a_2y_2& b_2x_2& b_2y_2\\ a_3x_3& a_3y_3& b_3x_3& b_3y_3\\ a_4x_4& a_4y_4& b_4x_4& b_4y_4\end{array}\right).$$
Hereafter we use the following notation for $`n`$-tuples $`X=(x_1,\mathrm{},x_n)`$ and $`Y=(y_1,\mathrm{},y_n)`$ of variables:
$$X+Y=(x_1+y_1,\mathrm{},x_n+y_n),XY=(x_1y_1,\mathrm{},x_ny_n),$$
and, for integers $`k`$ and $`l`$,
$$X^k=(x_1^k,\mathrm{},x_n^k),X^kY^l=(x_1^ky_1^l,\mathrm{},x_n^ky_n^l).$$
Let $`\mathrm{๐}`$ denote the $`n`$-tuple $`(1,\mathrm{},1)`$. For any subset $`I=\{i_1,\mathrm{},i_r\}\left(\genfrac{}{}{0pt}{}{[n]}{r}\right)`$, let $`X_I`$ denote the $`r`$-tuple $`(x_{i_1},\mathrm{},x_{i_r})`$.
###### Theorem 5.2.
Let $`q=abcd`$. If $`n`$ is an even integer, then we have
$`\xi (a,b,c,d;X_n)`$ $`={\displaystyle \underset{r=0}{\overset{n/2}{}}}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}}{\displaystyle \frac{(1)^{|I|\left(\genfrac{}{}{0pt}{}{r+1}{2}\right)}a^rq^{\left(\genfrac{}{}{0pt}{}{r}{2}\right)}}{_{iI}(1abx_i^2)}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,jI}{i<j}}{}}{\displaystyle \frac{x_i+x_j}{(x_ix_j)(1qx_i^2x_j^2)}}`$
$`\times detU^r(X_I^2,\mathrm{๐}+qX_I^4,X_I+bX_I^2,\mathrm{๐}b(a+c)X_I^2abcX_I^3).`$ (5.11)
If $`n`$ is an odd integer, then we have
$`\xi (a,b,c,d;X_n)={\displaystyle \underset{m=1}{\overset{n}{}}}{\displaystyle \frac{1+ax_m}{1abx_m^2}}{\displaystyle \underset{r=0}{\overset{(n1)/2}{}}}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]\{m\}}{2r}\right)}{}}{\displaystyle \frac{(1)^{|I|\left(\genfrac{}{}{0pt}{}{r+1}{2}\right)}a^rq^{\left(\genfrac{}{}{0pt}{}{r}{2}\right)}}{_{iI}(1abx_i^2)}}{\displaystyle \underset{iI}{}}{\displaystyle \frac{x_m+x_i}{x_mx_i}}`$
$`\times {\displaystyle \underset{\genfrac{}{}{0pt}{}{i,jI}{i<j}}{}}{\displaystyle \frac{x_i+x_j}{(x_ix_j)(1qx_i^2x_j^2)}}detU^r(X_I^2,\mathrm{๐}+qX_I^4,X_I+bX_I^2,\mathrm{๐}b(a+c)X_I^2abcX_I^3).`$ (5.12)
###### Theorem 5.3.
Let $`q=abcd`$. If $`n`$ is an even integer, then $`\zeta (a,b,c,d;z;X_n)`$ is equal to
$`{\displaystyle \underset{r=0}{\overset{n/2}{}}}z^{2r}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}}{\displaystyle \frac{(1)^{|I|\left(\genfrac{}{}{0pt}{}{r+1}{2}\right)}(abc)^rq^{\left(\genfrac{}{}{0pt}{}{r}{2}\right)}_{iI}x_i}{_{iI}(1abx_i^2)}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,jI}{i<j}}{}}{\displaystyle \frac{x_i+x_j}{(x_ix_j)(1qx_i^2x_j^2)}}`$
$`\times detU^r(X_I^2,\mathrm{๐}+qX_I^4,X_I+aX_I^2,\mathrm{๐}a(b+d)X_I^2abdX_I^3)`$
$`+{\displaystyle \underset{r=0}{\overset{n/2}{}}}z^{2r1}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{k<l}{k,lI}}{}}{\displaystyle \frac{(1)^{|I|\left(\genfrac{}{}{0pt}{}{r}{2}\right)1}a^rb^{r1}c^{r1}q^{\left(\genfrac{}{}{0pt}{}{r1}{2}\right)}\{1+b(x_k+x_l)+abx_kx_l\}_{iI^{}}x_i}{_{iI}(1abx_i^2)}}`$
$`\times {\displaystyle \frac{_{\genfrac{}{}{0pt}{}{i,jI}{i<j}}(x_i+x_j)detU^{r1}(X_I^{}^2,\mathrm{๐}+qX_I^{}^4,X_I^{}+aX_I^{}^2,\mathrm{๐}a(b+d)X_I^{}^2abdX_I^{}^3)}{_{\genfrac{}{}{0pt}{}{i,jI^{}}{i<j}}(x_ix_j)(1qx_i^2x_j^2)}},`$ (5.13)
where $`I^{}=I\{k,l\}`$.
Note that we can obtain a similar formula when $`n`$ is odd by expanding the Pfaffian in 5.3 along the first row/column.
To obtain the sum of this type we need a generalization of Lemma 3.7, in which the row/column indices always contain say the set $`\{1,2,\mathrm{},n\}`$, for some fixed $`n`$.
###### Lemma 5.4.
Let $`n`$ and $`N`$ be nonnegative integers. Let $`A=(a_{ij})`$ and $`B=(b_{ij})`$ be skew symmetric matrices of size $`(n+N)`$. We divide the set of row/column indices into two subsets, i.e. the first $`n`$ indices $`I_0=[n]`$ and the last $`N`$ indices $`I_1=[n+1,n+N]`$. Then
$`{\displaystyle \underset{\genfrac{}{}{0pt}{}{t0}{n+t\text{ even}}}{}}z^{(n+t)/2}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{I_1}{t}\right)}{}}\gamma ^{|I_0I|}\mathrm{Pf}\left(\mathrm{\Delta }_{I_0I}^{I_0I}(A)\right)\mathrm{Pf}\left(\mathrm{\Delta }_{I_0I}^{I_0I}(B)\right)`$
$`=\mathrm{Pf}\left(\begin{array}{cc}J_{n+N}{}_{}{}^{t}AJ_{n+N}& K_{n,N}\\ {}_{}{}^{t}K_{n,N}^{}& C\end{array}\right),`$ (5.14)
where $`C=(C_{ij})_{1i,j,n+N}`$ is given by $`C_{ij}=\gamma ^{i+j}b_{ij}z`$ and $`K_{n,N}=J_{n+N}\stackrel{~}{E}_{n,N}`$ with
$$\stackrel{~}{E}_{n,N}=\left(\begin{array}{cc}O_n& O_{n,N}\\ O_{N,n}& E_N\end{array}\right).$$
* Let $`V=\{(n+N)^{},\mathrm{},(n+1)^{},n^{},\mathrm{},1^{},1,\mathrm{},n,n+1,\mathrm{},n+N\}`$ be vertices arranged in this order on the $`x`$-axis. Put $`V_0=\{n^{},\mathrm{},1^{}\}`$ and $`V_1^{}=\{(n+N)^{},\mathrm{},(n+1)^{}\}`$, $`V_0=\{1,\mathrm{},n\}`$ and $`V_1=\{n+1,\mathrm{},N\}`$. From 2.9, the Pfaffian on the right-hand side of 5.14 is equal to
$$\underset{\sigma }{}\mathrm{sgn}\sigma \underset{(i,j)\sigma }{}a_{ij}\underset{(i,j)\sigma }{}C_{ij}$$
summed over all perfect matching $`\sigma `$ on $`V`$, in which there is a set $`I=\{i_1,\mathrm{},i_t\}V_1`$ of $`t`$ vertices such that $`\sigma |_{V_0V_1}`$ is a perfect matching on $`V_0I`$ and $`\sigma |_{V_0^{}V_1^{}}`$ is a perfect matching on $`V_0^{}I^{}`$, where $`I^{}=\{i_1^{},\mathrm{},i_t^{}\}`$, and each $`jV_1I`$ is adjoint to $`j^{}V_1^{}I^{}`$ in $`\sigma `$. Thus the summand vanishes unless $`n+t`$ is even, and this sum is equal to
$$\underset{t}{}z^{(t+n)/2}\underset{I\left(\genfrac{}{}{0pt}{}{V_1}{t}\right)}{}\gamma ^{n+|I|}\underset{(\sigma _1,\sigma _2)}{}\mathrm{sgn}\sigma _1\mathrm{sgn}\sigma _2\underset{(i,j)\sigma _1}{}a_{ij}\underset{(i,j)\sigma _2}{}b_{ij},$$
where the third sum runs over all pairs $`(\sigma _1,\sigma _2)`$ where $`\sigma _1`$ is a perfect matching on $`V_0I`$ and $`\sigma _2`$ is a perfect matching on $`V_0^{}I^{}`$. This is equal to the left-hand side of 5.14. $`\mathrm{}`$
For a nonnegative integer $`N`$, let $`\mu ^N=(N,\mathrm{},1,0)`$, and let $`\mathrm{\Gamma }_{\mu ^N}`$ denote the $`n\times (N+1)`$ matrix $`\left(x_i^{Nj}\right)_{1in,0jN}`$. Let
$$๐_{n,N}=\left(\begin{array}{cc}A_n& \mathrm{\Gamma }_{\mu ^N}J_{N+1}\\ J_{N+1}{}_{}{}^{t}\mathrm{\Gamma }_{\mu ^N}^{}& O_{N+1}\end{array}\right)$$
which is a skew-symmetric matrix of size $`n+N+1`$. For example, if $`n=4`$ and $`N=3`$, then
$$๐_{4,3}=\left(\begin{array}{cccccccc}0& \frac{x_1x_2}{x_1+x_2}& \frac{x_1x_3}{x_1+x_3}& \frac{x_1x_4}{x_1+x_4}& 1& x_1& x_{1}^{}{}_{}{}^{2}& x_{1}^{}{}_{}{}^{3}\\ \frac{x_2x_1}{x_1+x_2}& 0& \frac{x_2x_3}{x_2+x_3}& \frac{x_2x_4}{x_2+x_4}& 1& x_2& x_{2}^{}{}_{}{}^{2}& x_{2}^{}{}_{}{}^{3}\\ \frac{x_3x_1}{x_1+x_3}& \frac{x_3x_2}{x_2+x_3}& 0& \frac{x_3x_4}{x_3+x_4}& 1& x_3& x_{3}^{}{}_{}{}^{2}& x_{3}^{}{}_{}{}^{3}\\ \frac{x_4x_1}{x_1+x_4}& \frac{x_4x_2}{x_2+x_4}& \frac{x_4x_3}{x_3+x_4}& 0& 1& x_4& x_{4}^{}{}_{}{}^{2}& x_{4}^{}{}_{}{}^{3}\\ 1& 1& 1& 1& 0& 0& 0& 0\\ x_1& x_2& x_3& x_4& 0& 0& 0& 0\\ x_{1}^{}{}_{}{}^{2}& x_{2}^{}{}_{}{}^{2}& x_{3}^{}{}_{}{}^{2}& x_{4}^{}{}_{}{}^{2}& 0& 0& 0& 0\\ x_{1}^{}{}_{}{}^{3}& x_{2}^{}{}_{}{}^{3}& x_{3}^{}{}_{}{}^{3}& x_{4}^{}{}_{}{}^{3}& 0& 0& 0& 0\end{array}\right).$$
Let $`\beta _{ij}`$ be as in 2.10. Let $`B_N`$ denote the $`(N+1)\times (N+1)`$ matrix $`(\beta _{ij})_{0i,jN}`$ and let $`B_N^{}`$ denote the $`(N+2)\times (N+2)`$ matrix $`(\beta _{ij})_{1i,jN}`$.
###### Theorem 5.5.
Let $`n`$ and $`N`$ be integers such that $`nN0`$. Then
$`\zeta _N(a,b,c,d;z;X_n)=\mathrm{Pf}\left(๐_{n,N}\right)/\mathrm{Pf}_{\mathrm{}}(X_n),`$ (5.15)
where
$`๐_{n,N}=\left(\begin{array}{ccc}O_{N+1}& {}_{}{}^{t}\mathrm{\Gamma }_{\mu ^N}^{}J_n& J_{N+1}\\ J_n\mathrm{\Gamma }_{\mu ^N}& J_n{}_{}{}^{t}A_{n}^{}J_n& O_{n,N+1}\\ J_{N+1}& O_{N+1,n}& B_N\end{array}\right),`$ (5.16)
if $`n`$ is even, and
$`๐_{n,N}=\left(\begin{array}{ccc}O_{N+1}& {}_{}{}^{t}\mathrm{\Gamma }_{\mu ^N}^{}J_n& J_{N+1}^{}\\ J_n\mathrm{\Gamma }_{\mu ^N}& J_n{}_{}{}^{t}A_{n}^{}J_n& O_{n,N+2}\\ {}_{}{}^{t}J_{N+1}^{}& O_{N+2,n}& B_N^{}\end{array}\right)`$ (5.17)
where $`J_{N+1}^{}=\left(\begin{array}{cc}O_{N+1,1}& J_{N+1}\end{array}\right)`$ if $`n`$ is odd.
* Let $`_{n,N}`$ be the skew-symmetric matrix of size $`(n+N+1)`$ defined by
$$_{n,N}=\left(\begin{array}{cc}S_n& O_{n,N+1}\\ O_{N+1.n}& B_N\end{array}\right)$$
if $`n`$ is even, and
$$_{n,N}=\left(\begin{array}{cc}S_{n1}& O_{n,N+2}\\ O_{N+2.n}& B_N^{}\end{array}\right)$$
if $`n`$ is odd. Fix a strict partition $`\mu =(\mu _1,\mathrm{},\mu _l)`$ such that $`\mu _1>\mathrm{}>\mu _l0`$, and let $`K_n(\mu )=\{n+\mu _l,\mathrm{},n+\mu _1\}`$. From the definition of $`_{n,N}`$ and Theorem 2.2, we have
$`\mathrm{Pf}\left(\mathrm{\Delta }_{[n]K_n(\mu )}^{[n]K_n(\mu )}\left(_{n,N}\right)\right)=\omega (\mu )z^{\mathrm{}(\mu )}`$
if $`n+l`$ is even. Thus Lemma 5.4 immediately implies that $`\mathrm{Pf}_{\mathrm{}}(X_n)\zeta _N(a,b,c,d;z;X_n)`$ is equal to
$$\mathrm{Pf}\left(\begin{array}{cc}J_{n+N+1}{}_{}{}^{t}๐_{n,N}^{}J_{n+N+1}& K_{n,N+1}\\ {}_{}{}^{t}K_{n,N+1}^{}& _{n,N}\end{array}\right).$$
(5.18)
By simple elementary transformations on rows and columns, we obtain the desired results 5.16 and 5.17. $`\mathrm{}`$
For instance, if $`n=4`$ and $`N=2`$, then $`๐_{4,2}`$ looks as follows:
$$\left(\begin{array}{cccccccccc}0& 0& 0& x_4^2& x_3^2& x_2^2& x_1^2& 0& 0& 1\\ 0& 0& 0& x_4& x_3& x_2& x_1& 0& 1& 0\\ 0& 0& 0& 1& 1& 1& 1& 1& 0& 0\\ x_4^2& x_4& 1& 0& \frac{x_3x_4}{x_3+x_4}& \frac{x_2x_4}{x_2+x_4}& \frac{x_1x_4}{x_1+x_4}& 0& 0& 0\\ x_3^2& x_3& 1& \frac{x_4x_3}{x_4+x_3}& 0& \frac{x_2x_3}{x_2+x_3}& \frac{x_1x_3}{x_1+x_3}& 0& 0& 0\\ x_2^2& x_2& 1& \frac{x_4x_2}{x_4+x_2}& \frac{x_3x_2}{x_3+x_2}& 0& \frac{x_1x_2}{x_1+x_2}& 0& 0& 0\\ x_1^2& x_1& 1& \frac{x_4x_1}{x_4+x_1}& \frac{x_3x_1}{x_3+x_1}& \frac{x_2x_1}{x_2+x_1}& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0& 0& 0& az& abz\\ 0& 1& 0& 0& 0& 0& 0& az& 0& abcz^2\\ 1& 0& 0& 0& 0& 0& 0& abz& abcz^2& 0\end{array}\right)$$
###### Corollary 5.6.
Let $`n`$ and $`N`$ be integers such that $`nN0`$. Then
$`\zeta _N(a,b,c,d;z;X_n)=\mathrm{Pf}\left(๐_{n,N}\right)/\mathrm{Pf}_{\mathrm{}}(X_n),`$ (5.19)
where
$`๐_{n,N}=\left({\displaystyle \frac{x_ix_j}{x_i+x_j}}+{\displaystyle \underset{0k,lN}{}}\beta _{kl}x_i^lx_j^k\right)_{1i,jn},`$ (5.20)
if $`n`$ is even, and
$`๐_{n,N}=\left(\begin{array}{cc}0& {\displaystyle \underset{k=0}{\overset{N}{}}}\beta _{1,k}x_j^k\\ & \\ {\displaystyle \underset{k=0}{\overset{N}{}}}\beta _{k,1}x_i^k& {\displaystyle \frac{x_ix_j}{x_i+x_j}}+{\displaystyle \underset{0k,lN}{}}\beta _{kl}x_i^lx_j^k\end{array}\right)_{0i,jn},`$ (5.23)
if $`n`$ is odd.
* When $`n`$ is even, annihilate the entries in $`{}_{}{}^{t}\mathrm{\Gamma }_{\mu ^N}^{}J_n`$ of 5.16 by elementary transformation of columns, and annihilate the entries in $`J_n\mathrm{\Gamma }_{\mu ^N}`$ of 5.16 by elementary transformation of columns. Then expand the Pfaffian $`\mathrm{Pf}\left(๐_{n,N}\right)`$ along the first $`N+1`$ rows. The case when $`n`$ is similar. Perform the same operation on 5.17. $`\mathrm{}`$
* Perform the summations
$$\underset{0k<l}{}\beta _{kl}det\left(\begin{array}{cc}x_i^l& x_i^k\\ x_j^l& x_j^k\end{array}\right)$$
and
$$\underset{k=0}{\overset{\mathrm{}}{}}\beta _{1,k}x_j^k,$$
and apply Corollary 5.6. The details are left to the reader (cf. Proof of Theorem 2.1 in ). $`\mathrm{}`$
To prove these theorems, we need to cite a lemma from . (See Corollary 3.3 of and Theorem 3.2 of .)
###### Lemma 5.7.
Let $`n`$ be a non-negative integer. Let $`X=(x_1,\mathrm{},x_{2n})`$, $`A=(a_1,\mathrm{},a_{2n})`$, $`B=(b_1,\mathrm{},b_{2n})`$, $`C=(c_1,\mathrm{},c_{2n})`$ and $`D=(d_1,\mathrm{},d_{2n})`$ be $`2n`$-tuples of variables. Then
$`\mathrm{Pf}\left[{\displaystyle \frac{(a_ib_ja_jb_i)(c_id_jc_jd_i)}{(x_ix_j)(1tx_ix_j)}}\right]_{1i<j2n}`$
$`={\displaystyle \frac{V^n(X,\mathrm{๐}+tX^2;A,B)V^n(X,\mathrm{๐}+tX^2;C,D)}{_{1i<j2n}(x_ix_j)(1tx_ix_j)}},`$ (5.24)
where $`\mathrm{๐}+tX^2=(1+tx_1^2,\mathrm{},1+tx_n^2)`$.
In particular, we have
$`\mathrm{Pf}\left[{\displaystyle \frac{a_ib_ja_jb_i}{1tx_ix_j}}\right]_{1i<j2n}=(1)^{\left(\genfrac{}{}{0pt}{}{n}{2}\right)}t^{\left(\genfrac{}{}{0pt}{}{n}{2}\right)}{\displaystyle \frac{V^n(X,\mathrm{๐}+tX^2;A,B)}{_{1i<j2n}(1tx_ix_j)}}.\mathrm{}`$ (5.25)
* First, assume $`n`$ is even. Using the formula
$$\mathrm{Pf}(A+B)=\underset{r=0}{\overset{n/2}{}}\underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}(1)^{|I|r}\mathrm{Pf}(A_I^I)\mathrm{Pf}(B_{\overline{I}}^{\overline{I}}),$$
(5.26)
where $`\overline{I}`$ denotes the complementary set of $`I`$, we see that $`\xi (a,b,c,d;X_n)`$ is equal to
$$\underset{r=0}{\overset{n/2}{}}\underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}(1)^{|I|r}\underset{\genfrac{}{}{0pt}{}{i,jI}{i<j}}{}\frac{x_i+x_j}{x_ix_j}\mathrm{Pf}(\stackrel{~}{v}_{ij})_{i,jI}.$$
Apply Lemma 5.7 to obtain 5.11. When $`n`$ is odd, first expand the Pfaffian along the first row/column and repeat the same argument. $`\mathrm{}`$
* Note that the rank of the matrix $`(u_{ij})_{1i,jn}`$ is at most two. Thus we have
$$\mathrm{Pf}(u_{ij})_{1i,jn}=\{\begin{array}{cc}\frac{a(x_1x_2)\{1+b(x_1+x_2)+abx_1x_2\}}{(1abx_1^2)(1abx_2^2)}\hfill & \text{ if }n=2\text{,}\hfill \\ 0\hfill & \text{ otherwise.}\hfill \end{array}$$
Using 5.26, we obtain
$`\mathrm{Pf}\left(\gamma _{ij}\right)_{1i,jn}=\mathrm{Pf}({\displaystyle \frac{x_ix_j}{x_i+x_j}}+v_{ij}z^2)_{1i,jn}`$
$`+{\displaystyle \underset{1k<ln}{}}(1)^{k+l1}{\displaystyle \frac{az(x_kx_l)\{1+b(x_k+x_l)+abx_kx_l\}}{(1abx_k^2)(1abx_l^2)}}\mathrm{Pf}({\displaystyle \frac{x_ix_j}{x_i+x_j}}+v_{ij}z^2)_{\genfrac{}{}{0pt}{}{1i,jn}{i,jk,l}}.`$
Use 5.26 again to see that $`\zeta (a,b,c,d;z;X_n)`$ is equal to
$`{\displaystyle \underset{r=0}{\overset{n/2}{}}}z^{2r}{\displaystyle \underset{I\left(\genfrac{}{}{0pt}{}{[n]}{2r}\right)}{}}(1)^{|I|r}{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,jI}{i<j}}{}}{\displaystyle \frac{x_i+x_j}{x_ix_j}}\mathrm{Pf}(v_{ij})_{i,jI}`$
$`+{\displaystyle \underset{1k<ln}{}}(1)^{k+l1}{\displaystyle \frac{az(x_kx_l)\{1+b(x_k+x_l)+abx_kx_l\}}{(1abx_k^2)(1abx_l^2)}}`$
$`\times {\displaystyle \underset{r=1}{\overset{n/2}{}}}z^{2r2}{\displaystyle \underset{I^{}\left(\genfrac{}{}{0pt}{}{[n]\{k,l\}}{2r2}\right)}{}}(1)^{|I^{}|r+1}{\displaystyle \underset{\genfrac{}{}{0pt}{}{i,jI^{}}{i<j}}{}}{\displaystyle \frac{x_i+x_j}{x_ix_j}}\mathrm{Pf}(v_{ij})_{i,jI^{}}.`$
Put $`I=I^{}\{k,l\}`$ and apply Lemma 5.7 to obtain 5.13. $`\mathrm{}`$
Acknowledgment: The authors would like to express their gratitude to Dr. Yasushi Kajihara for his helpful comments and suggestions.
|
warning/0506/astro-ph0506174.html
|
ar5iv
|
text
|
# 1 Time evolution of the thermal, kinetic, potential and total energy for the adiabatic collapse of an initially โcoldโ gas sphere. The different lines corresponds to the different gas particle numbers. We also present the energy error results of the GADGET code with โstandardโ parameters for 32,000 particles.
SIMULATION OF THE GRAVITATIONAL COLLAPSE AND FRAGMENTATION OF ROTATING MOLECULAR CLOUDS
P. Berczik<sup>1,2</sup>, M.I. Petrov<sup>1</sup>
<sup>1</sup>Main Astronomical Observatory, National Academy of Sciences of Ukraine
27 Akademika Zabolotnoho Str., Kiev, 03680, Ukraine
e-mail: berczik@mao.kiev.ua; petrov@mao.kiev.ua
<sup>2</sup>Astrophysics Group, Department of Physics, Rochester Institute of Technology
54 Lomb Memorial Drive, Rochester, NY 14623, USA
e-mail: berczik@cis.rit.edu
In this paper we study the process of the subsequent (runaway) fragmentation of the rotating isothermal Giant Molecular Cloud (GMC) complex. Our own developed Smoothed Particle Hydrodynamics (SPH) gas-dynamical model successfully reproduce the observed Cloud Mass-distribution Function (CMF) in our Galaxy (even the differences between the inner and outer parts of our Galaxy). The steady state CMF is established during the collapse within a free-fall timescale of the GMC. We show that one of the key parameters, which defines the observed slope of the present day CMF, is the initial ratio of the rotational (turbulent) and gravitational energy inside the fragmented GMC.
<sup> </sup><sup> </sup>footnotetext: ยฉ P. Berczik, M.I. Petrov, 2004
INTRODUCTION
SPH based 3D hydrodynamical codes, starting with the series of pioneering works by Monaghan and Lattanzio , are always very successfully applied to the study of evolution and fragmentation in molecular clouds and molecular cloud complexes. These early simulations have been usually performed with a few hundred to a few thousand of SPH particles and with a fixed (few parsec) spatial resolution.
Nowadays the most up to date simulations of molecular cloud evolution (e.g. ) are performed using a few tens of thousands of SPH particles with variable smoothing lengths. These simulations also include the details of cooling and heating in the complex gas mixtures of H, H<sub>2</sub>, CO and HII species.
Our present high resolution (64,000 SPH particle) simulations, with highly flexible and adaptive smoothing lengths, study the runaway collapse and the subsequent isothermal fragmentation of the isolated GMC complex with different rotational (turbulent) energy parameters of the clouds. The resulting CMF is compared with the recent observational distributions ($`d\mathrm{N}/d\mathrm{M}\mathrm{M}^\gamma `$, where the slope of the power law $`\gamma `$ is in the range 1.4 to 1.8) of the molecular cloud complexes derived from the different CO data for the different parts of our Galaxy .
METHOD
Continuous hydrodynamic fields in SPH are described by the interpolation functions constructed from the known values of these functions at randomly positioned $`N`$ โsmoothโ particles with individual masses $`m_i`$ . To achieve the same level of accuracy for all points in the fluid it is necessary to use a spatially variable smoothing length. In this case each particle has an individual value of the smoothing length - $`h_i`$.
A more detailed and complete description of the basic numerical equations of SPH can be found in many of our previous publications (e.g. and the references herein). Therefore we just briefly repeat the skeleton SPH equations of the code here. The density at the position of the particle $`i`$ can be defined as:
$$\rho _i=\underset{j=1}{\overset{N}{}}m_jW_{ij},$$
The equations of motion for a particle $`i`$:
$$\frac{d๐ซ_i}{dt}=๐ฏ_i,$$
$$\frac{d๐ฏ_i}{dt}=\underset{j=1}{\overset{N}{}}m_j\left(\frac{P_i}{\rho _i^2}+\frac{P_j}{\rho _j^2}+\stackrel{~}{\mathrm{\Pi }}_{ij}\right)_iW_{ij}_i\mathrm{\Phi }_i.$$
where $`P_i`$ is the pressure, $`\mathrm{\Phi }_i`$ is the self gravitational potential and $`\stackrel{~}{\mathrm{\Pi }}_{ij}`$ is an artificial viscosity term.
The internal energy equation has the form:
$$\frac{du_i}{dt}=\frac{1}{2}\underset{j=1}{\overset{N}{}}m_j\left(\frac{P_i}{\rho _i^2}+\frac{P_j}{\rho _j^2}+\stackrel{~}{\mathrm{\Pi }}_{ij}\right)(๐ฏ_i๐ฏ_j)_iW_{ij}+\frac{\mathrm{\Gamma }_i\mathrm{\Lambda }_i}{\rho _i}.$$
Here $`u_i`$ is the specific internal energy of the particle $`i`$. The term $`(\mathrm{\Gamma }_i\mathrm{\Lambda }_i)/\rho _i`$ accounts for non adiabatic processes not associated with the artificial viscosity. We present the radiative cooling in the form proposed by (see case โBโ) using the MAPPINGS III software :
$$\mathrm{\Lambda }=\mathrm{\Lambda }(\rho ,u,\mathrm{Z},\mathrm{})\mathrm{\Lambda }^{}(T,[\mathrm{Fe}/\mathrm{H}])n_i^2,n_i=\rho _i/(\mu m_p),$$
where $`n_i`$ is the hydrogen number density, $`T_i`$ the temperature and $`\mu `$ the molecular weight.
The equation of state must be added to close the system:
$$P_i=(\gamma 1)\rho _iu_i,$$
where $`\gamma `$ is the adiabatic index.
In SPH one of the basic tasks is to find the nearest neighbors of each SPH particle, i.e. to construct an interaction list for each particles. Basically we need to find all particles with $`๐ซ_{ij}2\mathrm{max}(h_i,h_j)`$ in order to estimate the density and also calculate the hydrodynamical forces.
In our code we keep the number of neighbors exactly constant by defining $`2h_i`$ to be the distance to the $`N_B`$ \- nearest particle. The value of $`N_B`$ is chosen such that a certain fraction of the total number of โgasโ particles $`N`$ affects the local flow characteristics. From these we need to SELECT the closest $`N_B`$ particles. Fast algorithms for doing this exist . For computational reasons, if the defined $`h_i`$ becomes smaller than the selected minimal smoothing length $`h_{min}`$, we set the value $`h_i=h_{min}`$.
To calculate the self gravitational potential $`\mathrm{\Phi }_i`$ and self gravitational force $`_i\mathrm{\Phi }_i`$ we use the Mitaka Underground Vineyard (MUV) GRAPE6 computer system at the National Astronomical Observatory of Japan \[http://www.cc.nao.ac.jp/muv/\]. For a more detailed description of the GRAPE6 board and fot links to publication about the GRAPE6, we refer the reader to the official homepage of Jun Makino at Tokyo University \[http://grape.astron.s.u-tokyo.ac.jp/$``$makino/grape6.html\].
For the time integration of the system of hydrodynamical equations we use the second order Runge-Kutta-Fehlberg scheme. The time step $`\mathrm{\Delta }t_i`$ for each particle depends on the particleโs acceleration $`๐_i`$ and velocity $`๐ฏ_i`$, as well as on the sound speed $`c_i`$ and the heating vs. cooling balance:
$$\mathrm{\Delta }t=C_n\underset{i}{\mathrm{min}}[\sqrt{\frac{2h_i}{๐_i}};\frac{h_i}{๐ฏ_i};\frac{h_i}{c_i};\frac{u_i}{\dot{u}_i}],$$
where $`C_n`$ is the Courantโs number $`=0.1`$. For computational reasons we fix the minimal integration time step $`\mathrm{\Delta }t_{min}`$.
The main aim of our current work is a detailed study of the isothermal fragmentation processes inside the collapsing โcoldโ molecular cloud complex. For the purpose of finding the fragments and its physical parameters (mass and size) we use our own cluster finding algorithms. In our algorithms we modify the well known and โstandardโ friend-of-friend (FOF) method . Instead of just using the particle positions in the process of โconstructingโ or finding the clusters (fragments) we also use the information about the density distribution inside each potential cluster. On the basis of the density distribution analysis we can finally select in the more accurate way the members of our fragments (clusters). In this sense our method is more close to the so called SKID method, which is well described at the homepage of the Washington University โN-body Shopโ \[http://www-hpcc.astro.washington.edu/tools/skid.html\]. Here the reader can find a more detailed description of this density base method which is specially designed to find the gravitationally bound groups of particles in the N-body like simulations.
One of the features of our cluster finding routine is in the setting of the minimum limit of gaseous particles to 5, in order for a fragment to form. In other words we donโt count as โrealโ a cluster where the number of particles is less then 5.
CODE TESTING
The self gravitating collapse of an initially isothermal โcoldโ gas sphere has been a common test problem for different SPH codes . Following these authors, for the testing of our code, we calculate the adiabatic evolution of the spherically symmetric gas cloud of total mass M and radius R. For the initial internal energy per unit mass we set the value: $`u=0.05\frac{\mathrm{G}\mathrm{M}}{\mathrm{R}}`$. The initial density profile of the cloud calculates as:
$$\rho (r)=\frac{\mathrm{M}}{2\pi \mathrm{R}^2}\frac{1}{r}.$$
We distribute randomly the gas particles inside the set of spherical shells in a manner that reproduces the initial density profile. At the start of the simulation the gas particles are at rest. For the presentation of the results we use a system of units where G = M = R = 1.
In Fig. 1 we show the time evolution of the different types of energy and the relative total energy error during the calculation. For comparison of our test results we also plot the energy error results from the serial variant of the GADGET public access SPH-TREE code with โstandardโ parameters for the 32,000 particles \[http://www.mpa-garching.mpg.de/gadget/\].
During the central bounce around $`t1.1`$ most of the kinetic energy is converted into heat, and a strong shock wave travels outward. For all of these runs the number of neighbors was set $`N_B=50`$ and the gravitational softening was set $`\epsilon =0.01`$. For the integration of the system of equations we use the second order Runge-Kutta-Fehlberg scheme with a fixed time step $`\mathrm{\Delta }t=10^4`$.
The results presented in Fig. 1 agree very well with those of and . The maximum relative total energy error is around 0.05 % even for moderate (8,000) particle numbers. The largest adiabatic test calculation (with 64,000 gas particles up to $`t2.2`$) on an Intel Pentium 4 (3.4 GHz) host machine with a GRAPE6 board took $``$ 3.67 days of total CPU time.
INITIAL CONDITIONS
As an initial condition for our molecular cloud fragmentation study we use a model in which the parameters are comparable with the largest GMC complexes in our Galaxy . For the mass of the system we set M<sub>cloud</sub> = 10<sup>7</sup> M. For the radius of the cloud we set R<sub>cloud</sub> = 100 pc. For an initial density distribution we use the previous formula where $`\rho (r)`$ $``$ $`\frac{1}{r}`$. For the purpose of checking the possible โresolutionโ effects we carry out two sets of runs with 32,000 and 64,000 gas particles (with the corresponding indexes โlowโ and โhighโ). The total gravitational energy of the system in such a case can be easily calculated using the simple formula:
$$\mathrm{E}_{\mathrm{GRA}}^0=\frac{2}{3}\mathrm{G}\mathrm{M}_{cloud}^2/\mathrm{R}_{cloud}.$$
For the initial temperature we set the value which produced the overall ratio of the thermal energy to the gravitational energy of the system at the fixed level $`\alpha `$ $``$ E$`{}_{\mathrm{THE}}{}^{}{}_{}{}^{0}`$/$``$E$`{}_{\mathrm{GRA}}{}^{}{}_{}{}^{0}`$$``$ = 0.075. For the previous fixed mass and radius of the system this condition produced an initial temperature of the cloud T<sub>cloud</sub> $``$ 2200 K. The corresponding sound speed was $`c3.8`$ km/sec. This is consistent with the typical measured โkineticโ temperatures for such GMC complexes .
With these parameters we have an initial central concentration of $`n_0`$ $``$ 10<sup>3</sup> cm<sup>-3</sup>, and a free-fall time in the cloud center of $`\tau _{ff}`$ $``$ 1 Myr. The central Jeans radius is R<sub>J</sub> $``$ 10 pc with the corresponding Jeans mass of M<sub>J</sub> $``$ 10<sup>5</sup> M. Initially we give the whole system a rigid rotational velocity distribution with an angular velocity value $`\mathrm{\Omega }_{cloud}`$ which we set to the unity of $`\mathrm{\Omega }_0`$ = V<sub>0</sub>/ R<sub>cloud</sub> where:
$$\mathrm{V}_0\sqrt{\mathrm{G}\mathrm{M}_{cloud}/\mathrm{R}_{cloud}}.$$
Using our parameters this velocity is equal to V<sub>0</sub> $``$ 21 km/sec.
The main rotational parameters ($`\omega `$ and $`\beta `$) for our two sets of models are listed in Table 1.
As we can see from Table 1 the initial ratio of the rotational (or kinetic) energy of the motion the the fragments ($`\beta `$), even in a last models with โhighโ rotational parameter, donโt exceed more than a few % from the gravitational bounding energy of the cloud. This is even less for the initial ratio of the systems thermal energy to the gravitational energy ($`\alpha `$ = 7.5 %). In all models, usually after the first few Myr of evolution, these situations have changed. The ratio $`\beta `$ is rising to the approximate value of 0.5 or even more. Its mean is when the cloud starts the process of intensive isothermal fragmentation and the whole system of fragments becomes almost fully โrotationallyโ supported.
RESULTS
In Fig. 2 we show the time evolution of the total number of clusters N<sub>c</sub> and the total mass fraction inside these clusters $`\varphi `$ during the simulations. Starting from $``$ 5 Myr more than 80 % of the total mass is already concentrated inside the fragments. At around 6 Myr already almost 95 % of the total mass is inside the clusters. Fig. 2a shows the results for the โslowโ rotating model with $`\omega =0.1`$. In Fig. 2b we show the evolution of the โfastโ rotating model with $`\omega =0.3`$.
In Fig. 3 we show four different snapshots of the integrated cluster distribution function (ICMF) for two selected moments of time with two different rotational parameters $`\omega `$. For practical numerical reasons we use the ICMF instead of the CMF (which is sometimes in the literature also called the Differential Cloud Mass Function DCMF). Here is a simple definition of the ICMF:
$$\mathrm{ICMF}=_0^M๐\mathrm{N}/๐\mathrm{M}.$$
Basically it shows how many clouds we have from zero mass to any fixed mass (M). Because the CMF is usually approximated with the power law: $`\mathrm{M}^\gamma `$ in this case the ICMF we can be simply derived as $`\mathrm{M}^{\gamma +1}`$.
The reason for using the ICMF instead of the CMF (DCMF) is that the averaging and slope definition is mathematically better due to the integrated CMF (which is a monotone function) and because the histograms in this case donโt have any โholesโ. Of course when we compare our results with the observed (differential) CMF slope we need to subtract one from the ICMF slope to get the corresponding CMF slope.
In Fig. 3 we can see that in most cases the ICMF slope lies between -0.5 and -0.7 (the corresponding CMF is -1.5 and -1.7). The models with slow rotation always have a significantly lower value of the slope.
The ICMF slope time evolution for the set of our models with different rotation parameters are presented in Fig. 4. The models with initial โslowโ rotational parameters give the ICMF average slope a level of -0.8 (which corresponds to the CMF slope -1.8). The โfastโ rotating models give the ICMF slope a level of -0.4 (CMF slope -1.4).
On average all models show very close values of the CMF slopes in comparison with the observed values of $`\gamma `$ in the different parts of our Galaxy.
The slow rotation models systematically show the slope more close to the observed values in the outer part of our Galaxy ($`\gamma `$ -1.8$`\pm `$0.03). In contrast, the fast model CMF slopes is more consistent with the observations from the central part of our Galaxy $`\gamma `$ -1.5$`\pm `$0.1.
Of course our simulations are time and also resolution limited, but even in this case we can derive a statement about the two significantly different types of โpopulationโ in the molecular cloud distributions. The key parameter which produces the different CMF slopes is the initial rotational parameter of the forming (and subsequently fragmenting) GMC.
CONCLUSIONS
In this paper we present a study of the subsequent (runaway) fragmentation of the rotating isothermal GMC complex. Our own developed GRAPE based Smoothed Particle Hydrodynamics (SPH) gas-dynamical model successfully reproduced the observed Cloud Mass-distribution Function (CMF) in our Galaxy. The steady state CMF is quickly established during the collapse approximately on a scale of a few free-fall time in the central parts of the modeled GMC.
One of the key points in our model is that using our results we can naturally explain the source of possible differences between the observed slope on molecular clouds mass distribution function in the Galactic center and the outer regions of our Galaxy.
The basic idea, is what if the GMC formed as a result of the galactic disk instability on the scale of the disk height ($``$100 pc). In such a case the initial angular momentum of the forming GMC can be defined by the Coriolis force during the formation inside the differentially rotating disk. Therefore the central GMC has a bigger $`\beta `$ and the external GMC has a smaller rotational parameter.
According to our models this produces the different slopes of the resulting CMF during the runaway fragmentation process inside the system. The observed CMF gives to the central parts of Galaxy a slope well approximated with the value $`\gamma `$ -1.5$`\pm `$0.1 and for the outer parts of the Galaxy the approximate value $`\gamma `$ -1.8$`\pm `$0.03 .
Our results for the โslowโ and โfastโ rotating models give us exactly the same slopes with very good agreement with the recent observations. The โslowโ models corresponds to the initially more slowly rotating GMC in the outer parts of the Galaxy. The โfastโ rotating models corresponds to the GMC in the central part of the Galaxy. The central GMC can initially get more angular momentum from the differential rotation of the galactic disk during the process of GMC formation itself.
Our numerical investigation clearly shows that one of the key parameters, which determines the observed slope of the present day molecular CMF in different parts of our Galaxy, is the initial ratio of the rotational (turbulent) and gravitational energy inside the forming GMC.
ACKNOWLEDGEMENTS
P.B. wish to express his thanks for the support of his work to the German Science Foundation (DFG) under the grant SFB-439 (sub-project B5). P.B. work was also supported by the following grants: NNG04GJ48G from NASA, AST-0420920 from NSF and by HST-AR-09519.01-A from STScI. He is very grateful for the hospitality of the Astronomisches Rechen-Institut (Heidelberg, Germany) where the part of these work has been done. The work of the authors was also supported by the Ukrainian State Fund of Fundamental Investigation under the project 02.07.00132.
The calculation has been computed with the Mitaka Underground Vineyard (MUV) GRAPE6 system of the National Astronomical Observatory of Japan. The authors are want to express the special thanks for our colleagues Naohito Nakasato (Computational Astrophysics Group, RIKEN) for his constant help and support in the process of using the NAOJ GRAPE6 computational facilities.
The authors are also very grateful to Dan Batcheldor (Astrophysics Group, RIT) for his constructive comments to the first variant of the paper.
|
warning/0506/quant-ph0506110.html
|
ar5iv
|
text
|
# Optical generation of matter qubit graph states
## I Introduction
Despite significant and exciting experimental progress in recent years, the physical realization of a full-scale quantum computer (QC) remains a tremedous challenge Nielsen . In many systems excellent single qubits have already been realized (notably, of ions in a trap Blatt ; Wineland , NV centres in diamond NVDiamond ; NVDiamond2 , etc). However, few systems have demonstrated controlled qubit-qubit coupling between pairs taken from more than four qubits, and achieving the necessary exquisite control remains highly problematic. In general it is difficult to simultaneously satisfy the two key requirements of coupling diffent subsystems in a controlled manner, while at the same time shielding the system from its environment Nielsen . In the majority of QC schemes, some direct physical interaction is supposed to generate the two-qubit operations (e.g., phonon modes among trapped ions Blatt ; Wineland , Fรถrster interactions between excitions in semiconduction quantum dots Forster , etc). Thus one calls for the qubits to strongly interact with selective parts of their environment (namely, other qubits and the control mechanisms) while avoiding interactions the rest of the environment to a near perfect degree. This is obviously an challenging prescription.
As an alternative to employing a direct physical interaction between qubits, one can exploit the entangling power of *measurements*. A suitable measurement, at least for certain outcomes, will have the effect of projecting previously separate qubits into a highly entangled state. This idea has been explored as a route to QC using photon qubits in a linear optical apparatus. Measurement-based gates have indeed been shown to be sufficient for universal gate-based quantum computation KLM . However, in order to achieve each logical gate with high probability, one must prepare and then consume large auxiliary resources. This necessity is essentially due to the small probability of success of the elementary quantum gates based on auxiliary systems and measurements Gates .
One way to reduce this overhead is to exploit the idea of one-way computing Old ; Long . In this approach one would prepare a certain multi-qubit entangled state, a cluster Old or a graph state Long ; Graphs , prior to the computation. This state has the property that the computation can then proceed purely by single-qubit measurement โ essentially consuming the graph entanglement as a resource. Recently there has been a successful proof-of-principle experiment realising a 4-qubit cluster state ZeilingerNat . A key advantage of the one-way computing strategy is that it introduces a degree of separation between the act of creating entanglement and the act of executing the computation. Thus we need not expend the effort needed to ensure that each entangling operation succeeds with high probability โ we can tolerate failures during the growth process simply by rebuilding the affected graph section, provided of course that failures are heralded. Indeed, in this spirit various recent schemes Reznik ; MikeCluster ; Terry have shown how to take gate operations that are fundamentally non-deterministic, and use them to construct an such an entangled resource state with certainty.
One particularly attractive possibility is to use matter qubits, with the obvious benefits that they are static and potentially long lived, together with an optical coupling mechanism that creates suitable entanglement. Based on earlier schemes that allow for generating entanglement or realizing quantum gates in matter qubits using flying optical qubits Cabrillo ; Plenio ; Grangier ; Simon ; Duan ; PlenioNew ; Zou ; Rempe , two recent publications Sean ; Almut in particular have explored precisely this possibility. The matter qubits can be completely separate, for example each within its own cavity apparatus, providing that suitable optical channels connect them to a mutual measurement apparatus. The simplest scheme is that of Barrett and Kok (BK) Sean , where one requires only a single beam splitter and two detectors in order to couple pairs of qubits. The elegant BK approach however suffers from the constraint that, even with ideal apparatus, the entangling operation must fail with a probability of $`p=1/2`$. Failures damage the nascent graph state, but because the failure is flagged, or โheraldedโ, the damaged parts can be removed and the growth can continue. Nevertheless, the high rate of destructive failures introduces a considerable overhead BenjComment , especially with certain types of target graph topology. The scheme due to Lim, Beige, and Kwek (LBK) Almut introduces the idea of โrepeat until successโ entanglement, meaning that while failures still occur with probability $`p1/2`$, these failures are essentially passive and one can simply try again. Thus one can construct graph states with a lower overhead, in terms of number of entangling operations, and any topology can be directly implemented. However, the cost for this advance is that the underlying coupling process is more complex: each matter qubit gives rise to a superposition of an โearlyโ and a โlateโ photon in time-bin encoding, which must subsequently enter a beam splitter apparatus simultaneously. This appears to be more challenging relative to the simpler BK scheme, so that it is an open question which scheme is the more practical.
Here our goal is to unite the more desirable features of both these schemes, in particular the simple static optical apparatus of the BK scheme and the non-destructive โrepeat-until-successโ aspect of the LBK approach. Moreover we introduce a vital feature which neither of these approaches possess: we demonstrate a graph growth mechanism which does not require local unitary operations (e.g., flips) to be performed on the matter qubits during the growth process. The growth then becomes purely a sequence of optical excitations, with a corresponding significant increase in speed and considerable reduction in complexity.
We intend that the present paper will form a self contained overview of the entire paradigm that we are advocating, and to this end we include compact analysis of the relevant properties of graph states. We make use of the idea of a minimal graph state (MGS), and make a comparison with the more limited โcluster statesโ which results when the geometry of physical qubits and their neighbors are fixed by experimental constraints. We conclude that there are dramatic savings, in terms of qubits and entangling operations, when one adopts an architecture that can build an MGS directly.
## II Graph States and Cluster States
Graph states Graphs ; Long ; SchlingeOld are multi-qubit entangled states, which can be conceived as having been entangled according to certain pattern of two-qubit phase gates. Formally, this pattern is specified by the adjacency matrix of an (undirected simple) graph $`G(V,E)`$, where $`V`$ denotes a set of $`n`$ vertices associated with the qubits, and edge set $`E`$ reflecting the phase gates (see Fig. 1 (d) for example). The graph state of the empty graph has the state vector $`|\mathrm{\Psi }=|+^n=((|0+|1)/\sqrt{2})^n`$. The state vector of a graph state including edges can then be written as
$`|G={\displaystyle \underset{(a,b)E}{}}P^{(a,b)}|+^n,`$ (1)
with $`P^{(a,b)}`$ corresponding to a phase gate $`P^{(a,b)}=(\mathrm{๐}+\sigma _z^{(a)}+\sigma _z^{(b)}\sigma _z^{(a)}\sigma _z^{(b)})/2`$ between qubits labeled $`a`$ and $`b`$, expressed in terms of Pauli operators. Such graph states are stabilizer states TheStabil , and in turn, every stabilizer state of $`n`$ qubits is locally equivalent to a graph state MaartenPhD ; SchlingeEquiv .
A cluster state (CS) Old is a particular graph state: it is one with an underlying cubic lattice of one, two or three dimensions (see Fig. 2 (a) for example). A cluster state of more than one dimension has the remarkable property that it forms a universal resource for measurement-based one-way computing: having created this state, the actual computation is executed simply by making local measurements Long ; Old . It is universal in the sense that the procedure amounts to effectively implementing an arbitrary unitary on the input qubits.
However, the measurements performed in order to implement some chosen algorithm will include two classes which it is important to distinguish Long ; Old . The first are the Pauli measurements, which we can denote as measurements along the $`X`$, $`Y`$, or $`Z`$ axis. Each such measurement maps a graph state onto another graph state for all outcomes. For example the $`Z`$ measurement effectively deletes the measured qubit (node) and its associated edges, while $`X`$ and $`Y`$ measurements alter the graph according to the rules given in Ref. Graphs . These measurements correspond to the Clifford-part of the computation, and the resulting map on the level of states can always efficiently be determined on a classical computer Long . Having performed all the prescribed Pauli measurements on a cluster, we are left with a minimal graph state (MGS) which is the graph containing the smallest number qubits that is capable of realizing our desired algorithm. The remaining measurements are of the second class: von Neumann measurements in tilted bases. Such measurements take the system out of the graph state prescription and generally cannot be efficiently simulated on a classical computer. In a sense one can think of the Pauli measurements as simply customizing the (initially universal) cluster state into the form that will implement our chosen algorithm, while the more general tilted measurements actually execute the algorithm.
Many physical systems that can generate graph states are in fact limited to cluster state generation, because the physical qubit interactions are limited to some kind of nearest-neighbor (or at any rate, local) form. This applies to implementations in electron spin lattices and optical lattices. However, we are under no such constraint since the physical qubits have no defined geometry OtherExceptions . Instead, we can directly โgrowโ an arbitrary graph, and hence we may prepare the graph state that forms the specific resource for a given quantum algorithm. We would therefore seek to directly build a MGS, shortcutting the creation of the cluster state with its redundant universality. This proves to have dramatic advantages in terms of the number of entanglement operations and qubits needed. In general one finds that a MGS will often exhibit a high vertex degree, and will be contain significantly fewer qubits compared to the graph state that is obtained from a cluster state after measurements along the $`Z`$ basis, essentially merely removing qubits (typically up to an order of magitude). Explicit examples are described later.
## III Description of the Physical Scheme
In this section we describe the physical requirements and processes involved in implementing our proposal. We start by describing the elementary physical systems required. We then outline the action of the beam splitter network, both on simple product states and more importantly when fusing graph states together. Finally we then show how to do this without any single qubit unitaries, and make some concluding comments.
### III.1 Physical Components
In Fig. 1 (a) we indicate the basic energy level scheme that our matter qubits should incorporate. Obviously, a real quantum system may have additional levels, but providing these 3-level dynamics are incorporated, then such a system is suitable. Candidates include an actual atom or ion in a trap, of course, but we may also consider any optically active solid state structure with a discrete spectrum, such as a quantum dot, or NV-diamond centre sta03 . The ground states, labeled with state vectors $`|0`$, $`|1`$, are the qubit basis states. The third level, labeled $`|e`$, provides a mechanism for producing a photon from the atom, conditional on the atom being in state labeled with state vector $`|1`$. That is, there is an externally driven transition from $`|1|e`$ by using a $`\pi `$-pulse, followed by the optical relaxation $`|e|1`$ which emits a single photon into the cavity mode, and eventually โleaksโ out to an external optical network. In a single-mode description, this is characterized by a coupling strength $`\mathrm{\Omega }`$ of the Jaynes-Cummings coupling between the transition $`|1|e`$ and the cavity mode, with decay rate $`\mathrm{\Gamma }`$ of the cavity mode. The system is continuously observed via the photon detectors placed behind the four beam splitter array (4BS). We note that, as remarked in Ref. Almut , if we have a fourth level accessible from $`|e`$, then we can potentially create a photon directly in the cavity without significantly populating $`|e`$ โ this may be advantageous in avoiding dephasing. The state labeled $`|e`$ is also exploited when we wish to make a measurement โ continuous illumination by a laser adjusted to the transition energy will result in a fluorescence conditional on the qubit state. This is a $`Z`$ measurement; measurement in the other directions is accomplished by an appropriate local rotation followed by this fluorescence.
The elementary multi-qubit operation in our proposal is based on a four-port beam splitter (4BS), which is a composite of four ordinary beam splitters, arranged so that every input crosses every other, and finally incident on photon counters. Two types of 4BS will be employed: one without additional phase shifts, the basic network, and one which includes a certain phase-shift corresponding to a factor of $`e^{i\pi /2}`$, the shifter network, for producing certain important cluster states. The action of the 4BS is essentially to โeraseโ information about which cavity a photon originated from, so that a given detector cannot differentiate between matter qubits. Ideally the frequencies of the modes are identical, and other sources of mode matching problems are to a high extent eliminated โ we analyze the effects of realistic imperfections in Section VI.
### III.2 Action on Product States
The analysis to determine the specific projections is straightforward. The most simple interesting case we examine is that of inputting four qubits in the product state corresponding to $`|+^4`$. The action of the optical excitation $`|1|e`$ applied to all qubits, followed by the emission into their local cavities, then results in an equal superposition of all basis vectors containing all binary words,
$`|\varphi ={\displaystyle \frac{1}{4}}{\displaystyle \underset{i,j,k,l=0}{\overset{1}{}}}|i,j,k,l(c_1^{})^i(c_2^{})^j(c_3^{})^k(c_4^{})^l|0.`$ (2)
where the annihilation operators of the respective cavity modes are denoted by $`c_1,\mathrm{},c_4`$.
As the photons propagate through a beam splitter into new modes we employ mappings such as $`(d_j^{},d_k^{})^T=B(c_j^{},c_k^{})^T`$, $`j,k=1,\mathrm{},4`$, where phases of the transmitted and reflected mode are chosen such that $`B`$ is given by
$$B=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}1& i\\ i& 1\end{array}\right).$$
(3)
Thus the network corresponds to a unitary manipulation of the photon states. In this way we eventually obtain the final generation of operators representing photons in the modes upon which the detectors act. Thus we determine the state that results from a particular detector reading. In fact a number of states can occur, including states that are locally equivalent to linear $`4`$-qubit graphs and $`3`$-nodes, as shown in Fig. 1 (c).
### III.3 Fusion of Graph States
The ability to generate such graphs directly from product states appears to be a promising characteristic. However, to properly differentiate the possible outcomes one would require either resolving photon detectors (capable of distinguishing a single photon from a pair, etc) or else one would need to resort to a lengthy asymptotic variant of the โdouble heraldingโ idea in Ref. Sean . These undesirable features result from the fact that the systems state, prior to photon measurement, was not an eigenstate of the total photon number operator: there are elements in the superposition corresponding to $`0,1,\mathrm{},4`$ photons. To avoid the problem we must contrive to introduce a known number of photons. This is the same issue faced in Ref. Almut , where the authors suggest resorting time-bin approach and local flips in order to guarantee that each matter qubit ultimately generates one photon.
We take a different route, based on the idea of fusing graphs together. We will find that we can regard EPR pairs as a kind of raw ingredient from which graphs of arbitrary complexity can be grown deterministically. Recall that an EPR pair with state vector $`|\text{EPR}=(|0,1|1,0)/\sqrt{2}`$ is already LU-equivalent to the simplest non-trivial graph state, the one consisting of two vertices connected by an edge. We use the term LU-equivalent to mean, equivalent up to local unitary operations on individual qubits. Our fusion process exploits existing entanglement within the graphs sections: certain vertex pairs within a graph state can be locally rotated to the subspace $`\text{span}\{|0,1`$,$`|1,0\}`$ โ two such pairs then generate precisely two photons.
Suppose that a โleafโ node exists, i.e., a certain vertex (associated with qubit $`a`$) is attached to only one other vertex (associated with qubit $`b`$) of the graph. This is shown in Fig. 1 (d). Then the state vector $`|\mathrm{\Psi }_G`$ of the entire graph state is of the following form
$$|\mathrm{\Psi }_G=\left(|0,0+|1,0+|0,1P|1,1P\right)|\psi $$
(4)
where the vectors $`|a,b`$ correspond to the qubits $`a`$, $`b`$, and $`|\psi `$ refers to the external, arbitrarily connected part of the graph state, shown inside green bubbles in Fig. 1 (d). We define
$$P=\underset{iN_b\backslash \{a\}}{}\sigma _z^{(i)},$$
(5)
so with index $`i`$ running over the neighbors $`N_b`$ of qubit $`b`$ lying within $`|\psi `$. This state vector is in turn equivalent, up to a unitary rotation $`(\mathrm{๐}i\sigma _y^{(a)})/\sqrt{2}`$ on $`a`$, to
$`|\mathrm{\Psi }_E=(|1,0+|0,1P)|\psi `$ (6)
Having made this transformation, we know that our qubit pair labeled $`a`$ and $`b`$ will emit precisely one photon. If we similarly prepare a second pair of qubits, associated with a different graph (or, a different part of the same graph) then the two pairs of qubits can be employed in our 4BS device and will generate precisely two photons. As indicated in Fig. 1, we have considered two variants of the beam splitter network: one with and one without a phase shifter. A simple analysis determines that in both cases there are two possible classes of outcome. The two photons may arrive in a single detector, in which case the effect is simply equivalent to applying local phase gates to the matter qubits. Alternatively, the two photons may enter different detectors, in which case the two pieces of graph are fused together in a fashion we specify presently. Identifying the various outcomes does not require counting photon detectors, since we know there are two photons in total (but such detectors may be useful in fighting errors when taking imperfections into account, see Section VI).
The two classes of outcome are equally probable. In the case of the former one can try again without pausing to correct the local phases, which can be fixed after the eventual successful fusion. The average number of attempts required is two. This is then a โrepeat-until-successโ scenario equivalent to the one first observed in Ref. Almut . The particular form for our fused graph depends on whether the phase shifter was employed. If we do employ the shifter, and supposing that we input two EPR pairs, then the resulting state is LU-equivalent to a linear four-qubit graph as shown in Fig. 1 (e). If instead we use our 4BS without the phase shifter, i.e. the basic network, we can couple arbitrary graph fragments according to the rule shown in Fig. 1 (f). For the example of joining two EPR states, we show the outcome states conditioned on which detectors click, and their probabilities in Table 1. These observations lead us to regard EPR pairs as the basic resource for graph growth: EPR pairs can be generated easily by a single beam splitter using the BK scheme, or equivalently we have observed that they can be obtained from our 4BS network by choosing to excite just two of the four qubits. The combination of the two coupling processes Fig. 1 (e) and Fig. 1 (f) then allows graphs of arbitrary complexity to be built. Recall that a graph can be โprunedโ, i.e., qubit nodes can simply be removed, by making a $`Z`$ axis measurement, while other useful transforms result from $`X`$ or $`Y`$ measurements Graphs . Indeed, a recent preprint newBrowne makes ingenious use of measurements on leaf structures, reminiscent of those occurring at the fusion point in Fig. 1(f), for qubit loss tolerance in graph states.
### III.4 Growing a Graph Without Employing Local Gates
We have seen that we can grow graphs by transforming selected qubits to an EPR-type basis prior to fusion, and then applying additional LU operations to transform the โrawโ resultant state back to a graph state. But, can we avoid these local transformations? It is evidently necessary to employ single qubit rotations at two stages: the very beginning in the entangling procedure, where we must take โfreshโ qubits and prepare them in $`|+`$ in order to synthesize the EPR pairs which we regard as our basic ingredient, and the very end where we will wish to rotate qubits prior to our fluorescence measurement, in order to synthesize measurements along some general axis. Remarkably, we can in fact omit the numerous local rotations during graph state growth. We find that, within a light constraint on the growth process, we can ensure the state remains LU-equivalent to a graph state at each growth step.
To see that this is possible consider the following argument. Suppose that we have some multi-qubit state vector $`|\mathrm{\Psi }`$ which meets the following two conditions:
(i) The state vector $`|\mathrm{\Psi }`$ is equivalent up to local unitary operations to $`|\mathrm{\Psi }_G`$ corresponding to a graph $`G`$ of the form as in Fig. 1 (d).
(ii) Regarding the pair of qubits labeled $`a`$ and $`b`$,
$$|0,1|\mathrm{\Psi }|=|1,0|\mathrm{\Psi }|,|0,0|\mathrm{\Psi }|=|1,1|\mathrm{\Psi }|=0.$$
(7)
From (i) and recalling Eqn. (6) we know that $`|\mathrm{\Psi }`$ is LU-equivalent to a state vector $`|\mathrm{\Psi }_E=(|1,0+|0,1P)|\psi `$ since that is itself LU-equivalent to the graph state vector corresponding to $`G`$. The additional constraint (ii) implies that our state vector can be written as
$$|\mathrm{\Psi }=(|1,0U+|0,1\stackrel{~}{U})|\psi $$
(8)
where $`U`$ is a product of local unitaries acting on $`|\psi `$, i.e., acting in the Hilbert spaces of the qubits other than those labeled $`a`$ and $`b`$, and $`\stackrel{~}{U}=\mathrm{exp}(i\varphi )UP`$ with $`\varphi [0,2\pi )`$ an arbitrary phase. Now let us apply the 4BS process of Fig. 1 (f) to $`a`$ and $`b`$, along with an equivalent pair from some analogous graph state (or, another part of the same graph state). On failure (with probability $`1/2`$) the process simply introduces some known local phases, which do not alter our prescription. On eventual success we generate a state vector
$$|\mathrm{\Psi }_{\mathrm{tot}}=(|XU_{\mathrm{tot}}+|\overline{X}\stackrel{~}{U}_{\mathrm{tot}})|\psi _{\mathrm{tot}}.$$
(9)
Here $`X`$ is a binary word with two zeros, two ones, and $`\overline{X}`$ is its complement. The vector $`|\mathrm{\Psi }_{\mathrm{tot}}`$ refers to the entire state vector of the fused system, and $`|\psi _{\mathrm{tot}}`$ is the state vector for all qubits except the four that coupled via the 4BS. Similarly, $`U_{\mathrm{tot}}`$ is some product of local unitaries on $`|\psi _{\mathrm{tot}}`$, and $`\stackrel{~}{U}_{\mathrm{tot}}=\mathrm{exp}(i\theta )P_{\mathrm{tot}}U_{\mathrm{tot}}`$, where
$$P_{\mathrm{tot}}=\underset{i}{}\sigma _z^{(i)}$$
(10)
with the index running over neighbors of either of the original vertices (but excluding mutual neighbors). The phase $`\theta [0,2\pi )`$ is determined by $`\varphi `$ and its counterpart in the second pair, together with phases introduced in any failures preceding the successful fusion. This state vector $`|\mathrm{\Psi }_{\mathrm{tot}}`$ is indeed LU-equivalent to the desired fused graph state of Fig. 1 (f). Moreover, because $`X`$ has two zeros and two ones, if we nominate one of those four qubits to be a new โvertexโ qubit $`b`$, two of the three remaining โleafโ qubits are available to be labeled as $`a`$ to satisfy (i) and (ii). Thus we can go on to perform further fusions using $`|\mathrm{\Psi }_{\mathrm{tot}}`$. To conclude the argument we need only observe that conditions (i) and (ii) are met by simple EPR pairs, and by the state resulting from fusing two EPR pairs via the process depicted in Fig. 1 (e), i.e. the state that is LU-equivalent to a linear four qubit graph state. Thus these simple states can act as the initial building blocks as we construct a complex graph.
Then provided we are prepared to measure out one in every three of the โleafโ nodes which occur at each fusion point, we can grow our entire graph from EPR pairs without the use of local unitary operations during the growth process. This constraint is extremely light: we would rarely wish to use all three leaves, and in any case the number of leaves can be increased by two simply by fusing an EPR pair, which adds one to the number of leaves that are eligible in the sense of property (ii). Of course, the LU operations needed to map the final state to the desired ultimate graph state can subsumed into the rotation which in any case precedes measurement. The growth process is therefore entirely one of optical excitation and detector monitoring. One can anticipate than in many systems, the cost in efficiency arising from following the constraint would be vastly outweighed by the increase in growth speed.
### III.5 Further Remarks on Graph Growth
There is one additional comment to make regarding the speed of our protocol: in the scheme of Ref. Sean it is always necessary to wait a period after the initial measurement to ensure there are no further photons in the apparatus. The fidelity of the entangled states is only high if this wait period is long compared to the typical time for a photon to be detected. This additional waiting time necessary in Ref. Sean is not necessarily long, given that photon emission from these sources is approximately exponential, governed by the time scale $`1/\mathrm{\Gamma }_{\text{Slow}}`$. By contrast, because we contrive to have precisely two photons in the apparatus, once we see two detection events (either in different detectors, or, given resolving detectors as discussed later, within one detector) we have no need for such a wait. One should hence expect a factor of about $`5`$ in difference concerning the speed of this step in the respective schemes.
In a mature form of the architecture described here, one would envisage coupling the many qubits by a form of $`N`$-port all-optical router, as used in integrated optics, so that our qubits can remain static and we can choose which of them will couple by suitably setting the router and optically exciting only that subset. Devices relevant to this technology have already been developed for classical optical communications BellLambda . This would permit direct growth of graphs with an arbitrary topology, and in particular the ability to directly entangle arbitrary qubits gives a non-local architecture NL1 ; NL2 ; NL3 . Such non-local architectures may prove to have an advantage in quantum fault tolerance and error correction Svore .
## IV Characteristics of the derived graph states
As discussed above and illustrated in Fig. 1, our protocol can generate graphs of arbitrary topology, including nodes of high degree. We argued earlier in Section II that, given such an architecture, one should aim to create minimal graph states (MGS). The advantage in terms of resource consumption when preparing appropriate MGS compared to standard cluster states (CS) can be quite significant.
### IV.1 Minimal Graph States
In the best known scheme for an $`n`$-qubit quantum Fourier transformation Long , the number of required qubits is $`C_{\text{Fourier}}(n)=8n^2+O(n)`$ for a cubic cluster, followed by first $`X`$, $`Y`$, and $`Z`$ measurements, then tilted measurements. The MGS embodies $`G_{\text{Fourier}}(n)=(3/2)n^2+O(n)`$ qubits. In turn, for the quantum adder,
$`C_{\text{Adder}}(n)=312n+O(1),G_{\text{Adder}}(n)=16n+O(1)`$ (11)
Long . Hence, one may gain more than an order of magnitude in resource consumption. For the $`3`$-qubit Toffoli gate that we use here as our illustration, we have $`G_{\text{Toffoli}}=13`$ versus $`C_{\text{Toffoli}}=65`$, so a factor of $`5`$ difference in the number of qubits.
### IV.2 Generation of Edges
Concerning the actual preparation of the graph states, we emphasise two points: firstly, when introducing edges with a physical interaction, one should always prepare the LU-equivalent graph state corresponding to the graph with the minimal number of edges. Or, more specifically, the graph state with the minimal number of edges that is equivalent up to local Clifford unitaries Remark , which merely amount to a local Clifford basis change. This has also been emphasised in Ref. Perdrix . Fortunately, an efficient algorithm is known to check full local Clifford equivalence MaartenEquivalent . Any graphs that correspond to local Clifford equivalent graph states can be related to each other with a successive application of local complementations Graphs ; MaartenEquivalent . Also, it is known how many different graph states are contained in an equivalence class with respect to local Clifford unitaries MaartenPhD .
Secondly, the present scheme seems particularly suitable to prepare graph states of graphs involving vertices with a high vertex degree in a single step. In a CS after measurement of the unused qubits along the $`Z`$ direction, it suffices to have vertices with a maximal vertex degree of $`3`$. This is obviously the lowest possible for a graph less trivial than a linear cluster state. From the example of Fig. 2 one would suspect that a typical MGS may need higher degree nodes, and this is an important question in considering how they can be efficiently constructed. To explore this point we consider the โmaximal vertex degreeโ, by which we mean, the highest degree of any vertex in the graph. The vertex degree in a MGS can in principle take any value. The maximal vertex degree is notably not invariant under local Clifford unitaries Remark . To render the notion of maximal vertex degree meaningful, we have to take its minimum value when minimized over all local Clifford unitaries. For a GHZ state of $`n`$ qubits, it can easily be shown that it has a smallest maximal vertex degree of $`n1`$. So we immediately see that it is meaningful to talk about โhighly connected graph statesโ. For the resource state for the $`[7,1,3]`$-CSS code as considered in Ref. Graphs we find that in the whole orbit under local Clifford unitaries the smallest maximal vertex degree is $`6`$, whereas the largest is $`34`$. In the three-qubit quantum Fourier transform the smallest maximal vertex degree is $`4`$. Thus we see that high degree vertices are indeed generally unavoidable in a MGS: any scheme that claims to be able to directly and efficiently construct such a state must be able to create graph states with vectices of high vertex degree. The BK scheme, for example, appears somewhat limited by the increased difficulty of making the high degree vertices associated with graph states; the high rate of destructive failures leads one to take an indirect approach as depicted in Fig. 2(e), with an associated cost in resources. The scheme presented here is among the few that generate high degree nodes directly OtherExceptions , as one can quickly see by considering the fusion rule depicted in Fig. 1(f).
### IV.3 Cluster States
Of course, if for some reason we wished to generate a conventional CS rather than the MGS specific to some given algorthim, then we can do so efficiently. As an exercise, let us conclude this section with a comment on the required number of steps in the preparation of a CS with an underlying two-dimensional cubic lattice. We will count resources in terms of the number of applications of the shifter $`N_{\text{Shifter}}`$, with two EPR pairs fed in in each instance, and of the basic network $`N_{\text{Basic}}`$. The basic building block can be taken to be a cross shape of length $`4`$, requiring four EPR pairs, and the application of two shifter and one basic networks. One row of width $`n`$ can be build using $`2n`$ invokations of the shifter network and $`2n`$ uses of the basic network. A two-dimensional cluster state on a $`n\times n`$ cubic lattice hence requires
$$N_{\text{Shifter}}(n)=2n^2,N_{\text{Basic}}(n)=3n^2n.$$
(12)
## V Just in Time Graph Creation
In a lattice system, one may be well advised to prepare the multi-particle cluster state in one step, exploiting a natural nearest-neighbor interaction. However, in a scheme as considered here, there is no motivation to prepare graph edges far in advance of the eventual measurement operations that will consume them. One should therefore avoid doing so since this gives rise to unnecessary errors due to the graduate degradation in phase integrity from decoherence. Instead, one can introduce new edges and vertices for our MGS shortly before it is needed, in a manner analogous to the block-by-block process of Ref. Terry but at a finer scale. By analogy to the term used in classical computing, this may be referred to as just in time graph state generation Prehist (see also Refs. Long ; Elham ). As noted earlier in Section III, although we may require local unitaries to create the EPR pairs which constitute our โraw ingredientโ, the remaining steps involved in generating new graph structure can take place without such manipulations.
One can easily confirm that this is possible, even though the measurements on earlier parts of the graph are tilted and therefore have taken the system to a non-graph state. Consider a graph state with graph $`G=(V,E)`$ corresponding to the whole computation: let us consider the state vector after $`k`$ measurements on vertices $`a_1,\mathrm{},a_k`$, forming a vertex set $`V_kV`$. The resulting state vector after measurements in direction $`r_k`$ โ depending on the measurement outcomes $`s_1,\mathrm{},s_k\{1,1\}`$, โ in this temporal order is given by $`P_k|G=_{j=1}^k(\mathrm{๐}+(1)^{s_j}r_j(s_1,s_2,\mathrm{},s_{j1})\sigma ^{(a_j)})/2)|G`$, where $`\sigma ^{(a_j)}`$ is the vector of Pauli matrices at vertex labeled $`a_j`$. Note that the appropriate measurement basis $`r_j(s_1,s_2,\mathrm{},s_{j1})`$ at step $`j`$ depends on the earlier measurement outcomes. Yet, at this point we could have just prepared
$`|G_k={\displaystyle \underset{(a,b)E_k}{}}P^{(a,b)}|+^n`$ (13)
before performing the above measurements, where $`E_k=\{(a,b)E:aV_k\text{ or }bV_k\}`$. Thus the only constraint on this just in time approach, is that one should ensure that all edges in $`E_k`$ are appropriately entangled in step $`k`$, see Fig. 2 (d).
## VI Error Analysis
A physical implementation of this scheme would be subject to a number of possible errors. Our protocol relies on the subsystems being identical, so that their outputs are indistinguishable. Thus, mismatching parameters will lead to a reduction in performance. Other errors include dephasing of the matter qubits, imperfect optical excitation, phase noise (or drift) in the optical apparatus, and photon loss. section the majority of our analysis will focus on errors due to mismatched subsystems; we will comment on the other error sources at the end.
Since the results described here involve the detection of two photons arriving from a source, the qualitative effect of errors will be similar to the results presented in Ref. sta03 , and we analyse the system using similar methods. For the purpose of this analysis, we assume that each atom is a three level system, with degenerate ground states, labeled $`|0`$ and $`|1`$, and a level $`|e`$ that is optically coupled to $`|1`$, with an energy $`\mathrm{}\omega _e`$. The cavity is taken to have a frequency $`\omega _c=\omega _e+\mathrm{\Delta }`$, where $`\mathrm{\Delta }`$ is nominally zero. The transition $`|e|1`$ couples to the cavity mode with a strength $`\mathrm{\Omega }`$, and the cavity mode decays with a rate $`\mathrm{\Gamma }`$. Thus, we consider here imperfections in $`\mathrm{\Delta }_j`$, $`\mathrm{\Omega }_j`$ and $`\mathrm{\Gamma }_j`$ for each atom-cavity subsystem, $`j=1,\mathrm{},4`$. A comparable analysis was performed in Ref. Sean , so that we can compare that two-qubit scheme with the present four-qubit protocol. Remarkable, we find that the sensitivity to defects in the apparatus is essentially the same.
### VI.1 Continuous Measurement Analysis
In the following we describe the dynamics of a three-level atom in a leaky cavity in the Schrรถdinger picture, continuously monitored by a photodetector. Its stochastic dynamics under continuous measurement can be described using a quantum-jump approach, leading to a piecewise deterministic classical stochastic process in the set of all pure states gar00 ; Knight ; Holevo . The continuous time evolution is governed by an effective Hamiltonian, interrupted by discontinuous โjumpsโ reflecting photon detection. This continuous part is described by the Schrรถdinger equation $`_t|\stackrel{~}{\psi }=i\stackrel{~}{H}|\stackrel{~}{\psi }`$ for the unnormalised state vector $`|\stackrel{~}{\psi }`$, with the non-Hermitian, effective Hamiltonian
$`\stackrel{~}{H}=\omega _e|ee|+(\omega _ci\mathrm{\Gamma }/2)c^{}c+\mathrm{\Omega }(c|eg|+c^{}|ge|),`$
where $`c`$ is an annihilation operator for the cavity mode. The decreasing vector norm of $`|\stackrel{~}{\psi }(t)`$ due to the non-unitary evolution, $`\stackrel{~}{U}(t)=e^{i\stackrel{~}{H}t}`$, leads to the cumulative density function for the time at which the photodetector registers a photo-count,
$$P(t)=1|\stackrel{~}{\psi }(t)^2.$$
(15)
This in turn governs the waiting time distribution in the stochastic process. Correspondingly, a detector click corresponds to a โjumpโ in the state of the system according to $`|\psi \gamma c|\psi `$, where $`\gamma =\mathrm{\Gamma }^{1/2}`$ gar00 .
For the system of four atoms in cavities, with four detectors following a beam splitter network, the only change to this prescription is that $`\stackrel{~}{U}(t)=_j\stackrel{~}{U}_j(t)`$ and a click in detector $`k`$ effects a โjumpโ in the state according to $`|\mathrm{\Psi }d_k|\mathrm{\Psi }`$, where $`d_k`$ is related to the cavity mode operators according to, $`d_k=_j\beta _{k,j}\gamma _jc_j`$, where $`\beta =(\beta _{k,j})`$ is the unitary induced by the beam splitter network. Note that we use $`j=1,\mathrm{},4`$ to label subsystems and $`k=1,\mathrm{},4`$ to label detectors.
We examine the effect of errors on the basic 4BS network which creates 4-GHZ states from two EPR pairs, so that the initial state vector is given by $`|\mathrm{\Psi }(0)=|\mathrm{EPR}_{1,2}|\mathrm{EPR}_{3,4}`$. We treat the initial excitation of the protocol, $`|1|e`$, as instantaneous and ideal, and examine the effect of mismatched system parameters $`\mathrm{\Gamma }_j`$, $`\mathrm{\Delta }_j`$ and $`\mathrm{\Omega }_j`$ on the subsequent emission and detection process. The attraction of our protocol is that the initial state is in the two-excitation subspace, so we expect to register exactly two detector counts. The state vector of the system at the end of the protocol is conditional upon which detectors clicked, $`k_1`$ and $`k_2`$, and at what times, $`t_1`$ and $`t_2=t_1+\mathrm{\Delta }t`$,
$$|\stackrel{~}{\mathrm{\Psi }}(t_1,k_1;t_2,k_2)=d_{k_2}\stackrel{~}{U}(\mathrm{\Delta }t)d_{k_1}\stackrel{~}{U}(t_1)|\mathrm{\Psi }(0).$$
(16)
This particular outcome occurs with a probability density function given by $`p(t_1,k_1;t_2,k_2)=|\stackrel{~}{\mathrm{\Psi }}(t_1,k_1;t_2,k_2)^2`$. Integrating $`p`$ over $`t`$ and $`\mathrm{\Delta }t`$ yields the probabilities in Table 1, as required. Note that for the *ideal* case, the distribution of $`p`$ does not actually depend on $`k_1`$ and $`k_2`$; the only dependence is an overall multiplicative factor such that Table 1 is satisfied. An example is shown in Fig. 3.
For a given combination of detectors and times, we calculate the fidelity of the resulting state with respect to the ideal outcomes, shown in Table 1, which depend only on the detectors, and not the times: $`f(t_1,k_1;t_2,k_2)=|\mathrm{\Psi }(k_1,k_2)|\mathrm{\Psi }(t_1,k_1;t_2,k_2)|^2`$ (note that we have renormalised the state, denoted by the lack of a tilde). In order to give a fair estimate of the expected fidelity, we compute the time-averaged fidelity for outcomes where different detectors click,
$`F`$ $`=`$ $`2{\displaystyle \underset{k_1k_2}{}}{\displaystyle _0^{\mathrm{}}}๐t_1{\displaystyle _{t_1}^{\mathrm{}}}๐t_2p(t_1,k_1;t_2,k_2)f(t_1,k_1;t_2,k_2),`$ (17)
$`=`$ $`2{\displaystyle \underset{k_1k_2}{}}{\displaystyle _0^{\mathrm{}}}๐t_1{\displaystyle _{t_1}^{\mathrm{}}}๐t_2\stackrel{~}{f}(t_1,k_1;t_2,k_2),`$
where $`\stackrel{~}{f}(t_1,k_1;t_2,k_2)=|\mathrm{\Psi }(k_1,k_2)|\stackrel{~}{\mathrm{\Psi }}(t_1,k_1;t_2,k_2)|^2`$, and the factor of 2 accounts for the fact that in the ideal case the probability of fusing the states is $`1/2`$. $`\stackrel{~}{f}`$ consists of a sum of exponentially decaying terms, so we compute $`F`$ analytically, though the expression is rather lengthy.
When $`\mathrm{\Omega }_j=\mathrm{\Omega }`$ and $`\mathrm{\Gamma }_j=\mathrm{\Gamma }`$, we find $`F=1`$, so that the protocol works perfectly. Otherwise, the process works with reduced fidelity, as shown in Fig. 4. Since we are considering three error parameters per subsystem, we cannot show the dependence of $`F`$ on all of them graphically, however for small perturbations to the parameters, we can straightforwardly compute the dependence to quadratic order. In what follows, we work in units for which $`\mathrm{\Omega }=1`$. We find that $`F1_jฯต_j^TM_sฯต_j_{j_1j_2}ฯต_{j_1}^TM_\times ฯต_{j_2}`$, where
$$ฯต_j=(\delta \mathrm{\Delta }_j,\delta \mathrm{\Gamma }_j,\delta \mathrm{\Omega }_j)$$
(18)
is the vector of parametric errors in subsystem $`j`$ and $`M_{s,\times }`$ are the coefficient matrices. For a critically damped atom-cavity system, $`\mathrm{\Gamma }=4\mathrm{\Omega }`$, the coefficient matrices are
$`M_s=\left[\begin{array}{ccc}\hfill \frac{5}{128}& \hfill 0& \hfill 0\\ \hfill 0& \hfill \frac{3}{32}& \hfill \frac{3}{16}\\ \hfill 0& \hfill \frac{3}{16}& \hfill \frac{9}{16}\end{array}\right],M_\times =\left[\begin{array}{ccc}\hfill \frac{3}{128}& \hfill 0& \hfill 0\\ \hfill 0& \hfill \frac{1}{32}& \hfill \frac{1}{16}\\ \hfill 0& \hfill \frac{1}{16}& \hfill \frac{3}{16}\end{array}\right].`$ (25)
### VI.2 Photon Loss
Given an ideal apparatus, without any photon loss or detector failure, our fusion process would merely require four simple non-photon-number-resolving detectors. However, in practice any near-future experiment will certainly suffer significant photon loss. This appears potentially very damaging to our scheme (and to that of Ref. Almut , but not to that of Ref. Sean ), because we may misinterpret a photon loss as two photons entering a single detector. In order to counter this issue, we would require a limited degree of photon resolution at the detectors - specifically, we must differentiate the three cases: $`0`$, $`1`$ and more than one photons. This suffices to detect a photon loss event โ we would then reset the associated matter qubits and rebuild the graph section (analogously to Ref. Sean ). Importantly, graph state fidelity will not be affected by undercounts, reflecting a non-unit detection efficiency, which is the dominant problem in real world photon detector technologies. Undercounting is equivalent to a lossy channel followed by a perfect photon counter, and therefore detector inefficiency is simply subsumed into the total photon loss rate. Detector over-counting, i.e. dark counts, are potentially harmful in the present scheme and those of Ref. Sean ; Almut . Fortunately, since we know that correct operation of the scheme generates precisely two photons, we will successfully identify any photon loss event unless the loss occurs at the same time that a detector is independently subjected to a dark count, a process that is expected to happen with a very small probability.
One could construct an adequate detector simply from two non-photon-resolving detectors, together with a fast switch. This would exploit that fact that when two photons are incident on one detector, they are typically separated by a time interval of order $`1/\mathrm{\Gamma }`$, see Fig. 3; this time can be made long enough to trigger a pockels cell to redirect a possible second photon into a second detector, obviating the need for a true number-resolving detector. On occasions when the two photons occur too close together for the second to be redirected, we simply undercount and assume photon loss has occurred.
To summarize, the present scheme is potentially more susceptible to photon loss than the โdouble heraldingโ scheme of Ref. Sean . However, the issue can be dealt with using detector technologies that remain relatively simple โ we do not require high fidelity photon number resolving detectors in order to generate high fidelity graph states.
### VI.3 Other Errors
This shows that the protocol is most sensitive to errors in the atom-cavity coupling rate, and less sensitive to detuning or the cavity leakage rate (the same hierarchy as observed in Ref. Sean ) . The method used here can be adapted to include dephasing errors, as it was in Ref. sta03 , however it is rather more cumbersome, so for brevity we do not analyse it in detail here. In Ref. sta03 , it was found that dephasing was minimised when $`\mathrm{\Gamma }\mathrm{\Omega }`$, since such a critically damped system has the shortest lifetime of excitations in the system. It was also noted that dephasing was negligible when $`\mathrm{\Omega }`$ and $`\mathrm{\Gamma }`$ are much larger than the dephasing rate. We expect these statements to hold true in this system as well, since the underlying physical processes are the same.
The issue of interferometric instability is relevant to any scheme in which terms in the matter qubit superposition become coupled to the presence/absence of a photon in a given channel. Any phase noise suffered by a photon in transit through the apparatus ultimately can be mapped onto the matter qubits. Fortunately, there has been enormous progress recently in the development of experimental techniques for phase locking, which should prove to be beneficial for a scheme of the proposed type Gilchrist . Imperfect optical excitation can be dealt with by noting that this simply reduces the initial state fidelity, which thus reduces the protocol fidelity by an equal amount.
## VII Summary
We have described a scheme that unifies some of the desirable features of previous work on matter qubits and graph states. It is able to achieve deterministic growth while using simple static linear optics and a โone shotโ excitation. Moreover, the presented scheme obviates the need for continual local operations on qubits during graph growth, which implies a dramatic speedup in many systems. The scheme proves to have properties that make it ideal for creating the most resource efficient form of graph state, the minimal graph state. These minimal graph states, which form the essential resource for a given quantum computation, without its classically efficiently tracktable Clifford-part, typically correspond to graphs with a high maximal vertex degree. For the preparation of such graph states this scheme is particularly suitable. We observe that the use of minimal graph states is completely compatible with the idea of โjust in timeโ entanglement generation. Our protocol is relatively robust to mismatch in the subsystems, and an accuracy of greater than $`1\%`$ in the parameters will provide a fidelity of around $`0.9999`$ in the final state. We hope that the scheme presented in this work can contribute to bringing full-scale graph state quantum computation closer to practical realization.
## VIII Acknowledgements
We would like to thank H.J. Briegel, D.E. Browne, W. Munro, and E. Solano for fruitful discussions, and S. Barrett, P. Kok, and M.B. Plenio for helpful comments on the manuscript. This work has been supported by the EPSRC (QIP-IRC), the EU (IST-2002-38877), the DFG (Schwerpunktprogramm QIV), the European Research Councils (EURYI) and the Royal Society.
|
warning/0506/quant-ph0506203.html
|
ar5iv
|
text
|
# Low dimensional bound entanglement with one-way distillable cryptographic key
## Abstract
We provide a class of bound entangled states that have positive distillable secure key rate. The smallest state of this kind is $`44`$, which shows that peculiar security contained in bound entangled states does not need high dimensional systems. We show, that for these states a positive key rate can be obtained by one-way Devetak-Winter protocol. Subsequently the volume of bound entangled key-distillable states in arbitrary dimension is shown to be nonzero. We provide a scheme of verification of cryptographic quality of experimentally prepared state in terms of local observables. Proposed set of $`7`$ collective settings is proven to be optimal in number of settings.
Quantum cryptography is one of the very interesting practical phenomena within quantum information theory Bouwmeester et al. (2000); Nielsen and Chuang (2000); Alber et al. (2001); Gruska (1999). There were in general two ideas to produce cryptographic key. The first was based on sending nonorthogonal states Bennett and Brassard (1984), the second - on specially chosen measurements of maximally entangled pairs Ekert (1991). They have been shown to be equivalent in general Shor and Preskill (2000) including most general eavesdropper attack. An important ingredient of the protocol was so called Quantum Privacy Amplification Deutsch et al. (1996) based on distillation of EPR paris Bennett et al. (1997a). Despite of natural expectations, that distillability of EPR pairs is a precondition of secure key it has been recently shown Horodecki et al. (2005a); Horodecki et al. that the class of states which contain ideal key (private states) is much wider than class of maximally entangled states. It has been shown that certain bound entangled (BE) states Horodecki et al. (1998) (from which no EPR pair can be distilled) can be distilled to (approximate) private states. In other words - surprisingly - there are bound entangled states with $`K_D>0`$.
However the BE states with nonzero distillable key $`K_D>0`$ of Ref. Horodecki et al. (2005a); Horodecki et al. require Hilbert space of quite large dimension, which makes impression that BE states useful for cryptography are rather exceptional, and far from experimental regime. In this paper we provide a general construction of class of BE states with $`K_D>0`$ and give examples of $`44`$ states having this property. To this end we consider binary mixture $`\{p_1,p_2\}`$ of two orthogonal private bits (private states with at least one bit of ideal key). We first show that any such non equal mixture has $`K_D>0`$, and can be distilled by - quite remarkably - one-way Devetak-Winter protocol Devetak and Winter (2003). Next we construct special pairs of private bits, and show that for certain probability their mixture is key distillable and bound entangled. To obtain this we assure that the state remains positive (and even invariant) under partial transposition (PPT) Peres (1996) which is sufficient condition for a state to be non distillable Horodecki et al. (1998). We then provide an example of the smallest state of our construction which resides on 4 qubits. Basing on this family of BE key distillable states we argue, that the volume of BE states with $`K_D>0`$ is nonzero in arbitrary dimension. We exploit their properties, and consider their experimental preparation. We then show how to verify that experimentally prepared state has nonzero distillable key. Finally the optimal decomposition of observables needed for veryfication if certain $`44`$ state has $`K_D>0`$ is given.
Let us first recall that private bit Horodecki et al. is a state $`\gamma _{AA^{}BB^{}}`$, where $`AA^{}`$ is hold by Alice, $`BB^{}`$ by Bob, and measurement of $`AB`$ in standard basis provides one bit of perfect key (ie. possible outcomes are either $`\{00\}`$ or $`\{11\}`$ with equal probability $`1/2`$ and they are completely independent on measurements on physical system different than $`AA^{}BB^{}`$). Following Horodecki et al. we will write $`\gamma `$ in matrix representation in $`ABA^{}B^{}`$ order of subsystems, so that its $`4\times 4`$ structure corresponds to two qubit subsystem $`AB`$.
We shall start the construction of key distillable BE states by providing a class of states with $`K_D>0`$:
###### Proposition 1
.- Consider two private bits $`\gamma _1`$, $`\gamma _2`$ and take any biased mixture of the form:
$$\rho =p_1\gamma _1+p_2\sigma _x^A\gamma _2\sigma _x^A$$
(1)
with, say, $`p_1>p_2`$ and $`\sigma _x^A=[\sigma _x]_AI_{A^{}BB^{}}`$. The distillable key $`K_D(\rho )`$ fulfills $`K_D(\rho )1h(p_1)`$ where $`h(p_1)`$ is the binary entropy of distribution $`\{p_1,p_2\}`$.
Before we prove the proposition, we have to recall a technique called โprivacy squeezingโ Horodecki et al. that allows to investigate privacy of states of type $`\rho _{AA^{}BB^{}}`$. The state is purified to $`\psi _{ABA^{}B^{}E}`$ so that Eve holds the subsystem $`E`$ of $`\psi `$. To draw key, Alice and Bob will measure systems $`AB`$ in standard basis, and will process the outcomes by public discussion. The systems $`A^{}B^{}`$ will not be actively used, so the relevant state is
$$\rho _{ABE}^{(ccq)}=\underset{ij}{}p_{ij}|ijij|_{AB}\rho _E^{(ij)}.$$
(2)
where $`\rho _E^{ij}`$ are Eveโs states given the outcome was $`ij`$. The state is called $`ccq`$ (two registers are classical). From such state by Devetak-Winter Devetak and Winter (2003) protocol one can get
$$K^{DW}=I(A:B)I(A:E)$$
(3)
bits of key where $`I(A:B)_\rho =S(A)+S(B)S(AB)`$, $`S(X)`$ being von Neumann entropy of $`X`$ subsystem of state $`\rho `$. We will now provide a different ccq state $`\sigma _{ABE}^{(ccq)}`$, which is no better, in the sense that Eve can obtain it from $`\rho _{ABE}^{(ccq)}`$ by some operation on her system. Clearly, such $`\sigma ^{(ccq)}`$ can give no more key than $`\rho ^{(ccq)}`$. The state $`\sigma ^{(ccq)}`$ we produce as follows. First, from $`\rho _{AA^{}BB^{}}`$ we obtain its privacy squeezed (p-squeezed) version $`\sigma _{AB}`$. To this end $`(i)`$ we apply so-called twisting Horodecki et al. (2005a); Horodecki et al. i.e. a unitary transformation controlled by standard basis on $`AB`$ $`U_\tau =_{ij}|ijij|_{AB}U_{A^{}B^{}}^{ij}`$ and subsequently $`(ii)`$ we trace over systems $`A^{}B^{}`$. One finds Horodecki et al. (2005a); Horodecki et al. that p-squeezed state $`\sigma _{AB}`$ has a property that the ccq state $`\sigma _{ABE}^{(ccq)}`$ emerging after measuring it in standard basis $`|ij`$ is no better than $`\rho _{ABE}^{(ccq)}`$. However, if twisting is properly chosen, then it may still produce much key. The privacy is now squeezed to solely two systems $`AB`$.
Proof of Proposition. Using the above method, we will apply p-squeezing to the state $`\rho `$, and show that from the $`\sigma ^{(ccq)}`$ the DW protocol gives $`1h(p_1)`$ rate of key. A basic fact about private bits Horodecki et al. (2005a) is that there exists twisting which brings them to the form $`\psi _0^{AB}\stackrel{~}{\rho }_{A^{}B^{}}`$ where $`\stackrel{~}{\rho }_{A^{}B^{}}`$ is some state, $`\psi _0`$ is one of four Bell states
$`|\psi _{0,1}={\displaystyle \frac{1}{\sqrt{2}}}(|00\pm |11)`$ (4)
$`|\psi _{2,3}={\displaystyle \frac{1}{\sqrt{2}}}(|01\pm |01)`$ (5)
and in the twisting $`U_{01}=U_{10}=I`$. Using this, we immediately obtain a twisting which, followed by partial trace over $`A^{}B^{}`$, turns the state $`\rho `$ of eq. (1) into a mixture of $`\psi _0`$ and $`\psi _2`$. I.e. the p-squeezed state of (1) is $`\sigma _{AB}=p_1|\psi _0\psi _0|+p_2|\psi _2\psi _2|`$. Its purification is of the form
$$\psi _{ABE}^{}=\sqrt{p_1}|\psi _0_{AB}|e_1_E+\sqrt{p_2}|\psi _2_{AB}|e_2_E$$
(6)
so that by measuring it in standard basis Alice and Bob obtain the ccq state
$$\sigma _{ABE}^{(ccq)}=\frac{p_1}{2}[P_{00}+P_{11}]P_{e_1}+\frac{p_2}{2}[P_{01}+P_{10}]P_{e_2}$$
(7)
with $`P_{ij}=|ijij|`$ and $`P_{e_i}=|e_ie_i|`$. For this state, we have $`I(A:B)I(A:E)=1h(p_1)`$ Thus by DW protocol Devetak and Winter (2003) we get this amount of key. Because this state is no better than our state of interest $`\rho _{ABE}^{(ccq)}`$ (the ccq state obtained by measuring $`AB`$ systems of initial state $`\rho `$) we obtain that the distillable key of $`\rho `$ satisfies $`K_D(\rho )1h(p_1)`$, which ends the proof of proposition.
Let us note that the key is here drawn by one-way protocol. We have found that applying two-way recurrence protocol will not increase the key rate.
Construction of small dimensional bound entangled states with $`K_D>0`$ First let us recall, that any private bit can be represented in its $`X`$-form Horodecki et al. :
$`\gamma _{ABA^{}B^{}}={\displaystyle \frac{1}{2}}\left[\begin{array}{cccc}\sqrt{XX^{}}& 0& 0& X\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ X^{}& 0& 0& \sqrt{X^{}X}\end{array}\right]`$ (12)
where $`X`$ is arbitrary operator with trace norm one, and which completely represents the pbit. Then, we obtain that state $`\rho `$ is of the form
$`\rho ={\displaystyle \frac{1}{2}}\left[\begin{array}{cccc}p_1\sqrt{X_1X_1^{}}& 0& 0& p_1X_1\\ 0& p_2\sqrt{X_2X_2^{}}& p_2X_2& 0\\ 0& p_2X_2^{}& p_2\sqrt{X_2^{}X_2}& 0\\ p_1X_1^{}& 0& 0& p_1\sqrt{X_1^{}X_1}\end{array}\right].`$ (17)
We assume now that $`A^{}`$ and $`B^{}`$ are systems both described by $`๐^d`$. Now the essential part of the construction is the following substitution: $`X_1=\frac{1}{W_U}W_U`$ where
$$W_U=\underset{ij}{}u_{ij}|ijji|$$
(18)
and $`u_{ij}`$ are unitary matrix elements of some matrix $`U`$ on $`๐^d`$. Note that $`W_U=_{ij}|u_{ij}|`$ (here we use the trace norm of the matrix). The second operator we choose $`X_2=\frac{W_U^\mathrm{\Gamma }}{W_U^\mathrm{\Gamma }}`$ with $`\mathrm{\Gamma }`$ being partial transposition on subsystem $`B^{}`$. In this case one has just $`W_U^\mathrm{\Gamma }=d`$
The corresponding mixing probabilities are
$`p_1={\displaystyle \frac{W_U}{W_U+W_U^\mathrm{\Gamma }}}p_2={\displaystyle \frac{W_U^\mathrm{\Gamma }}{W_U+W_U^\mathrm{\Gamma }}}`$ (19)
respectively.
Collecting two simple observations, namely that: (i) $`[W_UW_U^{}]^{\frac{1}{2}}=[W_U^{}W_U]^{\frac{1}{2}}=_{ij}|u_{ij}||ijij|`$ (which after normalisation by factor $`W_U`$ gives separable, PPT-invariant state),(ii) $`[W_UW_U^{}]^{\frac{1}{2}}=[W_U^{}W_U]^{\frac{1}{2}}=_i|iiii|`$, (again after normalisation giving PPT-invariant separable state) We get immediately that $`\rho `$ with parameters defined like above is PPT invariant. At the same time we have desired security condition $`p_1>p_2`$ if only
$$\frac{p_1}{p_2}=\frac{W_U}{W_U^\mathrm{\Gamma }}\frac{_{ij}|u_{ij}|}{d}>1.$$
(20)
The latter is satisfied for any unitary U which written in $`\{|ij\}`$ basis has more than $`d`$ nonzero entries.
Thus we have a large class of states that contain secure key and are at the some time PPT invariant states. Of course they are entangled since entanglement is a precondition of secure key distillation Curty et al. (2004).
###### Observation 1
The ratio of $`p_1`$ and $`p_2`$ in (20) which is related to $`K^{DW}`$ key rate achieves the highest value for unimodular unitaries $`U`$ (ie such that $`|u_{ij}|=\frac{1}{\sqrt{d}}`$ irrespectively of indices $`\{i,j\}`$). Then it amounts to $`[\frac{p_1}{p_2}]_{optimal}=\sqrt{d}`$.
Indeed, by use of Lagrange multipliers with slightly more general constraints $`_{ij}|u_{ij}|^2=d`$ we get that optimal $`U`$ is unimodular.
Example of small bound entangled states with $`K_D>0`$ on $`44`$ system. Putting $`d=2`$ we get the smallest secure BE states in our construction. An easy example is a state with $`U`$ equal to 1-qubit Hadamard gate (H). Note, that in this case, the state $`\gamma _2`$ coincides with the so called โflower stateโ with $`U=H`$, which exhibited locking of entanglement cost Horodecki et al. (2005b). The total state can be written as a mixture of Bell states on $`AB`$ subsystem of state, that are classically correlated with some other states on $`A^{}B^{}`$. Namely we have
$$\rho _H=\underset{i}{}q_i|\psi _i\psi _i|_{AB}\rho _{A^{}B^{}}^{(i)}$$
(21)
where the correlated states are the following:
$`\rho ^{(0)}={\displaystyle \frac{1}{2}}[P_{00}+P_{\psi _2}]`$
$`\rho ^{(1)}={\displaystyle \frac{1}{2}}[P_{11}+P_{\psi _3}]`$
$`\rho ^{(2,3)}=P_{\chi _\pm }`$ (22)
with $`P_{\psi _i}`$ being projectors onto corresponding Bell states and $`P_{\chi _\pm }`$ projectors onto pure states
$$\chi _\pm =\frac{1}{2}(\sqrt{2\pm \sqrt{2}}|00\pm \sqrt{2\sqrt{2}}|11)$$
(23)
respectively. The mixing distribution $`\{q_i\}_{i=0}^3`$ is $`\{\frac{p_1}{2},\frac{p_1}{2},\frac{p_2}{2},\frac{p_2}{2}\}`$. Since $`d=2`$, one has $`p_1=\frac{\sqrt{2}}{1+\sqrt{2}}`$, so by proposition (1) a positive key rate can be gained from this $`4`$-qubit PPT state. It reads:
$$K_D^{DW}(\rho _H)=1h(p_1)=0.0213399$$
(24)
per copy of $`\rho _H`$. Note again, that it automatically means, that $`\rho _H`$ is bound entangled. Indeed we have shown, that state of that construction is PPT invariant, and it can not be separable, as it has non-zero distillable key. We show now the next property of this state.
###### Observation 2
The state $`\rho _H`$ is extremal in the set of PPT states.
$`\rho _H`$ is a mixture of form (17) with $`X_1=W_H`$ and $`X_2=X_1^\mathrm{\Gamma }`$. It is straightforward to check, that any mixture with the weight of $`\gamma _1`$ different then value $`p_1=\frac{\sqrt{2}}{1+\sqrt{2}}`$ leads to NPT state. In fact the same argument proves extremity of the state from our construction with $`X=W_U`$, if only $`X`$ is hermitian operator, and either $`X`$ or $`X^\mathrm{\Gamma }`$ has some positive eigenvalue.
Although the states $`\rho _U`$ lay on the edge of the set of PPT states, basing on this family of states we are able to show, that the set of PPT key distillable states has non-zero volume in the set of all PPT $`44`$ states .
###### Observation 3
The set of PPT distillable key of the form $`\rho _U`$ has nonzero volume in $`44`$.
The proof of this observation bases on the fact, that Devetak-Winter lower bound (3) is continuous in $`\rho `$, as mutual information is continuous function of $`\rho `$. If we consider now $`p_{mix}\rho +(1p_{mix})\frac{I}{16}`$, for suitably small $`p_{mix}`$ there exists suitably small $`ฯต>0`$ for which the ball of all states with the center at $`\rho _U`$ and radius $`ฯต`$ lays within the set of PPT states. By continuity argument, one gets the thesis. In fact a similar argument gives this result in any dimension higher than $`44`$.
Upper bound for $`K_D`$.- Similarly as private bit is analogue of singlet from entanglement distillation theory, the state $`\rho `$ of (1) is analogue of mixture of two singlets (which is actually just the p-squeezed state $`\sigma `$). The distillable entanglement of the latter state is just $`1h(p_1)`$, Bennett et al. (1997b) and moreover it is achieved, by coherent application of the DW protocol. Therefore there is a question, if $`K_D(\rho )`$ is just equal to $`1h(p_1)`$. In Horodecki et al. (2005a); Horodecki et al. we show that relative entropy of entanglement $`E_r`$ is an upper bound for $`K_D`$. We have analysed the $`44`$ PPT state $`\rho _H`$, and have found a separable state which gives $`E_r(\rho _H)0.116`$.
Preparation of $`\rho _U`$ $`44`$ states.- As it is shown in example (21) $`\rho _U`$ $`44`$ states are only classically correlated along $`AB`$ versus $`A^{}B^{}`$ cut. Thus they can be created by preparing randomly according to $`\{q_i\}`$ distribution, separately two states: a Bell state $`\psi _i`$ (for $`AB`$ subsystem) and $`\rho ^{(i)}`$ (for $`A^{}B^{}`$ subsystem). According to observation(2) $`\rho _U`$ lays on the boundary of PPT set, any small perturbation can destroy this property. However in spirit of observation (3), one can construct PPT states that are to some extent robust against perturbations.
Key distillability verification for experimentally prepared state.- We now address the question of verification whether a state prepared experimentally in many copies has nonzero distillable key. In the spirit of proof of the proposition (1) instead of estimating whole $`\rho _{ABA^{}B^{}}`$ we suggest to estimate only few parameters of its privacy squeezed state $`\sigma _{AB}`$ and subsequently compute some lower bound on the value of DW rate (3) for the ccq state $`\sigma _{ABE}^{(ccq)}`$ of the latter.
Once we do not estimate whole state $`\rho _{ABA^{}B^{}}`$, the formula (3) can not be used directly to decide its quality. Instead we first consider a lower bound for distillable key from $`ccq`$ state. Namely
$$K_D(\rho ^{(ccq)})I(A:B)_{\rho ^{(ccq)}}S(E)_{\rho ^{(ccq)}},$$
(25)
which is a consequence of formula (3). Indeed, $`I(A:E)`$ for ccq state is equal to $`S(\rho _E)_ip_iS(\rho _E^i)`$, which is Holevo function of ansamble $`\{p_i,\rho _E^i\}`$. This however can not be greater then just entropy of Eveโs subsystem $`S(\rho _E)`$ and the assertion follows.
Using this lower bound we provide now the one which is a function of only diagonal and antidiagonal matrix elements of (2-qubit) privacy squeezed state $`\sigma _{AB}`$. Note, that $`S(\sigma _E)=S(\sigma _{AB})`$, as the total $`ABE`$ state is pure. To estimate $`S(\sigma _E)`$ we will consider state $`\sigma _{AB}`$ subjected to twirling Bennett et al. (1997a) which projects onto Bell basis. Twirling can not decrease the entropy, and commutes with measurement in computational basis, so one has that
$$K_D(\sigma _{ABE}^{(ccq)})I(A:B)S(\sigma _{AB}^{twirl})$$
(26)
This is desired lower bound as it is a function of only those parameters, which we suggest to estimate experimentally.
Although the formula (26) is useful for one-way key distillable states, knowing the diagonal and antidiagonal elements of $`\sigma _{AB}`$ is enough to decide if a two-way recurrence protocol can make key rate non-zero.
We give now the observables, which measured on $`\rho _{ABA^{}B^{}}`$ reveals desired elements of its p-squeezed state. These are:
$`O_1=U_\tau ^{}[\sigma _z\sigma _z]_{AB}I_{A^{}B^{}}U_\tau `$
$`R_{1,2}=U_\tau ^{}[P_{\psi _{0,2}}P_{\psi _{1,3}}]_{AB}I_{A^{}B^{}}U_\tau `$
$`I_{1,2}=U_\tau ^{}[\stackrel{~}{P}_{\psi _{0,2}}\stackrel{~}{P}_{\psi _{1,3}}]_{AB}I_{A^{}B^{}}U_\tau `$
where $`\stackrel{~}{P}_{\psi _k}`$ are Bell states with relative phase $`\pm i`$. The observable $`O_1`$ reveals the diagonal elements of state $`\sigma _{AB}`$. In fact, it is just equal to $`[\sigma _z\sigma _z]_{AB}I_{A^{}B^{}}`$ as any twisting (here $`U_\tau ^{}`$) commutes with the measurement in basis which it controls Horodecki et al. . $`O_1`$ needs therefore just one setting. The $`R_k`$ ($`I_k`$) observables reveals real (imaginary) parts of (possibly complex) coherences on antidiagonal of $`\sigma _{AB}`$. Indeed, one has for example
$`TrR_1\rho _{ABA^{}B^{}}=`$
$`TrU_\tau ^{}[P_{\psi _0}P_{\psi _1}]_{AB}I_{A^{}B^{}}U_\tau \rho _{ABA^{}B^{}}=`$
$`Tr[P_{\psi _0}P_{\psi _1}]_{AB}I_{A^{}B^{}}U_\tau \rho _{ABA^{}B^{}}U_\tau ^{}=`$
$`Tr[P_{\psi _0}P_{\psi _1}]Tr_{A^{}B^{}}[U_\tau \rho _{ABA^{}B^{}}U_\tau ^{}]=`$
$`Tr[P_{\psi _0}P_{\psi _1}]\sigma _{AB}.`$ (28)
where second equality is by property of trace, and third by definition of subsystem of a quantum state. Last equality uses definition of privacy squeezed state ($`\sigma _{AB}`$) which is obtained by acting on $`\rho _{ABA^{}B^{}}`$ with some twisting $`U_\tau `$ and tracing out $`A^{}B^{}`$ subsystem.
Local decomposition of verification observables for $`\rho _H`$. In case of $`\rho _H`$ twisting which realises privacy squeezing is equal to:
$$U_\tau =|0000|W_H+|0101|P_H+(|1010|+|1111|)I$$
(29)
where by $`P_H=_{ij=0}^1h_{ij}|iijj|+_{ij}|ijij|`$ for $`h_{ij}`$ elements of Hadamard and $`W_H=P_H^\mathrm{\Gamma }`$. For this $`U_\tau `$, the observables $`R_1`$ and $`I_1`$ can be decomposed into Pauli operators $`\{I,\sigma _x,\sigma _y,\sigma _z\}`$ in the following way:
$`R_1`$ $`={\displaystyle \frac{1}{4}}[\sigma _x\sigma _x\sigma _y\sigma _y]_{AB}`$
$`[(I\sigma _z+\sigma _zI)+(\sigma _x\sigma _x+\sigma _y\sigma _y)]_{A^{}B^{}}`$
$`I_1`$ $`={\displaystyle \frac{1}{4}}[\sigma _x\sigma _y+\sigma _y\sigma _x]_{AB}`$
$`[(I\sigma _z+\sigma _zI)+(\sigma _x\sigma _x+\sigma _y\sigma _y)]_{A^{}B^{}}`$
and the same for $`R_2`$ and $`I_2`$, which read
$`R_2={\displaystyle \frac{1}{4}}[\sigma _x\sigma _x+\sigma _y\sigma _y]_{AB}[(II\sigma _z\sigma _z)+`$
$`{\displaystyle \frac{1}{\sqrt{2}}}((I\sigma _z+\sigma _zI)+(\sigma _x\sigma _x+\sigma _y\sigma _y))]_{A^{}B^{}}`$
$`I_2={\displaystyle \frac{1}{4}}[\sigma _y\sigma _x\sigma _x\sigma _y]_{AB}[(II\sigma _z\sigma _z)+`$
$`{\displaystyle \frac{1}{\sqrt{2}}}((I\sigma _z+\sigma _zI)+(\sigma _x\sigma _x+\sigma _y\sigma _y))]_{A^{}B^{}}`$
Generalizing approach for two-qubit case of Guehne et al. (2003); Guehne and Hyllus (2003) to four qubit case, one can easily show, that decomposition for $`R_2`$ and $`I_2`$ is optimal in the sense, that it needs $`6`$ settings. The set of these 6 collective settings is enough for both $`R_2,I_2`$ and $`R_1,I_1`$ (for the latter it needs only different classical post-processing). Since this set of settings is optimal for determining $`R_2,I_2`$ and suffices for determining all $`R_i,I_i`$, we conclude, that it is optimal for our task. Together with one setting for $`O_1`$ one needs 7 different collective settings to verify via lower bound (26) if the state $`\rho _H`$ has non zero distillable key.
We thank Ryszard Horodecki for helpful comments and remarks. KH, MH and PH acknowledge the hospitality of Isaac Newton Institute for Mathematical Sciences during the QIS programme. The work is supported by Polish Ministry of Scientific Research and Information Technology under the (solicited) grant no. PBZ-MIN-008/P03/2003 and by EC grants RESQ, contract no. IST-2001-37559 and QUPRODIS, contract no. IST-2001-38877.
|
warning/0506/cs0506027.html
|
ar5iv
|
text
|
# Sorting a Low-Entropy Sequence
## 1 Introduction
Sorting in the comparison model is one of oldest problems in computer science, but it remains an important and active area. Previous research has shown how we can take advantage of various kinds of pre-sortedness, such as long runs, few inversions, or only a small number of elements out of place (see ); in this paper, we show how we can take advantage of low entropy to reduce comparisons.
Consider a fixed sequence $`S=s_1,\mathrm{},s_m`$ containing $`n`$ distinct elements drawn from a total order. For any non-negative integer $`\mathrm{}`$, the *$`\mathrm{}`$th-order empirical entropy* of $`S`$, denoted $`H_{\mathrm{}}(S)`$, is our expected uncertainty about $`s_i`$ (measured in bits) given a context of length $`\mathrm{}`$, as in the following experiment: we are given $`S`$; $`i`$ is chosen uniformly at random from $`\{1,\mathrm{},m\}`$; if $`i\mathrm{}`$, we are told $`s_i`$; if $`i>\mathrm{}`$, we are told $`s_i\mathrm{},\mathrm{},s_{i1}`$. Specifically,
$$H_{\mathrm{}}(S)=\{\begin{array}{cc}\underset{aS}{}\frac{\mathrm{\#}_a(S)}{m}\mathrm{log}\frac{m}{\mathrm{\#}_a(S)}\hfill & \text{if }\mathrm{}=0\text{;}\hfill \\ & \\ \frac{1}{m}\underset{\alpha A_{\mathrm{}}}{}|S_\alpha |H_0(S_\alpha )\hfill & \text{if }\mathrm{}>0\text{.}\hfill \end{array}$$
Here, $`aS`$ means $`a`$ occurs in $`S`$; $`\mathrm{\#}_a(S)`$ is the number of occurrences of $`a`$ in $`S`$; $`\mathrm{log}`$ means $`\mathrm{log}_2`$; $`A_{\mathrm{}}`$ is the set of $`\mathrm{}`$-tuples in $`S`$; and $`S_\alpha `$ is the sequence whose $`i`$th element is the one immediately following the $`i`$th occurrence of $`\alpha `$ in $`S`$. The length of $`S_\alpha `$ is the number of occurrences of $`\alpha `$ in $`S`$ unless $`\alpha `$ is a suffix of $`S`$, in which case it is 1 less.
Notice $`\mathrm{log}nH_0(S)\mathrm{}H_{m1}(S)=H_m(S)=\mathrm{}=0`$. For example, if $`S`$ is the string TORONTO, then $`\mathrm{log}n=2`$,
$`H_0(S)`$ $`=`$ $`{\displaystyle \frac{1}{7}}\mathrm{log}7+{\displaystyle \frac{3}{7}}\mathrm{log}{\displaystyle \frac{7}{3}}+{\displaystyle \frac{1}{7}}\mathrm{log}7+{\displaystyle \frac{2}{7}}\mathrm{log}{\displaystyle \frac{7}{2}}1.84,`$
$`H_1(S)`$ $`=`$ $`{\displaystyle \frac{1}{7}}\left(\text{}H_0(S_\mathrm{N})+2H_0(S_\mathrm{O})+H_0(S_\mathrm{R})+2H_0(S_\mathrm{T})\right)`$
$`=`$ $`{\displaystyle \frac{1}{7}}\left(\text{}H_0(\mathrm{T})+2H_0(\mathrm{RN})+H_0(\mathrm{O})+2H_0(\mathrm{OO})\right)`$
$`=`$ $`2/70.29`$
and all higher-order empirical entropies of $`S`$ are 0. This means, if someone chooses a character uniformly at random from TORONTO and asks us to guess it, then our uncertainty is about $`1.84`$ bits. If they tell us the preceding character before we guess, then on average our uncertainty is about $`0.29`$ bits; if they tell us the preceding two or more characters, then we are certain of the answer. The difference between 0th-order and higher-order empirical entropies can be of practical importance: the encodings produced by most older compression algorithms are only bounded in terms of the 0th-order empirical entropy of the input, whereas those produced by most modern compression algorithms are bounded in terms of higher-order empirical entropies. For example, Manzini proved Burrows and Wheelerโs algorithm encodes $`S`$ using at most
$$(8H_{\mathrm{}}+O(1))m+N^{\mathrm{}}(2N\mathrm{log}N+9)$$
bits, where $`\mathrm{}`$ is any non-negative integer, $`N`$ is the size of the alphabet and, depending on the implementation, the hidden constant is about $`2/25`$.
Suppose we want to sort $`S`$, that is, to put the elements of $`S`$ in non-decreasing order. Many familiar sorting algorithms already take advantage of low 0th-order empirical entropy: Munro and Spira proved MergeSort, TreeSort and HeapSort use $`(H_0(S)+O(1))m`$ ternary comparisons<sup>1</sup><sup>1</sup>1A ternary comparison of $`x`$ and $`y`$ tells us whether $`x<y`$, $`x=y`$ or $`x>y`$; a binary comparison only tells us whether $`xy`$ or $`x>y`$. Our algorithm uses binary comparisons, which is a slight advantage: while most instruction sets support ternary comparisons, most high-level languages do not; a ternary comparison is usually implemented as two binary comparisons .; by the Static Optimality Theorem , SplaySort uses $`O((H_0(S)+1)m)`$ comparisons; Sedgewick and Bentley recently proved QuickSort uses $`O((H_0(S)+1)m)`$ comparisons in the expected case.
In Section 2 we give a new algorithm that sorts $`S`$ using $`(H_0(S)+O(1))m`$ comparisons. In Section 3 we generalize it so that, given a non-negative integer $`\mathrm{}`$ with $`n^{\mathrm{}+1}\mathrm{log}nO(m)`$, it uses $`(H_{\mathrm{}}(S)+O(1))m`$ comparisons. Our algorithmโs main disadvantage is its slowness: it takes $`O((H_{\mathrm{}}(S)+1)m\mathrm{log}n+\mathrm{}m)`$ time, whereas the algorithms mentioned above take $`O((H_0(S)+1)m)`$ time. It works in models where, for $`tm`$, it takes $`O(\mathrm{log}t)`$ time to perform a standard operation on a balanced binary search tree with $`t`$ keys, each of $`O(\mathrm{log}m)`$ bits ; if such a tree takes $`O(t)`$ space, then our algorithm takes $`O(m)`$ space. We emphasize that we do not make assumptions about the source of $`S`$, nor do we use randomization or pointer arithmetic.
## 2 Sorting $`S`$ using $`(H_0(S)+O(1))m`$ Comparisons
If we are given a list of the distinct elements in $`S`$ and their frequencies, then we can easily sort $`S`$ using fewer than $`(H_0(S)+2)m`$ comparisons: we construct a nearly optimal leaf-oriented binary search tree $`T`$, as described in Subsection 2.1, and perform an insertion sort into $`T`$. A *leaf-oriented* binary search tree (LBST) is one in which the data are stored at the leaves.
Since we are not given that information, we instead start with an LBST $`T_1`$ on $`s_1`$; for $`i`$ from 2 to $`m`$, we search for $`s_i`$ in $`T_{i1}`$ and then โin effectโ construct a new LBST $`T_i`$ which is nearly optimal for $`s_1,\mathrm{},s_i`$. In Subsection 2.2 we prove this uses $`(H_0(S)+O(1))m`$ comparisons. Of course, actually constructing every $`T_i`$ would be very slow; in Subsection 2.3, we show how we can quickly โin effectโ construct them. We used a similar approach in for dynamic alphabetic coding.
### 2.1 Constructing a Nearly Optimal Leaf-Oriented Binary Search Tree
Let $`a_1,\mathrm{},a_n`$ be the distinct elements in $`S`$ in increasing order. By Shannonโs Noiseless Coding Theorem , if we search for $`s_1,\mathrm{},s_m`$ in an LBST on $`a_1,\mathrm{},a_n`$, then we use at least $`H_0(S)m`$ comparisons. Mehlhorn gave an $`O(n)`$-time algorithm that, given $`a_1,\mathrm{},a_n`$ and $`\mathrm{\#}_{a_1}(S),\mathrm{},\mathrm{\#}_{a_n}(S)`$, constructs an LBST with which we use fewer than $`(H_0(S)+2)m`$ comparisons; we follow Knuthโs presentation.
###### Theorem 2.1 (Mehlhorn, 1977)
We can construct a leaf-oriented binary search tree on $`a_1,\mathrm{},a_n`$ whose leaves have depths $`\mathrm{log}\frac{m}{\mathrm{\#}_{a_1}(S)}+1,\mathrm{},\mathrm{log}\frac{m}{\mathrm{\#}_{a_n}(S)}+1`$.
###### Proof
For $`1in`$, let
$$f_i=\underset{j=1}{\overset{i1}{}}\frac{\mathrm{\#}_{a_j}(S)}{m}+\frac{\mathrm{\#}_{a_i}(S)}{2m}.$$
Since $`|f_if_i^{}|>\frac{\mathrm{\#}_{a_i}(S)}{2m}`$ for $`i^{}i`$, the first $`\mathrm{log}\frac{m}{\mathrm{\#}_{a_i}(S)}+1`$ bits of $`f_i`$โs binary representation suffice to distinguish it; let $`\sigma _i`$ be this sequence of bits. Notice $`\sigma _1,\mathrm{},\sigma _n`$ are lexicographically increasing.
We construct a binary tree such that, for $`1in`$ and $`1k|\sigma _i|`$, the $`k`$th edge on the path from the root to the $`i`$th leaf is a left edge if the $`k`$th bit of $`\sigma _i`$ is a 0, and a right edge if it is a 1. We store $`a_1,\mathrm{},a_n`$ at the leaves. At each internal node $`v`$, if $`v`$ has two children, then we store a pointer to the rightmost leaf in $`v`$โs left subtree. โ
Consider the LBST this algorithm produces. When searching for $`s_i`$, we start at the root and descend to the leaf that stores $`s_i`$, as follows: at each internal node $`v`$, if $`v`$ has two children and the rightmost leaf in $`v`$โs left subtree stores element $`a`$, then we compare $`s_i`$ with $`a`$ and proceed to $`v`$โs left child or right child depending on whether $`s_ia`$; if $`v`$ has only one child, we proceed immediately to that child. Searching for $`s_1,\mathrm{},s_m`$, we use a total of at most
$$\underset{i=1}{\overset{n}{}}\mathrm{\#}_{a_i}(S)\left(\mathrm{log}\frac{m}{\mathrm{\#}_{a_i}(S)}+1\right)<(H_0(S)+2)m$$
comparisons.
### 2.2 Using a Sequence of Leaf-Oriented Binary Search Trees
Let $`F`$ be the set of indices $`i`$ such that $`s_i`$ is the first occurrence of that element in $`S`$; that is, $`F=\{i:s_is_1,\mathrm{},s_{i1}\}`$. For $`1im`$, let $`T_i`$ be the nearly optimal LBST Mehlhornโs algorithm constructs for $`s_1,\mathrm{},s_i`$, augmented so that, for $`as_1,\mathrm{},s_i`$, the leaf storing $`a`$ also stores a counter set to $`\mathrm{\#}_a(s_1,\mathrm{},s_i)`$ and a list containing the indices of $`a`$โs occurrences in $`s_1,\mathrm{},s_i`$. Consider the concatenation of the lists in $`T_m`$, as a permutation: its inverse sorts $`S`$. <sup>2</sup><sup>2</sup>2In fact, if each list in $`T_m`$ is in increasing order, then their concatenationโs inverse stably sorts $`S`$; a *stable* sort preserves equal elementsโ relative order. Thus, with regard to comparisons needed, constructing $`T_m`$ is equivalent to sorting $`S`$; the following lemmas show $`(H_0(S)+O(1))m`$ comparisons suffice.
###### Lemma 1
We can construct $`T_m`$ using at most
$$\underset{iF\{1\}}{}(\mathrm{log}(i1)+3)+\underset{iF}{}\left(\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}+3\right)$$
comparisons.
###### Proof
By induction. We can construct $`T_1`$ without using any comparisons. For $`2im`$, suppose we have $`T_{i1}`$ and want to construct $`T_i`$. To do this, we first search for $`s_i`$ in $`T_{i1}`$.
If $`s_is_1,\mathrm{},s_{i1}`$, that is, $`iF`$, then our search uses $`\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}+1`$ comparisons and ends at the leaf storing $`s_i`$. Otherwise, our search uses at most $`\mathrm{log}(i1)+1`$ comparisons and ends at a leaf storing either $`s_i`$โs predecessor or successor in $`T_{i1}`$.
Let $`a`$ be the element stored at the leaf $`v`$ where our search ends. We determine whether $`a`$ is $`s_i`$โs predecessor, $`s_i`$ itself, or $`s_i`$โs successor by checking whether $`as_i`$ and whether $`s_ia`$. If $`a`$ is $`s_i`$โs predecessor, then we insert a new leaf immediately to the right of $`v`$, that stores $`s_i`$, a counter set to 1 and a list containing $`i`$; if $`a=s_i`$, we increment $`v`$โs counter and add $`i`$ to $`v`$โs list; if $`a`$ is $`s_i`$โs successor, then we insert a new leaf immediately to the left of $`v`$, that stores $`s_i`$, a counter set to 1 and a list containing $`i`$.
Notice $`T_{i1}`$โs leaves now contain the same information as $`T_i`$โs, which is enough for us to construct $`T_i`$ without any further comparisons. In total, if $`iF`$, then we use $`\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}+3`$ comparisons to construct $`T_i`$; otherwise, we use at most $`\mathrm{log}(i1)+3`$ comparisons. โ
###### Lemma 2
$$\underset{iF\{1\}}{}\mathrm{log}(i1)+\underset{iF}{}\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}(H_0(S)+O(1))m.$$
###### Proof
Let
$`C`$ $`=`$ $`{\displaystyle \underset{iF\{1\}}{}}\mathrm{log}(i1)+{\displaystyle \underset{iF}{}}\mathrm{log}{\displaystyle \frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}}`$
$`<`$ $`\mathrm{log}(m!){\displaystyle \underset{iF}{}}\mathrm{log}\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1}).`$
For $`iF`$, if $`s_i`$ is the $`j`$th occurrence of $`a`$ in $`S`$, then $`j2`$ and $`\mathrm{log}\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})=\mathrm{log}(j1)`$. Thus,
$`C`$ $`<`$ $`\mathrm{log}(m!){\displaystyle \underset{aS}{}}{\displaystyle \underset{j=2}{\overset{\mathrm{\#}_a(S)}{}}}\mathrm{log}(j1)`$
$`=`$ $`\mathrm{log}(m!){\displaystyle \underset{aS}{}}\mathrm{log}(\mathrm{\#}_a(S)!)+{\displaystyle \underset{aS}{}}\mathrm{log}\mathrm{\#}_a(S)`$
$``$ $`\mathrm{log}(m!){\displaystyle \underset{aS}{}}\mathrm{log}(\mathrm{\#}_a(S)!)+n\mathrm{log}{\displaystyle \frac{m}{n}}`$
$`=`$ $`\mathrm{log}(m!){\displaystyle \underset{aS}{}}\mathrm{log}(\mathrm{\#}_a(S)!)+O(m).`$
By Stirlingโs Formula,
$$x\mathrm{log}xx\mathrm{ln}2<\mathrm{log}(x!)x\mathrm{log}xx\mathrm{ln}2+O(\mathrm{log}x).$$
Thus,
$$Cm\mathrm{log}mm\mathrm{ln}2\underset{aS}{}\left(\text{}\mathrm{\#}_a(S)\mathrm{log}\mathrm{\#}_a(S)\mathrm{\#}_a(S)\mathrm{ln}2\right)+O(m)$$
Since $`_{aS}\mathrm{\#}_a(S)=m`$,
$`C`$ $``$ $`{\displaystyle \underset{aS}{}}\mathrm{\#}_a(S)\mathrm{log}{\displaystyle \frac{m}{\mathrm{\#}_a(S)}}+O(m)`$
$`=`$ $`(H_0(S)+O(1))m.`$
### 2.3 Using a Statistics Data Structure
Let $`T_1,\mathrm{},T_m`$ be as defined in Subsection 2.2. Since Mehlhornโs algorithm takes $`O(n)`$ time time, sorting $`S`$ by constructing $`T_1,\mathrm{},T_m`$ takes $`O(mn)`$ time; this is faster than BubbleSort, for example, but still impractical. To save time, we implement all of the $`T_i`$s as a single dynamic *statistics* data structure: an augmented balanced binary search tree that stores a list of triples $`a_1,w_1,L_1,\mathrm{},a_t,w_t,L_t`$, each of which consists of a key $`a_j`$, a positive integer weight $`w_j`$ and a list $`L_j`$. None of the following operations compares keys and each takes $`O(\mathrm{log}t)`$ time :
return the smallest $`j`$ with $`_{k=1}^jw_kb`$;
return $`_{k=1}^jw_k`$;
return $`a_j,w_j,L_j`$;
increment $`w_j`$;
append $`i`$ to $`L_j`$;
insert $`a,1,i`$ into the $`j`$th position in the list of triples.
As an aside, we note there are faster statistics data structures on a word RAM (see ); we leave as future work investigating whether we can improve our algorithm with one of them.
###### Lemma 3
Suppose we have a statistics data structure whose keys and weights are, respectively, the distinct elements in $`s_1,\mathrm{},s_i`$ and their frequencies. Then given the path from the root to a node $`v`$ in $`T_i`$, we can determine the following in $`O(\mathrm{log}n)`$ time:
* if $`v`$ is a leaf, the element stored at $`v`$;
* whether $`v`$ has a left child;
* whether $`v`$ has a right child;
* if $`v`$ has two children, the element stored at the rightmost leaf in $`v`$โs left subtree.
###### Proof
Let $`a_1,\mathrm{},a_t`$ be the distinct elements in $`s_1,\mathrm{},s_i`$ in increasing order and, for $`1jt`$, let
$$f_j=\underset{k=1}{\overset{j1}{}}\frac{\mathrm{\#}_{a_k}(s_1,\mathrm{},s_i)}{i}+\frac{\mathrm{\#}_{a_j}(s_1,\mathrm{},s_i)}{2i}.$$
Given a binary string $`\rho `$, we can find the smallest $`j`$ such that $`f_j`$โs binary representation begins $`\rho `$, if one exists: let $`j^{}`$ be the value returned by $`\mathrm{๐ฌ๐๐๐ซ๐๐ก}((.\rho )i)`$ with $`.\rho `$ interpreted as a binary fraction; by $`T_i`$โs construction, we know the $`j`$ we seek is either $`j^{}`$ or $`j^{}+1`$; we use $`\mathrm{๐ฌ๐ฎ๐ฆ}(j^{})`$, $`\mathrm{๐ญ๐ซ๐ข๐ฉ๐ฅ๐}(j^{})`$ and $`\mathrm{๐ญ๐ซ๐ข๐ฉ๐ฅ๐}(j^{}+1)`$ to compute $`f_j^{}`$ and $`f_{j^{}+1}`$, if they are defined.
Let $`\sigma `$ be the path from the root to $`v`$ encoded as a binary string, with each 0 indicating a left edge and each 1 indicating a right edge. We can determine each of the following properties of $`v`$ in $`O(\mathrm{log}t)O(\mathrm{log}n)`$ time: if there is only one $`j`$ such that $`f_j`$โs binary representation begins $`\sigma `$, then $`v`$ is a leaf storing $`a_j`$; if $`v`$ is an internal node and there is a $`j`$ such that $`f_j`$โs binary representation begins $`\sigma 0`$, then $`v`$ has a left child; similarly, if $`v`$ is an internal node and there is a $`j`$ such that $`f_j`$โs binary representation begins $`\sigma 1`$, then $`v`$ has a right child; finally, if $`v`$ has two children, then there is a $`j`$ such that $`f_j`$โs binary representation begins $`\sigma 0`$ and $`f_{j+1}`$โs binary representation begins $`\sigma 1`$ โ the rightmost leaf in $`v`$โs left subtree stores $`a_j`$. โ
For $`2im`$, let $`a_1,\mathrm{},a_t`$ be the distinct elements in $`s_1,\mathrm{},s_{i1}`$. Suppose we have a statistics data structure $`D`$ implementing $`T_{i1}`$, that is, storing $`a_1,\mathrm{\#}_{a_1}(s_1,\mathrm{},s_{i1}),L_1,\mathrm{},a_t,\mathrm{\#}_{a_t}(s_1,\mathrm{},s_{i1}),L_t`$ with each $`L_j`$ containing the indices of $`a_j`$โs occurrences in $`s_1,\mathrm{},s_{i1}`$. Using $`D`$ and Lemma 3, if $`s_is_1,\mathrm{},s_{i1}`$, then searching for $`s_i`$ in $`T_{i1}`$ takes $`\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}+1`$ comparisons and $`O\left(\left(\mathrm{log}\frac{i1}{\mathrm{\#}_{s_i}(s_1,\mathrm{},s_{i1})}+1\right)\mathrm{log}n\right)`$ time, and returns $`j`$ such that $`a_j=s_i`$; otherwise, searching for $`s_i`$ takes at most $`\mathrm{log}(i1)+1`$ comparisons and $`O((\mathrm{log}(i1)+1)\mathrm{log}n)`$ time, and returns $`j`$ such that $`a_j`$ is either $`s_i`$โs predecessor or successor in $`T_{i1}`$. Determining whether $`a_j`$ is $`s_i`$โs predecessor, $`s_i`$ itself, or $`s_i`$โs successor takes two more comparisons and $`O(\mathrm{log}n)`$ time. If $`a_j`$ is $`s_i`$โs predecessor, then we use $`O(\mathrm{log}n)`$ time to insert $`s_i,1,i`$ into the $`(j+1)`$st position in the list of triples; if $`a_j=s_i`$, then we use $`O(\mathrm{log}n)`$ time to increment the weight in the triple $`a_j,\mathrm{\#}_{a_j}(s_1,\mathrm{},s_{i1}),L_j`$ and append $`i`$ to $`L_j`$; if $`a_j`$ is $`s_i`$โs successor, then we use $`O(\mathrm{log}n)`$ time to insert $`s_i,1,i`$ into the $`j`$th position in the list of triples. After this, $`D`$ implements $`T_i`$.
We can construct a statistics data structure implementing $`T_1`$ in $`O(1)`$ time without using any comparisons; by Lemmas 1 and 2, we can use $`(H_0(S)+O(1))m`$ comparisons and $`O((H_0(S)+1)m\mathrm{log}n)`$ time to construct a statistics data structure implementing $`T_m`$; from this we can obtain the concatenation of the lists in $`T_m`$, in $`O(m\mathrm{log}n)`$ time. Therefore, we can sort $`S`$ using $`(H_0(S)+O(1))m`$ comparisons and $`O((H_0(S)+1)m\mathrm{log}n)`$ time.
## 3 Sorting $`S`$ using $`(H_{\mathrm{}}(S)+O(1))m`$ Comparisons
To generalize our algorithm, given $`S`$ and $`\mathrm{}`$ with $`n^{\mathrm{}+1}\mathrm{log}nO(m)`$, we work from left to right and maintain a set of statistics data structures, one for each distinct $`\mathrm{}`$-tuple seen so far, and keep track of them using two dictionaries. In effect, we partition $`S`$, use the statistics data structures to sort each of the parts, and then merge them. This uses a total of $`(H_{\mathrm{}}(S)+O(1))m`$ comparisons and $`O((H_{\mathrm{}}(S)+1)m\mathrm{log}n+\mathrm{}m)`$ time.
### 3.1 Using a Set of Statistics Data Structures
As we work, we maintain a statistics data structure $`D_\alpha `$ for each distinct $`\mathrm{}`$-tuple $`\alpha `$ that has occurred so far. Assume we have a black box $`B`$ that works as follows: for $`\mathrm{}+1im`$, suppose we query $`B`$ immediately before we process $`s_i`$; if the $`\mathrm{}`$-tuple $`s_i\mathrm{},\mathrm{},s_{i1}`$ has occurred before, then $`B`$ returns a pointer to $`D_{s_{i\mathrm{}+1},\mathrm{},s_{i1}}`$; otherwise, $`B`$ creates $`D_{s_{i\mathrm{}+1},\mathrm{},s_{i1}}`$ and returns a pointer to it; in both cases, querying $`B`$ costs $`O(\mathrm{log}n)`$ comparisons and $`O(\mathrm{}\mathrm{log}n)`$ time.
We use $`B`$ to keep track of the statistics data structures, but we only query it after seeing a new distinct $`(\mathrm{}+1)`$-tuple; this way, the total cost of querying $`B`$ is $`O(n^{\mathrm{}+1}\mathrm{log}n)O(m)`$ comparisons and $`O(n^{\mathrm{}+1}\mathrm{}\mathrm{log}n)O(\mathrm{}m)`$ time. We augment the statistics data structures so that, instead of storing just triples, they store quadruples: each $`D_{b_1,\mathrm{},b_{\mathrm{}}}`$ stores a list of quadruples $`a_1,w_1,L_1,p_1,\mathrm{},`$ $`a_t,w_t,L_t,p_t`$, where $`p_j`$ is a pointer to $`D_{b_2,\mathrm{},b_{\mathrm{}},a_j}`$; as well, $`D_{b_1,\mathrm{},b_{\mathrm{}}}`$ stores the ranks of $`b_2,\mathrm{},b_{\mathrm{}}`$, which we define and use later.
To process $`S`$, first, we query $`B`$ to obtain a pointer to a new statistics data structure $`D_{s_1,\mathrm{},s_{\mathrm{}}}`$. For $`\mathrm{}+1im`$, we search for $`s_i`$ in the LBST that $`D_{s_i\mathrm{},\mathrm{},s_{i1}}`$ implements, as in Subsection 2.3. If $`s_i\mathrm{},\mathrm{},s_i`$ has occurred before, then this search returns a quadruple $`s_i,w,L,p`$; we increment $`w`$, append $`i`$ to $`L`$, and retrieve $`p`$, which points to $`D_{s_{i\mathrm{}+1},\mathrm{},s_i}`$. If $`s_i\mathrm{},\mathrm{},s_i`$ has not occurred before, then we query $`B`$ to obtain a pointer $`p`$ to $`D_{s_{i\mathrm{}+1},\mathrm{},s_i}`$ and insert $`s_i,1,i,p`$ into $`D_{s_i\mathrm{},\mathrm{},s_{i1}}`$.
As in Section 1, let $`A_{\mathrm{}}`$ be the set of $`\mathrm{}`$-tuples in $`S`$ and, for $`\alpha A_{\mathrm{}}`$, let $`S_\alpha `$ be the sequence whose $`j`$th element is the one immediately following the $`j`$th occurrence of $`\alpha `$ in $`S`$. In total, processing $`S`$ takes
$$O(m)+\underset{\alpha A_{\mathrm{}}}{}(H_0(S_\alpha )+O(1))|S_\alpha |=(H_{\mathrm{}}(S)+O(1))m$$
comparisons and
$$O(\mathrm{}m)+\underset{\alpha A_{\mathrm{}}}{}O\left(\text{}(H_0(S_\alpha )+1)|S_\alpha |\mathrm{log}n\right)=O((H_{\mathrm{}}(S)+1)m\mathrm{log}n+\mathrm{}m)$$
time. When we finish processing $`S`$, for each $`\alpha A_{\mathrm{}}`$ and each $`aS_\alpha `$, $`D_\alpha `$ contains a quadruple $`a,\mathrm{\#}_a(S_\alpha ),L,p`$ with $`L`$ containing the indices of occurrences of $`a`$ that immediately follow occurrences of $`\alpha `$ in $`S`$.
After we process $`S`$, the at most $`n^{\mathrm{}}`$ statistics data structures each contain at most $`n`$ quadruples. Consider all these quadruples, as well as โdummyโ quadruples $`s_1,1,1,\mathrm{null},\mathrm{},s_{\mathrm{}},1,\mathrm{},\mathrm{null}`$; we sort them all by their keys, which takes $`O((n^{\mathrm{}+1}+\mathrm{})\mathrm{log}n)O(m)`$ comparisons and time. Now consider the concatenation of their lists of indices: its inverse sorts $`S`$. To see why, notice the indices are in non-decreasing order by the elements they index. Thus, to prove the following theorem, it only remains for us to implement our black box $`B`$.
###### Theorem 3.1
Given a sequence $`S=s_1,\mathrm{},s_m`$ containing $`n`$ distinct elements and a non-negative integer $`\mathrm{}`$ with $`n^{\mathrm{}+1}\mathrm{log}nO(m)`$, we can sort $`S`$ using $`(H_{\mathrm{}}(S)+O(1))m`$ comparisons and $`O((H_{\mathrm{}}(S)+1)m\mathrm{log}n+\mathrm{}m)`$ time.
### 3.2 Using a Dictionary of Elements and a Dictionary of $`\mathrm{}`$-tuples
For our black box $`B`$, we use two dictionaries, both implemented as balanced binary search trees: $`B_1`$ contains the at most $`n`$ distinct elements seen so far, and $`B_2`$ stores $`O(\mathrm{log}m)`$-bit encodings of the at most $`n^{\mathrm{}}`$ distinct $`\mathrm{}`$-tuples seen so far. We search in each dictionary once per query to $`B`$, that is, once per distinct $`(\mathrm{}+1)`$-tuple in $`S`$; we use a total of $`O(n^{\mathrm{}+1}\mathrm{log}n)O(m)`$ comparisons and time searching in $`B_1`$; we use a total of $`O(n^{\mathrm{}+1}\mathrm{}\mathrm{log}n)O(\mathrm{}m)`$ time searching in $`B_2`$, but no comparisons between elements of $`S`$.
We maintain the invariant that, immediately before we process $`s_i`$, $`B_1`$ stores a set of pairs $`a_1,r_1,\mathrm{},a_t,r_t`$, each of which consists of a distinct element $`a_js_1,\mathrm{},s_{i1}`$ and $`a_j`$โs rank $`r_j`$. We say $`a_j`$ has *rank* $`r_j`$ if it is the $`r_j`$th distinct element to appear in $`S`$; that is, for some $`k`$, the first occurrence of $`a_j`$ is $`s_k`$ and there are $`r_j`$ distinct elements in $`s_1,\mathrm{},s_k`$. Notice operations on $`B_1`$ use $`O(\mathrm{log}n)`$ comparisons and time. To query our black box $`B`$ before processing $`s_i`$, we start by searching for $`s_i`$ in $`B_1`$. If this search succeeds, then we retrieve $`s_i`$โs rank; if it fails, then $`s_i`$ is a new distinct element and we insert $`s_i,r`$, where $`r`$ is the number of distinct elements seen so far, including $`s_i`$. After this we have the ranks $`r_1,\mathrm{},r_\mathrm{}1`$ of $`s_{i\mathrm{}+1},\mathrm{},s_{i1}`$, which $`D_{s_i\mathrm{},\mathrm{},s_{i1}}`$ stores, and the rank $`r_{\mathrm{}}`$ of $`s_i`$.
We also maintain the invariant that, immediately before we process $`s_i`$, $`B_2`$ stores a set of pairs $`g_1,p_1,\mathrm{},g_t^{},p_t^{}`$, each of which consists of an $`O(\mathrm{log}m)`$-bit encoding $`g_j`$ of a distinct $`\mathrm{}`$-tuple $`\alpha `$ in $`s_1,\mathrm{},s_{i1}`$ and a pointer $`p_j`$ to $`D_\alpha `$. We use the *gamma code* to encode each $`a\alpha `$, which encodes any positive integer $`x`$ as a binary string $`\gamma (x)`$ consisting of $`\mathrm{log}x`$ copies of 0 followed by the $`(\mathrm{log}x+1)`$-bit binary representation of $`x`$.
###### Lemma 4
We can encode any sequence $`x_1,\mathrm{},x_{\mathrm{}}`$ of positive $`O(\mathrm{log}n)`$-bit integers as a unique $`O(\mathrm{log}m)`$-bit binary string.
###### Proof
The gamma code is prefix-free and, hence, unambiguous: any binary string is the concatenation of at most one sequence of encoded integers. Thus, the encoding $`\gamma (x_1)\mathrm{}\gamma (x_{\mathrm{}})`$ is unique and has length $`O(\mathrm{}\mathrm{log}n)O(\mathrm{log}m)`$. โ
Notice operations on $`B_2`$ use $`O(\mathrm{}\mathrm{log}n)`$ time and comparisons between encodings of $`\mathrm{}`$-tuples, but no comparisons between elements of $`S`$. To query our black box $`B`$, after searching for $`s_i`$ in $`B_1`$, we search for $`\gamma (r_1)\mathrm{}\gamma (r_{\mathrm{}})`$ in $`B_2`$. If this search succeeds, then we retrieve a pointer to $`D_{s_{i\mathrm{}+1},\mathrm{},s_i}`$; if it fails, then $`s_{i\mathrm{}+1},\mathrm{},s_i`$ is a new distinct $`\mathrm{}`$-tuple, so we create a new statistics data structure $`D_{s_{i\mathrm{}+1},\mathrm{},s_i}`$ and insert $`\gamma (r_1)\mathrm{}\gamma (r_{\mathrm{}}),p`$ into $`B_2`$, where $`p`$ is a pointer to $`D_{s_{i\mathrm{}+1},\mathrm{},s_i}`$.
|
warning/0506/hep-ph0506132.html
|
ar5iv
|
text
|
# Limiting temperature from a parton gas with power-law tailed distribution
(June 21, 2005)
## Abstract
We combine Tsallis distributed massless partons to an effective thermal prehadron spectrum by folding. A limiting temperature and a mass spectrum combined of three exponentials emerge by this procedure. Meson and baryon resonance spectra have different polynomial prefactors.
The idea to treat the bulk of mesonic and baryonic resonances as a statistical system stems from Rolf HagedornHAGEDORN . The mass spectrum is exponential, multiplied with an originally negative power of the mass, which fact gives rise to a maximal, limiting temperature, $`T_H`$, when heating a hadron gas. The exponential factor with the main part of the hadron energy, i.e. the rest mass, is the most essential feature in this assumption, power factors and depending on them the quantitative fit to the Hagedorn temperature, $`T_H`$, vary. A recent compilation of hadron resonances by Broniowski, Florkowski and Glozman shows that this idea works well beyond the data used for original fits, although baryons and mesons seem to follow a slightly different lineRESONANCES . Explanations for the origin of this exponential mass spectrum date back to the MIT bag model, where this behavior was demonstrated by KapustaBAG-LIMIT . Bugaev and others point out that such a system is a perfect thermostat forcing the same temperature to any finite system in thermal contactBUGAEV . The picture of Hagedorn resonances also fit well to lattice QCD results on the equation of stateREDLICH and is under consideration in recent microscopic models of quark matter rehadronization in heavy-ion collisionsGREINER .
In this paper we demonstrate how another approach may support the occurence of an essentially exponential mass spectrum of hadrons. We consider massless thermal partons, but with a generalized equilibrium distribution with power-law tail asymptotics. This distribution will then be folded to mesonic or baryonic one-particle distributions of energy, or - at zero rapidity - transverse mass and transverse momenta, respectively. Such stationary distributions are conform with the basic laws of statistical physics and may be considered as a description of the intermediate ($`p_T14`$ GeV) part of the observed hadron spectraTSB-RHIC-SCHOOL ; T-TEAM-FITS . This approach delivers qualitatively interesting results, in particular a characteristic difference between the mesonic and baryonic mass spectrum.
There are numerous occurrences of power-law tailed statistical distributions in nature. In particular hadron transverse spectra stemming from elementary particle or heavy ion collisions can be well fitted at mid rapidity by a formula reflecting $`m_T`$-scalingPHENIX ; STAR ; ZEUS ; MAREK ; BECK ; TASSO ; MT-SCALING ; T-TEAM-FITS :
$$f(p_T)\left(1+m_T/E_c\right)^v.$$
(1)
Interpreting these spectra in terms of the single particle energy, one considers $`E=m_T=\sqrt{p_T^2+m^2}`$ for a relativistic particle with mass $`m`$. This formula describes a Tsallis distribution, which was conjectured earlier by using axiomatic thermodynamical arguments TSALLIS-ENTROPY . This differs from the traditional interpretation of such spectral tails in particle physics, when these are treated as non-equilibrium phenomena. The very high-$`p_T`$ part, expected to stem from jet fragmentation, may still allow for a statistical interpretation below $`p_T=68`$ GeV.
Distributions with a power-law like tail are encountered in several different statistical modelsMULT-NOISE ; WILK ; BIALAS ; FLORKOW ; KODAMA ; RAF-BOLTZMANN-EQ ; WAL-RAFELSKI ; TSB-RHIC-SCHOOL ; ASTRO . They are investigated as generic distributions in the non-extensive thermodynamicsTSALLIS-RULES ; TSALLIS-WANG ; TSALLIS-FOKKER , based on a generalization of the Boltzmann-entropy, encountered first in informatics problemsTSALLIS-ENTROPY ; OTHER-ENTROPIES . Without being able to exclude the non-equilibrium interpretation of the power-law tail in particle physics, we explore here some consequences of a stationary state with Tsallis distributed extreme relativistic particles (massless partons).
We consider a massless parton gas with binary collisions obeying a general, non-extensive energy composition ruleNEBE , $`E_{12}=h(E_1,E_2)`$. Whenever this rule is associative, the one-particle energy can be mapped onto an additive quasi-energy $`X(E_{12})=X(E_1)+X(E_2)`$, with the help of a strict monotonic function, obeying $`X(0)=0`$. For the Tsallis stationary distribution one uses
$$X(E)=\frac{1}{a}\mathrm{ln}\left(1+aE\right),$$
(2)
with an energy scale $`a=1/E_c`$ related to the microscopical dynamics. This gives rise to the following non-extensive composition rule:
$$h(x,y)=x+y+axy.$$
(3)
The stationary one-particle distribution under these conditions becomes
$$f(E_i)=\frac{1}{Z_1(\beta )}\mathrm{exp}\left(X(E_i)/T\right),$$
(4)
with a temperature $`T`$ determined by the conserved total (quasi) energy $`X(E_{tot})`$ and particle number. The one-particle partition function here is given by the phase space integral $`Z_1(\beta )=๐\mathrm{\Gamma }_jf(E_j)`$.
In this paper we are interested in the energy distribution of large subsystems, i.e. in the microcanonical distribution of the energy of $`N`$ particles each following a Tsallis distribution given by eq.(4). The general formula for the $`N`$-particle energy distribution, assuming a factorization of the one-particle distributions is given by
$$F_N(E)=\mathrm{\Delta }EX^{}(E)\underset{i=1}{\overset{N}{}}d\mathrm{\Gamma }_i\delta (X(E)\underset{i=1}{\overset{N}{}}X(E_i))f(E_i).$$
(5)
Here $`X^{}(E)`$ stands for the derivative of the strict monotonic mapping function $`X(E)`$, $`\mathrm{\Delta }E`$ is the width of the $`N`$-particle energy shell and the $`d\mathrm{\Gamma }_i`$ integration measures refer to the one-particle phase space factors. The factorization is usually a good approximation for values of $`N`$ being still negligible besides the total number of particles. A check of this formula for $`N=1`$ expresses the one-particle energy distribution as being proportional to the one-particle phase space distribution:
$$F_1(E)=\frac{V}{2\pi ^2}E^2\mathrm{\Delta }Ef(E).$$
(6)
Since $`F_N(E)`$ contains an $`X^{}(E)`$ factor, $`F_N(E)dE=g_N(X)dX`$ relates it to the distribution of the $`N`$particle quasi energy, $`X=X(E)`$. As it was shown in Ref.NEBE , in certain kinetic models leading to a stationary state of the non-extensive thermodynamics, $`X(E)=_iX(E_i)`$ is the conserved quantity and the quasi-energy can be regarded as the physical energy of composite $`N`$particle systems.
Using the Fourier-expansion of the constraint on the sum of the quasi-energies, $`X_i=X(E_i)`$, the expression (5) factorizes:
$$g_N(X)=\mathrm{\Delta }E_{\mathrm{}}^+\mathrm{}\frac{ds}{2\pi }e^{isX}\underset{j=1}{\overset{N}{}}\left[๐\mathrm{\Gamma }_jf(E_j)e^{isX(E_j)}\right].$$
(7)
Utilizing now the equilibrium one-particle quasi-energy distribution, (4) we obtain
$$๐\mathrm{\Gamma }_jf(E_j)e^{isX(E_j)}=\frac{Z_1(\beta +is)}{Z_1(\beta )}.$$
(8)
The $`N`$-particle quasi-energy distribution we are seeking for is then given in a form normalized to one in an energy shell of width $`\mathrm{\Delta }E`$ as
$$g_N(X)=\mathrm{\Delta }E_{\mathrm{}}^+\mathrm{}\frac{ds}{2\pi }e^{isX}\left(\frac{Z_1(\beta +is)}{Z_1(\beta )}\right)^N.$$
(9)
Such integrals may be evaluated in a saddle point approximation, which is good for large $`N`$ as long as no singularity has been encountered in the expansion of $`\mathrm{ln}Z_1(\beta +is)`$. The result is a Gaussian distribution in $`X(E)`$. Irrespective to this approximation, as long as the factorization assumption is valid, the exact expectation value is given by
$$X(E)=N\frac{}{\beta }\mathrm{ln}Z_1(\beta ),$$
(10)
and the square width by
$$\delta X(E)^2=N\frac{^2}{\beta ^2}\mathrm{ln}Z_1(\beta ).$$
(11)
In case of the Tsallis distribution using (2) for massless particles one obtains
$$Z_1(\beta )=\frac{V_d}{(2\pi )^d}(d1)!\underset{k=1}{\overset{d}{}}B_k^1$$
(12)
with spatial volume $`V_d`$ in $`d`$ dimensions and with the factors $`B_k=\beta ka`$. The expectation value of the quasi-energy per particle becomes
$$ฯต_1=\frac{X(E)}{N}=\underset{k=1}{\overset{d}{}}B_k^1,$$
(13)
while the unit square width contribution is given by
$$\delta _1=\frac{\delta X^2}{N}=\underset{k=1}{}B_k^2.$$
(14)
All these expressions (12,13,14) loose their conventional interpretability for $`\beta da`$ in $`d`$ dimensions. The value, $`T_H=1/(da)`$ is a limiting temperature for positive values of the parameter $`a`$, at which the physically relevant quasi-energy per particle diverges, so there is no use of further heating at this temperature. More and more energy given to the system would raise the temperature less and less. For $`a<0`$, i.e. for attractive correction, the energy per particle is limited by $`(1+1/2+1/3)E_c`$ from above, but any temperature may occur (cf. Fig.1).
The $`N`$-particle energy distribution can be obtained also exactly in this case. The Fourier-integral (9) has $`N`$-fold poles at the values $`s_k=iB_k`$ for each directional degree of freedom $`k=1,\mathrm{},d`$. The evaluation of such integrals is somewhat involved in a general number of dimensions, so we restrict our further analysis to the cases $`d=1`$ and $`d=3`$. For a one-dimensional Tsallis distribution we obtain
$$g_{N+1}(X)=\mathrm{\Delta }EB_1\frac{(B_1X)^N}{N!}\mathrm{exp}\left(X/E_cX/T\right).$$
(15)
In this case a Hagedorn exponential emerges with the limiting temperature $`T_H=E_c`$. Considering on the other hand $`F_{N+1}(E)=g_{N+1}(X)X^{}(E)`$, i.e. the $`N+1`$-particle distribution of the (not conserved) naive energy expression only, the exponetially growing factor does not occur. It is due to $`X^{}(E)=e^{aX}`$ for the Tsallis distribution:
$$F_{N+1}(E)=\mathrm{\Delta }E\frac{B_1^{N+1}\left(\mathrm{ln}(1+aE)\right)^N}{a^NN!}(1+aE)^{1/aT}.$$
(16)
However, still $`\beta >a`$ or $`T<T_H`$ must be satisfied, and for positive $`a`$ values increasing quasi-energy per particle does not raise the temperature above $`T_H`$.
The corresponding expression in $`d=3`$ dimensions is more involved. There occur three different exponential factors with lowest limiting temperature $`T_H=E_c/3`$. The general dependence on $`\beta `$ can be factorized out by shifting the integration variable $`s`$ to $`s+i\beta `$:
$$g_{N+1}(X)=\frac{V^{N+1}\mathrm{\Delta }E}{(\pi ^2Z_1(\beta ))^{N+1}}e^{\beta X}\mathrm{\Phi }_N(X),$$
(17)
with $`\mathrm{\Phi }_N(X)`$ depending on the dynamical input parameters only, but not on the temperature $`T=1/\beta `$:
$$\mathrm{\Phi }_N(X)=\frac{ds}{2\pi }e^{isX}\left(\underset{k=1}{\overset{3}{}}B_k^1(is)\right)^{N+1}.$$
(18)
Factorizing out the ideal thermal factor $`e^{\beta X}/Z_N`$, the rest can be regarded as the mass spectrum of the $`N`$-parton system at $`X=m`$ and $`\mathrm{\Delta }E=\mathrm{\Delta }m`$:
$$\rho _N(m)=\left(\frac{V}{\pi ^2}\right)^N\mathrm{\Phi }_{N1}(m).$$
(19)
The functions $`\mathrm{\Phi }_N(X)`$ obey the recursion rule
$$\mathrm{\Phi }_N(X)=_0^X๐t\mathrm{\Phi }_{N1}(t)\mathrm{\Phi }_0(Xt)$$
(20)
with the starting point of the recursion,
$$\mathrm{\Phi }_0(X)=\frac{1}{2a^2}\left(e^{3aX}2e^{2aX}+e^{aX}\right).$$
(21)
Particular important cases are $`N=0`$ (partons), $`N=1`$ (mesons or diquarks) and $`N=2`$ (direct baryons). Besides the already known $`\mathrm{\Phi }_0(X)`$ (eq. 21) we obtain
$$\mathrm{\Phi }_1(X)=\frac{(aX3)e^{3aX}+4aXe^{2aX}+(aX+3)e^{aX}}{4a^5}$$
(22)
and
$`\mathrm{\Phi }_2(X)`$ $`=`$ $`{\displaystyle \frac{1}{16a^8}}[((aX)^29aX+24)e^{3aX}`$ (23)
$`8\left((aX)^2+6\right)e^{2aX}`$
$`+((aX)^2+9aX+24)e^{aX}]`$
For any $`N`$ the result contains three exponentials giving rise to a lowest limiting temperature of $`T_H=E_c/3=1/3a`$ for positive values of the parameter $`a`$.
Fig.2 shows the non-degenerate, integrated mass spectra. Values published on the Particle Data Group homepage are summed up in mass histograms. The respective numbers are fitted as $`N_M=1+Af_1((mm_M)/3T_H)`$ and $`N_B=1+A^2f_2((mm_B)/3T_H)`$ with $`A=V/a^3\pi ^2`$ and the integral functions $`f_n(x)=_0^x\mathrm{\Phi }_n(t)๐t`$. The fits to the data are most sensitive to the value of $`T_H`$, which however may be compensated by changing the assumed volume, $`V`$. Keeping $`m_M=0.14`$ GeV and $`m_B=0.94`$ GeV, only $`A`$ and $`T_H`$ are varied. Our best fit gives a relatively high value, $`T_H=0.35`$ GeV (meaning $`E_c=1.05`$ GeV) and a volume of $`V=261`$ $`fm^3`$ (a sphere with a radius of $`4`$ fm, or a box with a length of $`6.4`$ fm). Above the masses where the data seem to deviate from the fast growing part, the fit cannot be followed any more. According to ref.RESONANCES , newest data raise the experimental curve higher. Our idea, different from both the string and bag model consideration, seem to agree with the difference between the meson and baryon mass spectra, as well as with a polynomial upcurving of the baryon spectrum.
In conclusion we pointed out that Tsallis distributed massless partons can be combined to an effective mesonic and baryonic mass spectrum by considering the conserved quasi-energy as the hadron energy. Besides an ideal thermal factor, $`e^{\beta X}/Z_N`$, a further energy dependent factor results from the folding of parton distributions. It can be regarded as a thermal (pre)hadron mass spectrum emerging from a statistical hadronization picture. The prediction of this folding, while having two parameters (a volume and the Hagedorn temperature), gives an acceptable qualitative agreement with the known hadronic mass spectrum. In this picture a natural difference emerges between mesonic and baryonic resonances due to their different foldness by parton coalescence. The characteristic temperature, $`T_H=1/3a350`$ MeV is a limiting temperature: one cannot increase the temperature above this value, not even with an infinite amount of energy. The parameter $`E_c=1/a=3T_H1.05`$ GeV provides at the same time the scale where the power-law tail of individual $`p_T`$-spectra starts to dominate the exponential part, and it is intimately related to the typical pair-interaction energy, due to $`h(E_1,E_2)E_1E_2=E_1E_2/E_c`$.
Acknowledgment
This work has been supported by BMBF, by the Hungarian National Science Fund OTKA (T49466) and by DFG due to a Mercator Professorship for T.S.B.
|
warning/0506/hep-ph0506007.html
|
ar5iv
|
text
|
# High energy neutrino spin light
## Abstract
The quantum theory of spin light (electromagnetic radiation emitted by a Dirac massive neutrino propagating in dense matter due to the weak interaction of a neutrino with background fermions) is developed. In contrast to the Cherenkov radiation, this effect does not disappear even if the medium refractive index is assumed to be equal to unity. The formulas for the transition rate and the total radiation power are obtained. It is found out that radiation of photons is possible only when the sign of the particle helicity is opposite to that of the effective potential describing the interaction of a neutrino (antineutrino) with the background medium. Due to the radiative self-polarization the radiating particle can change its helicity. As a result, the active left-handed polarized neutrino (right-handed polarized antineutrino) converting to the state with inverse helicity can become practically โsterileโ. Since the sign of the effective potential depends on the neutrino flavor and the matter structure, the spin light can change a ratio of active neutrinos of different flavors. In the ultra relativistic approach, the radiated photons averaged energy is equal to one third of the initial neutrino energy, and two thirds of the energy are carried out by the final โsterileโ neutrinos.
Moscow State University, Department of Theoretical physics, $`119992`$ Moscow, Russia
A Dirac massive neutrino has non-trivial electromagnetic properties. In particular, it possesses non-zero magnetic moment . Therefore a Dirac massive neutrino propagating in dense matter can emit electromagnetic radiation due to the weak interaction of a neutrino with background fermions . As a result of the radiation, neutrino can change its helicity due to the radiative self-polarization. In contrast to the Cherenkov radiation, this effect does not disappear even if the refractive index of the medium is assumed to be equal to unity. This conclusion is valid for any model of neutrino interactions breaking spatial parity. The phenomenon was called the neutrino spin light in analogy with the effect, related with the synchrotron radiation power depending on the electron spin orientation (see ).
The properties of spin light were investigated basing upon the quasi-classical theory of radiation and self-polarization of neutral particles with the use of the BargmannโMichelโTelegdi (BMT) equation and its generalizations . This theory is valid when the radiated photon energy is small as compared with the neutrino energy, and this narrows the range of astrophysical applications of the obtained formulas.
In the present paper, the properties of spin light are investigated basing upon the consistent quantum theory, and this allows the neutrino recoil in the act of radiation to be considered for. The above mentioned restriction is eliminated in this way.
On the other hand, the detailed analysis of the results of our investigations shows that the features of the effect depend on the neutrino flavor, helicity and the matter structure . This fact leads to the conclusion that the spin light can initiate transformation of a neutrino from the active state to a practically โsterileโ state, and the inverse process is also possible.
When the interaction of a neutrino with the background fermions is considered to be coherent, the propagation of a massive neutrino in the matter is described by the Dirac equation with the effective potential . In what follows, we restrict our consideration to the case of a homogeneous and isotropic medium. Then in the frameworks of the minimally extended standard model, the form of this equation is uniquely determined by the assumptions similar to those adopted in :
$$\left(i\widehat{}\frac{1}{2}\widehat{f}(1+\gamma ^5)m_\nu \right)\mathrm{\Psi }_\nu =0.$$
(1)
The function $`f^\mu `$ is a linear combination of fermion currents and polarizations. The quantities with hats denote scalar products of Dirac matrices with 4-vectors, i.e. , $`\widehat{a}\gamma ^\mu a_\mu .`$
If the medium is at rest and unpolarized then $`๐=0.`$ The component $`f^0`$ calculated in the first order of the perturbation theory is as follows :
$$f^0=\sqrt{2}G_\mathrm{F}\left\{\underset{f}{}\left(I_{e\nu }+T_3^{(f)}2Q^{(f)}\mathrm{sin}^2\theta _\mathrm{W}\right)(n_fn_{\overline{f}})\right\}.$$
(2)
Here, $`n_f,n_{\overline{f}}`$ are the number densities of background fermions and anti-fermions, $`Q^{(f)}`$ is the electric charge of the fermion and $`T_3^{(f)}`$ is the third component of the weak isospin for the left-chiral projection of it. The parameter $`I_{e\nu }`$ is equal to unity for the interaction of electron neutrino with electrons. In other cases $`I_{e\nu }=0.`$ Summation is performed over all fermions $`f`$ of the background.
Let us obtain a solution of equation (1). Since function $`f^\mu =\mathrm{const},`$ equation (1) commutes with operators of canonical momentum $`i_\mu .`$ However, the commonly adopted choice of eigenvalues of this operator as quantum numbers in this problem is not satisfactory. Kinetic momentum components of a particle, related to its group 4-velocity $`u^\mu `$ by the relation $`q^\mu =m_\nu u^\mu ,q^2=m_\nu ^2,`$ are more suitable to play the role of its quantum numbers. This choice can be justified, since it is the particle kinetic momentum that can be really observed.
The explicit form of the kinetic momentum operator for the particle with spin is not known beforehand, and hence, in order to find the appropriate solutions, we have to use the correspondence principle.
It was shown in that when the effects of the neutrino weak interaction are taken into account, the Lorentz invariant generalization of the BMT equation for spin vector $`S^\mu `$ is as follows:
$$\dot{S}^\mu =2\mu _0\left\{\left(F^{\mu \nu }+G^{\mu \nu }\right)S_\nu u^\mu u_\nu \left(F^{\nu \lambda }+G^{\nu \lambda }\right)S_\lambda \right\},$$
(3)
where
$$G^{\mu \nu }=\frac{1}{2\mu _0}e^{\mu \nu \rho \lambda }f_\rho u_\lambda ,$$
(4)
and a dot denotes the differentiation with respect to the proper time $`\tau `$.
Let us introduce the quasi-classical spin wave functions. Such wave functions can be constructed as follows . Suppose the Lorentz equation is solved, i. e. the dependence of particle coordinates on proper time is found. Then the BMT equation transforms to ordinary differential equation, whose resolvent determines a one-parametric subgroup of the Lorentz group. The quasi-classical spin wave function is represented by a spin-tensor, whose evolution is determined by the same one-parametric subgroup.
In the case when the effect of an external electromagnetic field can be neglected as compared with the effect of the neutrino interaction with the background matter, the equation for the neutrino quasi-classical wave function $`\mathrm{\Psi }(\tau )`$ is
$$\dot{\mathrm{\Psi }}(\tau )=i\mu _0\gamma ^5{}_{}{}^{}G_{}^{\mu \nu }u_\nu \gamma _\mu \widehat{u}\mathrm{\Psi }(\tau ),$$
(5)
where $`{}_{}{}^{}G_{}^{\mu \nu }=1\mathrm{/}2e^{\mu \nu \rho \lambda }G_{\rho \lambda }`$ is a tensor dual to $`G^{\mu \nu }.`$ Obviously, the quasi-classical density matrix of a polarized neutrino takes the form
$$\rho (\tau ,\tau ^{})=\frac{1}{2}U(\tau ,\tau _0)\left(\widehat{q}(\tau _0)+m\right)\left(1\gamma ^5\widehat{S}(\tau _0)\right)U^1(\tau ^{},\tau _0),$$
(6)
where $`U(\tau ,\tau _0)`$ is the resolvent of equation (5), and the equation for $`U(\tau ,\tau _0)`$ is
$$\dot{U}(\tau ,\tau _0)=\frac{i}{4}\gamma ^5\left(\widehat{f}\widehat{u}\widehat{u}\widehat{f}\right)U(\tau ,\tau _0).$$
(7)
We note that the operator $`U(\tau ,\tau _0)`$ is defined up to a phase factor $`e^{iF(x)},`$ with the derivative of the exponent with respect to the proper time is equal to zero:
$$\dot{F}(x)=0.$$
(8)
Let us choose the solution of equation (1) in the form
$$\mathrm{\Psi }(x)=U\left(\tau (x)\right)\mathrm{\Psi }_0(x),$$
(9)
where $`\mathrm{\Psi }_0`$ is a solution of the Dirac equation for a free particle
$$\mathrm{\Psi }_0(x)=e^{i(qx)}(\widehat{q}+m_\nu )(1\gamma ^5\widehat{S}_0)\psi ^0.$$
(10)
Here $`\psi ^0`$ is constant bispinor and $`\mathrm{\Psi }_0(x)`$ normalized by the condition
$$\overline{\mathrm{\Psi }}_0(x)\mathrm{\Psi }_0(x)=2m_\nu .$$
Substitution of the expression (9) in eq. (1) results in the relation
$$\left\{\widehat{q}+(\widehat{}F)\frac{1}{2}\widehat{f}+\frac{1}{2}\gamma ^5\widehat{f}+\frac{1}{4}\gamma ^5\widehat{N}\left(\widehat{f}\widehat{u}\widehat{u}\widehat{f}\right)m_\nu \right\}e^{iF(x)}U\left(\tau (x)\right)\mathrm{\Psi }_0=0,$$
(11)
where $`N^\mu =^\mu \tau .`$ Since the commutator $`[\widehat{q},U]=0,`$ and the matrix $`U`$ is nondegenerate, then for this relation to hold the following condition is required
$$(\widehat{}F)\frac{1}{2}\widehat{f}+\frac{1}{2}\gamma ^5\widehat{f}+\frac{1}{4}\gamma ^5\widehat{N}\left(\widehat{f}\widehat{u}\widehat{u}\widehat{f}\right)=0.$$
(12)
It is easy to find out that the equation (12) is valid only if
$$\begin{array}{c}^\lambda F=\frac{1}{2}f^\lambda ,e^{\mu \nu \rho \lambda }N_\mu f_\nu u_\rho =0,\left(1(Nu)\right)f^\lambda +(Nf)u^\lambda =0.\hfill \end{array}$$
(13)
From two latter equations it follows that
$$N^\mu =\frac{f^\mu (fu)u^\mu f^2}{(fu)^2f^2u^2}.$$
(14)
So $`f^\mu =\mathrm{const},`$ then
$$\tau =(Nx),F=\frac{1}{2}(fx),$$
(15)
and we can write
$$U(x)=e^{i(fx)/2}\underset{\zeta =\pm 1}{}e^{i\zeta \phi }\mathrm{\Lambda }_\zeta .$$
(16)
Here
$$\mathrm{\Lambda }_\zeta =\frac{1}{2}\left[1\zeta \gamma ^5\widehat{S}_{tp}\widehat{q}/m_\nu \right],\zeta \pm 1,$$
(17)
are spin projection operators with eigenvalues $`\zeta \pm 1`$ respectively, and
$$\phi =\frac{\tau }{2}\sqrt{(fq)^2f^2m_\nu ^2}=\frac{(fq)(fx)f^2(qx)}{2\sqrt{(fq)^2f^2m_\nu ^2}},S_{tp}^\mu =\frac{q^\mu (fq)/m_\nu f^\mu m_\nu }{\sqrt{(fq)^2f^2m_\nu ^2}}.$$
(18)
From the obtained formulas it follows that the eigenvalues of the operator of canonical momentum $`i^\mu `$ are
$$P^\mu =q^\mu \left(1+\frac{\zeta f^2}{2\sqrt{(fq)^2f^2m_\nu ^2}}\right)+\frac{f^\mu }{2}\left(1\frac{\zeta (fq)}{\sqrt{(fq)^2f^2m_\nu ^2}}\right).$$
(19)
The dispersion law follows from eq. (19) in the form
$$P^2=m_\nu ^2+(Pf)f^2/2\zeta \sqrt{\left((Pf)f^2/2\right)^2f^2m_\nu ^2}.$$
(20)
If the medium is at rest and unpolarized then the neutrino total energy and canonical momentum are determined by the formulas
$$\epsilon =q^0+f^0/2,๐=๐ช\mathrm{\Delta }_{q\zeta },$$
(21)
where $`\mathrm{\Delta }_{q\zeta }=1+\zeta f^0/2|๐ช|,`$ and
$$S_{tp}^\mu =\frac{1}{m_\nu }\{|๐ช|,q^0๐ช/|๐ช|\},$$
(22)
i. e. the eigenvalues $`\zeta =\pm 1`$ determine the helicity of the particle. Consequently, the dispersion law is
$$\epsilon =\sqrt{\left(\mathrm{\Delta }|๐|\zeta f^0/2\right)^2+m_\nu ^2}+f^0/2,$$
(23)
where $`\mathrm{\Delta }=\mathrm{sign}\left(\mathrm{\Delta }_{q\zeta }\right).`$ Obviously
$$\frac{\epsilon }{๐}=\frac{๐ช}{q^0}$$
is the particle group velocity.
The relation (23) differs those used in previous papers (see, for example,) by the multiplier $`\mathrm{\Delta }.`$ This is due to the fact that, in these papers the projection of the particle spin on the canonical momentum $`๐`$ and not the helicity of the particle was used as the spin quantum number $`\zeta `$. The helicity is the projection of the spin on the direction of its kinetic momentum , because the rest frame of the particle is determined by the condition that its group velocity is equal to zero. In our problem the directions of canonical and kinetic momenta, generally speaking, are different, and hence, the projection of particle spin on the canonical momentum does not coincide with its helicity.
From formulas (21), it is seen that if the energy is expressed in terms of the kinetic momentum, then it does not depend on the particle helicity, while the particle canonical momentum is a function of the helicity. Therefore, the statement of the authors of , i.e., that the radiation of photons in the process of the spin light emission takes place due to neutrino transitions from the โexitedโ helicity state to the low-lying helicity state in matter, is not correct.
Let us consider the process of emitting photons by a massive neutrino in unpolarized matter at rest. In this case, the orthonormalized system of solutions for equation (1) is:
$$\mathrm{\Psi }(x)=\frac{\left|\mathrm{\Delta }_{q\zeta }\right|}{\sqrt{2q^0}}e^{i(q^0+f^0/2)x^0}e^{i\mathrm{๐ช๐ฑ}\mathrm{\Delta }_{q\zeta }}(\widehat{q}+m_\nu )(1\zeta \gamma ^5\widehat{S}_{tp})\psi ^0.$$
(24)
The formula for the spontaneous radiation transition probability of a neutral fermion with anomalous magnetic moment $`\mu _0`$ is<sup>1</sup><sup>1</sup>1In the expression for the radiation energy $`,`$ the additional factor $`k`$, i.e. the energy of radiated photon, appears in the integrand.:
$$\begin{array}{c}P=\frac{1}{2p^0}d^4xd^4y\frac{d^4qd^4k}{(2\pi )^6}\delta (k^2)\delta (q^2m_\nu ^2)\times \\ \times \mathrm{Sp}\left\{\mathrm{\Gamma }_\mu (x)\varrho _i(x,y;p,\zeta _i)\mathrm{\Gamma }_\nu (y)\varrho _f(y,x;q,\zeta _f)\right\}\varrho _{ph}^{\mu \nu }(x,y;k).\end{array}$$
(25)
Here, $`\varrho _i(x,y;p),\varrho _f(y,x;q)`$ are density matrices of the initial $`(i)`$ and final $`(f)`$ states of the fermion, $`\varrho _{ph}^{\mu \nu }(x,y;k)`$ is the radiated photon density matrix, $`\mathrm{\Gamma }^\mu =\sqrt{4\pi }\mu _0\sigma ^{\mu \nu }k_\nu `$ is the vertex function. The density matrix of longitudinally polarized neutrino in the unpolarized matter at rest constructed with the use of the solutions of equation (1) has the form
$$\varrho (x,y;p,\zeta )=\frac{1}{2}\mathrm{\Delta }_{p\zeta }^2(\widehat{p}+m_\nu )(1\zeta \gamma ^5\widehat{S}_p)e^{i(x^0y^0)(p^0+f^0/2)+i(๐ฑ๐ฒ)๐ฉ\mathrm{\Delta }_{p\zeta }}.$$
(26)
After summing over photon polarizations<sup>2</sup><sup>2</sup>2We do not consider the polarization of spin light photons here. In the quasi-classical approximation, this problem was investigated in . and integrating with respect to coordinates we obtain the expression for the transition rate under investigation:
$$\begin{array}{c}W=\frac{\mu _0^2}{p^0}\frac{d^4qd^4k}{(2\pi )}\delta (k^2)\delta (q^2m_\nu ^2)\delta (p^0q^0k^0)\delta ^3(๐ฉ\mathrm{\Delta }_{p\zeta _i}๐ช\mathrm{\Delta }_{q\zeta _f}๐ค)T(p,q),\end{array}$$
(27)
where
$$\begin{array}{c}T(p,q)=4\mathrm{\Delta }_{p\zeta _i}^2\mathrm{\Delta }_{q\zeta _f}^2\left\{(pk)(qk)\zeta _i\zeta _f\left[k^0|๐ฉ|p^0(\mathrm{๐ฉ๐ค})/|๐ฉ|\right]\left[k^0|๐ช|q^0(\mathrm{๐ช๐ค})/|๐ช|\right]\right\}.\end{array}$$
(28)
After integrating over $`๐ค,`$ $`k^0,`$ $`|๐ช|`$ we obtain the spectral-angular distribution of the final neutrino
$$\begin{array}{c}W=\zeta _i\zeta _f\frac{\mu _0^2}{\pi p^0|๐ฉ|}\underset{m_\nu }{\overset{p^0}{}}dq^0\mathrm{\Delta }_{p\zeta _i}\mathrm{\Delta }_{q\zeta _f}dO\times \\ \times \delta ((p^0q^0)^2+2|๐ฉ||๐ช|\mathrm{\Delta }_{p\zeta _i}\mathrm{\Delta }_{q\zeta _f}\mathrm{cos}\vartheta _\nu |๐ฉ|^2\mathrm{\Delta }_{p\zeta _i}^2|๐ช|^2\mathrm{\Delta }_{q\zeta _f}^2)\times \\ \times \{(f^0/2)^2\left[\zeta _f|๐ฉ||๐ช|+\zeta _i(m_\nu ^2p^0q^0)\right]^2+\left[(f^0/2)(\zeta _iq^0|๐ฉ|\zeta _fp^0|๐ช|)+m_\nu ^2(p^0q^0)\right]^2\},\end{array}$$
(29)
where
$$|๐ช|=\sqrt{(q^0)^2m_\nu ^2}.$$
It is convenient to express the results of integrating over angular variables using dimensionless quantities. Introducing the notations
$$x=q^0/m_\nu ,\gamma =p^0/m_\nu ,d=|f^0|/2m_\nu ,\overline{\zeta }_{i,f}=\zeta _{i,f}\mathrm{sign}(f^0)$$
(30)
we have
$$\begin{array}{c}W_{\overline{\zeta }_f}=\frac{\mu _0^2m_\nu ^3}{\gamma (\gamma ^21)}\frac{dx}{\sqrt{x^21}}\{d^2[\overline{\zeta }_f\sqrt{\gamma ^21}\sqrt{x^21}\overline{\zeta }_i(\gamma x1)]^2+\\ +[\gamma x+d(\overline{\zeta }_ix\sqrt{\gamma ^21}\overline{\zeta }_f\gamma \sqrt{x^21})]^2\}.\end{array}$$
(31)
The integration bounds in the formula (31) are
$$\begin{array}{cc}x\mathrm{}\hfill & \gamma [1,\mathrm{}),\hfill \end{array}$$
(32)
if $`\overline{\zeta }_i=1,`$
$$\begin{array}{cc}x\mathrm{}\hfill & \gamma [1,\gamma _0),\hfill \\ x[\omega _1,\omega _2]\hfill & \gamma [\gamma _0,\gamma _1),\hfill \\ x[1,\omega _2]\hfill & \gamma [\gamma _1,\gamma _2),\hfill \\ x\mathrm{}\hfill & \gamma [\gamma _2,\mathrm{}),\hfill \end{array}$$
(33)
if $`\overline{\zeta }_i=1,\overline{\zeta }_f=1,`$ and
$$\begin{array}{cc}x\mathrm{}\hfill & \gamma [1,\gamma _1),\hfill \\ x[1,\omega _1]\hfill & \gamma [\gamma _1,\gamma _2),\hfill \\ x[\omega _2,\omega _1]\hfill & \gamma [\gamma _2,\mathrm{}).\hfill \end{array}$$
(34)
if $`\overline{\zeta }_i=1,\overline{\zeta }_f=1.`$
Here
$$\omega _1=\frac{1}{2}\left(z_1+z_1^1\right),\omega _2=\frac{1}{2}\left(z_2+z_2^1\right),$$
(35)
where
$$\begin{array}{c}z_1=\gamma +\sqrt{\gamma ^21}2d,\hfill \\ z_2=\gamma \sqrt{\gamma ^21}+2d,\hfill \end{array}$$
(36)
and
$$\begin{array}{cc}\gamma _0=\sqrt{1+d^2},\hfill & \\ \gamma _1=\frac{1}{2}\left\{\left(1+2d\right)+\left(1+2d\right)^1\right\},\hfill & \\ \gamma _2=\frac{1}{2}\left\{\left(12d\right)+\left(12d\right)^1\right\}\hfill & d<1/2,\hfill \\ \gamma _2=\mathrm{}\hfill & d1/2.\hfill \end{array}$$
(37)
The integration is carried out elementary. The transition rate under investigation is defined as
$$\begin{array}{c}W_{\overline{\zeta }_f}=\frac{\mu _0^2m_\nu ^3}{4}\{(1+\overline{\zeta }_f)[Z(z_1,1)\mathrm{\Theta }(\gamma \gamma _1)+Z(z_2,1)\mathrm{\Theta }(\gamma \gamma _2)]+\\ +(1\overline{\zeta }_f)[Z(z_1,1)\mathrm{\Theta }(\gamma _1\gamma )+Z(z_2,1)\mathrm{\Theta }(\gamma _2\gamma )]\mathrm{\Theta }(\gamma \gamma _0)\}(1\overline{\zeta }_i).\end{array}$$
(38)
Here
$$\begin{array}{c}Z(z,\overline{\zeta }_f)=\frac{1}{\gamma (\gamma ^21)}\{\mathrm{ln}z[\gamma ^2+d\sqrt{\gamma ^21}+d^2+1/2]+\\ +\frac{1}{4}\left(z^2z^2\right)\left[d^2\left(2\gamma ^21\right)+d\sqrt{\gamma ^21}+1/2\right]+\\ +\frac{\overline{\zeta }_f}{4}\left(zz^1\right)^2\left[2d\sqrt{\gamma ^21}+1\right]d\gamma \\ \left(zz^1\right)\left[d^2+d\sqrt{\gamma ^21}+1\right]\gamma \\ \overline{\zeta }_f(z+z^12)[d\sqrt{\gamma ^21}+\gamma ^2]d\}.\end{array}$$
(39)
Therefore, the transition rate after summation over polarizations of the final neutrino becomes
$$W_{\overline{\zeta }_f=1}+W_{\overline{\zeta }_f=1}=\frac{\mu _0^2m_\nu ^3}{2}(1\overline{\zeta }_i)\left\{Z(z_1,1)+Z(z_2,1)\right\}\mathrm{\Theta }(\gamma \gamma _0).$$
(40)
If $`d\gamma 1,`$ then expression (38) leads to the formula
$$W_{\overline{\zeta }_f}=\frac{16\mu _0^2m_\nu ^3d^3}{3\gamma }(\gamma ^21)^{3/2}(1\overline{\zeta }_i)(1+\overline{\zeta }_f),$$
(41)
obtained in the quasi-classical approximation in .
In the ultra-relativistic limit $`(\gamma 1,d\gamma 1),`$ the transition rate is given by the expression
$$W_{\overline{\zeta }_f}=\mu _0^2m_\nu ^3d^2\gamma (1\overline{\zeta }_i)(1+\overline{\zeta }_f).$$
(42)
Let us consider now the radiation power. If we introduce the function
$$\stackrel{~}{Z}(z,\overline{\zeta }_f)=\gamma Z(z,\overline{\zeta }_f)Y(z,\overline{\zeta }_f),$$
(43)
where
$$\begin{array}{c}Y(z,\overline{\zeta }_f)=\frac{1}{\gamma (\gamma ^21)}\{\mathrm{ln}z[d^2+d\sqrt{\gamma ^21}+1]\gamma \\ \frac{1}{4}\left(z^2z^2\right)\left[d^2+d\sqrt{\gamma ^21}+1\right]\gamma +\\ +\frac{1}{12}\left(zz^1\right)^3\left[d^2\left(2\gamma ^21\right)+d\sqrt{\gamma ^21}+1/2\right]+\\ +\frac{1}{2}\left(zz^1\right)\left[2d^2\gamma ^2+2d\sqrt{\gamma ^21}+\gamma ^2+1\right]+\\ +\frac{\overline{\zeta }_f}{12}\left(\left(z+z^1\right)^38\right)\left[2d\sqrt{\gamma ^21}+1\right]d\gamma \\ \frac{\overline{\zeta }_f}{4}(zz^1)^2[d\sqrt{\gamma ^21}+\gamma ^2]d\},\end{array}$$
(44)
then the formula for the total radiation power can be obtained from (38), (40) by the substitution $`Z(z,\overline{\zeta }_f)\stackrel{~}{Z}(z,\overline{\zeta }_f).`$ It can be verified that if $`d\gamma 1`$ then the radiation power is
$$I_{\overline{\zeta }_f}=\frac{32\mu _0^2m_\nu ^4d^4}{3}(\gamma ^21)^2(1\overline{\zeta }_i)(1+\overline{\zeta }_f).$$
(45)
This result was obtained in the quasi-classical approximation in . In the ultra-relativistic limit, the radiation power is equal to
$$I_{\overline{\zeta }_f}=\frac{1}{3}\mu _0^2m_\nu ^4d^2\gamma ^2(1\overline{\zeta }_i)(1+\overline{\zeta }_f).$$
(46)
It can be seen from equations (42) and (46) that in the ultra-relativistic limit the averaged energy of emitted photons is $`\epsilon _\gamma =\epsilon _\nu /3.`$ It should be pointed out that the obtained formulas are valid both for a neutrino and for an antineutrino. The charge conjugation operation leads to the change of the sign of the effective potential and the replacement of the left-hand projector by the right-hand one in the equation (1). Thus the sign in front of the $`\gamma ^5`$ matrix remains invariant.
Using eq. (27), it is possible to find the dependence of the radiated photon energy on the angle $`\vartheta _\gamma `$ between the direction of the neutrino propagation and the photon wave vector:
$$\frac{k^0}{m_\nu }=2d\frac{\beta Xd/\gamma }{\left(X+d/\gamma \right)\left(Xd/\gamma \right)}.$$
(47)
Here $`\beta =\sqrt{\gamma ^21}/\gamma `$ is the neutrino velocity and $`X=1\left(\beta d/\gamma \right)\mathrm{cos}\vartheta _\gamma .`$ In the quasi-classical approximation, this formula reduces to the relation
$$\frac{k^0}{m_\nu }=\frac{2d\beta }{1\beta \mathrm{cos}\vartheta _\gamma },$$
(48)
which follows from the results of after Lorentz transformation to the laboratory frame.
The following conclusions can be made from the obtained results. A neutrino (antineutrino) can emit photons due to coherent interaction with matter only when its helicity has the sign opposite to the sign of the effective potential $`f^0.`$ Otherwise, radiation transitions are impossible. In the case of low energies of the initial neutrino, only radiation without spin-flip is possible and the probability of the process is very low. At high energies, the main contribution to radiation is given by the transitions with the spin-flip, the transitions without spin-flip are either absent or their probability is negligible. This leads to total self-polarization, i. e. the initially left-handed polarized neutrino (right-handed polarized antineutrino) is transformed to practically โsterileโ right-handed polarized neutrino (left-handed polarized antineutrino). For โsterileโ particles, the situation is opposite. They can be converted to the active form in the medium โtransparentโ for the active neutrino.
With the use of the effective potential calculated in the first order of the perturbation theory (2), the following conclusions can be made. If the matter consists only of electrons then, in the framework of the minimally extended standard model in the ultra-relativistic limit (here we use gaussian units), we have for the transition rate
$$W_{\overline{\zeta }_f}=\frac{\alpha \epsilon _\nu }{32\mathrm{}}\left(\frac{\mu _0}{\mu _\mathrm{B}}\right)^2\left(\frac{\stackrel{~}{G}_\mathrm{F}n_e}{m_ec^2}\right)^2(1\overline{\zeta }_i)(1+\overline{\zeta }_f),$$
(49)
and for the total radiation power
$$I_{\overline{\zeta }_f}=\frac{\alpha \epsilon _\nu ^2}{96\mathrm{}}\left(\frac{\mu _0}{\mu _\mathrm{B}}\right)^2\left(\frac{\stackrel{~}{G}_\mathrm{F}n_e}{m_ec^2}\right)^2(1\overline{\zeta }_i)(1+\overline{\zeta }_f).$$
(50)
Here $`\epsilon _\nu `$ is the neutrino energy, $`\mu _\mathrm{B}=e/2m_e`$ is the Bohr magneton, $`\alpha `$ is the fine structure constant, $`m_e`$ is the electron mass and $`\stackrel{~}{G}_\mathrm{F}=G_\mathrm{F}(1+4\mathrm{sin}^2\theta _\mathrm{W}),`$ where $`G_\mathrm{F},\theta _\mathrm{W}`$ are the Fermi constant and the Weinberg angle respectively. Thus, after the radiative transition, two thirds of the initial active neutrino energy are carried away by the final โsterileโ one.
At the same time, as it can be seen from eq. (2), a muon neutrino in the electron medium does not emit any radiation. Moreover, a muon neutrino does not emit radiation in an electrically neutral medium, when the number density of protons is equal to the electron number density. And an electron neutrino can emit radiation if the electron number density is greater than the neutron number density. An example of such medium is provided by the Sun. Therefore the spin light can change the ratio of active neutrino of different flavors.
It is obviously that the above conclusions change to opposite if the matter consists of antiparticles. Therefore the neutrino spin light can serve as a tool for determination of the type of astrophysical objects, since neutrino radiative transitions in dense matter can result in radiation of photons of super-high energies, even exceeding the GZK cutoff. Indeed, the neutron medium is โtransparentโ for all active neutrinos, but an active antineutrino emits radiation in such a medium, the transition rate and the total radiation power can be obtained from equations (49) and (50) after substitution $`\stackrel{~}{G}_\mathrm{F}G_\mathrm{F},n_en_n.`$ If the density of the neutron star is assume to be $`n10^{38},`$ the transition rate is estimated as
$$W=10^{22}\frac{\epsilon _\nu }{\epsilon _{\mathrm{G}ZK}}\left(\frac{\mu _0}{\mu _\mathrm{B}}\right)^2,$$
(51)
where $`\epsilon _{\mathrm{G}ZK}=5\times 10^{19}`$eV is GZK cutoff energy. Although the transition rate determined by eq. (51) is extremely low, this effect can still serve as one of a possible explanations of the cosmic ray paradox.
The spin light can also be important for the understanding of the โdark matterโ formation mechanism in the early stages of evolution of the Universe.
When the present paper was already submitted for publication, we came across an article , where the spin light theory was also considered. The formulas of in the ultra-relativistic limit of physical interest reproduce the results for the transition rate and the total power of spin light already obtained in our earlier publication .
The author is grateful to V.G. Bagrov, A.V. Borisov, and V.Ch. Zhukovsky for fruitful discussions.
This work was supported in part by the grant of President of Russian Federation for leading scientific schools (Grant SS โ 2027.2003.2).
|
warning/0506/nucl-th0506053.html
|
ar5iv
|
text
|
# Comparative Regge analysis of ฮ,ฮฃโฐ,ฮโข(1520) and ฮโบ production in ๐พโข๐, ๐โข๐ and ๐โข๐ reactions
## 1 Introduction
Inspite of the belief that the structure of baryons in the octet and decuplet representation is roughly understood (and exhausted), recent claims on the discovery of the manifestly exotic baryon $`\mathrm{\Theta }^+`$ have opened a new chapter in hadron physics (see e.g. Refs. Nakano ; Barmin ; Stepanyan ; Barth ; HERMES ; Kubarovsky ; Asratyan ; Alt ; Kubarovsky2 ; Aleev ; Abdel-Bary ). Although the existence and properties of this exotic and long-lived baryonic state still need final experimental confirmation a variety of theoretical models have been set up with different assumptions about the internal structure of the $`\mathrm{\Theta }^+`$. Here the quark-soliton model of Diakonov, Petrov and Polyakov Diakonov was the first to claim the existence of an antidecuplet with rather narrow spectral width (actually prior to the experimental observations). However, the $`\mathrm{\Theta }^+`$ might also be โboundโ due to strong diquark correlations (in a relative $`p`$-wave) as proposed in Ref. Jaffe or due to a strong mixing with the octet Hyodo . Alternatively, it might even be explained on the basis of the constituent quark model involving clusters Oka . Further models have been proposed in the last 2 years that all claim a different dynamical origin of the pentaquark $`\mathrm{\Theta }^+`$ Beane ; Ioffe (cf. Jafferev and Refs. cited therein). Additionally, the properties of the exotic state have been analysed within the framework of QCD sum rules Sugiyama and even lattice QCD Fodor ; Bali ; LQCDlast ; Csikor .
However, for a better understanding of this exotic state and its wave function (in terms of the elementary degrees of freedom) it is very important to study the dynamics of $`\mathrm{\Theta }^+`$ production in comparison to the production of non-exotic strange baryons. In this respect exclusive reactions with strangeness ($`s\overline{s}`$) production are of interest, i.e. (starting with $`\gamma `$ induced reactions):
$$\gamma p\overline{K}^0\mathrm{\Theta }^+,$$
(1)
$$\gamma p\overline{K}^0\mathrm{\Theta }^+$$
(2)
and
$$\gamma d\mathrm{\Lambda }\mathrm{\Theta }^+.$$
(3)
The first two reactions (1) and (2) โ where the $`s`$ quark ends up in the mesonic final state โ can be compared with $`\mathrm{\Lambda }`$ production in the binary reactions
$$\gamma pK^+\mathrm{\Lambda },$$
(4)
$$\gamma pK^+\mathrm{\Lambda }$$
(5)
where the $`s`$ quark ends up in the hyperon. Note that very detailed measurements of the reaction (4) have been performed in the energy range from threshold up to a photon energy of 2.6 GeV Glander04 . The third reaction (3), furthermore, can be compared with the two-body deuteron photodisintegration reaction $`\gamma dpn`$ which has been studied recently at Jlab Rossi .
We recall that several studies of $`\mathrm{\Theta }^+`$ photoproduction have been performed within the framework of isobar models using the Born approximation Oh ; Nam ; Li ; Ko ; Liu2 ; Yu . Since these models involve a variety of uncertain parameters (coupling constants and cutoffโs) the resulting cross sections differ from several nb to almost 1 $`\mu `$b. On the other hand the Regge model has a substantial advantage that the amount of uncertain parameters is much lower and that the latter can be fixed by other reactions in a more reliable fashion Mart . Accordingly, in this work we will apply the Quark-Gluon Strings Model (QGSM) combined with Regge phenomenology to the analysis of the differential and total cross sections of the exclusive reactions $`\gamma pK^+\mathrm{\Lambda }`$, $`\gamma pK^+\mathrm{\Sigma }^0`$ and $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$. Similar final channels will be investigated also for pion and proton induced reactions.
We note that the QGSM was originally proposed by Kaidalov in Ref. Kaidalov for the description of binary hadronic reactions; as demonstrated in Refs. Kaidalov ; KaidalovSurveys the QGSM describes rather well the experimental data on exclusive and inclusive hadronic reactions at high energy. More recently this model has been also successfully applied to the description of the nucleon and pion electromagnetic form factors Tchekin as well as deuteron photodisintegration Rossi ; Grishina ; Grishina2 .
We recall that the QGSM is based on two ingredients: i) a topological expansion in QCD and ii) the space-time picture of the interactions between hadrons, that takes into account the confinement of quarks. The $`1/N`$ expansion in QCD (where $`N`$ is the number of colors $`N_c`$ or flavors $`N_f`$) was proposed by โt Hooft Hooft ; the behavior of different quark-gluon graphs according to their topology, furthermore, was analyzed by Veneziano Veneziano with the result that in the large $`N`$ limit the planar quark-gluon graphs become dominant. This approach โ based on the $`1/N_f`$ expansion Veneziano with $`N_cN_f`$ โ was used by Kaidalov Kaidalov ; KaidalovSurveys in formulating the QGSM. Again for sufficiently large $`N_f`$ the simplest planar quark-gluon graphs were found to give the dominant contribution to the amplitudes of binary hadronic reactions. Moreover, it can be shown that (in the space-time representation) the dynamics described by planar graphs corresponds to the formation and break-up of a quark-gluon string (or color tube) in the $`s`$-channel (see e.g. Casher ; Artru ; Casher2 ; Andersson ; Gurvich ). On the other hand an exchange of the $`u`$ and $`\overline{s}`$ quarks in the $`t`$-channel implies that the energy behavior of the amplitudes โ described by quark diagrams in Fig. 1 โ is given by the contribution of the $`K^{}`$ Regge trajectory. In this sense the QGSM can be considered as a microscopic model of Regge phenomenology. This in turn allows to obtain many relations between amplitudes of different binary reactions and residues of Regge poles which determine these amplitudes Kaidalov ; KaidalovSurveys ; Boreskov .
Our investigation is organized as follows: In Section 2 we outline our approach and present the results for the differential cross sections $`\gamma pK^+\mathrm{\Lambda },K^+\mathrm{\Sigma }^0,\overline{K}^0\mathrm{\Theta }^+`$. In Section 3 we compare total cross sections for the reactions $`\gamma pK^+\mathrm{\Lambda }(1520)`$ and $`\gamma p\overline{K^0}\mathrm{\Theta }^+`$ while in Section 4 we step on with the pion induced reactions $`\pi ^{}pK^0\mathrm{\Lambda }`$ and $`\pi ^{}pK^{}\mathrm{\Theta }^+`$. An analysis of $`\mathrm{\Theta }^+`$ production in exclusive and inclusive $`NN`$ collisions is presented in Section 5 while a summary of our studies is given in Section 6.
## 2 The reactions $`\gamma pK^+\mathrm{\Lambda }`$, $`\gamma pK^+\mathrm{\Sigma }^0`$ and $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$
We first concentrate on $`\gamma `$ induced reactions and work out the Regge model in more detail. The reaction $`\gamma pK^+\mathrm{\Lambda }`$ can be described by the exchange of two valence ($`u`$ and $`\overline{s}`$) quarks in the $`t`$-channel with any number of gluon exchanges between them (Fig. 1 a) ). Alternatively, in terms of the Regge phenomenology this diagram corresponds to the $`K^{}`$\- Reggeon exchange mechanism shown in Fig. 2 a).
Employing the Regge model further on we can write the $`\gamma pK^+\mathrm{\Lambda }`$ amplitude in the form
$`T(\gamma pK^+\mathrm{\Lambda }){\displaystyle \frac{e}{2\gamma _\rho }}T(\rho ^0pK^+\mathrm{\Lambda })=`$
$`{\displaystyle \frac{e}{2\gamma _\rho }}g_{\rho KK^{}}g_{pK^{}\mathrm{\Lambda }}F_1(t)(s/s_0^{K\mathrm{\Lambda }})^{\alpha _K^{}(t)}.`$ (6)
Here $`e^2/4\pi `$ is the fine structure constant, $`\gamma _\rho ^2/4\pi `$ = 0.55, $`\alpha _K^{}(t)`$ is the $`K^{}`$ Regge trajectory, $`s_0^{K\mathrm{\Lambda }}=(M_\mathrm{\Lambda }+m_K)^2`$, $`g_{\rho KK^{}}`$ and $`g_{pK^{}\mathrm{\Lambda }}`$ are the coupling constants describing the interaction of the $`K^{}`$ Reggeon with the $`\rho K`$ and $`p\mathrm{\Lambda }`$ systems. Within the Reggeized-Born-term model (see eg. Refs. Irving ; Levy ; Guidal ) it is assumed that the coupling constants $`g_i`$ in the Regge amplitude of Eq. (6) can be identified with the coupling constants in an effective Lagrangian model. However it is difficult to justify this assumption and we do not address this model here. We follow another approach for a โReggeization of the amplitudeโ as proposed in Refs. Tchekin ; Grishina ; Grishina2 , i.e. by using the $`s`$channel convolution representation in the QGSM. In this approach one can express the amplitude for the reaction $`\gamma pK^+\mathrm{\Lambda }`$ in terms of the $`s`$-channel convolution of two amplitudes: $`T(\gamma pq+qq)`$ and $`T(q+qqK^+\mathrm{\Lambda })`$ (see Fig. 1). Then โ using the Regge representation for the hadron-quark and quark-hadron transition amplitudes โ we can Reggeize the binary amplitude $`\gamma pK^+\mathrm{\Lambda }`$. Such a procedure was applied in Refs. Grishina ; Grishina2 to define the spin structure of the deuteron photoproduction amplitude. We point out that this approach is more general and gives a vertex structure of the amplitude at negative $`t`$ different from the Reggeized-Born term.
Assuming, furthermore, that the $`\mathrm{\Theta }^+`$ is a pentaquark of structure ($`uudd\overline{s}`$) we can use a similar strategy for the $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ reaction. The relevant quark diagrams for this reaction are shown in Fig. 1 b), c). It is obvious that in terms of the Regge phenomenology we can also use the $`K^{}`$\- Reggeon exchange model to describe the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ (cf. Fig. 2 b). The $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ amplitude reads accordingly
$`T(\gamma p\overline{K}^0\mathrm{\Theta }^+){\displaystyle \frac{e}{2\gamma _\rho }}T(\rho _0p\overline{K}^0\mathrm{\Theta }^+)=`$
$`{\displaystyle \frac{e}{2\gamma _\rho }}g_{\rho KK^{}}g_{pK^{}\mathrm{\Theta }}F_2(t)(s/s_0^{K\mathrm{\Theta }})^{\alpha _K^{}(t)}`$ (7)
with $`s_0^{K\mathrm{\Theta }}=(M_\mathrm{\Theta }+m_K)^2`$. In the following calculations the form factor squared $`|F_i|^2`$ in (6), (7) is chosen always in the form
$$|F_i|^2=(1B_it)\mathrm{exp}(2R_i^2t).$$
(8)
For the further developments it is important to recall that the QGSM originally was formulated for small scattering angles (or small negative 4-momentum transfer (squared) $`t`$). Thus the question arises about the extrapolation of the QGSM amplitudes to large angles (or large $`t`$). Here we adopt the same concept as in our previous works Grishina ; Grishina2 : following Coon et al. Coon we assume that only a single analytic Regge term with a logarithmic trajectory gives the dominant contribution to large momentum transfer processes. As shown in Coon such a model (denoted as โlogarithmic dual modelโ) can describe very well the differential cross section $`d\sigma /dt`$ for elastic $`pp`$ scattering in the energy range of $`524`$ GeV/c for $`t`$ up to 18 GeV<sup>2</sup>. The logarithmic Regge trajectory itself can be written in the form
$$\alpha (t)=\alpha (0)(\gamma \nu )\mathrm{ln}(1t/T_B).$$
(9)
with the parameters $`\alpha (0)=0.32`$ and $`T_B=6`$ GeV<sup>2</sup> that have been fixed in Refs.Volkovitsky94 ; Burak . To describe the energy dependence of the $`\gamma pK^+\mathrm{\Lambda }`$ differential cross section at fixed $`t`$ we have found $`\gamma \nu =2.75`$.
We note in passing that logarithmic Regge trajectories have also been discussed in Refs. Bugrij ; Ito ; Chikovani . The special limit $`\gamma \nu 0`$ at large $`t`$ corresponds to โsaturatedโ trajectories, i.e. all trajectories approach a constant asymptotically. Such a case leads to the โconstituent-interchange modelโ that can be considered as a predecessor of the โasymptotic quark counting rulesโ Brodsky ; Hiller . Moreover, the model with โsaturatedโ trajectories has also successfully been applied to the large-$`t`$ behavior of exclusive photon- and hadron-induced reactions in Refs. Guidal ; Fiore ; White ; Battaglieri .
Formally, the amplitude (6) does not contain spin variables. Nevertheless it can be used for a description of the differential cross section that is averaged over the spin states of the initial particles and summed up over the polarizations of the final particles,
$`{\displaystyle \frac{d\sigma _{\gamma pK^+\mathrm{\Lambda }}}{dt}}={\displaystyle \frac{1}{64\pi s}}{\displaystyle \frac{1}{(p_\pi ^{\mathrm{cm}})^2}}\times `$
$`{\displaystyle \frac{1}{4}}{\displaystyle \underset{\lambda _\gamma ,\lambda _p,\lambda _\mathrm{\Lambda }}{}}\left|\lambda _\mathrm{\Lambda }|T_{\gamma pK^+\mathrm{\Lambda }}(s,t)|\lambda _p,\lambda _\gamma \right|^2.`$ (10)
Here the amplitude squared can be written as
$`{\displaystyle \frac{1}{4}}{\displaystyle \underset{\lambda _\gamma ,\lambda _p,\lambda _\mathrm{\Lambda }}{}}\left|\lambda _\mathrm{\Lambda }|T_{\gamma pK^+\mathrm{\Lambda }}(s,t)|\lambda _p,\lambda _\gamma \right|^2=`$
$`{\displaystyle \frac{e^2}{4\gamma _\rho ^2}}g_{\rho KK^{}}^2g_{pK^{}\mathrm{\Lambda }}^2|F_1(t)|^2\left|s/s_0^{K\mathrm{\Lambda }}\right|^{2\alpha _K^{}(t)}.`$ (11)
Let us now discuss constraints that have to be fulfilled for the residues and coupling constants. In line with Refs. KaidalovSurveys ; Volkovitsky94 ; Nogteva we assume that for the planar quark diagrams with light quarks there is some kind of โuniversalityโ of the secondary Reggeon couplings to $`q\overline{q}`$ mesons involved in a binary reaction, i.e. in particular
$$g_{\rho KK^{}}g_{\pi KK^{}}g_{\rho \pi \pi }=g_0$$
(12)
with $`g_0`$ 5.8. Taking $`g_{\rho KK^{}}=5.8`$ and normalising the differential cross section of the reaction $`\gamma pK^+\mathrm{\Lambda }`$ at $`t=0`$ we find $`g_{pK^{}\mathrm{\Lambda }}`$ 3.5. This result shows that the Reggeon couplings to mesons and baryons might be, in general, different by up to a factor of 2.
We mention that the form factor $`F_i`$ โ determining the $`t`$-dependence of the residue โ was parametrized in Refs. KaidalovSurveys ; Volkovitsky94 as
$$F_i(t)=\mathrm{\Gamma }(1\alpha _i(t)).$$
(13)
Indeed, such a choice of the form factor is convenient for an analytical continuation of the amplitude to positive $`t`$ where the $`\mathrm{\Gamma }`$ function decreases exponentially with $`t`$. However, in the region of negative $`t`$ the parametrization (13) exhibits a factorial growth and is not acceptable (see e.g. the discussion in Ref. Cassing ). Accordingly we use the parametrization of the form factor (8), which decreases with $`t`$. To keep the same normalization of the amplitude at $`t=0`$ we have to change the coupling constant squared as
$$g_0^2g_M^2=g_0^2\mathrm{\Gamma }(1\alpha _i(0)),$$
(14)
where for the $`K^{}`$ trajectory we have $`\mathrm{\Gamma }(1\alpha _K^{}(0))1.32`$.
The differential cross section for the reaction $`\gamma pK^+\mathrm{\Lambda }`$ is presented in Figs. 3 โ 4 as a function of the laboratory photon energy at fixed values of $`t`$. The solid lines are calculated using our model with the following coupling constants and parameters of the form factor: $`g_{\rho KK^{}}=5.8,g_{pK^{}\mathrm{\Lambda }}3.5`$, $`B_1=5`$ GeV<sup>-2</sup>, $`R_1^2=1.13`$ GeV<sup>-2</sup>. The experimental data are from Refs.Glander04 (full triangles), Boyarski69 (full circles), Boyarski71 (full stars), Feller72 (empty triangles) and Anderson76 (empty circles). The agreement between the solid curves and the experimental data clearly supports the dominant role of the $`K^{}`$ Regge trajectory in the reaction $`\gamma pK^+\mathrm{\Lambda }`$. The dashed line in Fig. 3 โ calculated at $`t=0.165`$ GeV<sup>2</sup> โ describes the result for a $`K`$ Regge trajectory normalized to the data at $`E_\gamma =2`$ GeV. Definitely, it can not describe the energy dependence of the differential cross section and we may conclude that the $`K`$ Regge trajectory is subdominant.
We have applied our model also to the description of the reaction $`\gamma pK^+\mathrm{\Sigma }^0`$ adopting the same coupling constants and form factor as for the reaction $`\gamma pK^+\mathrm{\Lambda }`$, however, modifying the scaling factor to $`s_0^{K\mathrm{\Sigma }}=(M_\mathrm{\Sigma }+m_K)^2`$. The total cross sections of the reactions $`\gamma pK^+\mathrm{\Lambda }`$ and $`\gamma pK^+\mathrm{\Sigma }^0`$ as a function of the laboratory photon energy are shown in Fig. 5 (upper and lower parts describe the reactions $`\gamma pK^+\mathrm{\Lambda }`$ and $`\gamma pK^+\mathrm{\Sigma }^0`$, respectively) in comparison to the experimental data from Ref. Glander04 . In this context one has to note that the Regge model gives only the average cross section for a particular channel and misses resonant amplitudes at low energy. For example, according to recent data on the reaction $`\gamma pK^+\mathrm{\Sigma }^0`$ McNabb the $`s`$-channel resonance contributions are found to be important for photon beam energies at least up to 1.5 GeV. Since the $`K^+\mathrm{\Lambda }`$ and $`K^+\mathrm{\Sigma }^0`$ systems show strong resonances in the 1.3 to 2 GeV invariant mass region the latter cannot be described in the Regge approach. Nevertheless, the results of our model calculations (presented as the solid lines) are in a good agreement with the data โ except for the resonance structures mentioned above โ and support the โuniversalityโ of Reggeon couplings.
## 3 Total cross sections for the reactions $`\gamma pK^+\mathrm{\Lambda }(1520)`$ and $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$
In this Section we explore if the universality of the $`K^{}`$ trajectory coupling to baryons with constituent $`qqs`$ quarks also holds in the binary reactions
$$\gamma pK^+Y_i$$
(15)
with $`Y_1=\mathrm{\Lambda }(1116),Y_2=\mathrm{\Lambda }(1520),\mathrm{}`$. In case of the universality to hold we have
$$g_{pK^{}\mathrm{\Lambda }(1520)}=g_{pK^{}\mathrm{\Lambda }}3.5,F_{\mathrm{\Lambda }(1520)}(t)=F_1(t).$$
(16)
The resulting total cross section of the reaction $`\gamma pK^+\mathrm{\Lambda }(1520)`$ is presented in Fig. 6 in comparison to the experimental data from Barber (full squares) and Barth (empty square). The solid curve is calculated for the coupling constant $`g_{pK^{}Y}=3.5`$ which corresponds directly to the prediction from the universality principle. The dashed curve is calculated using $`g_{pK^{}Y}=4.14`$ and is in a good agreement with the data; the deviation between the two curves does not exceed 40%. Therefore, the data on the reactions $`\gamma pK^+\mathrm{\Lambda }`$ and $`\gamma pK^+\mathrm{\Lambda }(1520)`$ support the assumption on the universality of the $`K^{}`$ trajectory coupling to $`q\overline{q}`$ mesons as well as to baryons with constituent $`qqs`$ quarks (at least within a factor of 2). We, accordingly, consider (or define) the universality principle to hold if a variety of cross sections is described (predicted) within a factor better than 2.
We continue with the $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ reaction and explore if the unversality principle (in the sense defined above) also holds in this case. The cross section of the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ was estimated by the SAPHIR collaboration Barth as
$$\sigma _{\gamma p\overline{K}^0\mathrm{\Theta }^+}200\text{nb}$$
(17)
at an average photon energy of $``$2 GeV. Letโs first adopt 200 nb as an upper limit for the total cross section of the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ at 2 GeV. In this case equation (17) implies already a noticeable violation of the universality principle for $`\mathrm{\Theta }^+`$ photoproduction since โ assuming the universality principle to hold for $`\mathrm{\Theta }^+`$ โ we get
$$g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Lambda }(1520)}4.14,F_2(t)=F_1(t).$$
(18)
This leads to a total cross section of the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ shown by the solid line in Fig. 7. The dash-dotted curve in Fig. 7 is calculated assuming
$$g_{pK^{}\mathrm{\Theta }}^{\text{SAPHIR}}0.4g_{pK^{}\mathrm{\Lambda }(1520)}1.8,F_2(t)=F_1(t)$$
(19)
in order to match the quoted cross section in (17). However, the new preliminary results from the CLAS collaboration DeVita do not support the estimate (17)and indicate that the upper limit on the total cross section of the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ should be much lower:
$$\sigma _{\gamma p\overline{K}^0\mathrm{\Theta }^+}1รท4\text{nb}.$$
(20)
When taking 4 $`nb`$ as an upper limit we find
$$g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}0.06g_{pK^{}\mathrm{\Lambda }(1520)}0.25,F_2(t)=F_1(t).$$
(21)
Therefore, using the preliminary result from the CLAS collaboration we find a very strong suppression of the $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ cross section relative to the prediction from the universality principle for photoproduction of the lowest $`qqs`$ baryons. If the pentaquark exists, we may interpret this finding as a clear indication of a substantially different quark structure of the $`\mathrm{\Theta }^+`$.
## 4 Pion induced reactions: $`\pi ^{}pK^0\mathrm{\Lambda }`$ and $`\pi ^{}pK^{}\mathrm{\Theta }^+`$
We continue with $`\pi ^{}`$ induced reactions and assume that the amplitudes of the reactions $`\pi ^{}pK^0\mathrm{\Lambda }`$ and $`\pi ^{}pK^{}\mathrm{\Theta }^+`$ are also dominated by the contribution of the $`K^{}`$ Regge trajectory (see Fig. 8 a) and b)) such that the cross sections are fully determined. We directly step on with the results for the differential cross section of the $`\pi ^{}pK^0\mathrm{\Lambda }`$ reaction as a function of $`t`$ at $`p_{\mathrm{lab}}`$=4.5, 6, 8, 10.7 and 15.7 GeV/c shown in Fig. 9 in comparison to the experimental data from Foley73 ; Crennel . The results of our model are displayed by the dashed lines calculated for the following parameters:
$$g_{\pi K^{}K}=5.8,g_{pK^{}\mathrm{\Lambda }}4.5,B=0,R_1^2=2.13\mathrm{GeV}^2.$$
(22)
We see that the coupling constants $`g_{\pi K^{}K}`$ and $`g_{pK^{}\mathrm{\Lambda }}`$ โ in the case of the reaction $`\pi ^{}pK^0\mathrm{\Lambda }`$ โ are also in agreement with the universality assumption for the coupling constants. However, to describe the $`t`$-dependence of the differential cross section we have to employ different parameters for the form factor $`F(t)`$ as compared to the reaction $`\gamma pK^+\mathrm{\Lambda }`$. This can be explained, in particular, by the different relative contributions of the baryon spin-flip terms in the reactions $`\gamma pK^+\mathrm{\Lambda }`$ and $`\pi ^{}pK^0\mathrm{\Lambda }`$.
The total cross section of the reaction $`\pi ^{}pK^0\mathrm{\Lambda }`$ is presented in Fig. 10 where the dashed curve is the result of our calculations. As expected for a Regge model โ and discussed above โ we see some deviation of the QGSM from the data Landoldt in the resonance region. However, at higher energies, where the explicit resonance structure disappears, the theoretical calculations are in a good agreement with the data.
Now using the coupling constants $`g_{\pi K^{}K}=5.8`$, $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{SAPHIR}}`$ and $`g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$ and assuming $`R_1^2`$ and $`B`$ to be the same as for the reaction $`\pi ^{}pK^0\mathrm{\Lambda }`$ we can calculate the cross section for the reaction $`\pi ^{}pK^{}\mathrm{\Theta }^+`$ . The results are shown in Fig. 10 by the solid and dotted line, respectively; these cross sections reach about 10 (0.2) $`\mu `$b in their maximum.
## 5 $`\mathrm{\Theta }^+`$ production in exclusive and inclusive $`NN`$ collisions
Further constraints on the universality of amplitudes and cross sections for $`\mathrm{\Theta }^+`$ production are provided by $`NN`$ collisions. Here the amplitude of the reaction $`pp\mathrm{\Theta }^+\mathrm{\Sigma }^+`$ (cf. upper diagram in Fig. 12) has been calculated using the coupling constant $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{SAPHIR}}`$ ($`g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$). Of course, taking into account the results of Section 2 we can safely assume that $`g_{pK^{}\mathrm{\Sigma }}g_{pK^{}\mathrm{\Lambda }}`$. At the same time the coupling constant $`g_{pK^{}\mathrm{\Lambda }}`$ may vary from 3.5 (if we define it via the reaction $`\gamma pK^+\mathrm{\Lambda }`$) to 4.5 (if we define it via the $`\pi ^{}pK^0\mathrm{\Lambda }`$ reaction ). In this section we use $`g_{pK^{}\mathrm{\Lambda }}`$ = 4.5. The predictions for the total cross section of the $`pp\mathrm{\Theta }^+\mathrm{\Sigma }^+`$ reaction are shown by the solid (dotted) line in Fig. 11 using $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{SAPHIR}}`$ ($`g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$) . The cross section described by the dotted line is $``$ 20โ30 times smaller than the experimental value $`\sigma =0.4\pm 0.1(\text{stat})\pm 0.1(\text{syst})\mu `$Abdel-Bary measured at a beam momentum of $`2.95`$ GeV/c (open square) and clearly signals an incompatibility of the different measurements.
We note, that the first analysis of the reaction $`pp\mathrm{\Theta }^+\mathrm{\Sigma }^+`$ was performed by Polyakov et al. in Ref. Polyakov even before the $`\mathrm{\Theta }^+`$ baryon was โdiscoveredโ from the experimental side. Their estimation of the cross section โ within the kaon-exchange approximation โ was about 2 $`\mu `$b at the initial momentum $``$3 GeV/c. Approximately the same value of the total cross section of the reaction $`pp\mathrm{\Theta }^+\mathrm{\Sigma }^+`$ was found by Liu and Ko in Ref. LiuKo later on. Our calculated cross section is about 0.8 $`\mu `$b on the basis of the coupling (19) and 16 $`nb`$ for the coupling (21) at the excess energy $`Q=22`$ MeV ($`p_{\mathrm{lab}}=2.95`$ GeV/c). The maxima of cross sections are $`3.3\mu `$b and 66$`nb`$, respectively, at the excess energy $`Q=460`$ MeV ($`p_{\mathrm{lab}}=4.4`$ GeV/c). These cross sections are smaller by factors of 2 and 90, respectively, compared to the early estimates.
We continue with alternative production mechanisms for $`\mathrm{\Theta }^+`$ production in $`pp`$ collisions as described by the middle and lower diagrams in Fig. 12. In the following we use the method of Yao Yao to calculate the cross sections of the reactions:
1) $`ppp\overline{K}^0\mathrm{\Theta }^+`$ with the pion exchange (middle diagram in Fig. 12)
2) $`ppp\overline{K}^0\mathrm{\Theta }^+`$ with the kaon exchange (lower diagram in Fig. 12 with $`X=\overline{K}^0p`$),
3) $`pp\mathrm{\Theta }^+X`$ with the kaon exchange (lower diagram in Fig. 12).
In the case of pion exchange the expression for the total cross section can be written in the form:
$`\sigma (ppp\overline{K}^0\mathrm{\Theta }^+)=`$
$`{\displaystyle \frac{G_{\pi NN}^2}{8\pi ^2p_1s}}{\displaystyle _{W_{\mathrm{min}}}^{W_{\mathrm{max}}}}kW^2\sigma (\pi ^0p\overline{K}^0\mathrm{\Theta }^+,W)๐W`$
$`\times {\displaystyle _{t_{\mathrm{min}}(W)}^{t_{\mathrm{max}}(W)}}F_\pi ^4(t){\displaystyle \frac{1}{(tm_\pi ^2)^2}}tdt,`$ (23)
where $`W`$ is the invariant mass of the $`\overline{K}^0\mathrm{\Theta }^+`$ system, $`k`$ is defined as
$$k=\left((W^2(m_pm_\pi )^2)(W^2(m_p+m_\pi )^2)\right)^{1/2}/2W,$$
$`p_1`$ is the initial proton momentum in the c.m. system, $`t=(p_2p_4)^2`$, and $`G_{\pi NN}=13.45`$. Assuming that $`J^P(\mathrm{\Theta }^+)=\frac{1}{2}^+`$ we have the following expression for the kaon exchange contribution
$`\sigma (pp\mathrm{\Theta }^+X)=`$
$`{\displaystyle \frac{G_{\mathrm{\Theta }KN}^2}{8\pi ^2p_1s}}{\displaystyle _{W_{\mathrm{min}}}^{W_{\mathrm{max}}}}kW^2\sigma (\overline{K}^0pX,W)๐W`$
$`\times {\displaystyle _{t_{\mathrm{min}}(W)}^{t_{\mathrm{max}}(W)}}F_K^4(t){\displaystyle \frac{1}{(tm_K^2)^2}}(t\mathrm{\Delta }_{M43}^2)dt.`$ (24)
Here $`W`$ is the invariant mass of the system $`X`$ and $`\mathrm{\Delta }_{M43}^2=(m_\mathrm{\Theta }m_p)^2`$. The form-factors for the virtual pion and kaon exchange have been chosen of the monopole type
$$F_j(t)=\frac{\mathrm{\Lambda }_j^2m_j^2}{\mathrm{\Lambda }_j^2t}$$
(25)
with $`\mathrm{\Lambda }_\pi =1.3`$ GeV and $`\mathrm{\Lambda }_K=1`$ GeV. These parameters have been used in Ref. Gasparyan to describe the total cross section of the reaction $`ppK^+\mathrm{\Lambda }p`$. Obviously, the contribution of the kaon exchange to the reaction $`pp\mathrm{\Theta }^+X`$ depends on the coupling constant $`G_{\mathrm{\Theta }KN}`$, which can only be estimated (or fixed by upper limits). If the $`\mathrm{\Theta }^+`$ decay width is less than $`1`$ MeV PDG we have $`G_{\mathrm{\Theta }KN}1.4`$. The solid and dashed lines in Fig. 13 present our results for the inclusive $`pp\mathrm{\Theta }^+X`$ and exclusive $`pp\mathrm{\Theta }^+\overline{K}^0p`$ reactions as calculated within the kaon-exchange model for $`G_{\mathrm{\Theta }KN}=1.4`$ (as an upper limit). The inclusive cross section turns out to be about $`1.5\mu `$b at high energies while the exclusive $`pp\mathrm{\Theta }^+\overline{K}^0p`$ cross section is at least one order of magnitude smaller.
The reaction $`ppp\overline{K}^0\mathrm{\Theta }^+`$ may proceed also via $`\pi `$ exchange. The corresponding contribution to the total cross section calculated with the coupling constant $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$ from Eq. (21) is presented by the dash-dotted line in Fig. 13. It reaches about $`10`$ nb in its maximum. Therefore, we can conclude that the $`\mathrm{\Theta }^+`$ production cross section in $`pp`$-collisions is dominated by the $`K`$ exchange and should be about a few $`\mu `$b. Note that using the hadronic Lagrangian model Liu and Ko LiuKo found a considerably larger cross section for $`\mathrm{\Theta }^+`$ production, i.e. about 50 $`\mu `$b in pion-nucleon reactions and $``$20 $`\mu `$b in proton-proton reactions. From our point of view the latter results are essentially due to a large coupling constant $`G_{\mathrm{\Theta }KN}4.4`$ in LiuKo , which corresponds to $`\mathrm{\Gamma }_\mathrm{\Theta }20`$ MeV. Such large couplings, however, should be excluded according to the more recent analysis in Ref. Sibirtsev .
## 6 Conclusions
In this study we have analyzed $`\mathrm{\Lambda }`$, $`\mathrm{\Sigma }^0`$, $`\mathrm{\Lambda }(1520)`$ and $`\mathrm{\Theta }^+`$ production in binary reactions induced by photon, pion and proton beams in the framework of the Quark-Gluon Strings Model combined with Regge phenomenology. Starting with the existing experimental data on the $`\gamma pK^+\mathrm{\Lambda }`$ reaction we have demonstrated that the differential and total cross sections at photon energies $`116`$ GeV and $`t<2`$ GeV<sup>2</sup> can be described very well by the model with a dominant contribution of the $`K^{}`$ Regge trajectory. We stress that the rather good description of the large $`t`$ region was possible only due to the logarithmic form of the $`K^{}`$ Regge trajectory (9). It has been demonstrated, furthermore, that the data on the reactions $`\gamma pK^+\mathrm{\Lambda }`$, $`\gamma pK^+\mathrm{\Sigma }^0`$ and $`\gamma pK^+\mathrm{\Lambda }(1520)`$ โ at least within a factor of 2 โ support the assumption on the universality of the $`K^{}`$ trajectory coupling to $`q\overline{q}`$ mesons as well as to baryons with $`qqs`$ constituent quarks. This implies that โ using the same parameters as for $`\gamma pK^+\mathrm{\Lambda }`$ โ we are able to reproduce the total $`\gamma pK^+\mathrm{\Sigma }^0`$ and $`\gamma pK^+\mathrm{\Lambda }(1520)`$ cross sections (within an accuracy of 40%). On the other hand, as a consequence of the SAPHIR data Barth and the preliminary data from CLAS DeVita , there is an essential suppression of the $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ cross section relative to the prediction within the universality principle that was shown to hold (with reasonable accuracy) for the photoproduction of the lowest $`qqs`$ baryons. We conclude that this suppression indicates a substantially different quark structure and wave function of the $`\mathrm{\Theta }^+`$ (in case of its final experimental confirmation).
Moreover, we have suggested that the amplitudes of the reactions $`\pi ^{}pK^0\mathrm{\Lambda }`$ and $`\pi ^{}pK^{}\mathrm{\Theta }^+`$ are also dominated by the contribution of the $`K^{}`$ Regge trajectory (cf. Fig.8 a) and b)). Indeed, the differential and total cross sections of the $`\pi ^{}pK^0\mathrm{\Lambda }`$ reaction are found to be in a reasonable agreement with the universality principle. Using parameters defined by the analysis of the reactions $`\gamma pK^+\mathrm{\Lambda }`$, $`\gamma pK^+\mathrm{\Sigma }^0`$, $`\gamma pK^+\mathrm{\Lambda }(1520)`$, $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ and $`\pi ^{}pK^0\mathrm{\Lambda }`$ we have calculated the cross section for the reaction $`\pi ^{}pK^{}\mathrm{\Theta }^+`$ . We predicted a maximum cross section of about 200 nb (cf. Fig. 10) for the coupling constant $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$ extracted from the new preliminary CLAS result on the reaction $`\gamma p\overline{K}^0\mathrm{\Theta }^+`$ DeVita .
We extended our model additionally to the analysis of the binary reaction $`pp\mathrm{\Sigma }^+\mathrm{\Theta }^+`$. We found that the cross section of this reaction โ measured by the COSY-TOF collaboration Abdel-Bary โ is 20โ30 times larger than the value predicted by the model with the coupling constant $`g_{pK^{}\mathrm{\Theta }}=g_{pK^{}\mathrm{\Theta }}^{\text{CLAS}}`$. Furthermore, we have investigated the exclusive and inclusive $`\mathrm{\Theta }^+`$ production in the reactions $`ppp\overline{K}^0\mathrm{\Theta }^+`$ and $`pp\mathrm{\Theta }^+X`$ and found that the inclusive $`\mathrm{\Theta }^+`$ production in pp collisions at high energy should be on the level of 1 $`\mu `$b.
The systematic and comparative Regge analysis โ provided by our study โ will also allow in future to relate different data sets from $`\gamma `$, $`\pi `$ and proton induced reactions on $`s\overline{s}`$ pair production and finally should yield a transparent picture of the dynamics as well as the properties of (possible) exotic states.
## Acknowledgments
The authors are grateful to A.B. Kaidalov and E. De Sanctis for useful discussions. This work was partially supported by DFG (Germany) and INFN (Italy). One of us (V. G.) acknowledges the financial support from Forschungszentrum Jรผlich (FFE grant 41520739 (COSY - 071)).
|
warning/0506/astro-ph0506731.html
|
ar5iv
|
text
|
# 1 INTRODUCTION
## 1 INTRODUCTION
Since the beginning of the 1980s objects like radio-galaxies, quasi-stellar radio sources (Quasars), Seyfert Galaxies are simply classified as Active Galactic Nuclei (AGN), because the โenergy-engineโ is thought to be the same for all of them: A super-massive black hole of millions of solar masses accreting from its host galaxy (Fig. 1). AGN with radio-emitting lobes or jets are called radio-loud, the others radio-quiet (Ulrich et al. 1997).
The class of X-ray binary systems is very similar, the โenergy-engineโ is a compact object of only a few solar masses accreting from the companion star (Fig. 1). Up to now there are known 280 X-ray binary systems (Liu et al. 2000, 2001), but only 18 of them (Fig. 2) show evidence of a radio-jet and therefore are radio-loud applying the same definition as for the AGN.
The radio-loud subclass of X-ray binary systems includes together with the microquasars โ objects where high resolution radio interferometric techniques like VLBI have given direct evidence of the presence of collimated and relativistic jets (Mirabel et al. 1992) โ also unresolved radio sources with a flat spectrum, which give indirect evidence for continuous ejection.
I here review how the three important basic components of a microquasar (Fig. 3) \- a compact object, an accretion disk and a collimated relativistic jet - have been observed in gamma-rays, X-rays, optical and radio emission. After a basic introduction of the accretion-ejection processes presented in section 2, the following sections describe the astronomical methods: First, presenting their theory and, afterwards their application on the source LS I +61303. In detail: Optical observations, reported in Sect. 3, reveal the nature of the compact object: Neutron star or black hole. X-ray observations, discussed in Sect. 4, deliver information on the accretion disk, while radio observations allow to study the jet; the section describes how the simultaneous use of X-ray and radio tools allow to study โthe disk-jetโ connection. The application to LS I +61303 shows the limits of an approximation of the accretion theory which assumes a constant velocity of the accretor along the orbit. Section 5 shows that the observational results can be explained if one takes into account the geometry of the orbit. The better understanding of the physical processes motivates for new observations at higher energy. Gamma-ray observations are discussed in Sect. 6. The characteristics of the jet (morphology, velocity, etc) are derived from high resolution radio-astronomical observations and their typical procedures are described in Sect. 7. Finally, the conclusion of this multiband approach applied on LS I +61303 are presented in Sect. 8
## 2 THE ACCRETION-EJECTION PROCESS
### 2.1 Accretion
X-ray binaries are stellar systems formed by two stars of a very different nature: A normal star (acting as a mass donor) and a compact object (the accretor) that can either be a neutron star or a black hole (White et al. 1996).
Several mechanisms have been proposed to explain the presence of a compact object in a binary system, and they principally depend on the mass of the companion. If the companion is a low mass star (Low Mass X-ray Binary, LMXB) the theory assumes, first, the formation of the neutron star/BH, which later coupled with its companion in a (tidal) capturing process. As a matter of fact several LMXB are close to the core of globular clusters or near the center of the Galactic bulge (Verbunt & van den Heuvel 1996). The High Mass X-ray Binary (HMXB) systems, where the companion has a mass above 5$`M`$, have a galactic disc distribution characteristic that of young stars (population I). It is assumed that large scale mass transfer has occurred in the system before the supernova explosion: When the progenitor star of the compact object had evolved to a Red Giant and had filled its Roche lobe, the smaller companion accreted from it to a level that it survived after the explosion. In most HMXRB systems the massive companion of the compact object is a rapid rotating Be star, whose formation is explained by the large amount of angular momentum received together with matter from the initially more massive and therefore faster evolving companion (Verbunt & van den Heuvel 1996).
Since the binary pair is in orbital motion around the common center of gravity, the matter leaving the companion star has some angular momentum ($`J`$), which prevents it from directly falling into the accretor. The stream of matter orbits the compact object with a radius determined by $`J`$ and the mass of the compact object ($`M_X`$). The angular momentum is redistributed by the viscosity: Some of the material takes angular momentum and spreads outwards, whereas other material spirals inwards. In this way a disk is created from the initial ring of matter (King 1996; Longair 1994 p. 135). Gradually the matter drifts inwards until it reaches the last stable orbit, called โthe inner radiusโ of the accretion disk ($`R_{in}`$), which for a non rotating black hole is approximately three times the Schwarzschild radius ($`r_\mathrm{s}`$):
$$r_\mathrm{s}=\frac{2GM_\mathrm{X}}{c^2}$$
(1)
$$R_{\mathrm{in}}=3r_\mathrm{s}9(\frac{M_\mathrm{X}}{M})\mathrm{km}.$$
(2)
The viscosity has two effects: Besides the transport of angular momentum it also acts like a frictional force resulting in the dissipation of heat. The amount of friction depends on how fast the gas orbits around the compact object; the temperature ($`T_{\mathrm{in}}`$) reaches its maximum at the inner disk where it rises up to (Longair 1994 p. 141)
$$T_{\mathrm{in}}210^7(\frac{M_\mathrm{X}}{M_{}})^{1/4}\mathrm{K}.$$
(3)
On the basis of this equation we see that for a microquasar of about 1 solar mass the matter around the last stable orbit is heated up to tens of million degrees therefore emitting predominantly in the X-ray band. This led to the name โX-ray binariesโ for this class of objects. The Earthยดs atmosphere is opaque at these wavelengths; therefore it is understandable that there was an impasse (until recent developments in X-ray astronomy) in discovering such stellar sources and (besides SS433 discovered by chance) their subclass with relativistic jets. On the contrary, the temperature of the last stable orbit around a super massive black hole of an AGN with a mass of $`10^9M`$ is $`T_{in}=10^5K`$. Therefore the emission is in the ultraviolet band causing the โblue bumpโ associated with the โvisibleโ Quasars. This last fact is an example that the same laws are applied to both, AGN and X-ray binaries, deriving parameters only scaled with the mass (see Heinz & Sunyaev 2003; Merloni et al. 2003; Falcke et al. 2003).
The accretion luminosity can be written as:
$$L=\eta \dot{m}c^2,$$
(4)
where $`\eta `$, the efficency of energy conversion, expresses here how compact an object (with radius $`R`$) is: $`\eta =1/2(r_s/R)`$ (Longair 1994 p. 134). Whereas for a white dwarf $`\eta `$ is only 0.0001, for neutron stars $`\eta `$ is 0.1. As a comparison, the release of nuclear binding energy occuring in the conversion of four protons into helium has an $`\eta =\frac{4m_pm_{He}}{4m_p}=7x10^3`$. Thus, accretion in neutron stars already is an order of magnitude more efficient as an energy source as compared with nuclear energy generation (Longair 1994 p. 134).
However, there is a limit of energy that is possible to extract by accretion: If the force generated in the accretion disk by radiation pressure exceeds the gravitational force of the compact object a further accretion of gas ceases. The expression for that luminosity limit, the Eddington luminosity, is (Longair 1994 p. 137; Frank et al. 2002)
$$L_\mathrm{E}=1.310^{38}(\frac{M_X}{M_{}})\mathrm{erg}\mathrm{s}^1$$
(5)
### 2.2 Magnetohydrodynamic Jet Production
In the case of a small vertical magnetic field threading the disk the plasma pressure dominates the magnetic field pressure and the differentially rotating disk bends the magnetic field lines, which are passively wound up (Fig. 4) (Meier et al. 2001).
Due to the compression of the magnetic field lines the magnetic pressure may become larger than the gas pressure at the surface of the accretion disk, where the density is lower. At this point the gas starts to follow the twisted magnetic field lines, creating two spinning flows. This extracts angular momentum (magnetic braking) from the surface of the disk and enhances the radial accretion. The avalanching material further pulls the deformed magnetic field with it and afterwards magnetic reconnection may happen (Matsumoto et al 1996). The flux tubes open up and reconnect as is known from stellar flares (Massi et al. 2002).
The thickness of the disc is a fundamental parameter in this magneto-rotational process, or better the extent of the poloidal magnetic field frozen in the disc (Meier 2001; Meier et al. 2001; Maccarone 2004). No radio jet is associated with X-ray binaries in High/Soft states (Sect. 4.1), where the X-ray spectrum is dominated by a geometrically thin (optically thick) accretion disc (Shakura & Sunyaev 1973). In the contrary numerical results show a jet being launched from the inner geometrically thick portion of the accretion disc that is present (ADAF/Corona) when the X-ray binaries are in their Low/Hard state (Sect. 4.2) (Meyer et al. 2000; Meier 2001).
In conclusion, a better understanding of the transition from radio-quiet to radio-loud therefore seems to be possible through a better understanding of the X-ray states and their switch mechanism.
### 2.3 Strong Magnetic Fieds: the X-Ray Pulsars
A radio-loud X-ray binary system may contain either a black hole or a neutron star with a low ($`B<10^{10}`$ Gauss) magnetic field. Accreting neutron stars with a low magnetic field can give rise to jet production because of the following reason: As described in the previous section only for a low magnetic field can the plasma pressure dominate and bend the field. On the contrary, jet formation is prevented in presence of a strong magnetic field. In the case of B$`>10^{12}`$ Gauss, the plasma is forced to move along the magnetic field lines, converges onto the magnetic poles of the neutron star and there releases its energy creating two X-ray emitting caps that, in case of a misalignment of the rotation and the magnetic axis, produce X-ray pulses (Fig. 5). X-ray pulsars are not associated with microquasars. The lack of detected radio emission from X-ray pulsar systems is discussed in Fender et al. (1996).
## 3 OPTICAL OBSERVATIONS
### 3.1 The Nature of the Compact Object
The most reliable method to determine the nature of the compact object is the study of the Doppler shift of absorption lines in the spectrum of its companion. The study of the changing radial velocity during the orbital motion is a technique that has been applied for more than one hundred years to measure the masses of stars in binary-systems. The same method is applied for systems like X-ray binaries, where one component is โinvisibleโ. In this case the variations of the radial velocity of the normal companion during its orbit are studied.
The amplitude ( $`K_\mathrm{c}`$) of the radial velocity variations (Fig. 6) of the mass donor and the period (P<sub>orb</sub>) of the system applying Newtonยดs/Keplerโs third law define a quantity called the โmass functionโ (Charles & Wagner 1996), which is equal to:
$$f=\frac{P_{\mathrm{orb}}K_\mathrm{c}^3}{2\pi G}=\frac{M_\mathrm{X}^3sin^3i}{(M_\mathrm{X}+M)^2}$$
where $`M_\mathrm{X}`$ and $`M`$ are the masses of the compact object and of the companion, respectively, $`i`$ is the angle between the axis of the orbit and the line of sight and $`G`$ is the gravitational constant.
The mass function alone already provides a lower limit for $`M_\mathrm{X}`$ corresponding to a zero-mass companion ($`M`$=0) viewed at the maximum inclination angle (i=90). In the cases where the inclination $`i`$ and the mass of the companion $`M`$ are known one can solve for the mass $`M_X`$ of the invisible object.
Rhoades & Ruffini (1974), taking the most extreme equation of state that produces the maximum critical mass of a neutron star, established the very upper limit of 3.2 $`M`$ for a neutron star. This absolute maximum mass provides a decisive constraint to observationally distinguish between neutron stars and black holes.
In Fig. 7 a list of some X-ray binaries is given for which both mass function and $`M_\mathrm{X}`$ is available. All sources below GRS 1009-45, with $`f(M)=3.17M`$, can be defined black hole candidates on the basis of the mass function alone. On the contrary, for cases where $`f(M)<3`$ the determination of inclination and mass of the companion is mandatory to determine the type of object.
It is worth mentioning here that the accumulation of accreted material on the surface of a neutron star triggers thermonuclear bursts (see typical profile in Fig. 8). These are called bursts of Type I. No Type I burst has ever been observed from a compact object where optical observations resulted in a mass above 3 $`M`$. That fact might confirm that in black holes there is no surface where material can accumulate (Narayan & Heyl 2002). In conclusion: Observations of Type I bursts give a direct evidence for the existence of a neutron star.
### 3.2 The Nature of the Companion Star
The classification of the X-ray binaries into Low Mass X-ray Binary and High Mass X-ray Binary systems leaves unspecified the nature of the accreting object and is based on the mass of the companion star (van Paradijs & McClintock 1996).
A LMXB contains a late type (K,M) low mass donor star. The mass transfer takes place via Roche lobe overflow (Frank et al. 2002): Material streams through the inner Lagrangian point and will orbit the compact object at the radius determined by its specific angular momentum.
In HMXB systems, the companion is an OB star. OB stars have a substantial stellar wind (mass loss rates $`10^{10}10^5M\mathrm{yr}^1`$) with a velocity of $`v_{wind}v_{escape}=\sqrt{\frac{2GM}{R}}10^3`$ km/s. However, matter leaves the star in all directions, not only towards the accretor as in the case of Roche lobe overflow. This accretion therefore is less efficient (King 1996). The expression for the accretion rate is (Bondi 1952):
$$\dot{M}=\frac{4\pi \rho _{\mathrm{wind}}(GM_X)^2}{v_{\mathrm{rel}}^3}$$
(6)
where $`\rho _{\mathrm{wind}}`$ is the density, and $`v_{rel}`$ depends on the velocity along the orbit $`v_{orb}`$ and on the wind velocity ($`v_{wind}`$). The accretion therefore becomes more efficient for denser and slower winds present in Be stars. In these rapidly spinning stars together with a high velocity (1000 km s<sup>-1</sup>) low density wind at high latitudes there also exists a dense and slow (100 km s<sup>-1</sup>) disk-like wind around the equator having a power law density distribution (Waters et al. 1988).
### 3.3 LS I +61303: The Be-Star
The ultraviolet spectroscopy of LS I +61303 by Hutchings & Crampton (1981) indicates that the primary star is a main sequence B0-B0.5 star (L$`10^{38}`$ erg sec<sup>-1</sup>, T$`{}_{eff}{}^{}2.610^4`$K). Its distance is 2.0$`\pm `$0.2 Kpc (Frail and Hjellming 1991). The optical spectrum is that of a rapidly rotating star with V$`\mathrm{sin}`$ i = 360 $`\pm `$ 25 km s<sup>-1</sup>. The critical rotational velocity for a normal B0 V star is $``$600 km s<sup>-1</sup> and Be stars may rotate at a velocity that does not generally exceed 0.9 of this, i.e. 540 km s<sup>-1</sup> (Hutchings et al. 1979). Therefore the lower limit for the inclination of the orbit compatible with these data is 38. However, Hutchings & Crampton observed shell absorptions in the strong Balmer and He I lines. For a disk sufficiently flat this corresponds to a large inclination angle (i $`90^{}`$) (Kogure 1969). The result is a range of 38โ90 for the inclination of the orbit for LS I +61303 .
LS I +61303 is the only X-ray binary system showing variations compatible with the orbital period at X-rays (Paredes et al. 1997; Leahy 2001), at Gamma-rays (Massi 2004; Massi et al. 2004b), at optical wavelengths in both continuum (Maraschi & Treves 1981; Paredes & Figueras 1986; Mendelson & Mazeh 1989) and lines, (Zamanov & Martรญ 2000; Liu et al. 2000; Apparao 2000). The most accurate value for the orbital period is however from radio astronomical measurements resulting in 26.4960 $`\pm `$ 0.0028 days (Gregory & Taylor 1978; Taylor & Gregory 1982; Gregory 2002).
The range of the mass for a B-star is 5โ18 M$``$. Fits performed on near infrared data by Martรญ and Paredes (1995) result in an eccentricity of e$``$ 0.7-0.8 and mass in the range 10โ18 M$``$.
Finally, because of the high eccentricity, an important parameter of the system is the phase at the periastron passage, already determined by Hutchings & Crampton (1981) and very recently confirmed by Casares et al. (2004) to be $`\varphi `$=0.2. The zero phase by convention refers to the date t<sub>0</sub>=JD 2443366.775, the date of the first radio detection of the system (Gregory & Taylor 1978).
Figure 9 shows a sketch of the system with the compact object travelling (and accreting) through the dense, variable and structured wind of the Be-star along the quite eccentric orbit.
### 3.4 LS I +61303: Really a Neutron Star ?
The composite and variable nature of the spectral features of LS I +61303, its long period and its high eccentricity make it difficult to derive meaningful radial velocities for this source. The observations of $`K_c`$ made by Hutchings & Crampton (1981) imply a mass function in the range 0.0028$`<f<`$0.043 (see also Punsly 1999). Following the discussion in section 3.1 we see that the low value of the mass function is not enough to establish the real nature of the compact object. Knowledge of the values for inclination and mass of the companion star also is necessary for that. On the other hand, we have seen in the previous section that the inclination has a quite large range (38-90) and the possible range for the mass (M) of the companion can be 10โ18 M$``$.
Hutchings & Crampton (1981) have assumed $`f`$=0.02, an inclination of about 70 ($`\mathrm{sin}^3i`$=0.8) and M=10 M$``$ and derived M<sub>X</sub>=1.2 M$``$. For this reason in the literature it has generally been assumed that the compact object in LS I +61$`{}_{}{}^{}303`$ is a neutron star. Only Punsly (1999) discussed the possibility that the compact object in LS I +61$`{}_{}{}^{}303`$ could be a black hole and he presented a model for the high-energy emission based on it.
As discussed above and also in Massi (2004) the uncertainties in the parameters derived by optical measurements are rather large; changing the inclination to i=38 we already get M<sub>X</sub>=2.5 M$``$. If we assume a value M=18 M$``$ we obtain M<sub>X</sub>=3.4 M$``$. Recently Casares et al (2004) have determined an upper limit of $`f`$=0.027 which would correpond to M<sub>X</sub>=3.8 M$``$ (Massi et al. 2004b). In conclusion: Accounting for the uncertainties in inclination, the mass of the companion and the mass function it cannot be ruled out that the compact object in LS I +61303 is a black hole.
## 4 X-RAY AND RADIO OBSERVATIONS
An X-ray binary system is called transient, if at least one outburst occurs with a flux variation of more than 2-3 orders of magnitude greater than the normal flux (McClintock & Reimmllard 2004). This outburst, which may last for days to months, is directly related to a variation of the accretion disk and therefore is distinct from the outburst of Type I discussed in section 3.1 which lasts a few seconds.
Generally Microquasars have been discovered through high resolution radio observations immediately performed during new transients. However, LS 5039 was discovered (Paredes et al. 2000) on the basis of a cross-identification in catalogs (in optical, radio, X- and Gamma-rays) without any X-ray outburst calling attention to it. Nowadays one observes changes in the X-ray โstatesโ (Fig. 10). These states are defined by spectral (see below) and timing characteristics (van der Klis 2004): As soon as an X-ray binary is discovered in a โLow/Hard stateโ, it immediately is observed at radio wavelengths with high-resolution techniques.
### 4.1 High/Soft State and Multicolor Disk
X-ray binaries with a neutron star as compact object may have spectra that are completely different depending on whether the magnetic field of the neutron star is strong or weak.
The form of the spectrum of a binary X-ray pulsar, with a surface magnetic field $`>10^{12}`$G is a flat hard power-law function with a sharp cut-off above a few tens of keV (Tanaka 1997; White et al. 1996).
The spectrum of an X-ray binary with a weakly magnetized neutron star is typically formed by the properties of the accretion disk and the neutron star envelope. The neutron star envelope contributes to the harder part of the spectrum and has a temperature of $``$ 2.5 keV. Mitsuda and collaborators (1984), assuming an optically thick disk, where the energy generated by viscosity is locally dissipated in blackbody radiation, have represented the disk spectrum as a superposition of spectra with temperatures varying from a low value T<sub>out</sub> at the outer edge to a maximum T<sub>in</sub> at the inner edge (i.e. at the the inner radius $`R_{in}`$) of the disk. This is the reason why the disk is generally called a multi-temperature or multi-color disk. By means of this model T<sub>in</sub> and $`R_{in}`$ can be determined through the softer part of the observed spectrum ($`f(E)`$), which is represented by:
$$f(E)=\frac{8\pi R_{in}^2\mathrm{cos}i}{3D^2}_{T_{\mathrm{out}}}^{T_{\mathrm{in}}}(\frac{T}{T_{\mathrm{in}}})^{11/3}B(E,T)\frac{dT}{T_{\mathrm{in}}}$$
(7)
here $`i`$ is the inclination angle of the disk, $`D`$ is the distance and $`B(E,T)`$ is the Planck function (Shakura & Sunyaev 1973; Mitsuda et al. 1984; van Paradijs & McClintock 1996; Tanaka 1997).
X-ray binaries known to contain a black hole, proved by measurements of the mass function resulting in a mass $``$ 3 M, have spectra with a soft component accompanied by a hard power-law tail (Tanaka 1997). The soft component is described by the multicolor blackbody spectrum given above and therefore it is associated with the accretion disk around the black hole. This X-ray state is defined: High/Soft. In this respect it is quite interesting to compare the different values for T<sub>in</sub> and $`R_{in}`$ derived in the two cases of neutron stars and black holes (Tanaka 1997). In Fig. 11 the values of R$`{}_{in}{}^{}cos^{1/2}i`$ obtained from the fits for accretion discs around black holes and neutron stars are collected: The projected inner radius R<sub>in</sub> of accretion disks around neutron stars always results in values of $``$10km, while the values for black hole binaries all are larger by a factor of 3-4 than those for neutron stars. This shows that these compact objects are indeed more massive than 3$`M`$ as expected following the relationship $`R_{in}M_x`$ (Eq. 2).
The temperature T<sub>in</sub> for disks around black holes is always found to be less than $``$1 keV, significantly lower than that for disks around neutron stars with similar luminosities. Also this difference is understood in terms of the difference in the mass M<sub>X</sub> of the compact object: $`T(1/M_x)^{1/4}`$ keV (Eq. 3).
No blackbody component is present in the X-ray spectra of black hole X-ray binaries, which is consistent with the absence of a solid surface in a black hole. The second spectral component in black holes in the High/Soft state is a weak power-law with spectral index $`\mathrm{\Gamma }`$, defined by the photon flux $``$E (Fig. 13). A photon index of 2.0-2.5 has been determined by Tanaka (1997) for a sample of 5 black holes. Esin and collaborators quote a range from 2.2 to 2.7. The recent review by McClintock and Remillard (2004) gives a photon index ranging from 2.1 to 4.8 for 10 black holes.
The power law component in the X-ray spectra of accreting black holes in their High/Soft state is generally interpreted as the result of inverse Compton up-scattering of low-energy disc photons by electrons with a power-law or at least a hybrid distribution (consisting of both thermal and non-thermal electrons) that can be located in coronal regions (possibly flaring) above the disc (Coppi 2000; Zdziarski et al. 2001).
Evidence for the existence of an accretion disc corona comes from systems seen almost edge on: The strong central X-ray source (i.e. the inner disc) remains hidden behind the disc rim, but X-rays are still seen. The source of emission must be quite extended because the eclipse by the companion star is only partial (White et al. 1996). The origin of accretion disc coronae is described by buoyancy of magnetic fields amplified in the disk (see Miller & Stone 2000 and references therein).
### 4.2 The Disk-Jet Connection
Both X-ray binary systems containing weakly-magnetized neutron stars and those containing black holes change their spectral shapes (Tanaka 1997). The most drastic change is a transition from a spectrum with the accretion disk component as that discussed in the previous section (Fig. 13), to a spectrum without it and showing a single power-law component (Fig. 13).
In the sample discussed by Tanaka (1997) the photon index ($`\mathrm{\Gamma }`$) in this state varies in the range from 1.4 to 1.7 for systems containing black holes and is equal to 1.8 for systems with neutron stars. Esin and collaborators (1998) quote the range from 1.4 to 1.9. McClintock and Remillard in their recent review (2004) give $`\mathrm{\Gamma }`$=1.5โ1.9 (excluding GRS 1915+105) for 9 black holes. In conclusion, the power law in this state is definitely less steep than in the High/Soft state. This spectral state, present also in systems with neutron stars, is defined in the literature as Low/Hard only for systems with black holes.
It has been established that when an X-ray binary system is radio-loud and in particular with a flat or inverted radio spectrum (i.e. spectral index $`\alpha 0`$ with flux density $`S\nu ^\alpha `$) then it is always in its Low/Hard state (Fender 2004 and references therein). Emissions in the radio band and at hard X-rays are related by: $`L_{radio}L_X^{0.7}`$ ( Corbel et al 2003; Gallo et al. 2003).
Figure 14 shows a multiband monitoring of GX 339-4. At the beginning of the observations both radio emission and hard-X ray emission (Hard/Low state) are present, whereas the emission in the softer X-ray band is quite weak. When the system switches to the High/Soft state, then both radio and hard X-ray emission become quenched.
Finally, as GX 339-4 switches again into its Low/Hard state, radio emission is again observed (Fender et al. 1999). From the plot it is clear that the radio and the hard X-ray fluxes are strongly anticorrelated with the soft X-rays.
As stated above the radio emission during a Low/Hard state has a spectrum which is flat or inverted.
As shown in the sketch of Fig. 15 one can imagine a continuous jet in adiabatic expansion (conical jet) as formed by contiguous cylinders of increasing radii and decreasing B<sub>i</sub>. With each cylinder a canonical synchrotron spectrum is associated with an optically thin part $`S\nu ^{(1p)/2}`$ (where $`p`$ is the electron energy index) and an optically thick part $`S\nu ^{2.5}`$, with the two parts of the spectrum joining around $`\nu _{peak}B^{(p+2)/(p+4)}`$ (van der Laan 1966; Dulk 1985). Hence, the composite spectrum in case of a prolongated emission will have an optically thin part (that of cylinder 1) and a thick part (cylinder 3) at the two opposite ends, but will also develop a central part with an intermediate or even flat slope (Torricelli et al. 1998; Massi 1999). A flat spectrum therefore reveals a continuous jet. That has observationally been proven by direct imaging a radio jet in Cyg X-1 during its Low/Hard state (Stirling et al. 2001).
With sufficient sensitivity in the range of a few tenths of keV in the Low/Hard state it has been possible to observe a second component due to the accretion disc and therefore to measure the inner disk radius. The $`R_{in}`$ resulted to be $`50r_s`$ (McClintock et al 2001b; McClintock & Remillard 2004). In the High/Soft state (sect. 4.1 and Fig. 11), the measured inner radius of the accretion disk corresponds to $`R_{in}3r_s`$. Either the inner disk has been removed or during the transition High/Soft to Low/Hard the inner part of the disk has made a transition to a cooler state, which makes it effectively invisible in X-rays. The transition from a soft thermal state to a hard power-law state therefore corresponds to a change in the disk structure, evolving from a state mainly characterized by the emission from the inner part of the disk to a state, in which this inner-most region has been strongly modified (Tanaka 1997; Belloni et al. 1997). Since the magneto-hydrodynamic theory of jet production (Sec. 2.2) assumes a large vertical magnetic field component, the inner-most region is expected now to be geometrically thick and therefore strongly different from the geometrically thin (optically thick) case of the High/Soft state. However, how in detail is the space within the inner radius filled? What is the origin of the power law emission in the Low/Hard state and why is it so well correlated to the radio emission? This all is still matter of debate (Fig. 16) (see the reviews by McClintock & Remillard 2003 and Fender 2004).
Together with the two extreme states High/Soft (radio quiet) and Low/Hard (continuous ejection) there are two intermediate (concerning their hardness) states: the Very High State (VHS) and the Intermediate State (IS). Multiple recurrent oscillations in X-rays in the source GRS 1915+105 (Belloni et al. 2000) are due to different VHS-like states called A,B and C, where ejections- emitting in the radio band - occur in the hardest (C) state (reaching a Low/Hard state with $`\mathrm{\Gamma }1.8`$). Synchronized variations of the inner radius have been observed to occur during this oscillations (Belloni et al. 1997,1997b). Figure 17 shows the spectral change corresponding to a variation of $`R_{in}`$ from 20 km (value compatible with the last stable orbit around a rotating black hole) to more than 300 km.
During similar episodes of cyclic variations of the inner disk in GRS 1915+105, Mirabel and collaborators (1998) could follow the onset of a flare - first at infrared wavelenghts and then at radio wavelengths - with a delay consistent with synchrotron radiation from expanding magnetized clouds of relativistic particles (Fig.18). The straightforward interpretation was that during the disappearance of the inner disk a relativistic plasma cloud was expelled.
The mass of the ejected cloud has been estimated to $`10^{19}`$g (Mirabel 1998), while the matter which disappeared from the inner disk in one dip of similar time length (Belloni 1997), has been estimated to $`10^{21}`$g. This fact could imply that only a very small fraction of the mass is ejected, whereas is not clear which fraction fell indeed onto the compact object (Mirabel and Rodriguez 1999).
These small oscillations are now well established, they have recurrence time-scales of tens of minutes and have been observed in X-ray (interpreted as possible draining and refilling of the inner disk) and at infrared, millimeter and radio wavelengths (interpreted as repeated ejection events) (Fender et al. 1999: Eikenberry et al. 1998; Fender et al. 2002). Finally, Fender and coauthors (1999: 2002) have shown that these low ($``$ 40 mJy in radio) amplitude oscillations can happen during the decrease of flux in major flare events.
In conclusion:
1. A Soft state, characterized by disk emission (at temperatures $``$ 1 keV contributing mostly at 1-2.5 kev) and a power law component steeper than $`\mathrm{\Gamma }2`$ implies a radio-quiet X-ray binary system.
2. Quasi-periodic oscillations with time-scales of several minutes in X-ray and in the infrared or radio band are a signature of episodic disk-removal/variations and plasma bubble ejection. In case of such isolated small ejections, one can follow the adiabatic expansion of the cloud and monitor radiation becoming optically thin at progressively lower frequencies.
3. When the X-ray binary system is persistently emitting a radio jet the X-ray spectrum has a power law component with $`\mathrm{\Gamma }1.6`$. This X-ray state is called Low/Hard. The superposition (Fig. 15) of spectra of different contiguous jet regions with different self-absorption cutoffs result in a composite flat spectrum (i.e. S$`\nu ^\alpha `$, with $`\alpha 0`$) through and beyound radio wavelengths. Emissions in the radio band and at hard X-rays are related by: $`L_{radio}L_X^{0.7}`$
### 4.3 LS I +61303: Soft and Hard States
There are three X-ray observations of LS I +61303: with ROSAT (Taylor et al. 1996), ASCA (Leahy et al. 1997) and RXTE (Harrison et al. 2000). The ROSAT observation was performed over a total orbital cycle in the energy range from 0.07 to 2.48 keV. A single component fit was made, either a black body or a power-law and average fitted results were presented: A temperature of 0.26 keV and a power law index of $`\mathrm{\Gamma }`$=2.
The hardness-ratio is calculated all around the orbit and Taylor et al. (1996) notice the hardening of the X-ray emission during the onset of the second radio outburst (Fig. 19).
Greiner and Rau (2001) calculated the photon index around the orbit based on RXTE data (measured range 2.3โ25 keV) (Fig. 20). It is quite remarkable that the fit (only a single component, i.e. a power law) gives $`\mathrm{\Gamma }=`$2.0-2.4 all around the orbit (i.e. a High/Soft state) except for one point - simultaneous with the onset of a radio outburst -, where the photon index is $`\mathrm{\Gamma }`$= 1.6 (i.e. Hard state). This seems to be the typical case of the theory presented in the previous section for a Low/Hard state. The X-ray peak occured at phase $``$0.48 whereas the radio outburst came almost 6 days later (Fig. 20). However, in the light of developments in the theory of the disk-jet connection (reviewed in the previous sections) the spectral switch to the hard state is relevant for the ejection; this switch must precede the onset of the radio outburst and RXTE data actually show this for LS I +61303 (Massi 2004).
If a source remains in the Low/Hard state, it is radio-loud and the spectrum is flat. How does the spectrum of LS I +61303 behave taking into account that the Low/Hard state seems to be of a quite short duration ? Indeed, a flat spectrum has been measured by Taylor et al. (1996): in Fig. 19-middle the radio spectral index is equal to zero before the radio peak. When the ejection phase is terminated the figure very nicely shows that the flat spectrum evolves into an optically thin one (i.e. $`\alpha `$ changes from 0, due to the composite spectrum, to -0.3 related to the optically thin part of the โlastโemitted bubble or cylinder as in Fig. 15). As shown by Paredes et al. (1991) the flat spectrum in LS I +61303 can be reproduced by a model of an adiabatically expanding cloud of synchrotron-emitting relativistic electrons only if a continuos ejection of particles (lasting two days) is taken into account as well as adiabatic expansion losses.
Leahy et al. (1997) reported two ASCA observations, where the photon index $`\mathrm{\Gamma }`$ was 1.63-1.78 at orbital phase $`\varphi `$=0.2, which is the periastron passage (where an ejection is predicted), and $`\mathrm{\Gamma }`$=1.75-1.90 at orbital phase 0.42 coincident with the onset of a radio outburst (as shown in Fig. 21). In conclusions, both ASCA values give a photon index consistent with a Low/Hard state during predicted/observed ejections.
Harrison and collaborators (2000) have performed a periodicity analysis of the two ASCA observations. The result is a clear periodicity (Fig. 22) in the ASCA pointing related to the onset of a radio outburst while no periodicities are found in the other more sparcely sampled pointing. The X-ray oscillations occured at phase 0.42 before the radio peak, which (Fig. 21) occured at phase 0.5-0.6. The photon indexes given above reflect an average of all the data; therefore we cannot check, if the Low/Hard state is stable or if there is a continuos switching between a sort of Very High States (i.e. A,B,C) like for GRS 1915+105 (Belloni et al. 2000) or if the Low/Hard state is indeed reached only in the hardest interval. Such oscillations are also present in the radio band. Peracaula and collaborators (1997) have performed a period-analysis for three radio observations: two in a decreasing phase of large outbursts and one at a high, but quiescent flux level. While in the last data set there was no evidence for a periodicity, on the contrary, a period of P = 84 minutes and significant power also at harmonics of P/2 (i.e. $`40`$ minutes again) and 2P have been found in the two data sets related to the decay of radio outbursts (Fig. 23).
In conclusion:
1. There is no estimate of any inner radius for the accretion disk in LS I +61303 . The multicolor disk mostly emits below 2.5 keV and this range is excluded by the RXTE analysis.
2. The available X-ray observations for LS I +61303 reveal transitions to Hard/Low states at the onset of radio outbursts as is expected in the context of the disk-jet connection (Massi 2004).
3. A prolongated ejection of particles generates a flat spectrum in the late portion of the rise in flux.
4. X-ray oscillations and radio-oscillations are present at the early rise phase and during the decay of radio outbursts (Peracaula et al. 1997: Harrison et al. 2000)..
### 4.4 The Periodical Radio Outbursts of LS I +61303
The greatest peculiarity of LS I +61303 is its periodic radio outburst activity with P=26.496 days (Gregory 2002)
(see in Fig.24 Left-Top a typical radio light curve). The second peculiarity of LS I +61303 is that the amplitude of each outburst is not randomly varying, but itself periodic with a periodicity of 4.6 years (Fig. 24 Left-Bottom) (Gregory 1999, 2002).
The orbital phase $`\varphi `$ at which these outbursts occur is modulated (Gregory et al. 1999) and varies within the interval 0.5โ0.8 (Fig. 24 Right and Fig. 25)(Paredes et al. 1990).
The theory of accretion (Eq. 6) predicts maximum accretion where the density is highest. The maximum density is obviously at the periastron passage, because the density of the wind there is the largest and in addition there probably occurs direct accretion from the the star (Roche Lobe overflow). As shown in Fig. 24 Right the $`\varphi `$ at periastron passage is 0.2 and therefore practically is opposite to the orbital region, where the radio outbursts occur. Therefore one of the fundamental questions concerning the periodic radio outbursts of LS I +61303 has been for years: Why are the radio outbursts shifted with respect to the periastron passage ?
## 5 THEORY OF THE ACCRETION: THE TWO PEAK ACCRETION MODEL
Equation 6 gives only one accretion peak for variable density and constant velocity $`v_{rel}`$. However, the orbit of LS I +61303 is quite eccentric and therefore with a strong variation of the velocity along the orbit. Taylor et al. (1992) and Martรญ & Paredes (1995) have shown that in this case the accretion rate $`\dot{M}\frac{\rho _{\mathrm{wind}}}{v_{\mathrm{rel}}^3}`$ develops indeed two peaks: The first peak corresponds to the periastron passage (highest density), while the second peak occurs when the drop in the relative velocity $`v_{\mathrm{rel}}`$ compensates the decrease in density (because of the inverse cube dependence) (Fig. 26 Top Right).
This also implies that, while the first peak always occurs at periastron passage, the second peak may move to different points in the orbit, if variations in $`\rho _{\mathrm{wind}}`$ or $`v_{\mathrm{rel}}`$ occur. Figure 26 Right-Bottom shows how for increasing values of the wind-velocity the second peak shifts toward the first peak (Martรญ & Paredes 1995).
On the other hand variations in the mass loss of the Be star have been well established by H$`\alpha `$ emission line observations (Zamanov & Martรญ 2000). Gregory & Neish (2002) suggest a periodic outward moving density enhancement (i.e., shell ejection) in the Be star wind. The variation of H$`\alpha `$ emission line (Zamanov & Martรญ 2000) is periodic with a comparable scale (1584 d) as the radio modulation and it is in phase with the onset of the outbursts (Fig. 27). The orbital shift in the phase of the radio outbursts is therefore related to variations of the wind parameters.
Finally, Martรญ & Paredes (1995) have shown that during both peaks the accretion rate is above the Eddington limit and therefore one expects that matter is ejected twice within the 26.496 days interval.
In conclusion, radio outbursts displaced from periastron passage correspond to the second peak of the two-accretion/ejection peaks. The remaining problem therefore is: Why is the first outburst at periastron passage in the radio band missing ?
## 6 GAMMA-RAY OBSERVATIONS
### 6.1 EGRET Sources
The Third EGRET Catalog contains about 170 not yet identified high energy Gamma-ray sources (E $`>`$ 100 MeV) ((Fig. 28). The discovery of the coincidence of the microquasar LS5039 (Fig. 29) with an unidentified EGRET source by Paredes and collaborators (2000) has opened the possibility that other EGRET sources could be microquasars as well. Gamma-rays can be produced by external Compton scattering of stellar UV photons of the massive companion by the relativistic electrons of the jet. LS5039 is a persistent radio emitting source and the Gamma ray flux, with all uncertainties reflected by the poor sampling, still reflects this persistence (Fig. 29)(Paredes et al 2000).
Therefore, for a periodic source like LS I +61303 periodic Gamma-ray emission would be expected.
### 6.2 The Variable Gamma-Ray Source LS I +61303
Gregory and Taylor (1978) reported the discovery of a radio source (later on associated with LS I +61303 ) within the 1$`\sigma `$ error circle of the COS B $`\gamma `$-ray source 2CG 135+01. This association however remained controversial because of the presence of the quasar QSO 0241+622 within the relatively large COS B error-box. In the CGRO mission (1991 May- 1995 October) the source, given there as 2EG J0241+6119, was detected by EGRET with a significance of 17 $`\sigma `$, with a time averaged photon flux of $`9.2\pm 0.6\times 10^7`$ cm<sup>-2</sup> s<sup>-1</sup> for energies $``$100 MeV. This flux reported by Kniffen et al. (1997) is slightly different from those quoted in the EGRET Catalogs because of the addition of data after the 1993 September cutoff date for the catalog. The position of this gamma-ray source is $`l`$ = 135.58, $`b`$ = 1.13. As shown in Fig. 30 the contour position obtained with additional data is about 11ยด from the 2EG catalog position and about 40ยด from the old 2CG catalog position. The radius of the 95$`\%`$ confidence error contour is about 13ยด, ruling out the possible identification with QSO 0241+622 at l = 135.7, b = 2.2, which is 64ยด away. The position is only 8ยด distant from LS I +61303 (Kniffen et al. 1997). In 1998 Tavani and collaborators established the possibility of variability of 2CG 135+01 on timescales of days (Tavani et al. 1998). Massi (2004) examined the EGRET data as a function of the orbital phase and noticed the clustering of high flux values around periastron passage. Figure 31 shows (Massi et al. 2004) the follow-up of the EGRET gamma-ray emission along one full orbit. At epoch JD 2 450 334 (i.e. circles in the plot, with empty circles indicating upper limits) the orbit has been well sampled at all phases: A clear peak is centered at periastron passage 0.2 and 1.2. At a previous epoch (JD 2 449 045; triangles in the plot) the sampling is incomplete, but the data show an increase near periastron passage at $`\varphi `$0.3, and a peak at $`\varphi `$0.5. The 3 squares refer to a third epoch (JD 2 449 471).
In conclusion, the gamma-ray data strongly support the ejection at periastron passage predicted by the two-peak model (Massi 2004b). During the first ejection (because of the proximity of the Be star) stellar photons are upscattered by the inverse Compton process by the relativistic electrons of the jet (Bosch-Ramon & Paredes 2004). The inverse Compton losses are so severe that no electrons survive: radio outbursts indeed never have been observed at periastron passage in more than 20 years of radio flux measurements (Gregory 2002). At the second accretion peak the compact object is far enough away from the Be- star, so that energetic losses are smaller and electrons can propagate out of the orbital plane. At this point the gamma-ray peak at $`\varphi 0.5`$ is very interesting. It could originate from a second ejection which occurred still enough close to the Be-star. In fact, while the first ejection is always at periastron passage, the second ejection occurs at a varying point in the orbital phase interval 0.4โ0.8.
## 7 RADIO INTERFEROMETRY: IMAGING AT HIGH RESOLUTION
Nowadays it is possible to obtain images of jets at infrared wavelengths and in X-rays (Sams et al. 1996; Corbel et al. 2002). However, these jets at tenths of arcseconds are not related to the emitting regions close to the engine (with quite short lifetimes because of their large adiabatic/synchrotron losses) but require a re-acceleration mechanism. The study of the jet closest as possible to the โengineโ at a spatial resolution up to milliarcseconds (mas) is possible at radio wavelengths thanks to Very Long Baseline Radio Interferometry (VLBI) (Appendix).
### 7.1 The Jet Velocity
For symmetric ejection of two jets at a velocity $`\beta `$ (i.e. expressed as a fraction of $`c`$), the two (approaching and receding) jets move with an apparent velocity of $`\beta _{\mathrm{a},\mathrm{r}}`$ (Mirabel & Rodrรญguez 1994, Fender 2004):
$$\beta _{\mathrm{a},\mathrm{r}}=\frac{\beta \mathrm{sin}\theta }{1\beta \mathrm{cos}\theta },$$
(8)
$`\theta `$ is the angle between the direction of motion of the ejecta and the line of sight to the observer. Depending on the angle, for a jet with $`\beta 0.7`$ the apparent velocity $`\beta _\mathrm{a}`$ of the approaching jet can become greater than 1 (superluminal effect, see Fig. 32).
In order to show how the apparent velocity of the jet is derived, let us assume $`\theta `$=90. In this case the proper motion, $`\mu `$, of the jet on the sky plane is:
$$\mu =\frac{170\beta }{D},$$
(9)
where the distance $`D`$ is in kpc and $`\mu `$ is expressed in milliarcseconds per day (mas/day). The range of $`\beta `$ is about 0.15โ0.99 and the range of the distance is about 1-12.5 kpc. From the two extreme conditions, i.e $`\frac{\beta _{max}}{D_{min}}`$ and $`\frac{\beta _{min}}{D_{max}}`$, the proper motion ranges from 2 mas/day to 170 mas/day. In order to estimate $`\beta `$ from multi-epoch observations one has to select the proper radio network. One must take into account that at $`\lambda `$=6cm the VLBI provides a resolution of $`1`$ mas, MERLIN one of $``$ 50 mas and the VLA in the largest configuration one of $``$ 100 mas. Therefore high proper motions are best studied with MERLIN and the VLA.
Beside multi-epoch observations even with only one observation it is possible to recover the quantity $`\beta \mathrm{cos}\theta `$, if $`\theta `$ is significantly less than 90. In fact, the observed flux densities $`S_{\mathrm{a},\mathrm{r}}`$ from the approaching and receding jets,
$$S_{\mathrm{a},\mathrm{r}}=S\delta _{\mathrm{a},\mathrm{r}}^{k\alpha },$$
(10)
(where $`\alpha `$ is the spectral index of the emission $`S\nu ^\alpha `$ and $`k`$ is 2 for a continuous jet and 3 for discrete condensations) are governed by the Doppler factor,
$$\delta _{\mathrm{a},\mathrm{r}}=\frac{1}{\gamma (1\beta \mathrm{cos}\theta )},$$
(11)
(where $`\gamma =(1\beta ^2)^{1/2}`$ is the Lorentz factor) and therefore (Mirabel & Rodrรญguez 1994) one can determine the quantity $`\beta \mathrm{cos}\theta `$ by means of the ratio between the flux densities from the approaching and receding jet:
$$\frac{S_\mathrm{a}}{S_\mathrm{r}}=\left(\frac{1+\beta \mathrm{cos}\theta }{1\beta \mathrm{cos}\theta }\right)^{k\alpha }.$$
(12)
Let us assume an ejection nearly aligned to the line of sight with $`\theta `$=0.5 and with the other parameters: $`\alpha =0.5`$, k=2 and $`\beta 0.6`$. Using equation 10 one determines $`\delta _\mathrm{a}^{k\alpha }6`$ and $`\delta _\mathrm{r}^{k\alpha }=0.1`$. As a result, the counter-jet can be rather faint, and if $`S_\mathrm{r}`$ results to be lower than the noise limit of the radio image the counter-jet will completely disappear. In this case the image will show a one-sided jet (the approaching one) and only a lower limit for $`\beta \mathrm{cos}\theta `$ can be estimated using the noise limit of the image (Massi et al. 2001).
A constant ejection angle $`\theta `$ implies a constant ratio between the flux densities from the approaching and receding jet during the epochs. An obvious variation of this ratio is interpreted as a variation of the ejection angle $`\theta `$ , explained as jet precession.
### 7.2 The Precessing Jet of LS I +61303
The first VLBI observation resolving the source (Massi et al. 1993) in Fig. 33 reveals a complex morphology (Fig.33 Bottom): A structure at PA$`30^{}`$ formed by two components separated 0.9 mas (about 2 AU at the distance of 2 kpc) is surrounded by an envelope clearly rotated with respect to it. This envelope could be an older expanding jet, previously ejected at another angle (because of precession).
Taylor and collaborators (2000) performing VLBI observations in combination with the HALCA orbiting antenna of the VSOP mission, imaged a structure reminiscent of the precessing radio jet seen in SS433 (Fig. 34). On the other hand at a lower resolution with a scale up to tens of AU (with the EVN), Massi and collaborators (2001) obtained an image that for the first time showed an elongation in a clear direction without any ambiguity (see Fig. 33). The most interesting aspect of the EVN map is that, for the first time, we detected asymmetric emission in the southeast direction. Using the noise level ($`\sigma `$) of the map and the peak value of the approaching component ($`S_\mathrm{a}^{\mathrm{peak}}`$) for k=2 and $`\alpha `$=-0.5 we determine $`\beta \mathrm{cos}\theta >0.6`$. This would correspond to the two limits of $`\theta <53^{}`$ for $`\beta 1`$ and $`\beta 0.6`$ for $`\theta =0^{}`$. A value for the velocity well within the range $`0.1c`$ to $`0.9c`$ found for other Microquasars (Mirabel & Rodrรญguez 1999i).
Two observations at still lower resolution have been performed with MERLIN (see Table 1). The first MERLIN image shows a double S-shaped jet extending to about 200 AU on both sides of a central source.
The morphology of the MERLIN image (Fig. 35a) has a bent, S-like structure. In the small box in Fig. 35a we show the simulated radio emission from the Hjellming & Johnston 1988 model of the precessing jet of SS 433 (rotated here for comparison purposes). The similarity between the MERLIN image of LS I +61303 and the precessing model for SS 433 strongly suggests a precession of the jet of LS I +61303. The precession becomes evident in the second MERLIN image, shown in Fig. 35b, where a new feature oriented to Northeast at a position angle (PA) of 67 is present. The Northwest-Southeast jet of Fig. 35a has PA=124. Therefore, a quite large rotation has occurred in only 24 hours. This fast precession causes a deformation of the morphology during the second observation, and the one-sided jet appears bent in Fig 35b. Only 3$`\sigma `$ features can be associated with the double jet of the day before. The feature at 3$`\sigma `$ to the East is well compatible with a displacement of $`0.6c\times 24`$ hours.
The appearance of successive ejections of a precessing jet with ballistic motion of each ejection is, as shown in Fig. 36, a curved path that, depending on the modality of the expansion and therefore on the adiabatic losses, seems to be a โtwin-corkscrewโ or a simply S-shaped pattern (Hjellming & Johnston 1988; Crocker et al 2002). The last one seems to be the case of LS I +61303. Can we trace any ballistic motion of any jet component ?
We have splitted the data base of each MERLIN run into two subsets. The first map (Fig. 37-a) represents the first four hours of the first run. It shows an ejection โAโ already quite displaced from the core. In Fig.37-b, there is present a new ejection โBโ at another PA. The combination of these two maps together produce as a consequence a โbentโ jet, that is the southern jet of Fig. 35a. The two counter-jets for โAโ and โBโ are (Figs. 37-a and 37-b) too weak to be detected, and they become visible only in the more sensitive image of Fig. 35a, where all the jet and counter jets for A and B form together the โS-shapedโ jet. The โBโ component is still detectable after 9 hours in the third image (Fig.37-c). Its motion is ballistic: the PA is still the same (almost 90) as in Fig.37-b. A new component โCโ is present at another PA (Fig. 37-d) 6 hours later, little rotation of the PA is compatible with $`\mathrm{\Delta }`$ PA$`{}_{(\mathrm{B}\mathrm{C})}{}^{}/3`$ of the previous image. The Northern elongation in the higher sensitivity map of Fig. 35b therefore is the result of a set of little ejections of a rotating stream. The noise level in this image is still lower than that in Fig. 35a, nevertheless the counter-jet is not visible. This implies a decreased $`\theta `$ due to precession. In the case of the MERLIN image of April 22 we derive $`\beta \mathrm{cos}\theta =0.12`$, which for $`\beta =0.6`$ leads to an ejection angle of $`\theta =78^{}`$. This is an average of the ejection angles $`\theta _\mathrm{A}`$ and $`\theta _\mathrm{B}`$ of features A and B in Figs. 37a and 37b. A direct estimate of these angles is prevented because of the lack of the receding jets. Using the r.m.s. noise we derive $`\theta _\mathrm{A}<90^{}`$, $`\theta _\mathrm{B}<80^{}`$ and for the C ejection in Fig.37c, $`\theta _\mathrm{C}<68^{}`$.
Therefore, the angle between the jet and the line of sight, $`\theta `$, has decreased by more than 10 in 24 hours. It is this much narrower alignment of the jet with the line of sight, that causes the counter-jet to get further Doppler de-boosted with respect to the first image and lets it disappear below the sensitivity limit of the image.
## 8 CONCLUSIONS
The conclusions of this review of the astronomical methods used for the investigation of Microquasars, with an examplary view on the source LS I +61303, are:
1. It is still an open issue, whether the compact object in this system is a neutron star or a black hole. In fact, taking into account the uncertainty in inclination, mass of the companion and the mass function, the existence of a black hole cannot be ruled out.
2. The observational results from X-rays for LS I +61303 are consistent with transitions between X-ray spectral states typical for a variable accretion disk. These transitions are properly related to the onset of strong radio emission, as expected if the jet is โfedโ by the disk. Quasi-periodic oscillations at soft X-rays and radio wavelengths are present in a strong analogy with those observed in GRS 1915+105. They occur at the onset and decay of large radio outbursts. If confirmed, this fact might indicate that at the beginning the matter is ejected from the disk in the form of discrete condensations (i.e. blob-like), then follows a steady state where the matter-supply occurs at a higher/continuous rate (i.e. a continuous jet) and finally the ejection again ends up in a blob-like form. (Hujerat & Blandford 2003).
3. At a scale of hundreds of AU the radio jet quite strongly changes its morphology in short intervals ($`<`$ 24 hours), evolving from an initial double-sided jet into an one-sided jet. This variation corresponds to a reduction of more than 10 in the angle between the jet and the line of sight. This new alignment severely Doppler de-boosts the counter-jet. Further observational evidence for a precessing jet is recognizable even at AU scales.
4. The same population of relativistic electrons emitting radio-synchrotron radiation upscatters - by inverse Compton processes - ultraviolet stellar photons and produces Gamma-ray emission. Ejections near the periastron passage produce Gamma-ray flares but no radio flares, implying severe Compton losses.
We conclude that, because precession and variable Doppler boosting are the causes of the rapid changes in the radio-morphology, precession and variable Doppler boosting are likely to produce Gamma-ray variations at short time scales. The amplification due to the Doppler factor for Compton scattering of stellar photons by the relativistic electrons of the jet is $`\delta ^{32\alpha }`$ (where $`\alpha <0`$), and therefore is higher than that for synchrotron emission, i.e. $`\delta ^{2\alpha }`$ (Georganopoulos et al. 2001; Kaufman Bernadรณ et al. 2002). LS I +61303 becomes therefore an ideal laboratory to test the recently proposed model for Microblazars with INTEGRAL and MERLIN observations now and by AGILE and GLAST in the future.
## 9 SUMMARY
Because of their accretion disk super-massive black holes, with 10<sup>6</sup>-10<sup>9</sup> solar masses, in the heart of galaxies are the cause for the most energetic sources of emission in our Universe. The centers of such galaxies are called Active Galactic Nuclei (AGN). Some AGN, like the quasars, produce โjetsโ of subatomic particles with speeds approaching that of light. A microquasar - as its name suggests - is a miniature version of a quasar: A disc of a few thousand kilometers radius surrounds a black hole of a few solar masses and two relativistic jets are propelled out of the disk by the same process occurring in a Quasar. The Microquasars therefore can serve as a convenient โlaboratoryโ for studying the physics of jets. The Microquasars are objects very much closer to us than Quasars and the study of the evolution of relativistic jets can be done in a few days only, whereas on the contrary for far distant Quasars observations of many years apart are necessary to obtain appreciable proper motions of the radio jets. Moreover, concerning the intrinsic variability of Microquasars, these โsmallโ objects change more quickly than Quasars: Considering as a characteristic time scale for variations $`\tau R_{Schwarzschild}/cMass`$, phenomena of timescales of minutes connected with a Microquasar of 10 solar masses would take years in a AGN of $`10^7`$ solar masses. Such an enormous difference is the main reason why Microquasars got such a great interest and growth in Astrophysics in the last decade.
The Microquasars belong to the class of the X-ray binaries, where a compact object (black hole or neutron star) accretes from a normal companion star. Such systems are well known since the 1960s. The X-ray emission originates from the very hot accretion disk surrounding the compact object. However, it took a long time to discover that some of these systems also have relativistic radio jets like Quasars. For several years, after its discovery in 1979, SS 433 with its spectacular jets was thought to be a unique exotic case, a mere curiosity in our galaxy. Since the beginning of the 1990, after the discovery of other possible candidates of the same nature, several groups (including the author of this review) have begun a systematic research on X-ray binaries with radio jets.
Here I review the astronomical methods used from Gamma-rays over X-rays and optical to radio wavelengths for the investigation of these objects. The description of the methods is accompanied by directly applying them to the system LS I +61303 , one of the most enigmatic objects in our galaxy, because it is associated with a variable high-energetic Gamma-ray emission of unknown origin.
The nature of the accretor - a neutron star or a black hole - is determined by optical measurements of the Doppler shift of spectral lines of the normal star orbiting around the invisible companion. Observations at X-rays probably are the most spectacular ones, in respect to the progress in the knowledge of the accretion disk and the disk-jet connection. Fitting the X-ray spectra information of the size of the last stable orbit around the compact object can be derived and ejections of matter into relativistic jets can be related to variations of the disk. The results from X-ray observations for LS I +61303 are consistent with transitions between spectral states typical for a variable accretion disk. These transitions are properly related to the onset of strong radio emission as expected for a jet โfedโ by the disk.
Onset and decay of some large radio outbursts are modulated with quasi-periodic oscillations that correspond to repetitive ejections of discrete condensations (i.e. blob-like). Continuous ejections have a flat radio spectrum and Low/Hard X-ray state.
Speed and morphology of the ejections at high resolution are studied with radio interferometric techniques. The results of more than 10 years of VLBI/EVN and MERLIN observations of LS I +61303 are presented together with our discovery of the relativistic jet and its precession. Successive ejections are in ballistic motion; because of precession their projected path on the sky plane draw a bending jet.
The radio bursts occur around apastron passage, where the low velocity of the accretor enables it to capture more material of the wind from the companion star. However, no bursts are observed at periastron passage, where accretion theory predicts another super-accretion event. There the accretor is completely embedded in the densest part of the wind. I found that this great open question about LS I +61303 and the other enigma about the association of LS I +61303 with a variable Gamma-ray source are indeed not two separate questions, but on the contrary one is the answer to the other: We do not see a radio outburst at periastron passage, because we see a Gamma-ray outburst. With LS 5039 for the first time we identified a Microquasar with an high-energy (E$`>`$ 100 Mev) source. This fact opens the perspective that others of the more than one hundred still unidentified EGRET sources could belong to a new class of objects: Gamma-ray Microquasars.
In this review I show that the variable Gamma-ray emission of LS I +61303 is related to the orbit of the system, with peaks clustering where the companion star - a strong emitter of ultraviolet photons - is closest (at periastron). The suggested most probable explanation is that the ejected relativistic electrons are not able to emit synchrotron radiation at radio wavelengths, because at periastron passage they are embedded in such a strong UV-field of radiation that they loose completely their energy by inverse Compton process. During the second accretion peak, the compact object is much farther away from the companion star and inverse Compton losses are lower: The electrons can propagate out of the orbital plane and radio outbursts are observed.
## 10 ZUSAMMENFASSUNG
Supermassive Schwarze Lรถcher mit 10<sup>6</sup>-10<sup>9</sup> Sonnenmassen im Zentrum von Galaxien sind wegen ihrer Akkretionscheibe der Grund fรผr die stรคrksten Strahlungsquellen in unserem Universum. Die Zentren solcher Galaxien werden Aktive Galaktische Nuklei (AGN) genannt. Einige AGN wie z.B. Quasare produzieren โJetsโ von subatomaren Teilchen mit Geschwindigkeiten bis nahe an die Lichtgeschwindigkeit. Ein Mikroquasar ist, wie der Name schon sagt, die Miniatur eines Quasars: Eine Scheibe von einigen tausend Kilometern umgibt ein Schwarzes Loch von einigen Sonnenmassen und zwei relativistische Jets werden durch denselben Prozess wie bei Quasaren aus der Scheibe herausgeschleudert. Ein Mikroquasar kann deshalb als ein brauchbares โLaborโ zum Studium der Physik solcher Jets dienen. Die Mikroquasare sind Objekte, die wesentlich nรคher zu uns liegen als Quasare und die Untersuchung der Evolution von relativistischen Jets kann in nur ein paar Tagen geschehen,wohingegen man fรผr weit entfernte Quasare langjรคhrige Beobachtungen benรถtigt, um ausreichende Eigenbewegungen zu erhalten. Weiterhin, wenn man die intrinsische Variabilitรคt von Mikroquasaren betrachtet, รคndern sich diese โkleinenโ Objekte schneller als Quasare: Nimmt man als charakteristische Zeitskala fรผr Variationen $`\tau R_{Schwarzschild}/cMass`$, nehmen Phรคnomene mit einer Zeitskala von Minuten bei Mikroquasaren von 10 Sonnenmassen eine Zeit von Jahren bei einem Quasar von $`10^7`$ Sonnenmassen in Anspruch, wenn man es mit der Masse des Akkretors skaliert.Diese enorme Differenz ist der Hauptgrund, weshalb Mikroquasare in der letzten Dekade solch ein grosses Interesse und Wachstum in der Astrophysik auf sich gezogen haben.
Die Mikroquasare gehรถren zur Klasse der Rรถntgen-Doppelsterne, wo ein kompaktes Objekt ( Schwarzes Loch oder Neutronenstern ) von einem normalen Begleitstern einen Massenzuwachs erfรคhrt. Solche Systeme sind seit den 1960er-Jahren gut bekannt.Die Rรถntgenstrahlung stammt von der sehr heissen Akkretionsscheibe, die das kompakte Objekt umgibt. Allerdings dauerte es eine lange Zeit, bis man entdeckte, dass einige dieser Systeme ebenso wie Quasare relativistische Radio-Jets aussenden. Fรผr viele Jahre nach seiner Entdeckung 1979 galt SS 433 mit seinen spektakulรคren Jets als ein einzelner exotischer Fall, eine einzigartige Kuriositรคt in unserer Galaxie. Anfag der 1990 nach der Entdeckung von anderen mรถglischen Kandidaten derselben Art begannen einige Gruppen ( einschliesslich des Autors dieses Reviews) mit einer systematischen Forschungsarbeit an Rรถntgen-Doppelsternen mit Radio-Jets.
Ich gebe hier einen รberblick รผber die astronomischen Methoden, die im Bereich von Gamma-Strahlung รผber Rรถntgen-Strahlung und optischen bis hin zu Radio Wellenlรคngen zur Untersuchung dieser Objekte angewandt werden. Die Beschreibung dieser Methoden wird unmittelbar begleitet durch die Anwendung der Methoden auf das System LS I +61303 , einem der rรคtselhaftesten Objekte in unserer Galaxie,weil es mit einer variablen hochenergetischen Gamma- Strahlungsquelle unbekannten Ursprungs verbunden ist.
Die Natur des Akkretors - ein Neutronenstern oder eine Schwarzes Loch - wird durch optische Messungen anhand der Dopplerverschiebung von Spektrallinien des normalen Sterns, der sich um seinen unsichtbaren Begleiter bewegt, ermittelt. Beobachtungen im Rรถntgenbereich sind wahrscheinlich die spektakulรคrsten im Hinblick auf den Fortschritt bezรผglich der Kenntnisse รผber die Akkretionsscheibe und die Scheiben-Jet Verknรผpfung. Indem man die Rรถntgenspektren untersucht, erhรคlt man Informationen รผber die letzte stabile Bahn um das kompakte Objekt und ebenso kann man die Variation dieser Grรถsse mit dem Auswurf von Materie in die relativistischen Jets correlieren. Die Ergebnisse von Rรถntgenbeobachtugen von LS I +61303 sind konsistent mit รbergรคngen zwischen spektralen Zustรคnden im Rรถntgenbereich typisch fรผr eine verรคnderliche Akkretionsscheibe.Diese รbergรคnge sind passend verbunden mit einem Anstieg von starker Radiostrahlung, wie man es fรผr einen Jet, der von der Scheibe โgespeistโ wird, erwartet.
Anstieg und Abfall der starken Ausbrรผche sind moduliert mit quasi-periodischen Oszillationen, die wiederholten Auswรผrfen von diskreten Kondensationen entsprechen. Ununterbrochene Auswรผrfen haben Radiospektrum flach und Rรถntgenzustand โLow/Hardโ.
Die Geschwindigkeit und die Morphologie der Auswรผrfe wereden mithilfe der Radiointerferometrie-Technik untersucht. Die Ergebnisse รผber mehr als zehn Jahre von Beobachtungen mit VLBI/EVN und MERLIN von LS I +61303 werden hier zusammen mit unserer Entdeckung des relativistischen Jets und seiner Prรคzession dargestellt. Aufeinander folgende Auswรผrfe folgen ballistischer Bewegung; wegen der Prรคzession bildet ihr auf die Himmelsebene projezierter Weg einen gebogenen Jet.
Der Radioausbruch geschieht um den Apoastron-Durchgang herum, wobei die geringe Geschwindigkeit es dem Akkretor erlaubt, mehr Material vom Wind des Begleitstern einzufangen. Allerdings werden keine Ausbrรผche beim Periastron-Durchgang beobachtet, wo die Akkretions-Theorie ein weiteres super-akkretives Ereignis vorhersagt. Dort ist der Akkretor vollstรคndig vom dichtesten Teil des Windes umgeben. Ich fand heraus, dass diese grosse offene Frage รผber LS I +61303 und das andere Rรคtsel bezรผglich der Verbindung von LS I +61303 mit der verรคnderlichen Gamma-Strahlungsquelle in Wirklichkeit keine zwei getrennten Fragen sind, sondern im Gegenteil ist die eine die Antwort auf die andere: Wir beobachten keinen Radioausbruch beim Periastron- Durchgang, weil wir einen Gamma-Strahlungsausbruch sehen. Mit LS 5039 haben wir zum ersten Mal einen Mikroquasar mit einer hochenergetischen (E$`>`$ 100 Mev) Quelle identifiziert. Diese Tatsache erรถffnet die Perspektive, dass andere der mehr als hundert noch nicht identifizierten EGRET Quellen zu einer neuen Objektklasse gehรถren kรถnnen: Gamma-Strahlungs-Mikroquasare.
In diesem Review zeige ich, dass die verรคnderliche Gamma-Strahlung von LS I +61303 mit der Umlaufbahn des stellaren Systems verbunden ist, mit sich lรคufenden Spitzen dann, wenn der Begleitstern - ein starker Strahler von Ultraviolett-Photonen - am nรคchsten ist (beim Periastron-Durchgang). Die vorgeschlagene wahrscheinlichste Erklรคrung ist, dass die ausgeworfenen relativistischen Elektronen nicht in der Lage sind, Synchrotonstrahlung im Radiobereich auszusenden, weil sie beim Periastron-Durchgang in solch einem starken UV-Strahlungsfeld eingebettet sind, dass sie vollstandig ihre Energie wegen des inversen Compton Prozesses verlieren. Wรคhrend der zweiten Akkretions-Spitze ist das kompakte Objekt wesentlich weiter weg von seinem Begleitstern und so sind Compton-Verluste geringer: Die Elektronen kรถnnen aus der Bahnebene heraustreten und man beobachtet Radioausbrรผche.
## 11 APPENDIX: Theory of Very-Long-Baseline-Interferometry Data Analysis
The lack of phase information had prevented VLBI from being a true imaging technique until Rogers and his co-workers (1974) applied a phase closure relationship. The introduction of the closure phase concept marks the beginning of a new era in VLBI. Many authors developed methods, reviewed by Pearson and Readhead (1984), which explicitly or implicitly use this quantity. Massi (1989) showed how the methods explicitly using the closure phase can be unified in one equation. In an attempt to unify all methods together Massi & Comoretto (1990) found that all methods turn out to be particular cases of the method proposed by Schwab (1980) depending on a proper scheme of baseline weighing.
Using Schwabโs method, called self-calibration,a map of the radio source can be obtained by using an algorithm which includes fourier transform and CLEAN, following an iterative procedure first indicated by Readhead and Wilkinson (1978) and called Hybrid mapping (Fig. 38).
That this procedure can converge on wrong solutions has been pointed out by many authors in the past: Walker (1986) indicated the bias in the resultant data due to the use of a point source as starting model. Bรฅรฅth (1989) suggested the use of the original data set in each iteration of self calibration. Linfield (1986) analysed the role of the (u-v) coverage and lack of intermediate spacing. Generally, the full procedure to avoid false features is not clear and only experience with imaging helps the user to avoid them.
Massi & Aaron (1999) demonstrated that the problem is connected with the non-linear nature of self calibration which leaves initial wrong assumptions frozen in the final solution. We demonstrated that the general precondition to avoid false structures in the map is that the errors (or more precisely their cube) of the model should be smaller than the observed closure phases. This condition, generally satisfied for a standard earth based array, is violated if one telescope of the array is very displaced from the others, as it is for an array including a telescope mounted on a satellite. In this case one should avoid the use of a point like model as starting model. Moreover, one should at each iteration of self calibration, adopt the model derived by CLEAN directly on the original data and not on the corrected data biased by previous wrong solutions (Fig. 38).
Using self calibration it is assumed that the baseline based errors are negligible. In spite of the fact that these baseline errors are quite small their effect on the mapโs quality is rather serious. Tests were performed to determine at which level errors limit the obtainable dynamic range with the VLA (Perley 1986), with the VLBA (Briggs et al. 1994), with the Nobeyama Radioheliograph (Koshiishi et al., 1994). Massi and collaborators have performed such an analysis for the European VLBI Network (EVN). The result was that the instrumental polarization (D terms) of the telescopes of the network had an average value of 9 percent arriving at some telescopes at values of 20 percent. For comparison the values of VLBA telescopes were below 2 percent. The instrumental polarization was therefore the main reson for the lower performance of the EVN damaging the dynamic range of the images up a factor of 7 (Massi et al. 1991; 1996; 1997; 1997b; 1997c; 1997d; 1997e; 1998; Massi 1999b).
## 12 REFERENCES
Apparao, K.M.V. 2000, A&A, 356, 972
Bรฅรฅth, 1989, Very Long Baseline Interferometry. Techniques and Applications. ed. M. Felli and R. Spencer. Klewer Academic Publishers. NATO ASI Series C. Vol.283, pag.206
Belloni, T., Klein-Wolt, M., Mรฉndez, M., van der Klis, M. & van Paradijs, J. 2000, A&A, 355,27
Belloni, T., Mรฉndez,M., King, A.R., van der Klis, M. & van Paradijs, J. 1997a, ApJ., 479, L145
Belloni, T., Mรฉndez, M., King, A.R., van der Klis, M. & van Paradijs, J. 1997b, ApJ., 488, L109
Bondi, H, 1952, MNRAS, 112, 195
Bosch-Ramon, V. & Paredes, J. M. 2004, A&A, 425,1069
Briggs, D.S., Davis, R.J., Conway, J. E. & Walker, R.C. 1994, July 25, VLBA memo 697
Brocksopp, C., Fender, R. P., McCollough, M., Pooley, G. G., Rupen, M. P., Hjellming, R. M., de la Force, C. J., Spencer, R. E., Muxlow, T. W. B., Garrington, S. T. & Trushkin, S. 2002, MNRAS, 331, 765
Cadolle Bel, M., Goldwurm, A.,Rodriguez, J. et al. 2004. A&A, 426, 659
Casares, J., Ribas, I., Paredes, J. M., Martรญ, J. & C. Allende Prieto 2004, MNRAS, submitted
Charles, P.A., & Wagner, R.M. 1996, Sky & telescope, May, 38.
Coppi, P. S. 2002 Bulletin of the American Astronomical Society, Vol. 32, p.1217
Corbel, S., Fender, R. P., Tzioumis, A. K., Tomsick, J. A., Orosz, J. A., Miller, J. M., Wijnands, R., & Kaaret, P. 2002, Science, 298, 196
Crocker, M.,M., Davis, R., J., Spencer, R., E., Eyres, S., P., S., Bode, M., F.,& Skopal, A. 2002, MNRAS, 335 , 1100
Dulk, G. A., 1985, ARA&A, 23, 169
Eikenberry, S.S., Matthews, K., Morgan, E., Remillard, R.A., & Nelson, W. R. 1998, ApJ., 494, L61
Esin, A.A, Narayan, R.,Cui, W., Grove, J.E., & Zhang, S.N. 1998, ApJ, 505, 854
Falcke, H., Kรถrding, E., & Markoff, S. 2004, A&A, 414, 895
Fender, R.P 2004, Compact Stellar X-Ray Sources, W.H.G. Lewin & M. van der Klis (Ed.), Cambridge University Press, Cambridge, astro-ph/0303339
Fender, R.P. & Belloni, T. 2004, ARA&A, 42, 317
Fender, R.P., Garrington, S.T., McKay, D.J., Muxlow, T.W.B., Pooley, G.G., Spencer, R.E., Stirling, A.M., & Waltman, E.B. 1999, MNRAS, 304, 865
Fender, R. P., Hjellming, R. M., Tilanus, R. P. J., Pooley, G. G., Deane, J. R., Ogley, R. N. & Spencer, R. E. 2001, MNRAS, 322 , L23
Fender, R.P., Rayner, D., Trushkin, S.A., OโBrien, K., Sault, R.J., Pooley, 2002, Lect. Notes Phys. 589, 101 astro-ph/0109502
Fender, R.P., Rayner, D., Trushkin, S.A., OโBrien, K., Sault, R.J., Pooley, G.G., & Norris, R.P. 2002, MNRAS, 330, 212
Fender, R. P., Roche, P., Pooley, G. G., Chakrabarty, D., Tzioumis, A. K., Hendry, M. A., Spencer, R.E. 1996 Proceedings of 2nd INTEGRAL workshop : The Transparent Universeโ, ESA SP-382 astro-ph/9612080
Fender, R. P., Spencer, R. E., Newell, S. J. & Tzioumis, A. K. 1997, MNRAS, 286, L29
Filippenko, A. V., Leonard, D. C., Matheson, T., Li, W., Moran, E. C., Riess, A. G. 1999, PASP, 111, 969
Fomalont, E. B., Geldzahler, B. J. & Bradshaw, C. F. 2001, ApJ, 558, 283
Frail, D.A., & Hjellming, R.M. 1991, AJ, 101, 2126
Frank, J., King, A. & Raine, D.J. 2002, Accretion Power in Astrophysics, ARI, CUP
Gallo, E., Corbel, S., Fender, R. P., Maccarone, T. J. & Tzioumis, A. K 2004, MNRAS, 347, 52L
Gallo, E., Fender, R. P. & Pooley, G. G. 2003, MNRAS, 344, 60
Geldzahler, B. J., Johnston, K. J., Spencer, J. H., Klepczynski, W. J., Josties, F. J., Angerhofer, P. E., Florkowski, D. R., McCarthy, D. D., Matsakis, D. N.& Hjellming, R. M. 1983, ApJ, .273, 65L
Georganopoulos, M., Kirk, J.G.,& Mastichiadis, A. 2001, ApJ, 561, 111
Gregory, P.C. 2002, ApJ, 575, 427.
Gregory, P.C. & Neish, C. 2002, Ap. J. 580, 1133
Gregory, P.C., & Taylor, A.R. 1978, Nature, 272, 70
Greiner, J., & Rau, A. 2001, A&A 375,145
Hannikainen, D., Wu, K., Campbell-Wilson, D., Hunstead, R., Lovell, J., McIntyre, Vi., Reynolds, J., Soria, R. & Tzioumis, T. 2001, Exploring the gamma-ray universe:, Proc. A. Gimenez, V. Reglero & C. Winkler. (Ed.), ESA SP-459, Noordwijk: ESA Pub. Division, ISBN 92-9092-677-5, 291
Harrison, F.A., Ray, P.S., Leahy, D.A., Waltman, E.B.,& Pooley, G.G. 2000, ApJ, 528, 454
Hartman, R.C., Bertsch, D.L., Bloom, S.D., etal. 1999, ApJS, 123, 79
Heinz, S. & Sunyaev, R. A. 2003, MNRAS, 343, 59
Hjellming, R. M., & Johnston, K. J. 1988, ApJ, 328, 600.
Hjellming, R. M. & Rupen, M. P. 1995, Nature, 375, 464
Hjellming, R. M., Rupen, M. P., Hunstead, R. W., Campbell-Wilson, D., Mioduszewski, A. J., Gaensler, B. M., Smith, D. A., Sault, R. J., Fender, R. P., Spencer, R. E., de la Force, C. J., Richards, A. M. S., Garrington, S. T., Trushkin, S. A., Ghigo, F. D., Waltman, E. B. & McCollough, M. 2000, ApJ, 544, 977
Hjellming, R. M., Rupen, M. P., Mioduszewski, A. J., Smith, D. A., Harmon, B. A., Waltman, E. B., Ghigo, F. D. & Pooley, G. G. 1998, AAS, 193, 103.08 (Bull. AAS 30, 1405)
Hujeirat, A.,& Blandford, R. astro-ph/0307317
Hutchings, J.B., & Crampton, D. 1981, PASP, 93, 486
Hutchings, J.B., Nemec, J.M., & Cassidy, J. 1979, PASP, 91, 313
Kaufman Bernadรณ, M.M., Romero, G.E., & Mirabel, I.F. 2002, A&A, 385, L10
King, A. 1996, X-Ray Binaries, W.H.G. Lewin, J. van Paradijs & M. van der Klis (Ed.), Cambridge University Press, Cambridge, 419.
Kniffen, D. A., Alberts, W. C. K., Bertsch, D. L., Dingus, B. L., Esposito, J. A., Fichtel, C. E., Foster, R. S., Hartman, R. C., Hunter, S. D., Kanbach, G., Lin, Y. C., Mattox, J. R., Mayer-Hasselwander, H. A., Michelson, P. F., von Montigny, C., Mukherjee, R., Nolan, P. L., Paredes, J. M., Ray, P. S., Schneid, E. J., Sreekumar, P., Tavani, M. & Thompson, D. J. 1997,ApJ, 486, 126
Kogure, T. 1969, PASJ, 21, 71
Koshiishi, H., Enome, S., Nakajima, H., Shibasaki, K., Nishio, M., Takano, T., Hanaoka, Y., Torii, C., Sekiguchi, H., Kawashima, S., Bushimata, T., Shinohara, N., Irimajiri, Y. & Shiomi, Y. 1994, PASJ, 46, L33
Leahy, D. A. 2001 A&A, 380,516
Leahy, D.A., Harrison, F.A.,& Yoshida, A. 1997, ApJ, 475, 823
Linfield R. P., 1986, A. J. 92, 21
Liu, Q.Z., Hang, H.R., Wu, G.J., Chang, J.,& Zhu, Z.X. 2000, A&A, 359, 646
Liu, Q.,Z., van Paradijs, J., & van den Heuvel, E.P.J. 2000, A&AS, 147, 25
Liu, Q.,Z., van Paradijs, J.,& van den Heuvel, E.P.J. 2001, A&A, 368, 1021
Longair, M.S. 1994, High Energy Astrophysics, Vol. 2, Stars, the Galaxy and the interstellar medium, Cambridge University Press, Cambridge, 135.
Maccarone, T. J. 2004, MNRAS 351, 1049
Maraschi, L.,& Treves, A. 1981, MNRAS, 194, 18
Margon, B.A. 1979 IAUC, 3345, 1
Margon, B.A. 1980, Sci. Am. 243, 54
Margon, B.A. 1984, ARA&A, 22, 507
Martรญ, J., & Paredes, J.M. 1995, A&A, 298, 151
Massi, M. 1989, A&A, 208, 392
Massi, M. 1999, Dissertation, The Dynamo and the Emission Processes in the Stellar System UX Arietis. University Bonn
Massi, M. 1999b, EVN Doc. n.91
Massi, M. 2003, Recent Research Developments in Astronomy & Astrophysics, eds. A. Gayathri (Ed.), Kerala, India 700-712 (2003)
Massi, M. 2004, A&A, 422, 26
Massi, M. 2004b, proc. 7th EVN Symposium. Bachiller,Colomer,Desmurs,de Vicente (eds) October 12th-15 2004, Toledo, Spain astro-ph/0410502
Massi, M., & Aaron, S. 1997c, EVN Doc. n. 75
Massi, M., & Aaron, S. 1997d, EVN Doc. n. 77
Massi, M.& Aaron, S. 1999, A&AS, 136, 211
Massi, M., & Comoretto, G. 1990, A&A, 228,569
Massi, M., Comoretto, G., Rioja, M., & Tofani G. 1996, A&A Suppl., 116, 167
Massi, M., Menten, K.,& Neidhรถfer, J. 2002, A&A, 382, 152.
Massi, M., Paredes, J.M., Estalella, R.,& Felli, M. 1993, A&A, 269, 249
Massi, M., Ribรณ, M., Paredes, J.M., Peracaula, M., & Estalella, R. 2001, A&A, 376, 217
Massi, M., Ribรณ, M., Paredes, J.M., Garrington, S.T., Peracaula, M., & Martรญ, J. 2004, A&A, 414, L1
Massi, M., Ribรณ, M., Paredes, J.M., Garrington, S.T., Peracaula, M., & Martรญ, J. 2004b, proc. of the Symposium on High-Energy Gamma-Ray Astronomy, Heidelberg, July 26-30, 2004 (AIP Proceedings Series) astro-ph/0410504
Massi, M., Ribรณ, M., Paredes, J.M., Peracaula, M., Martรญ, J., & Garrington, S.T. 2002b, The 4th Microquasar Workshop, Ph. Durouchoux, Y. Fuchs & J. Rodriguez, (Ed.), Center for Space Physics, Kolkata, 238
Massi, M. Rioja, M., Gabuzda, D., Leppanen,K. Sanghera,H., Ruf, K., & Moscadelli L. 1997, A&A, 318, L32
Massi, M. Rioja, M. Gabuzda, D. Leppโanen, K. Sanghera, H. Ruf, & K. Moscad elli, L. 1997b, Vistas in astronomy, vol. 41, Part 2
Massi, M., Ruf, K., & Orfei S. 1998, EVN Doc. n.85
Massi, M. Tofani, G. & Comoretto, G. 1991,A&A, 251, 732
Massi, M., Tuccari, G., & Orfei, S. 1997e, EVN Doc 81
Matsumoto, R., Uchida, Y.,Hirose, S., Shibata, K., Hayashi, M.R., Ferrari, A., & Bodo,G. 1996, ApJ, 461, 115.
McClintock, J.E., & Remillard,R.A. 2004, Compact Stellar X-Ray Sources, W.H.G. Lewin & M. van der Klis (Ed.), Cambridge University Press, Cambridge, astro-ph/0306213
Meier, D. L. 2001 ApJ, 548, 9
Meier, D. L., Koide, S.,& Uchida, Y. 2001, Science, 291, 84.
Mendelson, H., & Mazeh, T. 1989 MNRAS 239, 733
Merloni, A., Fabian, A. C. & Ross, R. R. 2000, MNRAS, 313, 193
Merloni, A., Heinz, S., & di Matteo, T. 2003, MNRAS 345, 1057
Meyer, F., Liu, B.F & Meyer-Hofmeister, E. 2000, A&A 354, L67
Mioduszewski, A. J., Hjellming, R. M. & Rupen, M. P. 1998, AAS, 192, 7402
Mirabel, I.F., Dhawan, V., Chaty, S., Rodrรญguez, L.F., Martรญ, J., Robinson, C.R., Swank, J., & Geballe, T.R. 1998, A&A, 330, L9
Mirabel, I.F.,& Rodrรญguez, L.F. 1994, Nature, 371, 46
Mirabel, I.F., & Rodrรญguez, L.F. 1999, ARA&A, 37, 409
Mirabel, I. F., Rodriguez, L. F., Cordier, B.; Paul, J.;& Lebrun, F. 1992, Nature, 358, 215
Mitsuda, K., Inoue, H., Koyama, K., Makishima, K., Matsuoka, M., Ogawara, Y.,Shibazaki, N., Suzuki, K.,& Tanaka, Y. 1984, PASJ 36, 741
Narayan, R., & Heyl, J.S. 2002, ApJ, 574, 139
Paredes, J.M., Estalella, R. & Rius, A. 1990, A&A, 232, 377
Paredes, J.M. & Figueras, F. 1986 A&A, 154, L30
Paredes, J.M., Marti, J., Estalella, R. & Sarrate, J. 1991, A&A, 248, 124
Paredes, J.M., Marti, J., Peracaula & M., Ribo, M. 1997, A&A, 320 ,L25
Paredes, J. M., Martรญ, J., Ribรณ, M., & Massi, M. 2000, Science, 288, 2340
Paredes, J.M., Massi, M., Estalella, R., & Peracaula, M. 1998, A&A, 335, 539
Pearson, T. J. & Readhead, A. C. S. 1984, ARA&A, 22, 97
Peracaula, M., Gabuzda, D. C., & Taylor, A. R. 1998, A&A, 330, 612
Peracaula, M., Martรญ, J.,& Paredes, J.M. 1997, A&A, 328, 283
Perley, R. A. 1986, โSyntesis Imagingโ proc. NRAO eds R.A. Perley, F.R. Schwab & A. H. Bridle, p.290
Punsly, B. 1999, ApJ, 519, 336
Readhead, A. C. S. &Wilkinson, P. N. 1978, ApJ, 223, 25
Rhoades, C.E. & Ruffini, R. 1974 Physical Review Lett., 32, 324
Rodriguez, L. F., Mirabel, I. F., & Marti, J. 1992, ApJ, 401, L15
Rogers, A. E. E., Hinteregger, H. F., Whitney, A. R., Counselman, C. C., Shapiro, I. I., Wittels, J. J., Klemperer, W. K., Warnock, W. W., Clark, T. A. & Hutton, L. K. 1974, ApJ, 193, 293
Sams, B.,Eckart, A., & Sunyaev, R. 1996, Nature, 382, 47
Schalinski, C. J., Johnston, K. J., Witzel, A. Parsec-scale radio jets, Proc.
Schalinski, C. J., Johnston, K. J., Witzel, A., Spencer, R. E., Fiedler, R., Waltman, E., Pooley, G. G., Hjellming, R. & Molnar, L. A. 1995, ApJ, 447, 752S
Schwab F. R., 1980, Proc. Soc. Photo-Opt. Inst. Eng. 231,18
Shakura, N.I.,& Sunyaev, R.A. 1973,A&A,24, 337
Spencer, R. E. 1979 Nature, 282, 483
Spencer, R. E., Swinney, R. W., Johnston, K. J., & Hjellming, R. M. 1986, ApJ, 309, 694
Stewart, R. T., Caswell, J. L., Haynes, R. F. & Nelson, G. J. 1993, MNRAS. 261, 593
Stirling, A.M., Spencer, R.E., De la Force, C.J., et al. 2001, MNRAS, 327, 1273
Tanaka,Y. 1997, Accretion Disks-New Aspects, E. Meyer-Hofmeister & H. Spruits (Ed.), Lecture Notes in Physics 487. Springer-Verlag Berlin Heidelberg New York, 1
Taylor, A.R., Dougherty, S.M., Scott, W.K., Peracaula, M., & Paredes, J.M. 2000, Astrophysical Phenomena Revealed by Space VLBI, H. Hirabayashi, P.G. Edwards, & D.W. Murphy (Ed.), ISAS, 223
Taylor, A.R., & Gregory, P.C. 1982, ApJ, 255, 210.
Taylor, A.R., Kenny, H.T., Spencer, R. E.,& Tzioumis, A. 1992, ApJ, 395,268
Taylor, A.R., Young, G., Peracaula, M., Kenny, H.T.,& Gregory, P.C. 1996, A&A, 305, 817
Tavani, M., Kniffen, D., Mattox, J.R., Paredes, J.M., & Foster, R.S. 1998, ApJ, 497, L81
Tennant, A. F., Fabian, A. C., & Shafer, R. A. 1986, MNRAS, 221, 27
Tingay, S. J., Jauncey, D. L., Preston, R. A., Reynolds, J. E., Meier, D. L., Murphy, D. W., Tzioumis, A. K., McKay, D. J., Kesteven, M. J., Lovell, J. E. J., Campbell-Wilson, D., Ellingsen, S. P., Gough, R., Hunstead, R. W., Jones, D. L., McCulloch, P. M., Migenes, V., Quick, J., Sinclair, M. W. & Smits, D 1995, Nature, 374, 141
Torricelli, G. Franciosini, E., Massi, M., Neidhรถfer, J. 1998, A&A, 333, 970
Ulrich, M., Maraschi, L., & Urry, C.M. 1997, ARAA, 35, 445
Van der Klis, M. 2004, Compact Stellar X-Ray Sources, W.H.G. Lewin & M. van der Klis (Ed.), Cambridge University Press, Cambridge, astro-ph/0410551
Van der Laan, H. 1966, Nature, 211, 1131
Van Paradijs, J. & McClintock, J.E. 1996 1996, X-Ray Binaries, W.H.G. Lewin, J. van Paradijs & M. van der Klis (Ed.), Cambridge University Press
Verbunt, F., & van den Heuvel, E.P.J. 1996, X-Ray Binaries, W.H.G. Lewin, J. van Paradijs & M. van der Klis (Ed.), Cambridge University Press
Wallace, P.M., Griffis, N.J., Bertsch, D.L., Hartman, R.C., Thompson, D.J., Kniffen, & D.A., Bloom, S.D. 2000, ApJ, 540, 184
Waters, L.B.F.M., van den Heuvel, E.P., Taylor, A.R., Habets, G.M.H.J., & Persi, P. 1988, A&A, 198, 200.
White, N.E., Nagase, F., & Parmar, A.N. 1996, X-Ray Binaries, W.H.G. Lewin, J. van Paradijs & M. van der Klis (Ed.), Cambridge University Press, Cambridge, 1, 33, 6
Zamanov, R. K. 1995, MNRAS, 272, 308
Zamanov, R. K., Reig, P., Martรญ, J., Coe, M. J., Fabregat, J., Tomov, N. A., Valchev, T. 2001, A&A, 367, 884
Zamanov, R.K., Martรญ, J. 2000, A&A, 358, L55
Zdziarski, A. A., Grove, J. E., Poutanen, J., Rao, A. R. & Vadawale, S. V. 2001, ApJ, 554, 45L
Zhang, S. N., Cui, W., Harmon, B. A., Paciesas, W. S., Remillard, R. E., & van Paradijs, J. 1997, ApJ, 477, L95
## 13 DANKSAGUNG
Mein besonderen Dank gilt Prof. Ulrich Mebold fรผr die Gelegenheit diese Arbeit durchzufรผhren und Prof. Karl Menten fรผr seine stetige Unterstรผtzung und sein kontinuierliches Interesse an diesem Projekt.
Die Resultate dieser Arbeit wurden รผber die letzten Jahre in Zusammenarbeit mit vielen Kollegen gewonnen; fรผr die Arbeit mit dem VLBI-, EVN- und MERLIN- Netzwerk sei besonders Marc Ribรณ, Prof. Josep Paredes, Prof. Josep Martรญ, Simon Garrigton and Marta Peracaula gedankt.
Ich mรถchte Prof. Ralph Spencer fรผr seine hilfreichen Kommentare, Prof. Rolf Chini fรผr seine Unterstรผtzung und Jรผrgen Kerp fรผr alle seine wichtigen praktischen Ratschlรคge danken.
Meinem Mann, Jรผrgen Neidhรถfer gilt mein Dank fรผr die zahlreichen Diskussionen, die unseren gemeinsamen Interessen an den physikalischen Prozessen von Doppelstern-System and ihren Periodizitรคten galten, und fรผr die kritische Durchsicht des Manuskripts.
Danken mรถchte ich auch meinen beiden Sรถhnen Guido und Claudio fรผr ihre Geduld, wenn ich einige โsekundรคreโ Sachen wie โdie Wรคschebรผgelnโ wochen lang verschoben habe.
## 14 ABSTRACT
The Astrophysics of microquasars - galactic miniatures of the far distant quasars - has become one of the most active fields of modern Astronomy in recent years. Here I review the astronomical methods used for the investigation of these objects, from Gamma-rays over X-rays and optical to radio wavelengths. The description of each astronomical method is always followed by an examplary application on the source LS I+61 303, one of the most observed Be/X-ray binary systems because of its periodical radio emission and strong, variable Gamma-ray emission.
|
warning/0506/nucl-ex0506006.html
|
ar5iv
|
text
|
# Dominant (๐_{9/2})ยฒ neutron configuration in the 4โบโ state of 68Zn based on new ๐ factor measurements
## 1 Introduction
Measurements of nuclear magnetic dipole and electric quadrupole moments provide valuable insights into the nuclear structure based on the intriguing interplay of single particle and collective degrees of freedom. In particular, nuclei in the $`sd`$\- and $`fp`$-shell model space have recently received considerable attention through many new experiments inspired by the unique possibility to compare extensive data with results from large-scale shell model calculations. This progress is also related with the observation that general features of the nuclear medium can be studied in lighter nuclei as well. For instance, superdeformation which had been exclusively investigated for a long time on nuclei in the rare earth mass region and beyond has recently been observed in <sup>36</sup>Ar , <sup>40</sup>Ca , <sup>60</sup>Zn and <sup>62</sup>Zn . The obvious advantage in all these cases is the possibility to directly compare data with structure calculations based on highly developed nuclear shell model codes thus providing new features of this phenomenon and collectivity in general on a microscopic non-phenomenological level.
In the same context measurements of $`g`$ factors and B(E2) values have been carried out for even-A Zn isotopes with $`A=6270`$ . Two sets of large-scale shell model calculations were applied, based either on a <sup>40</sup>Ca core and an $`fp`$-shell model space or a <sup>56</sup>Ni core and an $`fpg`$-shell configuration space with the inclusion of the $`0g_{9/2}`$ orbital; the latter was considered to be particularly important for the heavier Zn isotopes. In these measurements Coulomb excitation of 160 MeV Zn projectiles was achieved in collisions with a carbon target employing inverse kinematics. At this beam energy essentially the first $`2^+`$ states were strongly excited whereas higher-lying states were only weakly populated.
This deficiency was overcome in a succeeding experiment aiming at higher excitation energies of both <sup>64</sup>Zn and <sup>68</sup>Zn by increasing the projectile energies to 180 MeV close to the Coulomb barrier . In these new measurements the $`g`$ factors and the B(E2) values of the now accessible $`4_1^+`$ states were determined for the first time. Whereas the g($`4_1^+`$) value of <sup>64</sup>Zn was found to be consistent with predictions of both the collective and the spherical shell model, the value for <sup>68</sup>Zn of g($`4_1^+`$) = $`0.4(2)`$ turned out to be in severe conflict with both model predictions. The negative sign of the $`g`$ factor was a clear indication for a dominant $`0g_{9/2}`$ neutron component in the wave function, whereby the Schmidt value is $`g_{\mathrm{Schmidt}}=0.467`$ which, however, was not verified by the calculations. On the other hand, the newly determined B(E2) values were very well explained by these shell model calculations based on a <sup>56</sup>Ni core plus $`0g_{9/2}`$ neutrons. Evidently, in this case the explicit inclusion of the $`0g_{9/2}`$ orbital seems to be crucial for <sup>68</sup>Zn, as the alternative calculations, assuming an inert <sup>40</sup>Ca core but excluding the $`0g_{9/2}`$ orbital, underestimate the experimental E2 strength.
In order to set the $`g`$ factor result of the <sup>68</sup>Zn($`4_1^+`$) state on more solid grounds new measurements have been performed. As emphasized in such an effort is required as in the previous measurements the energy resolution of the NaI(Tl) scintillators used for $`\gamma `$ detection did not allow to fully separate the relevant ($`4_1^+2_1^+`$) $`\gamma `$ line from a neighbouring ($`2_3^+2_1^+`$) 1261 keV line. As the latter was also strongly Doppler-shifted due to the short nuclear lifetime, the separation of the two $`\gamma `$ lines was particularly crucial for the detector pair placed in the forward hemisphere. On the other hand, estimates of an eventual admixture of the $`2_3^+`$ state to the measured precession of the $`4_1^+`$ state, based on the observed line intensities and the angular correlations of the corresponding $`\gamma `$ transitions, could not explain the deduced $`g`$ factor, even under the assumption of a negative $`g`$ value for the $`2_3^+`$ state.
## 2 Experimental details
In the present experiment a beam of isotopically pure <sup>68</sup>Zn ions was accelerated to an energy of 180 MeV at the Munich tandem accelerator providing intensities of 20 $`e`$nA on a multilayered target. The latter consisted of 0.44 mg/cm<sup>2</sup> natural carbon deposited on a 3.34 mg/cm<sup>2</sup> Gd layer, which was evaporated on a 1.4 mg/cm<sup>2</sup> Ta foil, backed by a 4.49 mg/cm<sup>2</sup> Cu layer. Between C and Gd as well as between Ta and Cu thin layers of natural titanium ($`0.005`$ mg/cm<sup>2</sup>) provided good adherence being very crucial for the precession experiments. The same target had been used in former measurements under almost identical conditions . The target was cooled to liquid nitrogen temperature and magnetized to saturation by an external field of 0.06 T. The relevant level scheme of <sup>68</sup>Zn for Coulomb excitation is shown in Fig. 1 . The excited Zn nuclei move through the Gd layer at mean velocities of $`5.9v_0`$ ($`v_0=e^2/\mathrm{}`$) experiencing spin precessions in the transient field and are ultimately stopped in the hyperfine-interaction-free environment of the Cu backing.
The de-excitation $`\gamma `$ rays were measured in coincidence with the forward scattered carbon ions detected in a 100 $`\mu `$m Si detector at $`0^{}`$. A 5 $`\mu `$m thick Ta foil between target and particle detector served as a beam stopper which, however, was transparent to the carbon recoils. As in the previous experiments the detector was operated at a very low bias of $`5`$ V to establish a thin depletion layer for separating the energies of carbon ions from those of light particles as protons and $`\alpha `$ particles resulting from sub-Coulomb fusion and transfer reactions. Under these conditions very clean $`\gamma `$-coincidence spectra were obtained. Intrinsic Ge detectors of $`40\%`$ relative efficiency were used for $`\gamma `$ detection. Fig. 2 shows a typical coincidence spectrum with emphasis on the ($`4_1^+2_1^+`$) $`\gamma `$ line. Evidently, all relevant $`\gamma `$ lines were well resolved allowing a rigorous determination of their intensities required for the angular correlations as well as for the precessions of the nuclear states in question.
Particle-$`\gamma `$ angular correlations $`W(\mathrm{\Theta }_\gamma )`$ have been measured for determining the slope $`S`$ = \[1/W($`\mathrm{\Theta }_\gamma `$)\]$``$\[dW($`\mathrm{\Theta }_\gamma `$)/d$`\mathrm{\Theta }_\gamma `$\] in the rest frame of the $`\gamma `$ emitting nuclei at $`\mathrm{\Theta }_\gamma ^{lab}`$ = $`\pm 65^{}`$ and $`\pm 115^{}`$, where the sensitivity to the precessions was optimal for all transitions of interest. Precession angles, $`\mathrm{\Phi }^{exp}`$, were derived from counting-rate ratios โRโ for โupโ and โdownโ directions of the external magnetizing field which can be expressed as ,
$`\mathrm{\Phi }^{exp}={\displaystyle \frac{1}{S}}{\displaystyle \frac{\sqrt{R}1}{\sqrt{R}+1}}=g{\displaystyle \frac{\mu _N}{\mathrm{}}}{\displaystyle _{t_{in}}^{t_{out}}}B_{TF}(v_{ion}(t))e^{\frac{t}{\tau }}๐t`$ (1)
where $`g`$ is the $`g`$ factor of the nuclear state and $`B_{TF}`$ the transient field acting on the nucleus during the time interval $`(t_{\mathrm{out}}t_{\mathrm{in}}`$) which the ions spend in the gadolinium layer of the target; the exponential accounts for nuclear decay with lifetime $`\tau `$ in the Gd layer.
Simultaneously with the precessions the lifetimes of several excited states have been redetermined using the Doppler-Shift-Attenuation-Method (DSAM). For the analysis of the Doppler-broadened lineshapes, which were observed with a Ge detector placed at $`0^{}`$ to the beam direction, the computer code LINESHAPE has been used. Specific details of the analysis procedure are given in .
## 3 Results and discussion
The $`g`$ factors have been determined from the measured precession angles by calculating the effective transient field $`B_{TF}`$ on the basis of the empirical linear parametrization (see ):
$$B_{\mathrm{TF}}(v_{\mathrm{ion}})=G_{\mathrm{beam}}B_{\mathrm{lin}},$$
(2)
with
$$B_{\mathrm{lin}}=a(Gd)Z_{\mathrm{ion}}v_{\mathrm{ion}}/v_0,$$
(3)
where the strength parameter $`a(Gd)=17(1)T`$ , and $`G_{\mathrm{beam}}=0.61(6)`$ is the attenuation factor accounting for the demagnetization of the Gd layer induced by the Zn beam (see ). The same scheme has been successfully applied in many former measurements.
Precession and lifetime data from a single run are summarized in Table 1 together with the deduced $`g`$ factors which are compared with previous results . Evidently, all newly determined lifetimes and $`g`$ factors are in good agreement with earlier data . Furthermore, the $`g`$ factor of the $`4_1^+`$ state with its negative sign is confirmed whereby the relatively large error in the magnitude is of purely statistical origin due to the small excitation cross-section of the nuclear state and the low $`\gamma `$-detection efficiency of the Ge detectors. The striking difference in the $`g`$ factors of the $`4_1^+`$ states between <sup>64</sup>Zn and <sup>68</sup>Zn is shown in Fig. 3.
In Table 2, the $`g`$ factors and the B(E2)โs, both improved in accuracy by averaging with the data of , are compared with results from large-scale shell model calculations.
In order to study the importance of the $`0g_{9/2}`$ orbit we have performed shell model calculations using <sup>56</sup>Ni as closed shell core, with a model space defined by protons and neutrons occupying the single-particle orbitals $`0f_{5/2}`$, $`1p_{3/2}`$, $`1p_{1/2}`$ and $`0g_{9/2}`$ ($`0f_{5/2}1pg_{9/2}`$). To determine the effective interaction we use the recent charge-dependent potential model of Machleidt, the so-called CD-Bonn interaction . The final effective two-body interaction is obtained via many-body perturbation theory to third order, employing a renormalized nucleon-nucleon interaction defined for <sup>56</sup>Ni as closed shell core and including folded-diagrams to infinite order. For details, see for example Ref. . A harmonic oscillator basis was used, with an oscillator energy determined via $`\mathrm{}\mathrm{\Omega }=45A^{1/3}25A^{2/3}`$ = 10.1 MeV, A = 56 being the mass number. For the single-particle energies we employ values adapted from Grawe in Ref. , resulting in the energy differences $`ฯต_{0g_{9/2}}ฯต_{1p_{3/2}}=3.70`$ MeV, $`ฯต_{0f_{5/2}}ฯต_{1p_{3/2}}=0.77`$ MeV, $`ฯต_{1p_{1/2}}ฯต_{1p_{3/2}}=1.11`$ MeV for neutrons and $`ฯต_{0g_{9/2}}ฯต_{1p_{3/2}}=3.51`$ MeV, $`ฯต_{0f_{5/2}}ฯต_{1p_{3/2}}=1.03`$ MeV, $`ฯต_{1p_{1/2}}ฯต_{1p_{3/2}}=1.11`$ MeV for protons. The effective interaction has not been corrected for any eventual monopole changes. This means that the only parameters which enter our calculations are those defining the nucleon-nucleon interaction fitted to reproduce the scattering data, the experimental single-particle energies and the oscillator basis. Furthermore, for the computation of the B(E2)โs and $`g`$ factors, we have used unrenormalized magnetic moments and the canonical unrenormalized charges for protons and neutrons. The latter entails charges of $`1.5e`$ and $`0.5e`$ for protons and neutrons, respectively. The free-nucleon operator for the magnetic moment is defined as
$$\mu _{\mathrm{free}}=g_l๐ฅ+g_s๐ฌ,$$
(4)
with $`g_l(\mathrm{proton})=1.0`$, $`g_l(\mathrm{neutron})=0.0`$, $`g_s(\mathrm{proton})=5.586`$, $`g_s(\mathrm{neutron})=3.826`$. Note, however, that the magnetic moment operator in finite nuclei is modified from the free-nucleon operator due to core-polarization and meson-exchange current (MEC) corrections . The effective operator is defined as
$$\mu _{\mathrm{eff}}=g_{l,\mathrm{eff}}๐ฅ+g_{s,\mathrm{eff}}๐ฌ+g_{p,\mathrm{eff}}[Y_2,๐ฌ],$$
(5)
where $`g_{x,\mathrm{eff}}=g_x+\delta g_x`$, $`x=l`$, $`s`$ or $`p`$, with $`g_x`$ the free-nucleon, single-particle $`g`$ factors ($`g_p=0`$) and $`\delta g_x`$ the calculated correction to it. Note the presence of a new term $`[Y_2,๐ฌ]`$, absent from the free-nucleon operator, which is a spherical harmonic of rank $`\lambda ^{}=2`$ coupled to a spin operator to form a spherical tensor of multipolarity $`\lambda =1`$. In the calculations below we limit ourselves to a calculation with free operators only. An obvious modification of the latter is to use renormalized moments, as done in the recent work of Ref. for nuclei in the <sup>132</sup>Sn mass region.
The results of the calculations for the energies of the low-lying excited states for both <sup>64</sup>Zn and <sup>68</sup>Zn in Table 2, while $`B(E2;2_1^+0_1^+`$), $`B(E2;2_2^+0_1^+`$) and $`B(E2;4_1^+2_1^+`$) and the $`g`$ factors $`g(2_1^+)`$, $`g(2_2^+)`$ and $`g(4_1^+)`$ are presented and compared with the available data.
Two types of shell-model results are discussed and shown in this Table, SM-1 and SM-2. In SM-1 we limit the number of neutrons which can be excited to the $`0g_{9/2}`$ orbit to two, whereas SM-2 is the full shell-model calculation. The latter basis is, however, unnecessarily large since the results are converged with four neutrons at most in the $`0g_{9/2}`$ orbit. One sees clearly that with at most two neutrons in the $`0g_{9/2}`$ orbit the spectrum is rather poorly reproduced. However, even for the fully converged calculation the first excited $`0_2^+`$ state at 1.655 MeV is badly reproduced ($`E^{SM2}`$=2.406 MeV), indicating most likely the need for particle-hole excitations, especially from the $`0f_{7/2}`$ orbit. The typical occupation probabilities in the SM-2 calculation of the $`0g_{9/2}`$ neutron single-particle orbit span from 2.2 to 2.4. The $`1p_{3/2}`$ neutron orbit has an average occupancy of 3.5 particles whereas the low-lying $`1f_{5/2}`$ neutron orbit has occupancies around three. We note that for the $`2_1^+`$, $`2_2^+`$ and the $`4_1^+`$ states there is a satisfactory agreement with the data.
For <sup>68</sup>Zn the B(E2;2$`{}_{1}{}^{+}`$0$`{}_{1}{}^{}{}_{}{}^{+}`$) exhibits a good agreement with data using unrenormalized effective charges while the B(E2;4$`{}_{1}{}^{}{}_{}{}^{+}`$ $``$2$`{}_{1}{}^{}{}_{}{}^{+}`$) is overestimated. This could be ascribed to deficiencies in the two-body Hamiltonian and/or omitted degrees of freedom in the model space.
For <sup>64</sup>Zn we get 170.4 $`e^2`$fm<sup>4</sup> for the B(E2;2$`{}_{1}{}^{+}`$0$`{}_{1}{}^{}{}_{}{}^{+}`$), to be compared with the experimental value of 307 $`e^2`$fm<sup>4</sup>, hinting at the need of larger effective charges. This means in turn that particle-hole excitations involving the $`0f_{7/2}`$ orbit may be more important for <sup>64</sup>Zn than for <sup>68</sup>Zn, in good agreement with previous shell model calculations including this degree of freedom (see for example the discussions in Refs. ). This is also reflected in the $`g`$ factors for <sup>64</sup>Zn, which tend to be larger than the experimental values. If we, however, introduce effective magnetic moments by assuming a renormalization factor of $`0.70.8`$ for $`g_s`$ of protons and neutrons, we obtain $`g`$ factors closer to experiment. However, the $`0f_{7/2}`$ orbit, if coupled with the $`1p_{3/2}`$ orbit can yield a negative contribution to the $`g`$ factors. The latter analysis is obviously performed at the level of a simple two-body configuration (see for example discussions in Refs. ), however, together with the $`(0g_{9/2})^2`$ configuration these are the only two-body configurations of interest here which can yield a negative $`g`$ factor. For <sup>64</sup>Zn the occupancy of the $`0g_{9/2}`$ orbit is less than one, and plays therefore a negligible role in the calculation of $`g`$ factors. Since excitations from the <sup>56</sup>Ni closed shell core are not included in the present model space, this reduction cannot be accounted for. Thus, the theoretical values should be larger than the experimental ones.
For <sup>68</sup>Zn and its $`g`$ factors we note that for the $`2^+`$ states there is a fair agreement with data, confirming the previous shell-model analysis presented in Refs. . For $`g(4_1^+)`$ we see that there is again a change from the calculation with only two neutrons in the $`0g_{9/2}`$ orbit to the full calculation. This displays the role played by the $`0g_{9/2}`$ orbit, which for the $`4_1^+`$ has an average occupation probability of 2.4 in the full SM-2 calculation, much larger than we have for <sup>64</sup>Zn. For SM-1 the corresponding occupation probability is 1.99 and this difference is clearly reflected in the change of $`g(4_1^+)`$ from 0.066 to 0.008 demonstrating thereby the dominating role played by the $`(0g_{9/2})^2`$ admixture in the wave function. However, for SM-2 the $`g`$ factor although very small, is still positive. Introducing effective magnetic moments with a scaling factor of $`0.7`$ reduces the theoretical SM-2 value to $`g(4_1^+)=+0.003`$. We speculate again whether particle-hole excitations involving the $`0f_{7/2}`$ orbit could yield further reductions and eventually a negative contribution. Furthermore, another possibility is to include the effect of meson-exchange currents, as done in Ref. . Meson-exchange current corrections arise because nucleons in nuclei are interacting through the exchange of mesons, which can be disturbed by the electromagnetic field. In terms of effective one-body operators this leads to a correction term $`[Y_2,๐ฌ]`$ in Eq. (5). In spite of these omitted degrees of freedom, we see that our model space displays the important role played by the $`(0g_{9/2})^2`$ neutron configuration when going from the SM-1 to the full SM-2 shell-model calculation, lending thereby support to the experimental analysis.
The role of the neutron $`g_{9/2}`$ orbit is also seen in the $`g`$ factors of low-lying $`9/2^+`$ states in odd $`Zn`$ isotopes with a large negative value, although smaller in absolute value than the corresponding Schmidt value (see Ref. ).
## 4 Summary and conclusions
We have remeasured the $`g`$ factor of the $`4_1^+`$ state in <sup>68</sup>Zn, with an improved energy resolution of the detectors used. The experimental value is $`g(4_1^+)=0.37(17)`$ consistent with our previous measurements. In addition, the accuracy of the $`g`$ factors of the $`2_1^+`$, $`2_2^+`$ and $`3_1^{}`$ states has been improved and their lifetimes were well reproduced. The experimental results for <sup>64</sup>Zn and <sup>68</sup>Zn have been compared with large-scale shell model calculations using <sup>56</sup>Ni as a closed shell core and a model space consisting of protons and neutrons occupying the $`0f_{5/2}1pg_{9/2}`$ single-particle orbits. The agreement between theory and experiment is rather good and the calculations reproduce well the experimental trends for the $`g`$ factors from <sup>64</sup>Zn to <sup>68</sup>Zn, although our theoretical approach is not capable of reproducing the negative sign of the $`g(4_1^+)`$ value in <sup>68</sup>Zn. This deficiency may be ascribed to particle-hole excitations, with the $`0f_{7/2}`$ orbit playing a major role. The effect of meson-exchange currents may also play a role and will be investigated in future works, together with the inclusion of the $`0f_{7/2}`$ orbit. In view of the present results similar measurements for the Ge isotones <sup>66</sup>Ge and <sup>70</sup>Ge are highly desirable to search for corresponding effects in the nuclear wave functions.
The authors are thankful to the operating staff of the Munich tandem accelerator. Support by the BMBF and the Deutsche Forschungsgemeinschaft is acknowledged. The work of TE and MHJ has been supported by the Research Council of Norway (Program for Supercomputing) through a grant of computing time. Discussions with Alex Brown (MSU) and Georgi Georgiev (CERN) are gratefully acknowledged.
|
warning/0506/cs0506095.html
|
ar5iv
|
text
|
# Deriving a Stationary Dynamic Bayesian Network from a Logic Program with Recursive LoopsA preliminary version appears in the 15th International Conference on Inductive Logic Programming.
## 1 Introduction
Probabilistic logic programming (PLP) is a framework that extends the expressive power of Bayesian networks with first-order logic . The core of the PLP framework is a backward-chaining procedure, which generates a Bayesian network graphic structure from a logic program in a way quite like query evaluation in logic programming. Therefore, existing PLP methods use a slightly adapted SLD- or SLDNF-resolution as the backward-chaining procedure.
Recursive loops in a logic program are SLD-derivations of the form
$$A_1\mathrm{}A_2\mathrm{}A_3\mathrm{}$$
(1)
where for any $`i1`$, $`A_i`$ is the same as $`A_{i+1}`$ up to variable renaming.<sup>1</sup><sup>1</sup>1The left-most computation rule is assumed in this paper. Such loops present a challenging problem to the PLP framework. On the one hand, they loop forever so that the PLP backward-chaining inferences would never stop. On the other hand, they may generate cyclic influences, which are disallowed in Bayesian networks.
Two representative approaches have been proposed to avoid recursive loops. The first one is by Ngo and Haddawy and Kersting and De Raedt , who restrict to considering only acyclic logic programs . The second approach, proposed by Glesner and Koller , uses explicit time parameters to avoid occurrence of recursive loops. It enforces acyclicity using time parameters in the way that every predicate has a time argument such that the time argument in the clause head is at least one time step later than the time arguments of the predicates in the clause body. In this way, each predicate $`p(X)`$ is changed to $`p(X,T)`$ and each clause $`p(X)q(X)`$ is rewritten into $`p(X,T1)T2=T11,q(X,T2)`$, where $`T`$, $`T1`$ and $`T2`$ are time parameters.
In this paper, we propose a solution to the problem of recursive loops under the PLP framework. Our method is not restricted to acyclic logic programs, nor does it rely on explicit time parameters. Instead, it makes use of recursive loops to derive a stationary dynamic Bayesian network. We will make two novel contributions. First, we introduce the well-founded semantics of logic programs to the PLP framework; in particular, we use the well-founded model of a logic program to define the direct influence relation and apply SLG-resolution (or SLTNF-resolution ) to make the backward-chaining inferences. As a result, termination of the PLP backward-chaining process is guaranteed. Second, we observe that under the PLP framework recursive loops (cyclic influences) define feedbacks, thus implying a time sequence. For instance, the clause $`aids(X)aids(Y),contact(X,Y)`$ introduces recursive loops
$$aids(X)aids(Y)\mathrm{}aids(Y1)\mathrm{}$$
Together with some other clauses in a logic program, these recursive loops may generate cyclic influences of the form
$$aids(p1)\mathrm{}aids(p1)\mathrm{}aids(p1)\mathrm{}$$
Such cyclic influences represent feedback connections, i.e., that $`p1`$ is infected with aids (in the current time slice $`t`$) depends on whether $`p1`$ was infected with aids earlier (in the last time slice $`t1`$). Therefore, recursive loops of form (1) imply a time sequence of the form
$$A\underset{t}{\underset{}{\mathrm{}}}A\underset{t1}{\underset{}{\mathrm{}}}A\underset{t2}{\underset{}{\mathrm{}}}A\mathrm{}$$
(2)
where $`A`$ is a ground instance of $`A_1`$. It is this observation that leads us to viewing a logic program with recursive loops as a special temporal model. Such a temporal model corresponds to a stationary dynamic Bayesian network and thus can be compactly represented as a two-slice dynamic Bayesian network.
The paper is structured as follows. In Section 2, we review some concepts concerning Bayesian networks and logic programs. In Section 3, we introduce a new PLP formalism, called Bayesian knowledge bases. A Bayesian knowledge base consists mainly of a logic program that defines a direct influence relation over a space of random variables. In Section 4, we establish a declarative semantics for a Bayesian knowledge base based on a key notion of influence clauses. Influence clauses contain only ground atoms from the space of random variables and define the same direct influence relation as the original Bayesian knowledge base does. In Section 5, we present algorithms for building a two-slice dynamic Bayesian network from a Bayesian knowledge base. We describe related work in Section 6 and summarize our work in Section 7.
## 2 Preliminaries and Notation
We assume the reader is familiar with basic ideas of Bayesian networks and logic programming . In particular, we assume the reader is familiar with the well-founded semantics as well as SLG-resolution . Here we review some basic concepts concerning dynamic Bayesian networks (DBNs). DBNs are introduced to model the evolution of the state of the environment over time . Briefly, a DBN is a Bayesian network whose random variables are subscripted with time steps (basic units of time) or time slices (i.e. intervals). In this paper, we use time slices. For instance, $`Weather_{t1}`$, $`Weather_t`$ and $`Weather_{t+1}`$ are random variables representing the weather situations in time slices $`t1`$, $`t`$ and $`t+1`$, respectively. We can then use a DBN to depict how $`Weather_{t1}`$ influences $`Weather_t`$.
A DBN is represented by describing the intra-probabilistic relations between random variables in each individual time slice $`t`$ ($`t>0`$) and the inter-probabilistic relations between the random variables of each two consecutive time slices $`t1`$ and $`t`$. If both the intra- and inter-probabilistic relations are the same for all time slices (in this case, the DBN is a repetition of a Bayesian network over time; see Figure 1), the DBN is called a stationary DBN ; otherwise it is called a flexible DBN . As far as we know, most existing DBN systems reported in the literature are stationary DBNs.
In a stationary DBN as shown in Figure 1, the state evolution is determined by random variables like $`C`$, $`B`$ and $`A`$, as they appear periodically and influence one another over time (i.e., they produce cycles of direct influences). Such variables are called state variables. Note that $`D`$ is not a state variable. Due to the characteristic of stationarity, a stationary DBN is often compactly represented as a two-slice DBN.
###### Definition 2.1
A two-slice DBN for a stationary DBN consists of two consecutive time slices, $`t1`$ and $`t`$, which describes (1) the intra-probabilistic relations between the random variables in slice $`t`$ and (2) the inter-probabilistic relations between the random variables in slice $`t1`$ and the random variables in slice $`t`$.
A two-slice DBN models a feedback system, where a cycle of direct influences establishes a feedback connection. For convenience, we depict feedback connections with dashed edges. Moreover, we refer to nodes coming from slice $`t1`$ as state input nodes (or state input variables).<sup>2</sup><sup>2</sup>2When no confusion would occur, we will refer to nodes and random variables exchangeably.
###### Example 2.1
The stationary DBN of Figure 1 can be represented by a two-slice DBN as shown in Figure 3, where $`A`$, $`C`$ and $`B`$ form a cycle of direct influences and thus establish a feedback connection. This stationary DBN can also be represented by a two-slice DBN starting from a different state input node such as $`C_{t1}`$ or $`B_{t1}`$. These two-slice DBN structures are equivalent in the sense that they model the same cycle of direct influences and can be unrolled into the same stationary DBN (Figure 1).
Observe that in a two-slice DBN, all random variables except state input nodes have the same subscript $`t`$. In the sequel, the subscript $`t`$ is omitted for simplification of the structure. For instance, the two-slice DBN of Figure 3 is simplified to that of Figure 3.
In the rest of this section, we introduce some necessary notation for logic programs. Variables begin with a capital letter, and predicate, function and constant symbols with a lower-case letter. We use $`p(.)`$ to refer to any predicate/atom whose predicate symbol is $`p`$ and use $`p(\stackrel{}{X})`$ to refer to $`p(X_1,\mathrm{},X_n)`$ where all $`X_i`$s are variables. There is one special predicate, $`true`$, which is always logically true. A predicate $`p(\stackrel{}{X})`$ is typed if its arguments $`\stackrel{}{X}`$ are typed so that each argument takes on values in a well-defined finite domain. A (general) logic program $`P`$ is a finite set of clauses of the form
$$AB_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n$$
(3)
where $`A`$, the $`B_i`$s and $`C_j`$s are atoms. We use $`HU(P)`$ and $`HB(P)`$ to denote the Herbrand universe and Herbrand base of $`P`$, respectively, and use $`WF(P)=<`$$`I_t,I_f`$$`>`$ to denote the well-founded model of $`P`$, where $`I_t,I_fHB(P)`$, and every $`A`$ in $`I_t`$ is true and every $`A`$ in $`I_f`$ is false in $`WF(P)`$. By a (Herbrand) ground instance of a clause/atom $`C`$ we refer to a ground instance of $`C`$ that is obtained by replacing all variables in $`C`$ with some terms in $`HU(P)`$.
A logic program $`P`$ is a positive logic program if no negative literal occurs in the body of any clause. $`P`$ is a Datalog program if no clause in $`P`$ contains function symbols. $`P`$ is an acyclic logic program if there is a mapping $`map`$ from the set of ground instances of atoms in $`P`$ into the set of natural numbers such that for any ground instance $`AB_1,\mathrm{},B_k,\neg B_{k+1},\mathrm{},\neg B_n`$ of any clause in $`P`$, $`map(A)>map(B_i)`$ $`(1in)`$ . $`P`$ is said to have the bounded-term-size property w.r.t. a set of predicates $`\{p_1(.),\mathrm{},p_t(.)\}`$ if there is a function $`f(n)`$ such that for any $`1it`$ whenever a top goal $`G_0=p_i(.)`$ has no argument whose term size exceeds $`n`$, no atoms in any SLDNF- (or SLG-) derivations for $`G_0`$ have an argument whose term size exceeds $`f(n)`$ (this definition is adapted from ).
## 3 Definition of a Bayesian Knowledge Base
In this section, we introduce a new PLP formalism, called Bayesian knowledge bases. Bayesian knowledge bases accommodate recursive loops and define the direct influence relation in terms of the well-founded semantics.
###### Definition 3.1
A Bayesian knowledge base is a triple $`<`$$`PBCB,T_x,CR`$$`>`$, where
* $`PBCB`$ is a logic program, each clause in $`PB`$ being of the form
$`p(.)\underset{directinfluences}{\underset{}{p_1(.),\mathrm{},p_l(.)}},true,\underset{context}{\underset{}{B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n}},`$
$`\underset{typeconstraints}{\underset{}{member(X_1,DOM_1),\mathrm{},member(X_s,DOM_s)}}`$ (4)
where (i) the predicate symbols $`p,p_1,\mathrm{},p_l`$ only occur in $`PB`$ and (ii) $`p(.)`$ is typed so that for each variable $`X_i`$ in it with a finite domain $`DOM_i`$ (a list of constants) there is an atom $`member(X_i,DOM_i)`$ in the clause body.
* $`T_x`$ is a set of conditional probability tables (CPTs) of the form $`๐(p(.)|p_1(.),\mathrm{},`$ $`p_l(.))`$, each being attached to a clause (3.1) in $`PB`$.
* $`CR`$ is a combination rule such as noisy-or, min or max .
A Bayesian knowledge base contains a logic program that can be divided into two parts, $`PB`$ and $`CB`$. $`PB`$ defines a direct influence relation, each clause (3.1) saying that the atoms $`p_1(.),\mathrm{},p_l(.)`$ have direct influences on $`p(.)`$ in the context that $`B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n,`$ $`member(X_1,DOM_1),\mathrm{},`$ $`member(X_s,DOM_s)`$ is true in $`PBCB`$ under the well-founded semantics. Note that the special literal $`true`$ is used in clause (3.1) to mark the beginning of the context; it is always true in the well-founded model $`WF(PBCB)`$. For each variable $`X_i`$ in the head $`p(.)`$, $`member(X_i,DOM_i)`$ is used to enforce the type constraint on $`X_i`$, i.e. the value of $`X_i`$ comes from its domain $`DOM_i`$. $`CB`$ assists $`PB`$ in defining the direct influence relation by introducing some auxiliary predicates (such as $`member(.)`$) to describe contexts.<sup>3</sup><sup>3</sup>3The predicate $`true`$ can be defined in $`CB`$ using a unit clause. Clauses in $`CB`$ do not describe direct influences.
Recursive loops are allowed in $`PB`$ and $`CB`$. In particular, when some $`p_i(.)`$ in clause (3.1) is the same as the head $`p(.)`$, a cyclic direct influence occurs. Such a cyclic influence models a feedback connection and is interpreted as $`p(.)`$ at present depending on itself in the past.
In this paper, we focus on Datalog programs, although the proposed approach applies to logic programs with the bounded-term-size property (w.r.t. the set of predicates appearing in the heads of clauses in $`PB`$) as well. Datalog programs are widely used in database and knowledge base systems and have a polynomial time complexity in computing their well-founded models . In the sequel, we assume that except for the predicate $`member(.)`$, $`PBCB`$ is a Datalog program.
For each clause (3.1) in $`PB`$, there is a unique CPT, $`๐(p(.)|p_1(.),\mathrm{},p_l(.))`$, in $`T_x`$ specifying the degree of the direct influences. Such a CPT is shared by all instances of clause (3.1).
A Bayesian knowledge base has the following important property.
###### Theorem 3.1
(1) All unit clauses in $`PB`$ are ground. (2) Let $`G_0=p(.)`$ be a goal with $`p`$ being a predicate symbol occurring in the head of a clause in $`PB`$. Then all answers of $`G_0`$ derived from $`PBCB\{G_0\}`$ by applying SLG-resolution are ground.
Proof: (1) If the head of a clause in $`PB`$ contains variables, there must be atoms of the form $`member(X_i,DOM_i)`$ in its body. This means that clauses whose head contains variables are not unit clauses. Therefore, all unit clauses in $`PB`$ are ground.
(2) Let $`A`$ be an answer of $`G_0`$ obtained by applying SLG-resolution to $`PBCB\{G_0\}`$. Then $`A`$ must be produced by applying a clause in $`PB`$ of form (3.1) with a most general unifier (mgu) $`\theta `$ such that $`A=p(.)\theta `$ and the body $`(p_1(.),\mathrm{},p_l(.),true,B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n,`$ $`member(X_1,DOM_1),\mathrm{},member(`$ $`X_s,DOM_s))\theta `$ is evaluated true in the well-founded model $`WF(PBCB)`$. Note that the type constraints $`(member(X_1,DOM_1),\mathrm{},member(X_s,DOM_s))\theta `$ being evaluated true by SLG-resolution guarantees that all variables $`X_i`$s in the head $`p(.)`$ are instantiated by $`\theta `$ into constants in their domains $`DOM_i`$s. This means that $`A`$ is ground. $`\mathrm{}`$
For the sake of simplicity, in the sequel for each clause (3.1) in $`PB`$, we omit its type constraints $`member(X_i,DOM_i)`$ ($`1is`$). Therefore, when we say that the context $`B_1,\mathrm{},B_m,`$ $`\neg C_1,\mathrm{},\neg C_n`$ is true, we assume that the related type constraints are true as well.
###### Example 3.1
We borrow the well-known AIDS program from (a simplified version) as a running example to illustrate our PLP approach. It is formulated by a Bayesian knowledge base $`KB_1`$ with the following logic program:<sup>4</sup><sup>4</sup>4This Bayesian knowledge base $`KB_1=<`$$`PB_1CB_1,T_{x_1},CR_1`$$`>`$ may well contain contexts that describe a personโs background information. The contexts together with $`CB_1`$, $`T_{x_1}`$ and $`CR_1`$ are omitted here for the sake of simplicity.
| $`PB_1:`$ | 1. $`aids(p1).`$ |
| --- | --- |
| | 2. $`aids(p3).`$ |
| | 3. $`aids(X)aids(X).`$ |
| | 4. $`aids(X)aids(Y),contact(X,Y).`$ |
| | 5. $`contact(p1,p2).`$ |
| | 6. $`contact(p2,p1).`$ |
Note that both the 3rd and the 4-th clause produce recursive loops. The 3rd clause also has a cyclic direct influence. Conceptually, the two clauses model the fact that the direct influences on $`aids(X)`$ come from whether $`X`$ was infected with aids earlier (the feedback connection induced from the 3rd clause) or whether $`X`$ has contact with someone $`Y`$ who is infected with aids (the 4-th clause).
## 4 Declarative Semantics
In this section, we formally describe the space of random variables and the direct influence relation defined by a Bayesian knowledge base $`KB`$. We then define probability distributions induced by $`KB`$.
### 4.1 Space of Random Variables and Influence Clauses
A Bayesian knowledge base $`KB`$ defines a direct influence relation over a subset of $`HB(PB)`$. Recall that any random variable in a Bayesian network is either an input node (with no parent nodes) or a node on which some other nodes (i.e. its parent nodes) in the network have direct influences. Since an input node can be viewed as a node whose direct influences come from an empty set of parent nodes, we can define a space of random variables from a Bayesian knowledge base $`KB`$ by taking all unit clauses in $`PB`$ as input nodes and deriving the other nodes iteratively based on the direct influence relation defined by $`PB`$. Formally, we have
###### Definition 4.1
The space of random variables of $`KB`$, denoted $`๐ฎ(KB)`$, is recursively defined as follows:
1. All unit clauses in $`PB`$ are random variables in $`๐ฎ(KB)`$.
2. Let $`AA_1,\mathrm{},A_l,true,B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ be a ground instance of a clause in $`PB`$. If the context $`B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ is true in the well-founded model $`WF(PBCB)`$ and $`\{A_1,\mathrm{},A_l\}`$ $`๐ฎ(KB)`$, then $`A`$ is a random variable in $`๐ฎ(KB)`$. In this case, each $`A_i`$ is said to have a direct influence on $`A`$.
3. $`๐ฎ(KB)`$ contains only those ground atoms satisfying the above two conditions.
###### Definition 4.2
For any random variables $`A`$, $`B`$ in $`๐ฎ(KB)`$, we say $`A`$ is influenced by $`B`$ if $`B`$ has a direct influence on $`A`$, or for some $`C`$ in $`๐ฎ(KB)`$ $`A`$ is influenced by $`C`$ and $`C`$ is influenced by $`B`$. A cyclic influence occurs if $`A`$ is influenced by itself.
###### Example 4.1 (Example 3.1 continued)
The clauses 1, 2, 5 and 6 are unit clauses, thus random variables. $`aids(p2)`$ is then derived applying the 4-th clause. Consequently, $`๐ฎ(KB_1)=\{aids(p1),aids(p2),aids(p3),contact(p1,p2),contact(p2,p1)\}`$. $`aids(p1)`$ and $`aids(p2)`$ have a direct influence on each other. There are three cyclic influences: $`aids(pi)`$ is influenced by itself for each $`i=1,2,3`$.
Let $`WF(PBCB)=<`$$`I_t,I_f`$$`>`$ be the well-founded model of $`PBCB`$ and let $`I_{PB}=\{p(.)I_t|p`$ occurs in the head of some clause in $`PB\}`$. The following result shows that the space of random variables is uniquely determined by the well-founded model.
###### Theorem 4.1
$`๐ฎ(KB)=I_{PB}`$.
Proof: First note that all unit clauses in $`PB`$ are both in $`๐ฎ(KB)`$ and in $`I_{PB}`$. We prove this theorem by induction on the maximum depth $`d0`$ of backward derivations of a random variable $`A`$.
($``$) Let $`A๐ฎ(KB)`$. When $`d=0`$, $`A`$ is a unit clause in $`PB`$, so $`AI_{PB}`$. For the induction step, assume $`BI_{PB}`$ for any $`B๐ฎ(KB)`$ whose maximum depth $`d`$ of backward derivations is below $`k`$. Let $`d=k`$ for $`A`$. There must be a ground instance $`AA_1,\mathrm{},A_l,true,B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ of a clause in $`PB`$ such that the $`A_i`$s are already in $`๐ฎ(KB)`$ and $`B_1,\mathrm{},B_m,\neg C_1,\mathrm{},`$ $`\neg C_n`$ is true in the well-founded model $`WF(PBCB)`$. Since the head $`A`$ is derived from the $`A_i`$s in the body, the maximum depth for each $`A_i`$ must be below the depth $`k`$ for the head $`A`$. By the induction hypothesis, the $`A_i`$s are in $`I_{PB}`$. By definition of the well-founded model, $`A`$ is true in $`WF(PBCB)`$ and thus $`AI_{PB}`$.
($``$) Let $`AI_{PB}`$. When $`d=0`$, $`A`$ is a unit clause in $`PB`$, so $`A๐ฎ(KB)`$. For the induction step, assume $`B๐ฎ(KB)`$ for any $`BI_{PB}`$ whose maximum depth $`d`$ of backward derivations is below $`k`$. Let $`d=k`$ for $`A`$. There must be a ground instance $`AA_1,\mathrm{},A_l,true,\mathrm{}`$ of a clause in $`PB`$ such that the body is true in $`WF(PBCB)`$. Note that the predicate symbol of each $`A_i`$ occurs in the head of a clause in $`PB`$. Since the head $`A`$ is derived from the literals in the body, the maximum depth of backward derivations for each $`A_i`$ in the body must be below the depth $`k`$ for the head $`A`$. By the induction hypothesis, the $`A_i`$s are in $`๐ฎ(KB)`$. By Definition 4.1, $`A๐ฎ(KB)`$. $`\mathrm{}`$
Theorem 4.1 suggests that the space of random variables can be computed by applying an existing procedure for the well-founded model such as SLG-resolution or SLTNF-resolution. Since SLG-resolution has been implemented as the well-known $`XSB`$ system , in this paper we apply it for the PLP backward-chaining inferences. SLG-resolution is a tabling mechanism for top-down computation of the well-founded model. For any atom $`A`$, during the process of evaluating a goal $`A`$, SLG-resolution stores all answers of $`A`$ in a space called table, denoted $`๐ฏ_A`$.
Let $`\{p_1,\mathrm{},p_t\}`$ be the set of predicate symbols occurring in the heads of clauses in $`PB`$, and let $`GS_0=\{p_1(\stackrel{}{X_1}),\mathrm{},p_t(\stackrel{}{X_t})\}`$.
Algorithm 1: Computing random variables.
1. $`๐ฎ^{}(KB)=\mathrm{}`$.
2. For each $`p_i(\stackrel{}{X_i})`$ in $`GS_0`$
1. Compute the goal $`p_i(\stackrel{}{X_i})`$ by applying SLG-resolution to $`PBCB\{p_i(\stackrel{}{X_i})\}`$.
2. $`๐ฎ^{}(KB)=๐ฎ^{}(KB)๐ฏ_{p_i(\stackrel{}{X_i})}`$.
3. Return $`๐ฎ^{}(KB)`$.
###### Theorem 4.2
Algorithm 1 terminates, yielding a finite set $`๐ฎ^{}(KB)=๐ฎ(KB)`$.
Proof: Let $`WF(PBCB)=<`$$`I_t,I_f`$$`>`$ be the well-founded model of $`PBCB`$. By the soundness and completeness of SLG-resolution, Algorithm 1 will terminate with a finite output $`๐ฎ^{}(KB)`$ that consists of all answers of $`p_i(\stackrel{}{X_i})`$ ($`1it`$). By Theorem 3.1, all answers in $`๐ฎ^{}(KB)`$ are ground. This means $`๐ฎ^{}(KB)=I_{PB}`$. Hence, by Theorem 4.1 $`๐ฎ^{}(KB)=๐ฎ(KB)`$. $`\mathrm{}`$
We introduce the following principal concept.
###### Definition 4.3
Let $`AA_1,\mathrm{},A_l,true,B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ be a ground instance of the $`k`$-th clause in $`PB`$ such that its body is true in the well-founded model $`WF(PBCB)`$. We call
$$k.AA_1,\mathrm{},A_l$$
(5)
an influence clause.<sup>5</sup><sup>5</sup>5The prefix โ$`k.`$โ would be omitted sometimes for the sake of simplicity. All influence clauses derived from all clauses in $`PB`$ constitute the set of influence clauses of $`KB`$, denoted $`_{clause}(KB)`$.
The following result is immediate from Definition 4.1 and Theorem 4.1.
###### Theorem 4.3
For any influence clause (5), $`A`$ and all $`A_i`$s are random variables in $`๐ฎ(KB)`$.
Influence clauses have the following principal property.
###### Theorem 4.4
For any $`A_i`$ and $`A`$ in HB(PB), $`A_i`$ has a direct influence on $`A`$, which is derived from the $`k`$-th clause in $`PB`$, if and only if there is an influence clause in $`_{clause}(KB)`$ of the form $`k.AA_1,\mathrm{},A_i,\mathrm{},A_l`$.
Proof: ($``$) Assume $`A_i`$ has a direct influence on $`A`$, which is derived from the $`k`$-th clause in $`PB`$. By Definition 4.1, the $`k`$-th clause has a ground instance of the form $`AA_1,\mathrm{},A_i,\mathrm{},A_l,true,B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ such that $`B_1,\mathrm{},B_m,`$ $`\neg C_1,\mathrm{},\neg C_n`$ is true in $`WF(PBCB)`$ and $`\{A_1,\mathrm{},A_i,\mathrm{},A_l\}`$ $`๐ฎ(KB)`$. By Theorem 4.1, $`A_1,\mathrm{},A_i,\mathrm{},A_l`$ is true in $`WF(PBCB)`$. Thus, $`k.AA_1,\mathrm{},A_i,\mathrm{},A_l`$ is an influence clause in $`_{clause}(KB)`$.
($``$) Assume that $`_{clause}(KB)`$ contains an influence clause $`k.AA_1,\mathrm{},`$ $`A_i,\mathrm{},A_l`$. Then the $`k`$-th clause in $`PB`$ has a ground instance of the form $`AA_1,\mathrm{},A_i,\mathrm{},A_l,true,B_1,`$ $`\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ such that its body is true in $`WF(`$ $`PBCB)`$ and (by Theorem 4.3) $`\{A_1,\mathrm{},A_i,\mathrm{},A_l\}๐ฎ(KB)`$. By Definition 4.1, $`A๐ฎ(KB)`$ and $`A_i`$ has a direct influence on $`A`$. $`\mathrm{}`$
The following result is immediate from Theorem 4.4.
###### Corollary 4.5
For any atom $`A`$, $`A`$ is in $`๐ฎ(KB)`$ if and only if there is an influence clause in $`_{clause}(KB)`$ whose head is $`A`$.
Theorem 4.4 shows the significance of influence clauses: they define the same direct influence relation over the same space of random variables as the original Bayesian knowledge base does. Therefore, a Bayesian network can be built directly from $`_{clause}(KB)`$ provided the influence clauses are available.
Observe that to compute the space of random variables (see Algorithm 1), SLG-resolution will construct a proof tree rooted at the goal $`p_i(\stackrel{}{X_i})`$ for each $`1it`$ . For each answer $`A`$ of $`p_i(\stackrel{}{X_i})`$ in $`๐ฎ(KB)`$ there must be a success branch (i.e. a branch starting at the root node and ending at a node marked with success) in the tree that generates the answer. Let $`p_i(.)A_1,\mathrm{},A_l,true,\mathrm{}`$ be the $`k`$-th clause in $`PB`$ that is applied to expand the root goal $`p_i(\stackrel{}{X_i})`$ in the success branch and let $`\theta `$ be the composition of all mgus along the branch. Then $`A=p_i(.)\theta `$ and the body $`A_1,\mathrm{},A_l,true,\mathrm{}`$ is evaluated true, with the mgu $`\theta `$, in $`WF(PBCB)`$ by SLG-resolution. This means that for each $`1jl`$, $`A_j\theta `$ is an answer of $`A_j`$ that is derived by applying SLG-resolution to $`PBCB\{A_j^{}\}`$ where $`A_j^{}`$ is $`A_j`$ or some instance of $`A_j`$. By Theorem 3.1, all $`A_j\theta `$s are ground atoms. Therefore, $`k.p_i(.)\theta A_1\theta ,\mathrm{},A_l\theta `$ is an influence clause. Hence we have the following result.
###### Theorem 4.6
Let $`B_r`$ be a success branch in a proof tree of SLG-resolution, $`p_i(.)A_1,\mathrm{},A_l,`$ $`true,\mathrm{}`$ be the $`k`$-th clause in $`PB`$ that expands the root goal in $`B_r`$, and $`\theta `$ be the composition of all mgus along $`B_r`$. $`B_r`$ produces an influence clause $`k.p_i(.)\theta A_1\theta ,\mathrm{},A_l\theta `$.
Every success branch in a proof tree for a goal in $`GS_0`$ produces an influence clause. The set of influence clauses can then be obtained by collecting all influence clauses from all such proof trees in SLG-resolution.
Algorithm 2: Computing influence clauses.
1. For each goal $`p_i(\stackrel{}{X_i})`$ in $`GS_0`$, compute all answers of $`p_i(\stackrel{}{X_i})`$ by applying SLG-resolution to $`PBCB\{p_i(\stackrel{}{X_i})\}`$ while for each success branch starting at the root goal $`p_i(\stackrel{}{X_i})`$, collecting an influence clause from the branch into $`_{clause}^{}(KB)`$.
2. Return $`_{clause}^{}(KB)`$.
###### Theorem 4.7
Algorithm 2 terminates, yielding a finite set $`_{clause}^{}(KB)=_{clause}(KB)`$.
Proof: That Algorithm 2 terminates is immediate from Theorem 4.2, as except for collecting influence clauses, Algorithm 2 makes the same derivations as Algorithm 1. The termination of Algorithm 2 then implies $`_{clause}^{}(KB)`$ is finite.
By Theorem 4.6, any clause in $`_{clause}^{}(KB)`$ is an influence clause in $`_{clause}(KB)`$. We now prove the converse. Let $`k.AA_1,\mathrm{},A_l`$ be an influence clause in $`_{clause}(KB)`$. Then the $`k`$-th clause in $`PB`$ $`A^{}A_1^{},\mathrm{},A_l^{},true,\mathrm{}`$. has a ground instance of the form $`AA_1,\mathrm{},A_l,true,\mathrm{}`$ whose body is true in $`WF(PBCB)`$. By the completeness of SLG-resolution, there must be a success branch in the proof tree rooted at a goal $`p_i(\stackrel{}{X_i})`$ in $`GS_0`$ where (1) the root goal is expanded by the $`k`$-th clause, (2) the composition of all mgus along the branch is $`\theta `$, and (3) $`AA_1,\mathrm{},A_l,true,\mathrm{}`$ is an instance of $`(A^{}A_1^{},\mathrm{},A_l^{},true,\mathrm{})\theta `$. By Theorem 4.6, $`k.A^{}\theta A_1^{}\theta ,\mathrm{},A_l^{}\theta `$ is an influence clause. Since any influence clause is ground, $`k.A^{}\theta A_1^{}\theta ,\mathrm{},A_l^{}\theta `$ is the same as $`k.AA_1,\mathrm{},A_l`$. This influence clause from the success branch will be collected into $`_{clause}^{}(KB)`$ by Algorithm 2. Thus, any clause in $`_{clause}(KB)`$ is in $`_{clause}^{}(KB)`$. $`\mathrm{}`$
###### Example 4.2 (Example 4.1 continued)
There are two predicate symbols, $`aids`$ and $`contact`$, in the heads of clauses in $`PB_1`$. Let $`GS_0=\{aids(X),contact(Y,Z)\}`$. Algorithm 2 will generate two proof trees rooted at $`aids(X)`$ and $`contact(Y,Z)`$, respectively, as shown in Figures 4 and 5. In the proof trees, a label $`๐_i`$ on an edge indicates that the $`i`$-th clause in $`PB`$ is applied, and the other labels like $`X=p1`$ on an edge show that an answer from a table is applied. Each success branch yields an influence clause. For instance, expanding the root goal $`aids(X)`$ by the 3rd clause produces a child node $`aids(X)`$ (Figure 4). Then applying the answers of $`aids(X)`$ from the table $`๐ฏ_{aids(X)}`$ to the goal of this node leads to three success branches. Applying the mgu $`\theta `$ on each success branch to the 3rd clause yields three influence clauses of the form 3. $`aids(pi)aids(pi)`$ ($`i=1,2,3`$). As a result, we obtain the following set of influence clauses:
| $`_{clause}(KB_1):`$ | 1. $`aids(p1).`$ |
| --- | --- |
| | 2. $`aids(p3).`$ |
| | 3. $`aids(p1)aids(p1).`$ |
| | 3. $`aids(p2)aids(p2).`$ |
| | 3. $`aids(p3)aids(p3).`$ |
| | 4. $`aids(p2)aids(p1),contact(p2,p1).`$ |
| | 4. $`aids(p1)aids(p2),contact(p1,p2).`$ |
| | 5. $`contact(p1,p2).`$ |
| | 6. $`contact(p2,p1).`$ |
For the computational complexity, we observe that the cost of Algorithm 2 is dominated by applying SLG-resolution to evaluate the goals in $`GS_0`$. It has been shown that for a Datalog program $`P`$, the time complexity of computing the well-founded model $`WF(P)`$ is polynomial . More precisely, the time complexity of SLG-resolution is $`O(|P|N^{\mathrm{\Pi }_P+1}logN)`$, where $`|P|`$ is the number of clauses in $`P`$, $`\mathrm{\Pi }_P`$ is the maximum number of literals in the body of a clause, and $`N`$, the number of atoms of predicates in $`P`$ that are not variants of each other, is a polynomial in the number of ground unit clauses in $`P`$ .
$`PBCB`$ is a Datalog program except for the $`member(X_i,DOM_i)`$ predicates (see Definition 3.1). Since each domain $`DOM_i`$ is a finite list of constants, checking if $`X_i`$ is in $`DOM_i`$ takes time linear in the size of $`DOM_i`$. Let $`K_1`$ be the maximum number of $`member(X_i,DOM_i)`$ predicates used in a clause in $`P`$ and $`K_2`$ be the maximum size of a domain $`DOM_i`$. Then the time of handling all $`member(X_i,DOM_i)`$ predicates in a clause is bounded by $`K_1K_2`$. Since each clause in $`P`$ is applied at most $`N`$ times in SLG-resolution, the time of handling all $`member(X_i,DOM_i)`$s in all clauses in $`P`$ is bounded by $`|P|NK_1K_2`$. This is also a polynomial, hence SLG-resolution computes the well-founded model $`WF(PBCB)`$ in polynomial time. Therefore, we have the following result.
###### Theorem 4.8
The time complexity of Algorithm 2 is polynomial.
### 4.2 Probability Distributions Induced by $`KB`$
For any random variable $`A`$, we use $`pa(A)`$ to denote the set of random variables that have direct influences on $`A`$; namely $`pa(A)`$ consists of random variables in the body of all influence clauses whose head is $`A`$. Assume that the probability distribution $`๐(A|pa(A))`$ is available (see Section 5.2). Furthermore, we make the following independence assumption.
###### Assumption 1
For any random variable $`A`$, we assume that given $`pa(A)`$, $`A`$ is probabilistically independent of all random variables in $`๐ฎ(KB)`$ that are not influenced by $`A`$.
We define probability distributions induced by $`KB`$ in terms of whether there are cyclic influences.
###### Definition 4.4
When no cyclic influence occurs, the probability distribution induced by $`KB`$ is $`๐(๐ฎ(KB))`$.
###### Theorem 4.9
$`๐(๐ฎ(KB))=_{A_i๐ฎ(KB)}๐(A_i|pa(A_i))`$ under the independence assumption.
Proof: When no cyclic influence occurs, the random variables in $`๐ฎ(KB)`$ can be arranged in a partial order such that if $`A_i`$ is influenced by $`A_j`$ then $`j>i`$. By the independence assumption, we have $`๐(๐ฎ(KB))`$ $`=`$ $`๐(_{A_i๐ฎ(KB)}A_i)`$ $`=`$ $`๐(A_1|_{i=2}A_i)๐(_{i=2}A_i)`$ $`=`$ $`๐(A_1|pa(A_1))๐(A_2|_{i=3}A_i)๐(_{i=3}A_i)`$ $`=`$ $`\mathrm{}`$ $`=`$ $`_{A_i๐ฎ(KB)}๐(A_i|pa(A_i))`$ $`\mathrm{}`$
When there are cyclic influences, we cannot have a partial order on $`๐ฎ(KB)`$. By Definition 4.2 and Theorem 4.4, any cyclic influence, say โ$`A_1`$ is influenced by itself,โ must be resulted from a set of influence clauses in $`_{clause}(KB)`$ of the form
$`A_1`$ $``$ $`\mathrm{},A_2,\mathrm{}`$
$`A_2`$ $``$ $`\mathrm{},A_3,\mathrm{}`$ (6)
$`\mathrm{}\mathrm{}`$
$`A_n`$ $``$ $`\mathrm{},A_1,\mathrm{}`$
These influence clauses generate a chain (cycle) of direct influences
$`A_1A_2A_3\mathrm{}A_nA_1`$ (7)
which defines a feedback connection. Since a feedback system can be modeled by a two-slice DBN (see Section 2), the above influence clauses represent the same knowledge as the following ones do:
$`A_1`$ $``$ $`\mathrm{},A_2,\mathrm{}`$
$`A_2`$ $``$ $`\mathrm{},A_3,\mathrm{}`$ (8)
$`\mathrm{}\mathrm{}`$
$`A_n`$ $``$ $`\mathrm{},A_{1_{t1}},\mathrm{}`$
Here the $`A_i`$s are state variables and $`A_{1_{t1}}`$ is a state input variable. As a result, $`A_1`$ being influenced by itself becomes $`A_1`$ being influenced by $`A_{1_{t1}}`$. By applying this transformation (from influence clauses (4.2) to (4.2)), we can get rid of all cyclic influences and obtain a generalized set $`_{clause}(KB)_g`$ of influence clauses from $`_{clause}(KB)`$.<sup>6</sup><sup>6</sup>6Depending on starting from which influence clause to generate an influence cycle, a different generalized set containing different state input variables would be obtained. All of them are equivalent in the sense that they define the same feedbacks (cycles of direct influences) and can be unrolled into the same stationary DBN.
###### Example 4.3 (Example 4.2 continued)
$`_{clause}(KB_1)`$ can be transformed to the following generalized set of influence clauses by introducing three state input variables $`aids(p1)_{t1}`$, $`aids(p2)_{t1}`$ and $`aids(p3)_{t1}`$.
| $`_{clause}(KB_1)_g:`$ | 1. $`aids(p1).`$ |
| --- | --- |
| | 2. $`aids(p3).`$ |
| | 3. $`aids(p1)aids(p1)_{t1}.`$ |
| | 3. $`aids(p2)aids(p2)_{t1}.`$ |
| | 3. $`aids(p3)aids(p3)_{t1}.`$ |
| | 4. $`aids(p2)aids(p1)_{t1},contact(p2,p1).`$ |
| | 4. $`aids(p1)aids(p2),contact(p1,p2).`$ |
| | 5. $`contact(p1,p2).`$ |
| | 6. $`contact(p2,p1).`$ |
When there is no cyclic influence, $`KB`$ is a non-temporal model, represented by $`_{clause}(KB)`$. When cyclic influences occur, however, $`KB`$ becomes a temporal model, represented by $`_{clause}(KB)_g`$. Let $`๐ฎ(KB)_g`$ be $`๐ฎ(KB)`$ plus all state input variables introduced in $`_{clause}(KB)_g`$.
###### Definition 4.5
When there are cyclic influences, the probability distribution induced by $`KB`$ is $`๐(๐ฎ(KB)_g)`$.
By extending the independence assumption from $`๐ฎ(KB)`$ to $`๐ฎ(KB)_g`$, we obtain the following result.
###### Theorem 4.10
$`๐(๐ฎ(KB)_g)=_{A_i๐ฎ(KB)_g}๐(A_i|pa(A_i))`$ under the independence assumption.
Proof: Since $`_{clause}(KB)_g`$ produces no cyclic influences, the random variables in $`๐ฎ(KB)_g`$ can be arranged in a partial order such that if $`A_i`$ is influenced by $`A_j`$ then $`j>i`$. The proof then proceeds in the same way as that of Theorem 4.9. $`\mathrm{}`$
## 5 Building a Bayesian Network from a Bayesian Knowledge Base
### 5.1 Building a Two-Slice DBN Structure
From a Bayesian knowledge base $`KB`$, we can derive a set of influence clauses $`_{clause}(KB)`$, which defines the same direct influence relation over the same space $`๐ฎ(KB)`$ of random variables as $`PBCB`$ does (see Theorem 4.4). Therefore, given a probabilistic query together with some evidences, we can depict a network structure from $`_{clause}(KB)`$, which covers the random variables in the query and evidences, by backward-chaining the related random variables via the direct influence relation.
Let $`Q`$ be a probabilistic query and $`E`$ a set of evidences, where all random variables come from $`๐ฎ(KB)`$ (i.e., they are heads of some influence clauses in $`_{clause}(KB)`$). Let $`TOP`$ consist of these random variables. An influence network of $`Q`$ and $`E`$,<sup>7</sup><sup>7</sup>7Note the differences between influence networks and influence diagrams. Influence diagrams (also known as decision networks) are a formalism introduced in decision theory that extends Bayesian networks by incorporating actions and utilities . denoted $`_{net}(KB)_{Q,E}`$, is constructed from $`_{clause}(KB)`$ using the following algorithm.
Algorithm 3: Building an influence network.
1. Initially, $`_{net}(KB)_{Q,E}`$ has all random variables in $`TOP`$ as nodes.
2. Remove the first random variable $`A`$ from $`TOP`$. For each influence clause in $`_{clause}(KB)`$ of the form $`k.AA_1,\mathrm{},A_l`$, if $`l=0`$ then add to $`_{net}(KB)_{Q,E}`$ an edge $`A\stackrel{k}{}`$. Otherwise, for each $`A_i`$ in the body
1. If $`A_i`$ is not in $`_{net}(KB)_{Q,E}`$ then add $`A_i`$ to $`_{net}(KB)_{Q,E}`$ as a new node and add it to the end of $`TOP`$.
2. Add to $`_{net}(KB)_{Q,E}`$ an edge $`A\stackrel{k}{}A_i`$.
3. Repeat step 2 until $`TOP`$ becomes empty.
4. Return $`_{net}(KB)_{Q,E}`$.
###### Example 5.1 (Example 4.2 continued)
To build an influence network from $`KB_1`$ that covers $`aids(p1)`$, $`aids(p2)`$ and $`aids(p3)`$, we apply Algorithm 3 to $`_{clause}(`$ $`KB_1)`$ while letting $`TOP=\{aids(p1),aids(p2),aids(p3)\}`$. It generates an influence network $`_{net}(KB_1)_{Q,E}`$ as shown in Figure 6.
An influence network is a graphical representation for influence clauses. This claim is supported by the following properties of influence networks.
###### Theorem 5.1
For any $`A_i,A_j`$ in $`_{net}(KB)_{Q,E}`$, $`A_j`$ is a parent node of $`A_i`$, connected via an edge $`A_i\stackrel{k}{}A_j`$, if and only if there is an influence clause of the form $`k.A_iA_1,\mathrm{},A_j,\mathrm{},A_l`$ in $`_{clause}(KB)`$.
Proof: First note that termination of Algorithm 3 is guaranteed by the fact that any random variable in $`๐ฎ(KB)`$ will be added to $`TOP`$ no more than one time (line 2a). Let $`A_i,A_j`$ be nodes in $`_{net}(KB)_{Q,E}`$. If $`A_j`$ is a parent node of $`A_i`$, connected via an edge $`A_i\stackrel{k}{}A_j`$, this edge must be added at line 2b, due to applying an influence clause in $`_{clause}(KB)`$ of the form $`k.A_iA_1,\mathrm{},A_j,\mathrm{},A_l`$ (line 2). Conversely, if $`_{clause}(KB)`$ contains such an influence clause, it must be applied at line 2, with edges of the form $`A_i\stackrel{k}{}A_j`$ added to the network at line 2b. $`\mathrm{}`$
###### Theorem 5.2
For any $`A_i,A_j`$ in $`_{net}(KB)_{Q,E}`$, $`A_i`$ is a descendant node of $`A_j`$ if and only if $`A_i`$ is influenced by $`A_j`$.
Proof: Assume $`A_i`$ is a descendant node of $`A_j`$, with a path
$$A_i\stackrel{k}{}B_1\stackrel{k_1}{}\mathrm{}B_m\stackrel{k_m}{}A_j$$
(9)
By Theorem 5.1, $`_{clause}(KB)`$ must contain the following influence clauses
$`k.`$ $`A_i\mathrm{},B_1,\mathrm{}`$
$`k_1.`$ $`B_1\mathrm{},B_2,\mathrm{}`$ (10)
$`\mathrm{}\mathrm{}`$
$`k_m.`$ $`B_m\mathrm{},A_j,\mathrm{}`$
By Theorem 4.4 and Definition 4.2, $`A_i`$ is influenced by $`A_j`$. Conversely, if $`A_i`$ is influenced by $`A_j`$, there must be a chain of influence clauses of the form as above. Since $`A_i,A_j`$ are in $`_{net}(KB)_{Q,E}`$, by Theorem 5.1 there must be a path of form (9) in the network. $`\mathrm{}`$
###### Theorem 5.3
Let $`V`$ be the set of nodes in $`_{net}(KB)_{Q,E}`$ and let $`W=\{A_j๐ฎ(KB)|`$ for some $`A_iTOP`$, $`A_i`$ is influenced by $`A_j\}`$. $`V=TOPW`$.<sup>8</sup><sup>8</sup>8This result suggests that an influence network is similar to a supporting network introduced in .
Proof: That $`_{net}(KB)_{Q,E}`$ covers all random variables in $`TOP`$ follows from line 1 of Algorithm 3. We first prove that if $`A_jW`$ then $`A_jV`$. Assume $`A_jW`$. There must be a chain of influence clauses of form (5.1) with $`A_iTOP`$. In this case, $`B_1,B_2,\mathrm{},B_m,A_j`$ will be recursively added to the network (line 2). Thus $`A_jV`$. We then prove that if $`A_jV`$ and $`A_jTOP`$ then $`A_jW`$. Assume $`A_jV`$ and $`A_jTOP`$. $`A_j`$ must not be added to $`V`$ at line 1. Instead, it is added to $`V`$ at line 2a. This means that for some $`A_iTOP`$, $`A_i`$ is a descendant of $`A_j`$. By Theorem 5.2, $`A_i`$ is influenced by $`A_j`$. Hence $`A_jW`$. $`\mathrm{}`$
Theorem 4.9 shows that the probability distribution induced by $`KB`$ can be computed over $`_{clause}(KB)`$. Let $`_{net}(KB)_{๐ฎ(KB)}`$ denote an influence network that covers all random variables in $`๐ฎ(KB)`$. We show that the same distribution can be computed over $`_{net}(KB)_{๐ฎ(KB)}`$. For any node $`A_i`$ in $`_{net}(KB)_{๐ฎ(KB)}`$, let $`parents(A_i)`$ denote the set of parent nodes of $`A_i`$ in the network. Observe the following facts: First, by Theorem 5.1, $`parents(A_i)=pa(A_i)`$. Second, by Theorem 5.2, $`A_i`$ is a descendant node of $`A_j`$ in $`_{net}(KB)_{๐ฎ(KB)}`$ if and only if $`A_i`$ is influenced by $`A_j`$ in $`_{clause}(KB)`$. This means that the independence assumption (Assumption 1) applies to $`_{net}(KB)_{๐ฎ(KB)}`$ as well, and that $`_{clause}(KB)`$ produces a cycle of direct influences if and only if $`_{net}(KB)_{๐ฎ(KB)}`$ contains the same (direct) loop. Combining these facts leads to the following immediate result.
###### Theorem 5.4
When no cyclic influence occurs, the probability distribution induced by $`KB`$ can be computed over $`_{net}(KB)_{๐ฎ(KB)}`$. That is, $`๐(๐ฎ(KB))`$ $`=`$ $`_{A_i๐ฎ(KB)}๐(A_i|pa(A_i))`$ $`=`$ $`_{A_i๐ฎ(KB)}๐(A_i|parents(A_i))`$ under the independence assumption.
Theorem 5.4 implies that an influence network without loops is a Bayesian network structure. Let us consider influence networks with loops. By Theorem 5.2, loops in an influence network are generated from recursive influence clauses of form (4.2) and thus they depict feedback connections of form (7). This means that an influence network with loops can be converted into a two-slice DBN, simply by converting each loop of the form
into a two-slice DBN path
$$A_1\stackrel{k_1}{}A_2\stackrel{k_2}{}\mathrm{}\stackrel{k_{n1}}{}A_n\stackrel{k_n}{}A_{1_{t1}}$$
by introducing a state input node $`A_{1_{t1}}`$.
As illustrated in Section 2, a two-slice DBN is a snapshot of a stationary DBN across any two time slices, which can be obtained by traversing the stationary DBN from a set of state variables backward to the same set of state variables (i.e., state input nodes). This process corresponds to generating an influence network $`_{net}(KB)_{Q,E}`$ from $`_{clause}(KB)`$ incrementally (adding nodes and edges one at a time) while wrapping up loop nodes with state input nodes. This leads to the following algorithm for building a two-slice DBN structure, $`2๐ฎ_{net}(KB)_{Q,E}`$, directly from $`_{clause}(KB)`$, where $`Q`$, $`E`$ and $`TOP`$ are the same as defined in Algorithm 3.
Algorithm 4: Building a two-slice DBN structure.
1. Initially, $`2๐ฎ_{net}(KB)_{Q,E}`$ has all random variables in $`TOP`$ as nodes.
2. Remove the first random variable $`A`$ from $`TOP`$. For each influence clause in $`_{clause}(KB)`$ of the form $`k.AA_1,\mathrm{},A_l`$, if $`l=0`$ then add to $`2๐ฎ_{net}(KB)_{Q,E}`$ an edge $`A\stackrel{k}{}`$. Otherwise, for each $`A_i`$ in the body
1. If $`A_i`$ is not in $`2๐ฎ_{net}(KB)_{Q,E}`$ then add $`A_i`$ to $`2๐ฎ_{net}(KB)_{Q,E}`$ as a new node and add it to the end of $`TOP`$.
2. If adding $`A\stackrel{k}{}A_i`$ to $`2๐ฎ_{net}(KB)_{Q,E}`$ produces a loop, then add to $`2๐ฎ_{net}(KB)_{Q,E}`$ a node $`A_{i_{t1}}`$ and an edge $`A\stackrel{k}{}A_{i_{t1}}`$, else add an edge $`A\stackrel{k}{}A_i`$ to $`2๐ฎ_{net}(KB)_{Q,E}`$.
3. Repeat step 2 until $`TOP`$ becomes empty.
4. Return $`2๐ฎ_{net}(KB)_{Q,E}`$.
###### Example 5.2 (Example 5.1 continued)
To build a two-slice DBN structure from $`KB_1`$ that covers $`aids(p1)`$, $`aids(p2)`$ and $`aids(p3)`$, we apply Algorithm 4 to $`_{clause}(`$ $`KB_1)`$ while letting $`TOP=\{aids(p1),aids(p2),aids(p3)\}`$. It generates $`2๐ฎ_{net}(KB_1)_{Q,E}`$ as shown in Figure 7. Note that loops are cut by introducing three state input nodes $`aids(p1)_{t1}`$, $`aids(p2)_{t1}`$ and $`aids(p3)_{t1}`$. The two-slice DBN structure concisely depicts a feedback system where the feedback connections are as shown in Figure 8.
Algorithm 4 is Algorithm 3 enhanced with a mechanism for cutting loops (item 2b), i.e. when adding the current edge $`A\stackrel{k}{}A_i`$ to the network forms a loop, we replace it with an edge $`A\stackrel{k}{}A_{i_{t1}}`$, where $`A_{i_{t1}}`$ is a state input node. This is a process of transforming influence clauses (4.2) to (4.2). Therefore, $`2๐ฎ_{net}(KB)_{Q,E}`$ can be viewed as an influence network built from a generalized set $`_{clause}(KB)_g`$ of influence clauses. Let $`๐ฎ(KB)_g`$ be the set of random variables in $`_{clause}(KB)_g`$, as defined in Theorem 4.10. Let $`2๐ฎ_{net}(KB)_{๐ฎ(KB)}`$ denote a two-slice DBN structure (produced by applying Algorithm 4) that covers all random variables in $`๐ฎ(KB)_g`$. We then have the following immediate result from Theorem 5.4.
###### Theorem 5.5
When $`_{clause}(KB)`$ produces cyclic influences, the probability distribution induced by $`KB`$ can be computed over $`2๐ฎ_{net}(KB)_{๐ฎ(KB)}`$. That is, $`๐(๐ฎ(KB)_g)`$ $`=`$ $`_{A_i๐ฎ(KB)_g}๐(A_i`$ $`|pa(A_i))`$ $`=`$ $`_{A_i๐ฎ(KB)_g}๐(A_i|parents(A_i))`$ under the independence assumption.
###### Remark 5.1
Note that Algorithm 4 produces a DBN structure without using any explicit time parameters. It only requires the user to specify, via the query and evidences, what random variables are necessarily included in the network. Algorithm 4 builds a two-slice DBN structure for any given query and evidences whose random variables are heads of some influence clauses in $`_{clause}(KB)`$. When no query and evidences are provided, we may apply Algorithm 4 to build a complete two-slice DBN structure, $`2๐ฎ_{net}(KB)_{๐ฎ(KB)}`$, which covers the space $`๐ฎ(KB)`$ of random variables, by letting $`TOP`$ consist of all heads of influence clauses in $`_{clause}(KB)`$. This is a very useful feature, as in many situations the user may not be able to present the right queries unless a Bayesian network structure is shown.
Also note that when there is no cyclic influence, Algorithm 4 becomes Algorithm 3 and thus it builds a regular Bayesian network structure.
### 5.2 Building CPTs
After a Bayesian network structure $`2๐ฎ_{net}(KB)_{Q,E}`$ has been constructed from a Bayesian knowledge base $`KB`$, we associate each (non-state-input) node $`A`$ in the network with a CPT. There are three cases. (1) If $`A`$ (as a head) only has unit clauses in $`_{clause}(KB)`$, we build from the unit clauses a prior CPT for $`A`$ as its prior probability distribution. (2) If $`A`$ only has non-unit clauses in $`_{clause}(KB)`$, we build from the clauses a posterior CPT for $`A`$ as its posterior probability distribution. (3) Otherwise, we prepare for $`A`$ both a prior CPT (from the unit clauses) and a posterior CPT (from the non-unit clauses). In this case, $`A`$ is attached with the posterior CPT; the prior CPT for $`A`$ would be used, if $`A`$ is a state variable, as the probability distribution of $`A`$ in time slice 0 (only in the case that a two-slice DBN is unrolled into a stationary DBN starting with time slice 0).
Assume that the parent nodes of $`A`$ are derived from $`n`$ ($`n1`$) different influence clauses in $`_{clause}(KB)`$. Suppose these clauses share the following CPTs in $`T_x`$: $`๐(A_1|B_1^1,\mathrm{},B_{m_1}^1)`$, โฆ, and $`๐(A_n|B_1^n,\mathrm{},B_{m_n}^n)`$. (Recall that an influence clause prefixed with a number $`k`$ shares the CPT attached to the $`k`$-th clause in $`PB`$.) Then the CPT for $`A`$ is computed by combining the $`n`$ CPTs in terms of the combination rule $`CR`$ specified in Definition 3.1.
###### Example 5.3 (Example 5.2 continued)
Let CPT<sub>i</sub> denote the CPT attached to the $`i`$-th clause in $`PB_1`$. Consider the random variables in $`2๐ฎ_{net}(KB_1)_{Q,E}`$. Since $`aids(p1)`$ has three parent nodes, derived from the 3rd and 4-th clause in $`PB_1`$ respectively, the posterior CPT for $`aids(p1)`$ is computed by combining CPT<sub>3</sub> and CPT<sub>4</sub>. $`aids(p1)`$ has also a prior CPT, CPT<sub>1</sub>, derived from the 1st clause in $`PB_1`$. For the same reason, the posterior CPT for $`aids(p2)`$ is computed by combining CPT<sub>3</sub> and CPT<sub>4</sub>. The posterior CPT for $`aids(p3)`$ is CPT<sub>3</sub> and its prior CPT is CPT<sub>2</sub>. $`contact(p1,p2)`$ and $`contact(p2,p1)`$ have only prior CPTs, namely CPT<sub>5</sub> and CPT<sub>6</sub>. Note that state input nodes, $`aids(p1)_{t1}`$, $`aids(p2)_{t1}`$ and $`aids(p3)_{t1}`$, do not need to have a CPT; they will be expanded, during the process of unrolling the two-slice DBN into a stationary DBN, to cover the time slices involved in the given query and evidence nodes. If the resulting stationary DBN starts with time slice 0, the prior CPTs, CPT$`_{aids(p1)_0}`$ and CPT$`_{aids(p3)_0}`$, for $`aids(p1)`$ and $`aids(p3)`$ are used as the probability distributions of $`aids(p1)_0`$ and $`aids(p3)_0`$.
Note that $`aids(p2)`$ is a state variable, but there is no unit influence clause available to build a prior CPT for it. We have two ways to derive a prior CPT, CPT$`_{aids(p2)_0}`$, for $`aids(p2)`$ from some existing CPTs. (1) CPT$`_{aids(p2)_0}`$ comes from averaging CPT$`_{aids(p1)_0}`$ and CPT$`_{aids(p3)_0}`$. For instance, let the probability of $`aids(p1)=yes`$ be $`0.7`$ in CPT$`_{aids(p1)_0}`$ and the probability of $`aids(p3)=yes`$ be $`0.74`$ in CPT$`_{aids(p3)_0}`$. Then the probability of $`aids(p2)=yes`$ is $`(0.7+0.74)/2=0.72`$ in CPT$`_{aids(p2)_0}`$. (2) CPT$`_{aids(p2)_0}`$ comes from averaging the posterior probability distributions of $`aids(p2)`$. For instance, let $`\{0.9,0.7,0.4,0.8\}`$ be the posterior probabilities of $`aids(p2)=yes`$ in the posterior CPT for $`aids(p2)`$. Then the probability of $`aids(p2)=yes`$ is $`(0.9+0.7+0.4+0.8)/4=0.7`$ in CPT$`_{aids(p2)_0}`$.
## 6 Related Work
A recent overview of existing representational frameworks that combine probabilistic reasoning with logic (i.e. logic-based approaches) or with relational representations (i.e. non-logic-based approaches) is given by De Raedt and Kersting . Typical non-logic-based approaches include probabilistic relational models (PRM), which are based on the entity-relationship (or object-oriented) model , and relational Markov networks, which combine Markov networks and SQL-like queries . Representative logic-based approaches include frameworks based on the KBMC (Knowledge-Based Model Construction) idea , stochastic logic programs (SLP) based on stochastic context-free grammars , parameterized logic programs based on distribution semantics (PRISM) , and more. Most recently, a unifying framework, called Markov logic, has been proposed by Domingos and Richardson . Markov logic subsumes first-order logic and Markov networks. Since our work follows the KBMC idea focusing on how to build a Bayesian network directly from a logic program, it is closely related to three representative existing PLP approaches: the context-sensitive PLP developed by Haddawy and Ngo , Bayesian logic programming proposed by Kersting and Raedt , and the time parameter-based approach presented by Glesner and Koller . In this section, we make a detailed comparison of our work with the three closely related approaches.
### 6.1 Comparison with the Context-Sensitive PLP Approach
The core of the context-sensitive PLP is a probabilistic knowledge base (PKB). In order to see the main differences from our Bayesian knowledge base (BKB), we reformulate its definition here.
###### Definition 6.1
A probabilistic knowledge base is a four tuple $`<`$$`PD,PB,CB,CR`$$`>`$, where
* $`PD`$ defines a set of probabilistic predicates (p-predicates) of the form $`p(T_1,\mathrm{},`$ $`T_m,V)`$ where all arguments $`T_i`$s are typed with a finite domain and the last argument $`V`$ takes on values from a probabilistic domain $`DOM_p`$.
* $`PB`$ consists of probabilistic rules of the form
$$P(A_0|A_1,\mathrm{},A_l)=\alpha B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n$$
(11)
where $`0\alpha 1`$, the $`A_i`$s are p-predicates, and the $`B_j`$s and $`C_k`$s are context predicates (c-predicates) defined in $`CB`$.
* $`CB`$ is a logic program, and both $`PB`$ and $`CB`$ are acyclic.
* $`CR`$ is a combination rule.
In a probabilistic rule (11), each p-predicate $`A_i`$ is of the form $`q(t_1,\mathrm{},t_m,v)`$, which simulates an equation $`q(t_1,\mathrm{},t_m)=v`$ with $`v`$ being a value from the probabilistic domain of $`q(t_1,\mathrm{},t_m)`$. For instance, let $`D_{color}=\{red,green,blue\}`$ be the probabilistic domain of $`color(X)`$, then the p-predicate $`color(X,red)`$ simulates $`color(X)=red`$, meaning that the color of $`X`$ is $`red`$. The left-hand side $`P(A_0|A_1,\mathrm{},A_l)=\alpha `$ expresses that the probability of $`A_0`$ conditioned on $`A_1,\mathrm{},A_l`$ is $`\alpha `$. The right-hand side $`B_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n`$ is the context of the rule where the $`B_j`$s and $`C_k`$s are c-predicates. Note that the sets of p-predicate and c-predicate symbols are disjoint. A separate logic program $`CB`$ is used to evaluate the context of a probabilistic rule. As a whole, the above probabilistic rule states that for each of its (Herbrand) ground instances
$$P(A_0^{}|A_1^{},\mathrm{},A_l^{})=\alpha B_1^{},\mathrm{},B_m^{},\neg C_1^{},\mathrm{},\neg C_n^{}$$
if the context $`B_1^{},\mathrm{},B_m^{},\neg C_1^{},\mathrm{},\neg C_n^{}`$ is true in $`CB`$ under the program completion semantics, the probability of $`A_0^{}`$ conditioned on $`A_1^{},\mathrm{},A_l^{}`$ is $`\alpha `$.
PKB and BKB have the following important differences.
First, probabilistic rules of form (11) in PKB contain both logic representation (right-hand side) and probabilistic representation (left-hand side) and thus are not logic clauses. The logic part and the probabilistic part of a rule are separately computed against $`CB`$ and $`PB`$, respectively. In contrast, BKB uses logic clauses of form (3.1), which naturally integrate the direct influence information, the context and the type constraints. These logic clauses are evaluated against a single logic program $`PBCB`$, while the probabilistic information is collected separately in $`T_x`$.
Second, logic reasoning in PKB relies on the program completion semantics and is carried out by applying SLDNF-resolution. But in BKB, logic inferences are based on the well-founded semantics and are performed by applying SLG-resolution. The well-founded semantics resolves the problem of inconsistency with the program completion semantics, while SLG-resolution eliminates the problem of infinite loops with SLDNF-resolution. Note that the key significance of BKB using the well-founded semantics lies in the fact that a unique set of influence clauses can be derived, which lays a basis on which both the declarative and procedural semantics for BKB are developed.
Third, most importantly PKB has no mechanism for handling cyclic influences. In PKB, cyclic influences are defined to be inconsistent (see Definition 9 of the paper ) and thus are excluded (PKB excludes cyclic influences by requiring its programs be acyclic). In BKB, however, cyclic influences are interpreted as feedbacks, thus implying a time sequence. This allows us to derive a stationary DBN from a logic program with recursive loops.
Recently, Fierens, Blockeel, Ramon and Bruynooghe introduced logical Bayesian networks (LBN). LBN is similar to PKB except that it separates logical and probabilistic information. That is, LBN converts rules of form (11) into the form
$$A_0|A_1,\mathrm{},A_lB_1,\mathrm{},B_m,\neg C_1,\mathrm{},\neg C_n$$
where the $`A_i`$s are p-predicates with the last argument $`V`$ removed, and the $`B_j`$s and $`C_k`$s are c-predicates defined in $`CB`$. This is not a standard clause of form (3) as defined in logic programming . Like PKB, LBN differs from BKB in the following: (1) it has no mechanism for handling cyclic influences (see Section 3.2 of the paper ), and (2) although the well-founded semantics is also used for the logic contexts, neither declarative nor procedural semantics for LBN has been formally developed.
### 6.2 Comparison with Bayesian Logic Programming
Building on Ngo and Haddawyโs work, Kersting and De Raedt introduce the framework of Bayesian logic programs. A Bayesian logic program (BLP) is a triple $`<`$$`P,T_x,CR`$$`>`$ where $`P`$ is a well-defined logic program, $`T_x`$ consists of CPTs associated with each clause in $`P`$, and $`CR`$ is a combination rule. A distinct feature of BLP over PKB is its separation of probabilistic information ($`T_x`$) from logic clauses ($`P`$). According to , we understand that a well-defined logic program is an acyclic positive logic program satisfying the range restriction.<sup>9</sup><sup>9</sup>9A logic program is said to be range-restricted if all variables appearing in the head of a clause appear in the body of the clause. For instance, a logic program containing clauses like $`r(X)r(X)`$ (cyclic) or $`r(X)s(Y)`$ (not range-restricted) is not well-defined. BLP relies on the least Herbrand model semantics and applies SLD-resolution to make backward-chaining inferences.
BLP has two important differences from BKB. First, it applies only to positive logic programs. Due to this, it cannot handle contexts with negated atoms. (In fact, no contexts are considered in BLP.) Second, it does not allow cyclic influences. BKB can be viewed as an extension of BLP with mechanisms for handling contexts and cyclic influences in terms of the well-founded semantics. Such an extension is clearly nontrivial.
### 6.3 Comparison with the Time Parameter-Based Approach
The time parameter-based framework (TPF) proposed by Glesner and Koller is also a triple $`<`$$`P,T_x,CR`$$`>`$, where $`CR`$ is a combination rule, $`T_x`$ is a set of CPTs that are represented as decision trees, and $`P`$ is a logic program with the property that each predicate contains a time parameter and that in each clause the time argument in the head is at least one time step later than the time arguments in the body. This framework is implemented in Prolog, i.e. clauses are represented as Prolog rules and goals are evaluated applying SLDNF-resolution. Glesner and Koller state: โโฆ In principle, this free variable $`Y`$ can be instantiated with every domain element. (This is the approach taken in our implementation.)โ By this we understand that they consider typed logic programs with finite domains.
We observe the following major differences between TPF and BKB. First, TPF is a temporal model and its logic programs contain a time argument for every predicate. It always builds a DBN from a logic program even if there is no cyclic influence. In contrast, logic programs in BKB contain no time parameters. When there is no cyclic influence, BKB builds a regular Bayesian network from a logic program (in this case, BKB serves as a non-temporal model); when cyclic influences occur, it builds a stationary DBN, represented by a two-slice DBN (in this case, BKB serves as a special temporal model). Second, TPF uses time steps to describe direct influences (in the way that for any $`A`$ and $`B`$ such that $`B`$ has a direct influence on $`A`$, the time argument in $`B`$ is at least one time step earlier than that in $`A`$), while BKB uses time slices (implied by recursive loops of form (1)) to model cycles of direct influences (feedbacks). Time-steps based frameworks like TPF are suitable to model flexible DBNs, whereas time-slices based approaches like BKB apply to stationary DBNs. Third, most importantly TPF avoids recursive loops by introducing time parameters to enforce acyclicity of a logic program. A serious problem with this method is that it may lose and/or produce wrong answers to some queries. To explain this, let $`P`$ be a logic program and $`P_t`$ be $`P`$ with additional time arguments added to each predicate (as in TPF). If the transformation from $`P`$ to $`P_t`$ is correct, it must hold that for any query $`p(.)`$ over $`P`$, an appropriate time argument $`N=0,1,2,\mathrm{}`$ can be determined such that the query $`p(.,N)`$ over $`P_t`$ has the same set of answers as $`p(.)`$ over $`P`$ when the time arguments in the answers are ignored. It turns out, however, that this condition does not hold in general cases. Note that finding an appropriate $`N`$ for a query $`p(.)`$ such that evaluating $`p(.,N)`$ over $`P_t`$ (applying SLDNF-resolution) yields the same set of answers as evaluating $`p(.)`$ over $`P`$ corresponds to finding an appropriate depth-bound $`M`$ such that cutting all SLDNF-derivations for the query $`p(.)`$ at depth $`M`$ does not lose any answers to $`p(.)`$. The latter is the well-known loop problem in logic programming . Since the loop problem is undecidable in general, there is no algorithm for automatically determining such a depth-bound $`M`$ (rep. a time argument $`N`$) for an arbitrary query $`p(.)`$ . We further illustrate this claim using the following example.
###### Example 6.1
The following logic program defines a $`path`$ relation; i.e. there is a path from $`X`$ to $`Y`$ if either there is an edge from $`X`$ to $`Y`$ or for some $`Z`$, there is a path from $`X`$ to $`Z`$ and an edge from $`Z`$ to $`Y`$.
| $`P:`$ | 1. $`e(s,b1).`$ |
| --- | --- |
| | 2. $`e(b1,b2).`$ |
| | $`\mathrm{}\mathrm{}`$ |
| | 99. $`e(b98,b99).`$ |
| | 100. $`e(b99,g).`$ |
| | 101. $`path(X,Y)e(X,Y).`$ |
| | 102. $`path(X,Y)path(X,Z),e(Z,Y).`$ |
To avoid recursive loops, TPF may transform $`P`$ into the following program.
| $`P_t:`$ | 1. $`e(s,b1,0).`$ |
| --- | --- |
| | 2. $`e(b1,b2,0).`$ |
| | $`\mathrm{}\mathrm{}`$ |
| | 99. $`e(b98,b99,0).`$ |
| | 100. $`e(b99,g,0).`$ |
| | 101. $`e(X,Y,T1)T2=T11,e(X,Y,T2).`$ |
| | 102. $`path(X,Y,T1)T2=T11,e(X,Y,T2).`$ |
| | 103. $`path(X,Y,T1)T2=T11,path(X,Z,T2),e(Z,Y,T2).`$ |
$`P_t`$ looks more complicated than $`P`$. In addition to having time arguments and time formulas, it has a new clause, the 101st clause, formulating that $`e(X,Y)`$ being true at present implies it is true in the future.
Let us see how to check if there is a path from $`s`$ to $`g`$. In the original program $`P`$, we simply pose a query $`\mathrm{?}path(s,g)`$. In the transformed program $`P_t`$, however, we have to determine a specific time parameter $`N`$ and then pose a query $`\mathrm{?}path(s,g,N)`$, such that evaluating $`path(s,g)`$ over $`P`$ yields the same answer as evaluating $`path(s,g,N)`$ over $`P_t`$. Interested readers can practice this query evaluation using different values for $`N`$. The answer to $`path(s,g)`$ over $`P`$ is $`yes.`$ However, we would get an answer $`no`$ to the query $`path(s,g,N)`$ over $`P_t`$ if we choose any $`N<100`$.
## 7 Conclusions and Discussion
We have developed a novel theoretical framework for deriving a stationary DBN from a logic program with recursive loops. We observed that recursive loops in a logic program imply a time sequence and thus can be used to model a stationary DBN without using explicit time parameters. We introduced a Bayesian knowledge base with logic clauses of form (3.1). These logic clauses naturally integrate the direct influence information, the context and the type constraints, and are evaluated under the well-founded semantics. We established a declarative semantics for a Bayesian knowledge base and developed algorithms that build a two-slice DBN from a Bayesian knowledge base.
We emphasize the following three points.
1. Recursive loops (cyclic influences) and recursion through negation are unavoidable in modeling real-world domains, thus the well-founded semantics together with its top-down inference procedures is well suitable for the PLP application.
2. Recursive loops define feedbacks, thus implying a time sequence. This allows us to derive a two-slice DBN from a logic program containing no time parameters. We point out, however, that the user is never required to provide any time parameters during the process of constructing such a two-slice DBN. A Bayesian knowledge base defines a unique space of random variables and a unique set of influence clauses, whether it contains recursive loops or not. From the viewpoint of logic, these random variables are ground atoms in the Herbrand base; their truth values are determined by the well-founded model and will never change over time.<sup>10</sup><sup>10</sup>10 However, from the viewpoint of Bayesian networks the probabilistic values of these random variables (i.e. values from their probabilistic domains) may change over time. Therefore, a Bayesian network is built over these random variables, independently of any time factors (if any). Once a two-slice DBN has been built, the time intervals over it would become clearly specified, thus the user can present queries and evidences over the DBN using time parameters at his/her convenience.
3. Enforcing acyclicity of a logic program by introducing time parameters is not an effective way to handle recursive loops. Firstly, such a method transforms the original non-temporal logic program into a more complicated temporal program and builds a dynamic Bayesian network from the transformed program even if there exist no cyclic influences (in this case, there is no state variable and the original program defines a regular Bayesian network). Secondly, it relies on time steps to define (individual) direct influences, but recursive loops need time slices (intervals) to model cycles of direct influences (feedbacks). Finally, to pose a query over the transformed program, an appropriate time parameter must be specified. As illustrated in Example 6.1, there is no algorithm for automatically determining such a time parameter for an arbitrary query.
Promising future work includes (1) developing algorithms for learning BKB clauses together with their CPTs from data and (2) applying BKB to model large real-world problems. We intend to build a large Bayesian knowledge base for traditional Chinese medicine, where we already have both a large volume of collected diagnostic rules and a massive repository of diagnostic cases.
## Acknowledgements
We are grateful to several anonymous referees for their constructive comments, which greatly helped us improve the presentation.
|
warning/0506/cond-mat0506769.html
|
ar5iv
|
text
|
# How to Compute Loop Corrections to Bethe Approximation
## 1 Introduction
Mean field approximations are the among the most frequently used tools in Statistical Physics. Among them, Bethe approximation (BA) allows to treat with reasonable accuracy a large variety of lattice models. Recently it has been successfully applied (in an algorithmic form) to problems of inference , communications , and combinatorial optimization . Often in these cases the underlying lattice has few or no short loops, and BA (which is exact on trees) can become exact in the thermodynamic limit.
BA can be systematically improved using Kikuchi or cluster variational methods (CVM). These approaches take into account โexactlyโ of correlations up to some finite range $`r`$ and their complexity grows exponentially with $`r`$. Because of this feature, they are unsuited for understanding the effect of long length scale fluctuations. Furthermore, in lattices without short loops, no improvement is obtained unless $`r`$ is very large (which is of course unfeasible).
For models on $`d`$-dimensional lattices, mean field can be also regarded as the zeroth order term in a $`1/d`$ expansion.<sup>2</sup><sup>2</sup>2Generally, BA takes into account exactly also the first $`1/d`$ correction. Such an expansion is however close in spirit to CVM, in that it keeps into account only the effect of short loops in the lattice .
In the field-theoretical setting , mean field approximation is usually derived by retaining only tree-level Feynman diagrams. The usual loop expansion improves systematically over such an approximation by taking into account of fluctuations on all length scales order-by-order in a properly defined coupling parameter. When resummed using renormalization-group ideas, it gives an accurate description of many critical phenomena.
Often a simple (and correct) field-theoretical formulation of the problem is hard to derive. This is the case, for instance, of problems with quenched disorder, where one usually invoke the replica trick for averaging over the disorder . Also, field theoretical methods are usually unreliable for computing non-universal quantities. These can be on the other hand important for some of the applications (inference, communications, optimization) mentioned above. In this paper we present an approach for computing corrections to BA coming from fluctuations on all length scales.
To be concrete, we shall focus on spin models with pairwise interactions on general graphs, with Hamiltonian
$`E(\sigma )={\displaystyle \underset{(ij)๐ข}{}}J_{ij}\sigma _i\sigma _j{\displaystyle \underset{i=1}{\overset{N}{}}}H_i\sigma _i.`$ (1.1)
Here $`๐ข=(๐ฑ,)`$ is a graph with vertex set $`๐ฑ=\{1,\mathrm{},N\}`$ and edges $`(i,j)`$, $`๐ฑ\times ๐ฑ`$. The set of neighbors of the site $`i`$ is noted $`i`$. We shall use the letters $`a,b,\mathrm{}`$ to denote generic edges, and, whenever necessary write $`(i_a,j_a)=a`$. Finally, for any set of vertices $`A๐ฑ`$, $`\sigma _A\{\sigma _i:iA\}`$. Several families of graphs and choices of the couplings $`J_{ij},H_i`$ will be considered in Section 3, but for the time being we shall remain completely general.
The Bethe approximation for such a model is better described by introducing a field $`h_i^{(j)}`$ for each directed link $`ij`$ of $`๐ข`$. Such fields are required to satisfy the equations
$`h_i^{(j)}=H_i+{\displaystyle \underset{li\backslash j}{}}u_{J_{ij}}(h_l^{(i)}),u_J(h){\displaystyle \frac{1}{\beta }}\mathrm{atanh}[\mathrm{tanh}(\beta J)\mathrm{tanh}(\beta h)].`$ (1.2)
Once a solution of these equations is found, one can use the fields $`h_i^{(j)}`$ to estimate the thermal average of local operators. For instance
$`\sigma _i`$ $`\stackrel{\mathrm{Bethe}}{=}`$ $`{\displaystyle \frac{1}{w_i}}{\displaystyle \underset{\sigma =\pm 1}{}}\sigma \mathrm{exp}(\beta H_i\sigma ){\displaystyle \underset{ji}{}}{\displaystyle \frac{e^{\beta u(J_{ij},h_j^{(i)})\sigma }}{2\mathrm{cosh}(\beta u(J_{ij},h_j^{(i)}))}},`$ (1.3)
$`w_i`$ $`=`$ $`{\displaystyle \underset{\sigma =\pm 1}{}}\mathrm{exp}(\beta H_i\sigma ){\displaystyle \underset{ji}{}}{\displaystyle \frac{e^{\beta u(J_{ij},h_j^{(i)})\sigma }}{2\mathrm{cosh}(\beta u(J_{ij},h_j^{(i)}))}}.`$ (1.4)
The basic approximation involved in derived these equations is the following. Consider a spin $`\sigma _i`$ and set its interaction with the neighbors to $`0`$: $`J_{ij}=0`$ for all $`ji`$ (in other words $`\sigma _i`$ is โremovedโ from the system). Now look at the joint probability distribution of the neighboring spins $`\sigma _i`$ in the system without $`\sigma _i`$. Bethe approximation amount to saying that
$`P_i(\sigma _i)\stackrel{\mathrm{Bethe}}{=}{\displaystyle \underset{ji}{}}{\displaystyle \frac{e^{\beta h_j^{(i)}\sigma _j}}{2\mathrm{cosh}\beta h_j^{(i)}}},.`$ (1.5)
Our approach consists in deriving a set of exact equations for the โcavityโ distributions $`P_i()`$โs. When the form (1.5) is plugged in these equations, the Bethe equations (1.2) are derived. Corrections are computed by introducing correlations in $`P_i(\sigma _i)`$.
In Section 2 we shall explain the general method for computing corrections to BA. We then use it in Section 3 for computing the leading corrections to BA for two particular examples: the ferromagnet on cubic $`d`$-dimensional lattices and the spin glass on random graphs.
## 2 The general approach
In order to explain the general computation scheme, it is convenient to introduce some notation. We denote by $`E^{(i)}(\sigma )`$ a modified energy function in which the interactions between the spin $`i`$ and its neighbors have been canceled. Analogously, $`E^{(a)}(\sigma )`$, with $`a`$, is the energy function modified by eliminating the interaction along the edge $`a`$. In formulae:
$`E^{(i)}(\sigma )=E(\sigma )+{\displaystyle \underset{ji}{}}J_{ij}\sigma _i\sigma _j,E^{(a)}(\sigma )=E(\sigma )+J_{i_aj_a}\sigma _{i_a}\sigma _{j_a}.`$ (2.1)
We denote by $`^{(i)}`$ and $`^{(a)}`$ the Boltzmann averages with respect to these modified energy functions. As in the introduction, $`P_i(\sigma _i)`$ the marginal distribution of the neighbors of $`i`$ with respect to the system with energy $`E^{(i)}(\sigma )`$. Analogously, we define the distribution $`P_a(\sigma _{i_a},\sigma _{j_a})`$.
In order to have a concrete representation for the distributions $`P_i(\sigma _i)`$, we shall use the correlation functions
$`\stackrel{~}{C}_๐^{(i)}{\displaystyle \underset{j๐}{}}\sigma _j^{(i)}={\displaystyle \underset{\sigma _i}{}}P_i(\sigma _i){\displaystyle \underset{j๐}{}}\sigma _j.`$ (2.2)
for any non-empty subset $`๐i`$. In the special case $`๐=\{j\}`$ we shall also use the more conventional notation $`M_j^{(i)}=\stackrel{~}{C}_j^{(i)}`$. In Bethe approximation, the distribution $`P_i(\sigma _i)`$ is assumed to be factorized, cf. Eq. (1.5). In order to compute corrections, it is convenient to introduce the connected correlation functions $`C_๐^{(i)}`$. We have the usual relation
$`\stackrel{~}{C}_๐^{(i)}={\displaystyle \underset{[๐_1\mathrm{}๐_n]}{}}C_{๐_1}^{(i)}\mathrm{}C_{๐_n}^{(i)},`$ (2.3)
with $`[๐_1,\mathrm{},๐_n]`$ running over the partitions of $`๐`$. Finally $`C^{(i)}\{C_๐^{(i)}:๐i\}`$.
Let us now derive the basic relation between the $`P_i()`$โs to be exploited in the following. Consider two sites $`i`$ and $`j`$ which are joined by an edge in $`๐ข`$. We can construct the distribution $`P_{(ij)}(\sigma _i,\sigma _j)`$ in two ways:
$`P_{(ij)}(\sigma _i,\sigma _j)={\displaystyle \frac{1}{Z_j}}{\displaystyle \underset{\sigma _{j\backslash i}}{}}P_j(\sigma _j)\mathrm{exp}\left\{\beta H_j\sigma _j+\beta {\displaystyle \underset{lj\backslash i}{}}J_{jl}\sigma _j\sigma _l\right\},`$ (2.4)
and the equivalent one (let us call $`P_{ij}^{}(\sigma _i,\sigma _j)`$ the corresponding expression) which is obtained by interchanging $`i`$ and $`j`$. Here $`Z_j`$ is a constant which ensures the correct normalization of $`P_{(ij)}(\sigma _i,\sigma _j)`$. We can now marginalize the right hand side of Eq. (2.4) with respect to $`\sigma _j`$ (to $`\sigma _i`$) in order to compute the magnetizations on site $`i`$ (site $`j`$) with respect to the system with energy function $`E^{(ij)}(\sigma )`$. The same calculation can be performed using the expression $`P_{(ij)}^{}(\sigma _i,\sigma _j)`$. Since the result of these two calculations must be the same, we obtain two equations of the form:
$`๐_j^{(i)}(C^{(i)})=๐_i^{(j)}(C^{(j)}),๐_j^{(i)}(C^{(i)})=๐_i^{(j)}(C^{(j)}).`$ (2.5)
The function $`๐_j^{(i)}()`$ yields the magnetization at site $`j`$ when the distribution $`P_i(\sigma _i)`$ is modified through the addition of the interactions $`J_{il}`$, $`lj`$. Analogously, $`๐_j^{(i)}()`$ yields the magnetization at site $`i`$ for the same system. Elementary algebraic manipulations yields the explicit expressions:
$`๐_j^{(i)}(C)`$ $`=`$ $`{\displaystyle \frac{\underset{๐\text{even}}{}t_๐\stackrel{~}{C}_{๐j}+t(H_i)\underset{๐\text{odd}}{}t_๐\stackrel{~}{C}_{๐j}}{\underset{๐\text{even}}{}t_๐\stackrel{~}{C}_๐+t(H_i)\underset{๐\text{odd}}{}t_๐\stackrel{~}{C}_๐}},`$ (2.6)
$`๐_j^{(i)}(C)`$ $`=`$ $`{\displaystyle \frac{t(H_i)\underset{๐\text{even}}{}t_๐\stackrel{~}{C}_๐+\underset{๐\text{odd}}{}t_๐\stackrel{~}{C}_๐}{\underset{๐\text{even}}{}t_๐\stackrel{~}{C}_๐+t(H_i)\underset{๐\text{odd}}{}t_๐\stackrel{~}{C}_๐}},`$ (2.7)
where the sums over $`๐`$ run over all the subsets of neighbors of $`i`$, $`๐i`$, which do not include $`j`$. Furthermore, we used the shorthands $`t_๐_{l๐}t_{il}`$, $`t_{il}\mathrm{tanh}(\beta J_{il})`$ and $`t(H_i)\mathrm{tanh}(\beta H_i)`$. The above functions can be written in terms of the connected correlations $`C_๐^{(i)}`$ by using the relation (2.3).
Consider for instance the case depicted in Fig. 1 where $`i=0`$ and $`j=1`$ have both degree 3. We furthermore assume, for the sake of simplicity, $`H_0=H_1=0`$. Then
$`๐_1^{(0)}(C^{(0)})`$ $`=`$ $`M_1^{(0)}+t_{02}t_{03}{\displaystyle \frac{C_{21}^{(0)}M_3^{(0)}+C_{31}^{(0)}M_2^{(0)}+C_{123}^{(0)}}{1+t_{02}t_{03}M_2^{(0)}M_3^{(0)}+t_{02}t_{03}C_{23}^{(0)}}},`$ (2.8)
$`๐_1^{(0)}(C^{(0)})`$ $`=`$ $`{\displaystyle \frac{t_{02}M_2^{(0)}+t_{03}M_3^{(0)}}{1+t_{02}t_{03}M_2^{(0)}M_3^{(0)}+t_{02}t_{03}C_{23}^{(0)}}}.`$ (2.9)
The analogous expressions for $`๐_0^{(1)}(C^{(1)})`$ and $`๐_0^{(1)}(C^{(1)})`$ are obtained by interchanging $`01`$ and $`\{2,3\}\{4,5\}`$. It is therefore easy to write explicitly the equation $`๐_0^{(1)}(C^{(1)})=๐_1^{(0)}(C^{(0)})`$:
$`M_0^{(1)}+t_{14}t_{15}{\displaystyle \frac{C_{40}^{(1)}M_5^{(1)}+C_{50}^{(1)}M_4^{(1)}+C_{045}^{(1)}}{1+t_{14}t_{15}M_4^{(1)}M_5^{(1)}+t_{14}t_{15}C_{45}^{(1)}}}={\displaystyle \frac{t_{02}M_2^{(0)}+t_{03}M_3^{(0)}}{1+t_{02}t_{03}M_2^{(0)}M_3^{(0)}+t_{02}t_{03}C_{23}^{(0)}}}.`$
There are $`2||`$ equations of the form (2.5): one for each directed link in the graph. The number of unknowns is, on the other hand $`_i(2^{|i|}1)`$, with the sum running over the sites of the graph, and $`|i|`$ being their connectivity. Therefore, these equations are not sufficient to determine the correlation functions $`C^{(i)}`$. If, on the other hand, we neglect multi-spin connected correlation functions and only retain the cavity magnetizations $`M_j^{(i)}`$ (as in Bethe approximation), we are left with $`2||`$ variables to determine. In this case Eqs. (2.6) and (2.7) are considerably simplified:
$`๐_j^{(i)}(C^{(i)})`$ $`\stackrel{\mathrm{Bethe}}{=}`$ $`M_j^{(i)},`$ (2.10)
$`๐_j^{(i)}(C^{(i)})`$ $`\stackrel{\mathrm{Bethe}}{=}`$ $`\mathrm{tanh}\left\{\beta H_i+{\displaystyle \underset{ki\backslash j}{}}\mathrm{atanh}\left[t_{ik}M_k^{(i)}\right]\right\}.`$ (2.11)
By setting $`M_j^{(i)}=\mathrm{tanh}\beta h_j^{(i)}`$, it is easy to see that the equations (2.5) are in this case the Bethe equations<sup>3</sup><sup>3</sup>3This derivation of BA is in fact mentioned as a side remark in Ref. . (1.2). If, for instance, $`๐ข`$ is a tree, the connected cavity correlations $`C_๐^{(i)}`$ vanish if $`|๐|2`$. We thus proved recovered the well-known result that Bethe approximation is exact on tree graphs.
We want now to estimate the connected cavity correlations $`C_๐^{(i)}`$, for $`|๐|2`$, and then use Eq. (2.5) to improve the calculation of $`M_j^{(i)}`$. In synthesis, the correlations are estimated through the fluctuation-dissipation theorem:
$`{\displaystyle \frac{1}{\beta ^n}}{\displaystyle \frac{^nM_{i_1}}{H_{i_2}\mathrm{}H_{i_{n+1}}}}=C_{i_1\mathrm{}i_{n+1}},`$ (2.12)
where $`i_1\mathrm{}i_{n+1}`$ are $`n+1`$ distinct index sites. Here $`M_i`$ and $`C_{i_1\mathrm{}i_{n+1}}`$ are magnetizations and correlations with respect to an arbitrary Hamiltonian of the form (1.1). In other terms, one can obtain equations for the correlations by taking appropriate derivatives of the exact equations (2.5) with respect to an external field.<sup>4</sup><sup>4</sup>4One can see that the differentiation procedure is well-defined through the following (numerically imprecise but conceptually simple) implementation. Compute the magnetization $`M_i`$, then change slightly the external field on site $`j`$, $`H_jH_j+\delta H_j`$, and evaluate the correlation function between sites $`i`$ and $`j`$ as $`C_{ij}=lim_{\delta H_j0}(\delta M_i/\beta \delta H_j)`$. Higher-order correlations are computed analogously by considering variations of the external fields at several points.
For, the sake of clarity, let us compute the leading-order correction to BA. We neglect all connected cavity correlation functions $`C_๐^{(i)}`$ with $`|๐|3`$. Moreover, we treat two point correlation functions to the linear order. To this order, the expressions (2.6) and (2.7) become
$`๐_j^{(i)}(C)`$ $`=`$ $`M_j^{(i)}+{\displaystyle \underset{li\backslash j}{}}\mathrm{\Omega }_{j,l}^{(i)}t_{il}C_{jl}^{(i)}+O(C^2),`$ (2.13)
$`๐_j^{(i)}(C)`$ $`=`$ $`T_j^{(i)}+{\displaystyle \underset{(l_1,l_2)i\backslash j}{}}\mathrm{\Gamma }_{j,l_1l_2}^{(i)}t_{il_1}t_{il_2}C_{l_1l_2}^{(i)}+O(C^2),`$ (2.14)
where
$`\mathrm{\Omega }_{j,l}^{(i)}`$ $`=`$ $`{\displaystyle \frac{T_{jl}^{(i)}}{1+t_{il}M_l^{(i)}T_{jl}^{(i)}}},`$ (2.15)
$`\mathrm{\Gamma }_{j,l_1l_2}^{(i)}`$ $`=`$ $`{\displaystyle \frac{T_{jl_1l_2}^{(i)}T_j^{(i)}}{1+t_{il_1}t_{il_2}M_{l_1}^{(i)}M_{l_2}^{(i)}+t_{il_1}M_{l_1}^{(i)}T_{jl_1l_2}^{(i)}+t_{il_2}M_{l_2}^{(i)}T_{jl_1l_2}^{(i)}}},`$ (2.16)
and
$`T_{l_1l_2\mathrm{}}^{(i)}`$ $`=`$ $`\mathrm{tanh}\left\{\beta H_i+{\displaystyle \underset{ki\backslash l_1,l_2\mathrm{}}{}}\mathrm{atanh}\left[t_{ik}M_k^{(i)}\right]\right\}.`$ (2.17)
We can therefore proceed as follows. First solve the Bethe equations (1.2) for the original energy function (1.1). Then, for each site $`i๐ฑ`$, consider the energy function $`E^{(i)}(\sigma )`$ corresponding to the spin $`\sigma _i`$ being removed. Compute the two point connected correlation functions in the reduced system $`C_{j_1j_2}^{(i)}`$ using BA together with the fluctuation-dissipation relations (2.12). Finally write
$`M_j^{(i)}=\mathrm{tanh}(\beta h_j^{(i)})+\mathrm{\Delta }M_j^{(i)}+O(C^2).`$ (2.18)
The first order corrections $`\mathrm{\Delta }M_j^{(i)}`$ are computed by expanding the equations (2.5) up to first order in $`C`$. Using the expansion (2.13), (2.13), we get
$`\mathrm{\Delta }M_j^{(i)}+{\displaystyle \underset{li\backslash j}{}}t_{il}\mathrm{\Omega }_{j,l}^{(i)}C_{jl}^{(i)}={\displaystyle \underset{kj\backslash i}{}}Q_{i,k}^{(j)}\mathrm{\Delta }M_k^{(j)}+{\displaystyle \underset{(l_1,l_2)j\backslash i}{}}\mathrm{\Gamma }_{i,l_1l_2}^{(j)}t_{jl_1}t_{jl_2}C_{l_1l_2}^{(j)}.`$ (2.19)
where
$`Q_{i,k}^{(j)}={\displaystyle \frac{t_{jk}[1(T_i^{(j)})^2]}{1(t_{jk}\mathrm{tanh}(\beta h_k^{(j)}))^2}}\mathrm{\Delta }M_k^{(j)}`$ (2.20)
Here the coefficients $`\mathrm{\Omega }_{j,l}^{(i)}`$ and $`\mathrm{\Gamma }_{i,l_1l_2}^{(j)}`$ are computed through Eqs (2.15) and (2.16) by setting $`M_k^{(l)}\mathrm{tanh}(\beta h_k^{(l)})`$. We thus obtained one equation of the form (2.19) for each directed edge in the lattice. These completely determine the $`\mathrm{\Delta }M_j^{(i)}`$.
It is important to stress that the above procedure is not uniquely defined. One could consider equivalent calculations which differ from the above by higher order corrections. Here are two examples:
1. One could compute the connected cavity correlations without removing the spin $`i`$ from the system but rather using the relation
$`C_{jl}^{(i)}=(1\mathrm{tanh}^2\beta h_j^{(i)})\beta {\displaystyle \frac{h_j^{(i)}}{H_l}},`$ (2.21)
which in turns follows from $`M_j^{(i)}=\mathrm{tanh}\beta h_j^{(i)}`$. Equations for $`\frac{h_j^{(i)}}{H_l}`$ are easily obtained by differentiating Eq. (1.2).
2. Instead of expanding the equations (2.5) to the first order in $`C_{jl}^{(i)}`$ and $`\mathrm{\Delta }M_j^{(i)}`$, one could proceed as follows. Set to zero all the correlation functions $`C_๐^{(i)}`$, $`|๐|3`$, replace $`C_{jl}^{(i)}`$ by their Bethe approximation, and solve for $`M_j^{(i)}`$.
In particular, the last implementation is exact whenever the graph $`๐ข`$ contains (at most) a unique loop (thus improving over BA). Since the equations (2.5) are exact, in order to prove this claim it is enough to show that the procedure defined above computes correctly the correlation functions $`C_๐^{(i)}`$, $`|๐|2`$. If $`|๐|3`$ and $`๐ข`$ has a single loop, then $`C_๐^{(i)}=0`$, and the algorithm does not make any error in neglecting these correlations. Consider now the computation of $`C_{l_1l_2}^{(i)}`$ and distinguish two cases. In the first case, the graph obtained by removing the vertex $`i`$ is a tree (as e.g. for site $`1`$ in Fig. 2). Therefore BA is exact for the energy function $`E^{(i)}(\sigma )`$ and correctly computes the correlations $`C_{l_1l_2}^{(i)}`$. In the second case upon removing site $`i`$, the resulting graph, let us call it $`๐ข\backslash i`$, still contains a loop (as for site $`2`$ in Fig. 2). Notice that $`๐ข\backslash i`$ is disconnected, and each of the neighbors of $`i`$ belongs to a distinct connected component. Therefore $`C_{j_1j_2}^{(i)}=0`$ for any two neighbors $`j_1,j_2`$ of $`i`$. While BA is not exact for $`๐ข\backslash i`$, it is easy to see that it correctly gives yields vanishing correlations among sites belonging to different connected components.
At this point it should be clear how to improve the above first-order scheme. After removing the spin $`i`$, one can compute the correlations $`C_๐^{(i)}`$ within the first order scheme rather than Bethe approximation (and retain three points correlation as well) and then recompute the magnetizations $`M_j^{(i)}`$ using Eq. (2.5).
The general procedure can be explained as a recursive pseudocode. The code makes use of a routine Correlation( $`๐ข`$$`E()`$$`\{C^{(i)}\}`$ ) which takes as input a graph $`๐ข`$, an Hamiltonian $`E()`$ of the form (1.1), an estimation of the $`n`$-points cavity correlations $`\{C^{(i)}\}`$ for any $`i๐ข`$. The output consists of a new estimation of all $`n`$-points correlation functions. This is obtained (letโs repeat ourselves) by a joint solution of Eqs. (2.5) and (2.12). A particular case of the routine Correlation( $``$ ) is obtained when all the multi-point cavity correlations $`\{C^{(i)}\}`$ are set to $`0`$. This corresponds of course to BA. For the sake of clarity, we shall denote the corresponding routine Bethe( $`๐ข`$$`E()`$ ) instead of Correlation( $`๐ข`$$`E()`$$`0`$ ).
The recursive routine, $`\mathrm{๐ป๐๐๐}()`$, takes as input a graph $`๐ข`$, an Hamiltonian of the form (1.1), and the order of approximation $`\mathrm{}`$ to be achieved. The output consists of an estimation of all $`n`$-points connected correlation functions in the system.
1. Loop( $`๐ข`$, $`E()`$, $`\mathrm{}`$ )
2. If ($`\mathrm{}==1`$) Output Bethe( $`๐ข`$, $`E()`$ )
3. Else
* For ($`i๐ฑ`$) $`C^{\left(i\right)}:=`$ Loop( $`๐ข\backslash i`$, $`E^{\left(i\right)}()`$, $`\mathrm{}1`$ )
* Output Correlation( $`๐ข`$$`E()`$$`\left\{C^{\left(i\right)}\right\}`$ )
4. End
Let us stress that this algorithm deals with particular samples of the model, without need for an average over disorder realizations. Its complexity is (for graphs with bounded connectivity) $`O(N^{\mathrm{}})`$, i.e. polynomial for any fixed $`\mathrm{}`$. This makes its application to inference/optimization problems a viable research direction. The algorithm implements the strategy 2 above (actual elimination of a site): arguing as above, it can be proved that by induction that Loop( $``$, $``$, $`\mathrm{}`$ ) is exact on graphs with cyclic number not larger than $`\mathrm{}`$.
## 3 Applications
In this Section we apply the method developed so far to two simple problems: the spin glass on random graphs with general connectivity distributions, and the ferromagnetic Ising model on the cubic $`d`$-dimensional lattice. In both examples we will keep ourselves to the high temperature, no external field phase. The precise nature of the corrections computed within our approach is different for these two applications. In the first case, they correspond to higher orders in the loop expansion. While there is no formal parameter to order the loop expansion for the Ising model on cubic lattices, we are able to recover the well known Ginzburg criterion and the associated one-loop integral. In the second one, successive terms in our expansion correspond to higher powers in $`1/N`$ ($`N`$ being the number of spins).
### 3.1 Ising model on the cubic lattice
We take $`๐ข`$ to be the $`d`$โdimensional cubic lattice, i.e. $`^d`$ with edges joining neighboring vertices. Vertices of the lattices will be denoted in this Section by $`x,y,z,\mathrm{}^d`$, while unit vectors by $`\mu ,\nu ,\mathrm{}\{(1,0,\mathrm{}),(0,1,\mathrm{}),\mathrm{}\}`$. We consider the usual Ising ferromagnet, i.e. $`J_{xy}=1`$ and $`H_x=H`$.
Because of translational invariance, the Bethe equations admit an uniform solution $`h_y^{(x)}=h`$, with $`h`$ solving
$`h=H+{\displaystyle \frac{(2d1)}{\beta }}\mathrm{atanh}[\mathrm{tanh}\beta \mathrm{tanh}\beta h].`$ (3.1)
Analogously, the cavity magnetizations do not depend on the site: $`M_j^{(i)}=M^{\mathrm{cav}}`$, and we have $`M^{\mathrm{cav}}=\mathrm{tanh}\beta h+\mathrm{\Delta }M^{\mathrm{cav}}+O(C^2)`$. In order to write the results in a compact form, it is convenient to introduce the field
$`h_pH+{\displaystyle \frac{p}{\beta }}\mathrm{atanh}[\mathrm{tanh}\beta \mathrm{tanh}\beta h].`$ (3.2)
We shall furthermore use the shorthand $`t\mathrm{tanh}\beta `$. After some tedious but straightforward calculations, Eq. (2.19) yields
$`\mathrm{\Delta }M^{\mathrm{cav}}={\displaystyle \frac{A}{B}}{\displaystyle \underset{(\mu ,\nu )}{}}C_{\mu ,\nu }^{(0)},`$ (3.3)
where
$`A`$ $`=`$ $`{\displaystyle \frac{1}{d}}{\displaystyle \frac{t\mathrm{tanh}\beta h_{2d2}}{(1+t\mathrm{tanh}\beta h\mathrm{tanh}\beta h_{2d2})(1\mathrm{tanh}^2\beta h)}}+{\displaystyle \frac{d1}{d}}{\displaystyle \frac{2t^3\mathrm{tanh}\beta h}{1t^2\mathrm{tanh}^2\beta h}},`$ (3.4)
$`B`$ $`=`$ $`{\displaystyle \frac{1}{1\mathrm{tanh}^2\beta h}}(2d1){\displaystyle \frac{t}{1t^2\mathrm{tanh}^2\beta h}}.`$ (3.5)
Using these expressions, we can compute the (non-cavity) magnetization $`M=\sigma _x`$
$`M`$ $`=`$ $`M_0+M_1{\displaystyle \underset{(\mu ,\nu )}{}}C_{\mu ,\nu }^{(0)}+O(C^2),`$ (3.6)
$`M_0`$ $`=`$ $`\mathrm{tanh}\beta h_{2d},`$ (3.7)
$`M_1`$ $`=`$ $`t^2{\displaystyle \frac{\mathrm{tanh}\beta h_{2d2}\mathrm{tanh}\beta h_{2d}}{1+t^2\mathrm{tanh}^2\beta h+2t\mathrm{tanh}\beta h\mathrm{tanh}\beta h_{2d2}}}2dt{\displaystyle \frac{1\mathrm{tanh}^2\beta h_{2d}}{1t^2\mathrm{tanh}^2\beta h}}{\displaystyle \frac{A}{B}}.`$ (3.8)
By taking the $`H0`$ limit, we can compute the zero field susceptibility defined by $`M=\chi H+O(H^2)`$. This admits the expansion $`\chi =\chi _0+\chi _1_{(\mu ,\nu )}C_{\mu ,\nu }^{(0)}+O(C^2)`$, where
$`\chi _0=\beta {\displaystyle \frac{1+t}{1(2d1)t}},\chi _1=2\beta {\displaystyle \frac{t^2(1t^2)}{[1(2d1)t]^2}}.`$ (3.9)
We are left with the task of computing the parameter $`\overline{C}_{(\mu ,\nu )}C_{\mu ,\nu }^{(0)}`$. We will follow the strategy 1 described in the previous Section. By differentiating the Bethe equations, we get
$`{\displaystyle \frac{h_x^{(x+\mu )}}{H_z}}=\delta _{x,z}+t{\displaystyle \frac{1\mathrm{tanh}^2\beta h}{1t^2\mathrm{tanh}^2\beta h}}{\displaystyle \underset{\nu (\mu )}{}}{\displaystyle \frac{h_{x+\nu }^{(x)}}{H_z}}.`$ (3.10)
Using the fluctuation-dissipation relation (2.21) and translation invariance, we get the following equation for the cavity correlations
$`C_{z,\mu }^{(0)}=q\delta _{z,\mu }+r{\displaystyle \underset{\nu (\mu )}{}}C_{z+\mu ,\nu }^{(0)},`$ (3.11)
$`q1\mathrm{tanh}^2\beta h,r{\displaystyle \frac{t}{1t^2\mathrm{tanh}^2\beta h}}.`$ (3.12)
These equations are easily solved by introducing the Fourier transform
$`C_\mu ^{(0)}(p)={\displaystyle \underset{x}{}}e^{ipx}C_{x,\mu }^{(0)},C_{x,\mu }^{(0)}={\displaystyle \frac{\mathrm{d}^dp}{(2\pi )^d}C_\mu (p)e^{ipx}},`$ (3.13)
where the integral over $`p`$ runs over the Brillouin zone $`[\pi ,\pi ]^d`$. We obtain
$`C_\mu (p)={\displaystyle \frac{q(e^{ip\mu }r)}{12dr+(2d1)r^2+r\widehat{p}^2}},`$ (3.14)
where $`\widehat{p}^22d_\mu e^{ip\mu }`$. Therefore the correlation parameter entering in the corrections to BA is
$`\overline{C}={\displaystyle \frac{q}{2r^2}}+{\displaystyle \frac{(1r^2)q}{2r^2}}{\displaystyle \frac{\mathrm{d}^dp}{(2\pi )^d}\frac{1}{12dr+(2d1)r^2+r\widehat{p}^2}}.`$ (3.15)
A simple application of the above calculation consists in computing the critical temperature. We can do this by solving the equation $`\chi ^1=0`$, which to this order implies
$`1(2d1)\mathrm{tanh}^2\beta _\mathrm{c}+2\mathrm{tanh}^2\beta _\mathrm{c}(1\mathrm{tanh}\beta _\mathrm{c})\overline{C}=0,`$ (3.16)
By solving this equation to the first order in $`\overline{C}`$, we get
$`\mathrm{tanh}\beta _\mathrm{c}={\displaystyle \frac{1}{2d1}}{\displaystyle \frac{2d2}{(2d1)^3}}[(2d1)2d(2d2)I_d],`$ (3.17)
$`I_d{\displaystyle \frac{\mathrm{d}^dp}{(2\pi )^d}\frac{1}{\widehat{p}^2}}={\displaystyle _0^{\mathrm{}}}[e^{2t}I_0(2t)]^d,`$ (3.18)
where $`I_0()`$ is the Bessel function. The integral $`I_d`$ is convergent for $`d>2`$. A numerical calculation of $`I_d`$ yields the values of $`\beta _\mathrm{c}`$ reported in Table 1. This is compared with numerical simulations and with the second order $`1/d`$ expansion
$`\beta _\mathrm{c}^{1/d}={\displaystyle \frac{1}{2d1}}\left[1+{\displaystyle \frac{1}{3d^2}}+O(d^3)\right].`$ (3.19)
We can also derive the critical behavior of the susceptibility. From Eq. (3.9) we get $`\beta \chi ^1=K(\beta _\mathrm{c}\beta )+O((\beta _\mathrm{c}\beta )^2)`$ where
$`K=2d\left\{{\displaystyle \frac{2d}{2d1}}{\displaystyle \frac{4d(2d2)(2d^2d+1)}{(2d1)^2}}I_d+{\displaystyle \frac{4d^2(2d2)^2}{(2d1)^2}}J_d\right\},`$ (3.20)
$`J_d{\displaystyle \frac{\mathrm{d}^dp}{(2\pi )^d}\frac{1}{(\widehat{p}^2)^2}}={\displaystyle _0^{\mathrm{}}}t[e^{2t}I_0(2t)]^d.`$ (3.21)
The integral $`J_d`$ is infrared divergent for $`d4`$. We have therefore rederived the well known upper-critical dimension $`d_{\mathrm{up}}=4`$.
### 3.2 Spin glass on random graphs
We consider here the case in which $`๐ข`$ is an Erdรถs-Renyi random graph with $`N`$ vertices and average connectivity $`\gamma `$. Such a graph is constructed by drawing an edge between any couple $`(i,j)`$ of distinct vertices independently with probability $`\gamma /N`$. Spins joined by an edge interact via a coupling $`J_{ij}`$ which are i.i.d. symmetric random variables with probability density function $`p(J)=p(J)`$. This is also known as the Viana-Bray model and was first studied in Ref. . In the following $`๐ผ`$ will denote expectation with respect to the couplings and/or the graph realization. Finally we shall focus on the case of vanishing external field $`H_i=0`$.
The interest of such a simple model is that it can be easily treated by the replica method, thus providing an useful check of our approach.<sup>5</sup><sup>5</sup>5Some of the properties of the high temperature phase can be furthermore derived rigorously, see . For ideas on $`1/N`$ corrections in the low temperature phase see . Before applying the strategy outlined in the previous Section, let us briefly recall the replica results. Averaging over disorder, one gets the following representation for the moments of the partition function
$`๐ผZ^n={\displaystyle \mathrm{exp}\{NS[c]\}๐c},`$ (3.22)
where $`c(\stackrel{}{\sigma })=c(\sigma ^1,\mathrm{},\sigma ^n)`$, $`\sigma ^a\{\pm 1\}`$ is the replica order parameter. The integral is then evaluated using the saddle point method, the paramagnetic saddle point being $`c(\stackrel{}{\sigma })=1/2^n`$. The free energy density $`\beta f_N(\beta )=(1/N)๐ผ\mathrm{log}Z_N`$ is then obtained by taking the $`n0`$ limit:
$`\beta f_N(\beta )`$ $`=`$ $`\mathrm{log}2{\displaystyle \frac{\gamma }{2}}๐ผ\mathrm{log}\mathrm{cosh}\beta J+{\displaystyle \frac{1}{N}}\beta f^{(1)}+O(N^2),`$ (3.23)
$`\beta f^{(1)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\gamma ^2๐ผ\mathrm{log}\mathrm{cosh}\beta J+{\displaystyle \frac{1}{4}}\gamma ^2๐ผ\mathrm{log}\mathrm{cosh}\beta (J_1+J_2)`$
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\gamma ^k}{k}}๐ผ\mathrm{log}\left\{1+{\displaystyle \underset{i=1}{\overset{k}{}}}\mathrm{tanh}\beta J_i\right\}.`$
Here expectations are taken with respect to the $`J_1,J_2,\mathrm{}`$ which are i.i.d. with distribution $`p(J)`$. The $`O(1/N)`$ term is the contribution of Gaussian fluctuations around this saddle point. As expected, it diverges upon approaching the spin glass critical temperature $`\beta _\mathrm{c}`$ (defined by $`\gamma ๐ผ\mathrm{tanh}^2\beta _\mathrm{c}J=1`$). More precisely we have
$`f^{(1)}(\beta )={\displaystyle \frac{1}{4}}\mathrm{log}(\beta _\mathrm{c}\beta )+O(1).`$ (3.25)
We shall now rederive the above results by using the approach outlined in the previous Section. A serious shortcoming of this approach is that it does not provide an explicit expression for the free energy. One can circumvent this problem by considering the internal energy density $`u_N(\beta )=๐ผE(\sigma )/N`$. Since all the edges of the graph are equivalent,
$`u_N(\beta )={\displaystyle \frac{N1}{2N}}\gamma ๐ผ\left\{J_{ij}\sigma _i\sigma _j\right|(i,j)๐ข\}.`$ (3.26)
Here $`๐ผ\{|(i,j)G\}`$ denotes expectation conditional to the edge $`(i,j)`$ belonging to the graph $`๐ข`$. Consider now a particular graph $`๐ข`$ in which the link $`(i,j)`$ is present. It is simple to show that
$`\sigma _i\sigma _j=t_{ij}+(1t_{ij}^2){\displaystyle \frac{C_{ij}^{(ij)}}{1+t_{ij}C_{ij}^{(ij)}}},`$ (3.27)
where $`C_{ij}^{(ij)}`$ denotes the correlation $`\sigma _i\sigma _j`$ after link $`(i,j)`$ has been removed (notice that local magnetizations $`\sigma _i`$ vanish by symmetry), and $`t_{ij}=\mathrm{tanh}\beta J_{ij}`$. Notice that sampling $`๐ข`$ under the condition $`(i,j)๐ข`$, and then removing $`(i,j)`$ is equivalent (fro the Erdรถs-Renyi ensemble) to sampling $`๐ข`$ under the condition $`(i,j)๐ข`$. Substituting in Eq. (3.26), we get
$`u_N(\beta )={\displaystyle \frac{N1}{2N}}\gamma ๐ผ\{Jt_J\}{\displaystyle \frac{N1}{2N}}\gamma ๐ผ\left\{J(1t_J^2){\displaystyle \frac{C_{ij}}{1+t_JC_{ij}}}\right|(i,j)๐ข\},`$ (3.28)
where we used the shorthand $`t_J=\mathrm{tanh}\beta J`$. The second term vanishes as $`N\mathrm{}`$, since it behaves as the correlation between two uniformly random sites in the system. We will therefore estimate it to the leading non-trivial order. Moreover, we can expand $`(1+t_JC_{ij})^1`$ in an absolutely convergent series to get
$$\begin{array}{c}u_N(\beta )=\frac{\gamma }{2}๐ผ\{Jt_J\}+\frac{1}{N}\frac{\gamma }{2}๐ผ\{Jt_J\}\hfill \\ \hfill \frac{\gamma }{2}\underset{k=0}{\overset{\mathrm{}}{}}(1)^k๐ผ\{J(1t_J^2)t_J^k\}๐ผ\{C_{ij}^{k+1}|(i,j)๐ข\}+O(N^2),\end{array}$$
(3.29)
where we factorized the expectation thanks to the fact that $`๐ข`$ does not contain $`(i,j)`$ and therefore $`C_{ij}`$ is independent of $`J`$.
We are left with the task of computing the moments of $`C_{ij}`$. According to our general strategy, we use the identity $`C_{ij}=\frac{1}{\beta }\frac{M_i}{H_j}|_{H=0}`$. It is well known that, for the Erdรถs-Renyi random graph, $`M_l\stackrel{\mathrm{d}}{=}\mathrm{tanh}\beta h_i^{(j)}`$ up to corrections vanishing as $`N\mathrm{}`$ (here $`\stackrel{\mathrm{d}}{=}`$ denoted equality in distribution). Furthermore, by differentiating Bethe equations (1.2) we get
$`{\displaystyle \frac{h_i^{(m)}}{H_j}}=\delta _{ij}+{\displaystyle \underset{li\backslash m}{}}\mathrm{tanh}\beta J_{il}{\displaystyle \frac{h_l^{(i)}}{H_j}}.`$ (3.30)
By averaging over the graph and couplings and recalling that the degree of a site is, for large $`N`$, a Poisson random variable of mean $`\gamma `$, we obtain (here we use the shorthand $`_j`$ for partial derivative with respect to $`H_j`$):
$`๐ผ\left\{\left(_jh_i^{(m)}\right)^k\right\}`$ $`=`$ $`{\displaystyle \frac{1}{N}}+\gamma ๐ผ\{t_J^k\}๐ผ\left\{\left(_jh_l^{(i)}\right)^k\right\}+O(N^2).`$ (3.31)
Moreover if we condition on $`i`$, $`j`$ being distinct and not connected by an edge, the term $`\delta _{ij}`$ in Eq. (3.30) is surely missing for at least two iterations, leading to
$`๐ผ\left\{\left(_jh_i^{(m)}\right)^k\right|ij,(i,j)๐ข\}`$ $`=`$ $`(\gamma ๐ผ\{t_J^k\})^2๐ผ\left\{\left(_jh_l^{(i)}\right)^k\right\}+O(N^2).`$ (3.32)
By solving these equations and identifying $`๐ผ\{C_{ij}^{k+1}|(i,j)๐ข\}`$ with the left hand side of the last equation, we finally get
$`๐ผ\{C_{ij}^k|(i,j)๐ข\}={\displaystyle \frac{1}{N}}{\displaystyle \frac{(\gamma ๐ผ\{t_J^k\})^2}{1\gamma ๐ผ\{t_J^k\}}}+O(N^2),`$ (3.33)
for $`k`$ even (for $`k`$ odd the expectation vanishes by symmetry. We can now plug this into Eq. (3.29) to get the final result
$`u_N(\beta )`$ $`=`$ $`{\displaystyle \frac{\gamma }{2}}๐ผ\{Jt_J\}+{\displaystyle \frac{1}{N}}u^{(1)}(\beta )+O(N^2),`$ (3.34)
$`u^{(1)}(\beta )`$ $`=`$ $`{\displaystyle \frac{\gamma }{2}}๐ผ\{Jt_J\}+{\displaystyle \frac{\gamma }{2}}{\displaystyle \underset{k\text{odd}}{}}๐ผ\{J(1t_J^2)t_J^k\}{\displaystyle \frac{(\gamma ๐ผ\{t_J^{k+1}\})^2}{1\gamma ๐ผ\{t_J^{k+1}\}}}.`$ (3.35)
One can then compute the free energy density by integrating over the temperature with boundary condition $`\beta f_N(\beta )\mathrm{log}2`$ as $`\beta 0`$. It is easy to check that the resulting expression coincides with the replica result (3.23), (3.2).
One surprising feature of the above calculation is the behavior of correlations. One would naively assume that the correlation between two random spins is concentrated around a typical value of order $`N^\delta `$, with $`\delta >0`$, leading to $`๐ผ\{C_{ij}^k\}N^{k\delta }`$, in contradiction with the correct result, Eq. (3.33). It is therefore interesting to study the distribution of $`C_{ij}`$. For the sake of simplicity we shall admit the cases $`i=j`$ and $`(i,j)๐ข`$ which where excluded in the above calculation. In the $`N\mathrm{}`$ limit, the correlations satisfy the same distributional equation as the response functions, cf. Eq (3.30):
$`C\stackrel{\mathrm{d}}{=}{\displaystyle \underset{i=1}{\overset{k}{}}}(\mathrm{tanh}\beta J_i)C_i.`$ (3.36)
Here $`k`$ is a Poisson random variable of mean $`\gamma `$ and $`J_i`$ are i.i.d. with distribution $`P(J)`$. Notice that this equation can only fix the distribution of $`C`$, to be denoted by $`\rho (C)`$, up to a scaling factor. In fact, if $`C`$ is a random variable satisfying the above equation, also $`aC`$ does. We shall therefore write $`\rho (C)=(1/C_0)\rho _{}(C/C_0)`$, where the solution $`\rho _{}(C)`$ is fixed arbitrarily and $`C_0=C_0(N)`$ is the typical scale of correlations in a system of size $`N`$. The scale $`C_0`$ will be determined by a matching procedure.
Consider the characteristic function $`\varphi (s)\mathrm{exp}(isC)d\rho _{}(C)`$. This satisfies the equation
$`\varphi (s)=\mathrm{exp}\{\gamma [1๐ผ\varphi (ts)]\},`$ (3.37)
where expectation is taken with respect to $`t=\mathrm{tanh}\beta J`$. From this is easy to derive the small $`s`$ behavior $`\varphi (s)1\varphi _0|s|^\alpha `$ where we can fix the freedom in the scale of $`C`$ by setting $`\varphi _0=1`$. The exponent $`\alpha `$ is determined by
$`\gamma ๐ผ\{|\mathrm{tanh}\beta J|^\alpha \}=1.`$ (3.38)
Therefore $`\alpha `$ grows monotonously with $`\beta `$ and takes the value $`\alpha =2`$ at $`\beta =\beta _\mathrm{c}`$. The large $`C`$ behavior of the correlations distribution is $`\rho _{}(C)(\frac{1}{\pi }\mathrm{\Gamma }(1+\alpha )\mathrm{sin}(\frac{\pi \alpha }{2}))|C|^{1\alpha }`$. The physical reason of this power law tail is easily understood. Consider the neighborhood of a site $`i`$. Asymptotically, this will be a random tree with Poisson degree distribution (a Galton-Watson tree). For a site $`j`$ at distance $`r`$ from $`i`$, we have $`C_{ij}=\mathrm{tanh}\beta J_1\mathrm{}\mathrm{tanh}\beta J_r`$, where $`J_1,\mathrm{},J_r`$ are the couplings on the path joining $`i`$ to $`j`$. If we only consider these sites and sum over all finite $`r`$ as $`N\mathrm{}`$, we get
$`\rho _{\mathrm{neigh}}(C)={\displaystyle \frac{1}{N}}{\displaystyle \underset{r=0}{\overset{\mathrm{}}{}}}\gamma ^r๐ผ\delta \left(C{\displaystyle \underset{i=1}{\overset{r}{}}}\mathrm{tanh}\beta J_i\right).`$ (3.39)
The small $`C`$ asymptotics of this distribution is $`\rho ^{\mathrm{neigh}}(C)\rho _0^{\mathrm{neigh}}|C|^{1\alpha }`$. By computing $`\rho _0^{\mathrm{neigh}}`$ and matching this behavior with the large $`C`$ behavior of $`\rho (C)=(1/C_0)\rho _{}(C/C_0)`$, we determine
$`C_0(N)=\left\{{\displaystyle \frac{\pi }{\mathrm{\Gamma }(1+\alpha )\mathrm{sin}\frac{\pi \alpha }{2}}}{\displaystyle \frac{1}{(\gamma ๐ผ|t_J|^\alpha \mathrm{log}|t_J|)}}\right\}^{1/\alpha }N^{1/\alpha },`$ (3.40)
where $`t_J=\mathrm{tanh}\beta J`$. As the temperature decreases from $`\mathrm{}`$ to the critical temperature, $`\alpha `$ increases from $`0`$ to $`2`$ and therefore the typical correlation scale increases from $`N^{\mathrm{}}`$ to $`N^{1/2}`$. However, correlations are never concentrated around a particular value but have a power-law behavior at all temperature. Integer moments are therefore governed by the largest correlations in the system (in particular they are ruled by $`\rho ^{\mathrm{neigh}}(C)`$) and are always of order $`N^1`$. Finally notice that, because of the power-law behavior, there is no definite loop length responsible for corrections to BA.
Let us conclude with a comment. One could have been skeptical about the success of the present approach in computing $`1/N`$ effects in for spin models on random graphs. In fact, an average fraction $`1/N`$ of spins of such systems lies in a neighborhood of finite-size loops (e.g. triangles). For such spins the violation of BA is non-perturbative and our approach could have seemed a priori hopeless. However, the exactness of our method for uni-cyclic graph allows to overcome this problem. On the other hand, it was crucial not to neglect terms of order $`๐ผ\{C_{ij}^{k+1}\}`$, $`k>0`$ in Eq. (3.29), i.e. to follow the procedure 2 described in Section 2.
The same kind of argument suggest that the systematic expansion described in Section 2 corresponds in fact to the $`1/N`$ expansion for random graphs. The next correction is due to couples of joined closed loops, an event occurring in average near a fraction $`1/N^2`$ of the sites.
## Acknowledgments
While this paper was being completed, we discovered that Giorgio Parisi and Frantiลกek Slanina had been working at a similar loop expansion using a completely different approach . We are grateful to them for letting us know about their results.
This work was supported by EVERGROW, i.p. No. 1935 in the complex systems initiative of the Future and Emerging Technologies directorate of the IST Priority, EU Sixth Framework.
|
warning/0506/hep-th0506041.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Singularities of string backgrounds have attracted much attention and have been investigated using a variety of methods . One is to study the gauge theory on a $`D`$-brane probe of the singularity. While there has been much work done on extracting gauge theory data for various types of singularities (abelian and non-abelian orbifolds, conifolds , toric and generalized del Pezzo singularities), a general method for determining the superpotential has been lacking. In the superpotential was obtained from integrating the quiver relations for certain cases, with various ad-hoc methods being used to resolve ambiguities that arise in such an integration. In this paper, using previous work of , we present a general rigorous method for obtaining the superpotential of such quiver gauge theories from the quiver relations. We show that the superpotential is just the naive sum of terms of the form relation times the $`\mathrm{Ext}^2`$ field corresponding to the relation. We apply the method to the trivial example of a $`^2`$ as well as to a $`\mathrm{dP}_1`$, in which case we get a non-homogeneous superpotential. In principle we can apply it to a general del Pezzo singularity โ all we need is the del Pezzo quiver and relations.
We deal with flat compactifications $`M\times X`$ where $`M`$ is $`4`$ dimensional Minkowski space and $`X`$ is a Calabi-Yau manifold, and we probe the theory with space filling branes โ i.e., $`(n+3)`$-branes, where $`n`$ is the dimensionality of the brane within $`X`$. We expect such branes to be BPS and stable when probing a smooth point of $`X`$, but to marginally decay into a collection of so-called fractional branes when the point becomes singular. We consider singularities obtained when a complex surface (i.e., one that has 4 real dimensions) $`S`$ shrinks down to zero size in $`X`$ by varying the Kรคhler parameters. Assuming that $`S`$ is smooth and irreducible, it is known that $`S`$ must be a del Pezzo surface, i.e., $`^1\times ^1`$, or $`^2`$ blown up at $`m`$ points (denoted by $`\mathrm{dP}_m`$), where $`m`$ ranges from $`0`$ to $`8`$. The $`3+1`$ dimensional quiver gauge theory associated to this marginal decay into fractional branes is the one whose superpotential we are after. This set up has been studied extensively in the literature .
The moduli space of a D-brane is given by the space of critical points of the superpotential. Thus, knowing the moduli space one may make a guess at the form of the superpotential. In the case at hand, this leads to a natural conjecture for the superpotential. By using more rigorous methods we are able to show that this conjecture is correct.
For our analysis we will use the algebraic machinery of the derived category of coherent sheaves, developed in particular in . The fractional branes have tractable representations as elements of the derived category $`๐(X)`$ of coherent sheaves on $`X`$. In principle the method of can be applied to find the so-called $`A_{\mathrm{}}`$ products in the algebra of $`\mathrm{Ext}`$ groups. These $`A_{\mathrm{}}`$ products are determined by combinatorial relations that they have to satisfy (coming from Feynman diagrams in the associated topological theory) and they encode the superpotential. Applying the technique of directly is difficult and in order to make the problem tractable, we instead proceed in two steps. First, we use a spectral sequence argument to reduce the problem from one of studying sheaves on $`X`$ to the simpler one of studying sheaves on $`S`$. It is in this reduction that we show that each $`\mathrm{Ext}^2`$ appears linearly in the superpotential, multiplying a term that involves only $`\mathrm{Ext}^1`$s and is determined by $`A_{\mathrm{}}`$ relations over $`S`$. To compute these we exploit the well understood properties of $`๐(S)`$, and specifically its intimate relation with the derived category of quiver representations . We see that the terms involving the $`\mathrm{Ext}^1`$s are just the (possibly non-homogeneous) relations in the quiver. It is important to note that we obtain the superpotential only up to certain nonlinear field redefinitions (see ) โ this is the most that could be expected from such topological sigma model methods as we use.
The plan of the paper is as follows: in section 2 we review quivers and sheaves on del Pezzo surfaces, and how they relate. In section 3 we review $`A_{\mathrm{}}`$ algebras and the method of for using the topological $`B`$ model to compute $`A_{\mathrm{}}`$ structure and hence the superpotential. In section 4 we introduce the quiver gauge theory we want to study and prove that its superpotential is linear in the $`\mathrm{Ext}^2`$s, which multiply terms determined by the $`A_{\mathrm{}}`$ structure over $`S`$ โ this is the reduction from sheaves on $`X`$ to sheaves on $`S`$. In section 5 we solve the problem on $`S`$ by reframing it as a computation in the derived category of quiver representations, and apply the solution to the case $`S=^2`$ and the more nontrivial case $`S=\mathrm{dP}_1`$. This example illustrates the general algorithm that can be carried through for any quiver with known relations.
## 2 Quivers and Sheaves
### 2.1 Quivers
We now review the necessary mathematical notions relating to quivers and their representations. Further background can be found in . First, a quiver is a directed graph $`Q`$ that consists of nodes $`v_i`$ and arrows $`a_\alpha `$. Its path algebra $`A`$ is defined as follows: as a vector space, $`A`$ is generated by all of the paths constructed through concatenation of arrows in $`Q`$. The product structure of $`A`$ is defined on these generators as follows: if path $`1`$ ends on the same node that path $`2`$ begins on, the product is defined to be the obvious concatenation; otherwise it is defined to be $`0`$. Note that corresponding to each node $`v_i`$ we have a corresponding zero length path and hence an idempotent element $`e_i`$ of $`A`$.
In the remainder of the paper, we will deal with a slight generalization to a quiver with relations. This is just a quiver whose path algebra is defined as the above $`A`$ quotiented out by a subspace generated by linear combinations of paths called relations. We stipulate that any given relation must be a linear combination of paths between the same two nodes. It does not, however, need to be homogeneous. A simple example (with homogeneous relations) is the so-called Beilinson quiver, defined as:
(1)
with relations $`a_\alpha b_\beta a_\beta b_\alpha `$.
For a given quiver $`Q`$, we can consider the associated category $`A`$mod of left $`A`$-modules. For any left $`A`$-module $`V`$ we can form the vector spaces $`V_i=e_iV`$, of dimension $`N_i`$; if we then think of each $`V_i`$ as living on node $`v_i`$ then we see that multiplication by any arrow $`a_\alpha `$ acts as a linear transformation between the spaces $`V_i`$ at the tail and head of $`a_\alpha `$, and these linear transformations respect the relations. This structure is known as a representation of the quiver $`Q`$, of dimension $`(N_i)`$.
A map $`\varphi `$ between left $`A`$-modules $`V`$ and $`W`$ is simply a linear transformation that commutes with the action of $`A`$, i.e., $`\varphi (av)=a\varphi (v)`$. If we think of $`V`$ and $`W`$ as quiver representations then this condition is just the obvious constraint that the maps from $`V_i`$ to $`W_i`$ must commute with the linear transformations induced by the arrows, i.e., $`\varphi `$ is a map of representations. Thus we sometimes refer to the category of left $`A`$-modules as the category of quiver representations, and we use these terms interchangeably from now on.
Corresponding to each node of $`Q`$ there are two distinguished representations $`P_i`$ and $`L_i`$; when we make the connection to sheaves on del Pezzo surfaces below, these will correspond to sheaves in the exceptional collection and fractional branes, respectively, as we will see. $`L_i`$ is defined simply as the one dimensional representation with $`V_i=`$ and all other $`V_j=0`$. $`P_i`$ is defined as the subspace of $`A`$ generated by all paths that begin at $`v_i`$; it is trivially seen to be a subrepresentation. It may seem that $`P_i`$ is a rather large representation, and indeed, if the quiver has any loops there will be infinite dimensional $`P_i`$. However, in the case of quivers associated to del Pezzoโs there will be an ordering on the nodes that is respected by the arrows and hence all $`P_i`$ will be finite-dimensional. This finite dimensionality is in large part responsible for the tractability of the problem and motivates the reduction mentioned earlier from sheaves on the Calabi-Yau to sheaves on the del Pezzo. In the Beilinson quiver, for example, we see that $`P_0`$, $`P_1`$, and $`P_2`$ have dimensions $`(1,0,0)`$, $`(3,1,0)`$, and $`(6,3,1)`$ respectively.
When we make the connection between quivers and sheaves it will be through the derived category. Before we talk about that, however, let us first discuss some basic homological properties of quivers. First of all, one can show that the $`P_i`$ are projective objects in $`A`$mod. In fact, they form a complete set in the sense that any left $`A`$-module has a resolution by various direct sums of these $`P_i`$. These projective resolutions can be used to compute higher $`\mathrm{Ext}`$ groups. For example, the projective resolution of any $`L_i`$ is
(2)
Here $`n_{ij}`$ is the number of arrows from node $`i`$ to node $`j`$ and $`r_{ij}`$ is the number of independent relations imposed on paths from $`i`$ to $`j`$. In the case of the Beilinson quiver, the resolutions are:
(3)
Noting that $`\mathrm{Ext}^k`$ is the $`k`$th derived functor of $`\mathrm{Hom}`$ and that $`\mathrm{Hom}(P_i,L_j)=\delta _{ij}`$ we can compute $`\mathrm{Ext}^p(A,B)`$ by taking a projective resolution
(4)
where the $`\mathrm{\Pi }_i`$ are direct sums of $`P_i`$โs and from it constructing the complex
(5)
The cohomology of this complex in the $`p`$th position is then $`\mathrm{Ext}^p(A,B)`$. Using this method one can show that
$$\begin{array}{cc}\hfill dim\mathrm{Ext}^1(L_i,L_j)& =n_{ij}\hfill \\ \hfill dim\mathrm{Ext}^2(L_i,L_j)& =r_{ij}.\hfill \end{array}$$
(6)
In general, higher $`\mathrm{Ext}`$โs may also exist. For example, $`\mathrm{Ext}^3`$ represents relations amongst relations. However, for the purposes of quiver gauge theories, the appearance of higher $`\mathrm{Ext}`$โs is unphysical and so we assume
$$\mathrm{Ext}^k(L_i,L_j)=0,k3.$$
(7)
Finally, we note that we can recover the quiver path algebra from the projective representations $`P_i`$. Supposing $`Q`$ is a quiver with $`n`$ nodes, this is done as follows: we let
$$T=P_0P_1\mathrm{}P_{n1}.$$
(8)
Using the fact that $`\mathrm{Hom}(P_i,P_j)`$ is simply the vector space of paths from $`j`$ to $`i`$ we can verify that
$$A\mathrm{End}(T)^{\text{op}}.$$
(9)
In other words, $`A`$ is just the algebra $`\mathrm{End}(T)`$ with the product structure reversed.
### 2.2 The Derived Category
Having reviewed this preliminary material about quiver representations we move on to briefly discuss the derived category. As mentioned above, the derived category will form a bridge between the quiver representations that we have already discussed and the category of coherent sheaves introduced below. One source is ; here we just review the facts we will use.
Given any abelian category $`\mathrm{A}`$ (such as that of quiver representations, or that of coherent sheaves discussed below) we can define its derived category $`๐(\mathrm{A})`$ as follows. The objects in $`๐(\mathrm{A})`$ are complexes of objects in $`\mathrm{A}`$:
(10)
To construct the morphisms in $`๐(\mathrm{A})`$, we begin with the abelian group of all possible maps between complexes (not necessarily respecting the differential). These maps are graded by their degree $`p`$ and can be written as
$$\underset{n}{}f_{n,n+p}$$
(11)
where $`f_{m,n}`$ is a map from $`\mathrm{E}^m`$ to $`\mathrm{E}^n`$. We define a differential on this group by (abusing notation slightly):
$$(df)_{n,p+1}=d_{n+p}f_{n,p}(1)^pf_{n+1,p}d_n$$
(12)
The derived morphisms are now defined as the cohomology of this group, with formal inverses added in for all quasi-isomorphisms (that is, those chain maps which induce isomorphisms on cohomology).
Now we state some necessary results without proof. Given any object $`A`$ in $`\mathrm{A}`$, we can construct the associated one term complex whose only nonzero entry is $`A`$, at the zeroth position. For brevity we will henceforth refer to both the object and the associated one term complex by $`A`$. Then, for $`A,B`$ in $`\mathrm{A}`$, $`\mathrm{Ext}^{}(A,B)`$ is given by the group of derived morphisms between the complexes associated to $`A`$ and $`B`$, with the grading on $`\mathrm{Ext}`$ corresponding to the grading of the derived morphisms. In fact, this is the generalization of the notion of $`\mathrm{Ext}`$ to the arbitrary elements of $`๐(\mathrm{A})`$. Also, any $`A`$ in $`\mathrm{A}`$ is equivalent to its projective or injective resolution in the derived category. Further, if we represent either $`A`$ by its projective resolution or $`B`$ by its injective resolution then the generators of $`\mathrm{Ext}(A,B)`$ can be written as honest chain maps between these complexes.
### 2.3 Sheaves
We now turn to reviewing key aspects of the other relevant category, that of coherent sheaves. It turns out that (as we will see in more detail below) the derived category of coherent sheaves on a Calabi-Yau manifold $`X`$, denoted $`๐(X)`$, precisely describes $`D`$-branes in the topological B model defined on $`X`$. The open string modes stretching between them are described by the $`\mathrm{Ext}`$ groups of the sheaf homs between the relevant branes . These, in turn, describe the massless spectrum of the physical theory on $`M\times X`$. In fact $`๐(X)`$ contains enough information to determine the tree-level superpotential of the low energy effective theory, in the form of $`A_{\mathrm{}}`$ products. We will discuss all of this below, but for now let us start by introducing coherent sheaves on Calabi-Yauโs and del Pezzoโs.
The category of coherent sheaves on a space $`X`$ is an enlargement of the category of vector bundles (also referred to as โlocally free sheavesโ) on $`X`$ โ it contains vector bundles as well as all kernels and cokernels of maps of vector bundles. For a precise definition, starting from the general concept of a sheaf, see or . We can very roughly think of it as including, in addition to vector bundles over $`X`$, more exotic objects such as vector bundles over submanifolds of $`X`$.
In the physical problem we consider $`D`$-branes on a shrinking cycle $`S`$ which is embedded in $`X`$: $`i:SX`$. $`i`$ induces an embedding $`i_{}:๐(S)๐(X)`$, and it is no surprise that the branes weโll be interested in are in fact in the image of $`i_{}`$. Now, $`๐(S)`$ has been studied extensively by mathematicians and is well understood.
We proceed by first defining a complete strongly exceptional collection of sheaves on $`S`$ to be an ordered set $`\{\mathrm{F}_0,\mathrm{},\mathrm{F}_{n1}\}`$ that generates $`๐(S)`$ and satisfies $`\mathrm{Ext}_S^p(\mathrm{F}_i,\mathrm{F}_j)=0`$ for $`p0`$ and any $`i`$ and $`j`$, and $`\mathrm{Ext}_S^0(\mathrm{F}_i,\mathrm{F}_j)=\mathrm{Hom}_S(\mathrm{F}_i,\mathrm{F}_j)=0`$ for $`i>j`$ and $`\mathrm{Hom}_S(\mathrm{F}_i,\mathrm{F}_i)=`$. Given such a complete strongly exceptional collection, we can define
$$A=\mathrm{End}(\mathrm{F}_0\mathrm{F}_1\mathrm{}\mathrm{F}_{n1})^{\text{op}}$$
(13)
It turns out that $`A`$ is the path algebra of a quiver $`Q`$, and the $`\mathrm{F}_i`$ are isomorphic (as $`A`$-modules) to the projective representations $`P_i`$ defined above. Given this we can reconstruct the quiver uniquely simply by noting that $`\mathrm{Hom}_S(\mathrm{F}_i,\mathrm{F}_j)=\mathrm{Hom}(P_i,P_j)`$ is just the space of paths from node $`j`$ to node $`i`$. In fact, Bondal proved that the derived category of $`A`$-modules, $`๐(A\text{}\text{mod})`$ is equivalent to $`๐(S)`$.
As an example, consider $`S=^2`$. An exceptional collection is given by $`\{\mathrm{O},\mathrm{O}(1),\mathrm{O}(2)\}`$. We have $`\mathrm{Hom}(\mathrm{O},\mathrm{O}(1))^3`$, $`\mathrm{Hom}(\mathrm{O}(1),\mathrm{O}(2))^3`$. Denote these maps, which are just multiplication by the homogeneous coordinates, by $`x_i`$ and $`y_i`$ respectively, $`i=1,2,3`$. We also have $`\mathrm{Hom}(\mathrm{O},\mathrm{O}(2))^6`$ โ these maps are multiplication by homogeneous degree two polynomials in the homogeneous coordinates. All this implies that we have three arrows $`x_i`$ from node $`2`$ to node $`1`$, three arrows $`y_i`$ from node $`3`$ to node $`2`$, and that all paths from node $`3`$ to node $`1`$ are compositions of these arrows, with relations $`x_iy_jx_jy_i=0`$.
Another example which will be thoroughly dealt with below is $`S=\mathrm{dP}_1`$, which is $`^2`$ with one point blown up. Letting $`C_1`$ be the exceptional divisor, a complete strongly exceptional collection is $`\{\mathrm{O},\mathrm{O}(\mathrm{C}_1),\mathrm{O}(\mathrm{H}),\mathrm{O}(2\mathrm{H})\}`$ where $`H`$ is the hyperplane divisor. A slightly more involved analysis shows the quiver to be
(14)
with the relations $`b_0d_1b_1d_0=0`$, $`ab_0d_2cd_0=0`$, and $`ab_1d_2cd_1=0`$.
Now that we have Bondalโs theorem, we can use either $`๐(S)`$ or $`๐(A\text{}\text{mod})`$ to describe branes on $`S`$. We will call the branes that correspond to the representations $`L_i`$ fractional branes. Of course, we are actually interested in branes in $`X`$, i.e., in the image of $`๐(S)`$ in $`๐(X)`$ as noted above. Using a local model of the Calabi-Yau X, namely representing it as the total space of the normal bundle of $`S`$ in $`X`$ (which is isomorphic to the canonical bundle) one can determine the $`\mathrm{Ext}`$ groups of sheaves in $`i_{}๐(S)`$ in terms of the $`\mathrm{Ext}`$ groups in $`๐(S)`$. Namely, we find using a spectral sequence argument that
$$\mathrm{Ext}_X^p(i_{}L_i,i_{}L_j)=\mathrm{Ext}_S^p(L_i,L_j)\mathrm{Ext}_S^{3p}(L_j,L_i).$$
(15)
In fact it is also true that only one of the direct summands on the right hand side of the above equation is nonzero, and so we see that embedding $`S`$ in $`X`$ creates new open string degrees of freedom โ new $`\mathrm{Ext}^1`$โs corresponding to reversing $`\mathrm{Ext}^2`$โs in the del Pezzo quiver. We can add in arrows corresponding to these new $`\mathrm{Ext}^1`$โs to obtain the completed quiver. For example, the completion of the Beilinson quiver is
(16)
while the completion of the $`\mathrm{dP}_1`$ quiver becomes
(17)
## 3 Superpotentials from Topological Field Theory
### 3.1 Topological Field Theory
Having developed and reviewed the requisite mathematical machinery, let us get to the problem at hand, namely computing superpotentials for effective dimensionally reduced theories . Our setting is, as we said, $`M\times X`$ with $`M`$ being four dimensional Minkowski space and $`X`$ a Calabi-Yau threefold. In general, the object is to figure out how to obtain the superpotential for a specified distribution of space-filling branes โ the case of interest involves putting $`D3`$ branes (which look like points in $`X`$) on a collapsing del Pezzo cycle $`S`$ in $`X`$, but let us for the purpose of developing some formalism first tackle the case of a single space-filling and Calabi-Yau filling $`D9`$ brane, described by a complex line bundle $`EX`$ with a hermitian connection. (By itself this case is unphysical, in a sense, because of anomalies but the topological field theory makes perfect sense.)
In this case, the massless four dimensional field content is determined by the Dolbeault cohomology of $`X`$ valued in $`\mathrm{End}(E)`$, $`H_\overline{}^{0,q}(X,\mathrm{End}(E))`$. Specifically, the number of vector bosons is given by $`H_\overline{}^{0,0}(X,\mathrm{End}(E))=\mathrm{End}(E)`$, where by abuse of notation the second term refers to the space of global sections of $`\mathrm{End}(E)`$. We will work with simple line bundles, for which $`\mathrm{End}(E)=`$. We could of course also take $`N`$ copies of the brane, $`E^N`$, whereby we obtain a $`U(N)`$ gauge boson. Likewise, the number of chiral superfields is given by $`H_\overline{}^{0,1}(X,\mathrm{End}(E))`$. Again, these are in the adjoint of $`U(N)`$ when we take the bundle to be $`E^N`$.
In order to get a term in the tree-level superpotential, we have to compute a disk diagram with boundary insertions of vertex operators that correspond to the chiral superfields that appear in that term. What makes this problem computationally tractable is the fact that this disk diagram can be computed in a topological theory ; it is in some sense protected from $`\alpha ^{}`$ corrections. Specifically, the open string topological $`B`$-model on $`X`$ with a $`D`$-brane $`E`$ has open string spectrum given by $`A=H_\overline{}^{0,q}(X,\mathrm{End}(E))`$. Thus, if we define the disk correlation functions as:
$$B_{i_0,i_1,\mathrm{},i_k}=(1)^{\zeta _1+\zeta _2+\mathrm{}+\zeta _{k1}}\psi _{i_0}\psi _{i_1}P\psi _{i_2}^{(1)}\psi _{i_3}^{(1)}\mathrm{}\psi _{i_{k1}}^{(1)}\psi _{i_k},$$
(18)
Here the $`\psi _{i_m}`$ are vertex operators of ghost number one, i.e., they correspond to states in $`H_\overline{}^{0,1}(X,\mathrm{End}(E))`$. If we let $`Z_i`$ be the effective four dimensional superfield corresponding to the open string mode $`\psi _i`$, then the superpotential is
$$W=\mathrm{Tr}\left(\underset{k=2}{\overset{\mathrm{}}{}}\underset{i_0,i_1,\mathrm{},i_k}{}\frac{B_{i_0,i_1,\mathrm{},i_k}}{k+1}Z_{i_0}Z_{i_1}\mathrm{}Z_{i_k}\right).$$
(19)
What have we accomplished by reducing the problem to a computation in a topological sigma model? Heuristically, the situation is as follows : we have, by reducing to the topological theory, essentially gotten rid of the higher mode excitations of the string. Hence the disk diagram we want is really a sum of Feynman diagrams in a field theory, called holomorphic Chern-Simons theory. Because big Feynman diagrams can be built from smaller ones, we obtain from this way of looking at things combinatorial relations among the correlators, called $`A_{\mathrm{}}`$ relations, and it turns out that these determine the correlators uniquely (up to field redefinition). In fact, the $`A_{\mathrm{}}`$ relations give a specific algorithm for generating the correlators, and this algorithm generalizes to a more general setting where $`D`$-branes are represented as elements of the derived category of coherent sheaves.
We now proceed to flesh out the above heuristic and describe the algorithm in detail. First, we briefly review some mathematical background on $`A_{\mathrm{}}`$ products.
### 3.2 $`A_{\mathrm{}}`$ structure
Given a graded vector space $`B`$, such as the Dolbeault complex graded by $`q`$ defined above, an $`A_{\mathrm{}}`$ structure on $`B`$ is defined as a series of products $`m_k`$, $`k1`$, of degree $`2k`$
$$m_k:B^kB,$$
(20)
which satisfy the $`A_{\mathrm{}}`$ constraints:
$$\underset{r+s+t=n}{}(1)^{r+st}m_u(\mathrm{๐}^rm_s\mathrm{๐}^t)=0,$$
(21)
for any $`n>0`$, where $`u=n+1s`$. Here we assume the usual sign rule
$$(fg)(ab)=(1)^{|g|.|a|}f(a)g(b)$$
(22)
when moving arguments past operators.
The $`A_{\mathrm{}}`$ products can actually be rephrased in terms of a differential acting on a certain space, with the complicated and unnatural looking relations between them being just the condition that the differential squares to zero . We will not pursue this interpretation here however, except to note that it is useful to consider maps between spaces that commute with the differential. In terms of the $`A_{\mathrm{}}`$ products, such a map between two spaces $`B`$ and $`B^{}`$ is described as an $`A_{\mathrm{}}`$ morphism, which is to say it is given by a series of maps
$$f_k:B^kB^{},$$
(23)
for $`k1`$, which satisfy
$$\underset{r+s+t=n}{}(1)^{r+st}f_u(\mathrm{๐}^rm_s\mathrm{๐}^t)=\underset{\genfrac{}{}{0.0pt}{}{1rn}{i_1+\mathrm{}+i_r=n}}{}(1)^qm_r(f_{i_1}f_{i_2}\mathrm{}f_{i_r}),$$
(24)
for any $`n>0`$, $`u=n+1s`$, and $`q=(r1)(i_11)+(r2)(i_21)+\mathrm{}+(i_{r1}1)`$.
Now note that the $`A_{\mathrm{}}`$ relations give $`m_1m_1=0`$, so that $`B`$ has the structure of a graded differential complex, and we can take cohomology $`H^{}(B)`$. We now come to a theorem that forms the basis for the computational tractability of our results. Let $`B`$ be as above, except assume that all products $`m_k`$ are zero for $`k3`$ โ this structure is called a differential graded algebra (dga). Given an embedding $`i:H^{}(B)B`$ Kadeishvili shows that we may define an $`A_{\mathrm{}}`$ structure on $`H^{}(B)`$ that has $`m_1=0`$ and an $`A_{\mathrm{}}`$ morphism $`f`$ from $`H^{}(B)`$ to $`B`$ with $`f_1`$ equal to the embedding $`i`$. Furthermore if $`B`$ and $`B^{}`$ are quasi-isomorphic dgaโs (that is, there is a map from one to the other that induces an isomorphism on cohomology) then the two Kadeishvili $`A_{\mathrm{}}`$ structures on $`H^{}(B)`$ and $`H^{}(B^{})`$ are $`A_{\mathrm{}}`$-isomorphic.
There is in fact a well defined algorithm for determining the $`A_{\mathrm{}}`$ products of Kadeishviliโs theorem. The above condition for an $`A_{\mathrm{}}`$ morphism, for the case $`n=2`$, gives
$$im_2=(ii)+df_2.$$
(25)
The cohomology class of the right hand side of the above equation is just that of $`ii`$ and hence $`m_2`$ is uniquely determined. Therefore $`df_2`$ is also uniquely determined, and we can invert $`d`$ to obtain a (non-unique) choice of $`f_2`$. Now putting $`n=3`$ we have
$$im_3=f_2(\mathrm{๐}m_2)f_2(m_2\mathrm{๐})+(if_2)(f_2i)+df_3.$$
(26)
Once again, this equation uniquely determines $`m_3`$ and $`df_3`$, and allows us to make a choice of $`f_3`$. Continuing in this way, it is apparent that all $`A_{\mathrm{}}`$ products can be determined. The ambiguity in the choice of $`f_k`$ reflects the ambiguity in the uniqueness clause of the above theorem.
### 3.3 Holomorphic Chern-Simons Theory
The field theory that the topological $`B`$-model on $`X`$ reduces to is holomorphic Chern-Simons theory:
$$S=_X\mathrm{Tr}\left(๐ \overline{}๐ +\frac{2}{3}๐ ๐ ๐ \right)\mathrm{\Omega },$$
(27)
where the $`๐ `$ is a $`(0,1)`$-form on $`X`$ taking values in $`\mathrm{End}(E)`$, and $`\mathrm{\Omega }`$ is a holomorphic $`(3,0)`$-form on $`X`$. As mentioned above, computation of the disk correlator in holomorphic Chern-Simons theory reduces to a sum of Feynman diagrams (this reduction can be seen explicitly as localization of the supersymmetric path integral on Feynman fat-graph configurations arising from instantons at infinity, see ). The combinatorial relations which the Feynman diagram picture gives rise to are precisely the $`A_{\mathrm{}}`$ relations. To make a rigorous statement, first define a trace map
$$\gamma (a)=_X\mathrm{Tr}(a)\mathrm{\Omega },$$
(28)
$`\gamma `$ is a degree $`3`$ map in the sense that only $`(0,3)`$-forms $`a`$ have nonzero trace. Define $`m_1`$ to be $`\overline{}`$ and $`m_2`$ to be the wedge product together with composition in $`\mathrm{End}(E)`$ โ these give the Dolbeault complex the structure of a dga. The embedding of $`\overline{}`$ cohomology into the Dolbeault complex by harmonic forms then gives via Kadeishvili an $`A_{\mathrm{}}`$ structure to $`H_\overline{}^{0,q}(X,\mathrm{End}(E))`$. The correlation functions can then be written :
$$B_{i_0,i_1,\mathrm{},i_k}=\gamma \left(m_2(m_k(\psi _{i_0},\psi _{i_1},\mathrm{},\psi _{i_{k1}}),\psi _{i_k})\right),$$
(29)
They satisfy the cyclicity property :
$$B_{i_0,i_1,\mathrm{},i_k}=(1)^{\zeta _k(\zeta _0+\zeta _1+\mathrm{}+\zeta _{k1})}B_{i_k,i_0,i_1,\mathrm{},i_{k1}}.$$
(30)
which will be important to us later.
Up to now we have been dealing with a single Calabi-Yau filling $`D`$-brane. The advantage of working in the above framework is that it extends easily to more general $`D`$-brane configurations. For example, (still for a single brane $`E`$) we may replace the Dolbeault complex by a ฤech complex, thereby turning a difficult problem in analysis, namely inverting $`\overline{}`$, into a more manageable combinatorial one. The uniqueness theorem above guarantees that the two $`A_{\mathrm{}}`$ structures obtained are $`A_{\mathrm{}}`$-isomorphic. We could also use an injective resolution of a sheaf instead of the ฤech complex, and by appropriate abstraction reframe the entire discussion in terms of $`๐(X)`$. In fact, for now the most convenient complex for us to use is a hybrid of the ฤech complex and that obtained from locally free resolutions (i.e., resolutions by vector bundles). Specifically, we claim that, for a $`D`$-brane represented in the derived category by the locally free resolution
$$\mathrm{E}^{}=\left(\text{}\right).$$
(31)
the following complex has cohomology that gives the correct open string spectrum for the brane and induced $`A_{\mathrm{}}`$ structure that gives rise to the correct superpotential:
(32)
where
$$\begin{array}{cc}\hfill \mathrm{B}^n& =\underset{p+q=n}{}\mathrm{B}^{p,q}\hfill \\ \hfill \mathrm{B}^{p,q}& =\stackrel{ห}{C}^p(๐,\mathrm{H}\text{om}^q(\mathrm{E}^{},\mathrm{E}^{})).\hfill \end{array}$$
(33)
This is shown in . Two points need to be made here. First, the differential in (32) is $`d=\delta +(1)^p๐ก_q`$, with $`\delta `$ the ฤech differential and $`๐ก_q`$ given by
$$๐ก_nf_{n,p}=๐ฝ_{p+n}f_{n,p}(1)^nf_{p+1,n}๐ฝ_p.$$
(34)
where $`_pf_{n,p}`$, with $`f_{n,p}:\mathrm{E}^p\mathrm{E}^{p+n}`$, is an element of
$$\mathrm{H}\text{om}^n(\mathrm{E}^{},\mathrm{E}^{})=\underset{p}{}\mathrm{H}\text{om}(\mathrm{E}^p,\mathrm{E}^{p+n}).$$
(35)
Second, to rigorously show that $`\mathrm{B}`$ indeed reproduces the correct spectrum and superpotential is non-trivial and requires an analysis of elements of the derived category as boundary states of the worldsheet theory .
## 4 Superpotentials for del Pezzo singularities
### 4.1 Moduli Spaces
Before launching into a more rigorous discussion, let us first consider a heuristic argument that will lead to a conjecture for the form of the superpotential.
First we quickly review the connection of the mathematics of quivers to the physics of D-branes and stability. One should view a quiver as representing a decay of a D-brane. The nodes in the quiver correspond to the decay products, i.e., the so-called โfractional branesโ and the arrows correspond to open strings between these decay products. The D-brane we are particularly concerned with is the 3-brane corresponding to a point in $`X`$.
At the instant of decay, the open strings corresponding to the arrows should be exactly massless. In the case of B-branes, these masses are a function of the complexified Kรคhler form $`B+iJ`$. Here we assume that this masslessness occurs precisely when the del Pezzo surface is collapsed to a point. This assumption was justified in .
If one moves away from the critical point where the open strings are massless, then the D-brane may become stable or unstable with respect to the decay. If we deform the Kรคhler form to some generic value to give a nonzero size the del Pezzo surface (and all the curves within it) then we expect the 3-brane to be stable.
In this resolution one may compute the moduli space of the 3-brane, which should, of course, yield $`X`$ itself. We need not concern ourselves with the details of this process but we note the following. For more details we refer to . The moduli space of 3-branes is essentially given by the moduli space of representations of the quiver. One takes all possible quiver representations which satisfy โ$`\theta `$-stabilityโ and then divides by a gauge equivalence.
Physically this moduli space is given by the moduli space of chiral fields (given by the matrices associated to arrows in the quiver) corresponding to classical solutions of the field theory divided by gauge equivalence. Importantly for us, this must mean that the superpotential imposes conditions on the chiral fields equivalent to the relations in the quiver.
In other words, finding the critical points of the superpotential must be equivalent to imposing the quiver relations. This leads to an obvious proposition for the superpotential. Let $`A_i`$ be the chiral fields in the worldvolume gauge theory associated to the arrows in the (non-completed) quiver associated to a del Pezzo surface. The relations will be denoted $`r_k(A_1,A_2,\mathrm{})=0`$, where $`r_k`$ is some polynomial. We know from section 2.3 that each $`r_k`$ is associated to some arrow in the completed quiver, and so some chiral field $`R_k`$. It is believed (see , for example) that in terms of the moduli space, setting all $`R_k`$ equal to zero amounts to restricting the 3-brane is be on the del Pezzo surface $`S`$ itself. Giving nonzero expectation values to the $`R_k`$ fields moves the 3-brane off $`S`$.
If the superpotential is given by
$$W=\underset{k}{}R_kr_k(A_1,A_2,\mathrm{}),$$
(36)
then, on $`S`$, the equations of motion will yield precisely the correct constraints, at least for 3-branes on $`S`$. This, therefore, is our conjectured form for the superpotential.
### 4.2 Quiver Gauge Theories for del Pezzos
Let us now consider more systematically what happens when we put a $`D3`$-brane on a shrinking del Pezzo cycle $`S`$ in a Calabi-Yau $`X`$. Now, every BPS space-filling brane corresponds to a topological brane on $`X`$, but not vice versa. A point on $`X`$ is always a valid topological brane; when $`S`$ is of finite size, the $`D3`$-brane will be located on a smooth point of $`S`$ and as we said we expect it to be BPS. On the other hand, when $`S`$ shrinks, one can argue (see e.g. ) that the $`D3`$ is marginally stable against decay into the fractional branes introduced earlier. We think of these fractional branes as wrapping $`S`$ โ when $`S`$ shrinks the point-like $`D3`$ is allowed to marginally decay to them.
To get a precise description of these fractional branes, we recall that they correspond to the representations $`L_i`$, which have resolutions in terms of the projective representations $`P_i`$. If we replace the $`P_i`$ by the corresponding elements of the strongly exceptional collection (which are all vector bundles in the cases we consider) and use the equivalence between derived categories, we obtain locally free resolutions of the fractional branes. For example in the case of $`^2`$ and the Beilinson quiver, $`L_2`$ is represented in $`D(S)`$ as
(37)
The maps in the above complex are determined by the corresponding maps in the quiver resolution of $`L_2`$, using the fact that $`\mathrm{Hom}(\mathrm{F}_i,\mathrm{F}_j)`$ and $`\mathrm{Hom}(P_i,P_j)`$ are naturally isomorphic. It turns out that a D3-brane decays into a collection of fractional branes with each fractional brane occurring dim $`\mathrm{F}_i`$ times . The quiver gauge theory for a $`\mathrm{dP}_k`$ will thus have $`k+3`$ gauge groups, corresponding to each of the $`L_i`$, and massless matter in the bifundamental from strings stretching between $`L_i`$ and $`L_j`$. The proper setting for discussing the homological structure in this context is an $`A_{\mathrm{}}`$ category, but we do not need to get so abstract. We simply let $`\mathrm{M}^{}`$ be the direct sum of the locally free resolutions of the $`L_i`$, and use it as the starting point for the $`A_{\mathrm{}}`$ computations. We remind the reader that $`\mathrm{M}^{}`$ is a locally free resolution of sheaves over $`S`$, not $`X`$.
It will be convenient to represent $`X`$ as the total space of the normal bundle $`N`$ of $`S`$. Because $`S`$ is a del Pezzo, $`NK_S`$. We are allowed to take this limit in Kรคhler moduli space because the tree level superpotential does not depend on Kรคhler parameters.
### 4.3 From Branes on X to Branes on S
The actual superpotential is computed from the $`A_{\mathrm{}}`$ products of the ฤech complex associated to branes not on $`S`$, but on $`X`$, i.e., not from $`\mathrm{B}^,`$, but rather from the associated complex obtained from considering all the sheaves as embedded in $`X`$. The goal of this section is to show that the computation of $`A_{\mathrm{}}`$ products associated to branes on $`X`$ essentially reduces to the computation on $`S`$. Specifically, we recall that (as we will see in greater detail below) $`\mathrm{Ext}^1`$โs of sheaves on $`S`$ considered as sheaves on $`X`$ include all the $`\mathrm{Ext}^1`$โs of the sheaves on $`S`$ plus some extra $`\mathrm{Ext}^1`$โs, which, after a reversal of arrows, correspond to $`\mathrm{Ext}^2`$โs of the sheaves on $`S`$. We will prove that there is a choice of $`A_{\mathrm{}}`$ structure over $`X`$ such that all $`A_{\mathrm{}}`$ products that contain more than one of the โextraโ $`\mathrm{Ext}^1`$โs vanish. The products that contain no โextraโ $`\mathrm{Ext}^1`$โs are the same as they were over $`S`$, and the ones with one โextraโ $`\mathrm{Ext}^1`$ are determined uniquely by the requirement of cyclicity. This, together with the fact that there are no cycles in del Pezzo quivers, will show that the superpotential is linear in the โextraโ $`\mathrm{Ext}^1`$s, with these $`\mathrm{Ext}^1`$s multiplying terms that are just the quiver relations.
To proceed with the proof, let $`\pi :ES`$ be the projection from the total space of $`K_S`$ to the del Pezzo $`S`$. We have a canonical section $`\mathrm{O}\pi ^{}\mathrm{K}_\mathrm{S}`$, given as follows: to each point in $`E`$ we tautologically associate a point of $`S`$ and an element of the fiber of $`K_S`$ over that point; this element can be viewed as an element of the fiber of $`\pi ^{}K_S`$ over the original point in $`E`$. Dualizing, we get a canonical map which fits into an exact sequence of sheaves
$$0\pi ^{}(K_{S}^{}{}_{}{}^{})\mathrm{O}\mathrm{O}_\mathrm{S}0.$$
(38)
and we can think of the first two terms as a locally free resolution of $`\mathrm{O}_\mathrm{S}`$. The key point now is to tensor the resolution $`\mathrm{M}^{}`$ with the above resolution of $`\mathrm{O}_\mathrm{S}`$ in order to obtain a locally free resolution of $`i_{}M`$:
(39)
Collapsing the above double complex along the diagonal we get a free resolution, and the associated spectral sequence, which collapses at the $`E_2`$ term, shows that it is in fact a resolution of $`i_{}\mathrm{M}`$. This can also be viewed as the Cone construction . We will choose to retain the bi-grading, so let us represent the above resolution as $`\mathrm{M}^,`$ (the first index corresponds to the index of $`\mathrm{M}^{}`$, and the second is either $`0`$, for $`\pi ^1\mathrm{M}^{}`$, or $`1`$, for $`\pi ^{}(K_{S}^{}{}_{}{}^{})\pi ^1\mathrm{M}^{}`$). We now define
$$\mathrm{C}^{p,i,j}=\stackrel{ห}{C}^p(\pi ^1๐,\mathrm{H}\text{om}^{i,j}(\mathrm{M}^,,\mathrm{M}^,))$$
(40)
Here $`๐`$ is an affine open cover of $`S`$, and hence $`\pi ^1๐`$ is an affine open cover of $`E`$. The complex $`\mathrm{C}`$ is central in our analysis. There are several differentials we define on $`\mathrm{C}`$. First of all, in the locally free resolution $`\mathrm{M}^,`$ label the differentials that increase the first index by $`๐ฝ_{}^{0}{}_{i,j}{}^{}`$ and that which increases the second index by $`๐ฝ_{}^{1}{}_{i,j}{}^{}`$. The combination $`๐ฝ_{i,j}=๐ฝ_{i,j}^0+(1)^i๐ฝ_{i,j}^1`$ is the standard differential associated to the locally free resolution of $`\mathrm{M}`$ over $`E`$. Now, given a section of $`\mathrm{H}\text{om}^{i,j}(\mathrm{M}^{},\mathrm{M}^{})`$ over an open set $`U`$, i.e., a section of $`_{p,q}\mathrm{Hom}_U(\mathrm{M}^{p,q},\mathrm{M}^{p+i,q+j})`$, we can denote it by
$$\underset{p,q}{}f_{p,q}^{i,j},$$
(41)
where
$$f_{p,q}^{i,j}:\mathrm{M}^{p,q}(U)\mathrm{M}^{p+i,q+j}(U)$$
(42)
Then we can define differentials
$$๐ก_{}^{0}{}_{i,j}{}^{}f_{p,q}^{i,j}=๐ฝ_{}^{0}{}_{i,j}{}^{}f_{p,q}^{i,j}(1)^{(i+j)}f_{p+1,q}^{i,j}๐ฝ_{}^{0}{}_{i,j}{}^{}$$
(43)
$$๐ก_{}^{1}{}_{i,j}{}^{}f_{p,q}^{i,j}=๐ฝ_{}^{1}{}_{i,j}{}^{}f_{p,q}^{i,j}(1)^{(i+j)}f_{p,q+1}^{i,j}๐ฝ_{}^{1}{}_{i,j}{}^{}.$$
(44)
Finally, we have the ฤech differential $`\delta `$, and we define the total differential on $`\mathrm{C}`$ by $`d=\delta +(1)^p\left(๐ก^0+(1)^i๐ก^1\right)`$. Now, the sum $`๐ก^0+(1)^i๐ก^1`$ is the standard differential associated to the locally free resolution of $`\mathrm{M}`$ over $`E`$, so that collapsing on the $`(i,j)`$ indices yields the double complex in , showing that $`\mathrm{C}`$ does indeed correctly compute the $`A_{\mathrm{}}`$ products.
To actually get a handle on determining the $`A_{\mathrm{}}`$ algebra, it is useful to collapse $`\mathrm{C}`$ in a different way and leverage our knowledge of the $`A_{\mathrm{}}`$ structure for sheaves on the del Pezzo $`S`$. Specifically, let us collapse the complex on the $`(p,i)`$ indices:
$$\mathrm{D}^{q,j}=\underset{p+i=q}{}\mathrm{C}^{p,i,j}$$
(45)
$`\mathrm{D}^,`$ is a double complex with anticommuting differentials $`d_0=\delta +(1)^p๐ก^0`$, which increases the first index, and $`d_1=(1)^q๐ก^1`$, which increases the second one, that add up to $`d`$. The desired cohomology is computed using a spectral sequence associated to this double complex, which by arguments of degenerates at the $`E_2`$ term to give:
(46)
(In our exposition the $`E_1`$ term is given by taking cohomology with respect to $`d_1`$). Serre duality shows that $`\mathrm{Ext}_S^i(M,M)\mathrm{Ext}_S^{2i}(M,MK_S)`$, so that
$$\mathrm{Ext}_X^1(i_{}M,i_{}M)\mathrm{Ext}_S^1(M,M)\mathrm{Ext}_S^2(M,M).$$
(47)
The two terms on the right correspond to the $`\mathrm{Ext}^1`$s and โextraโ $`\mathrm{Ext}^1`$s, respectively.
Now, the bottom row in the above diagram reproduces the cohomology of the complex $`\mathrm{B}^{p,i}`$ โ the complex associated to branes on $`S`$ rather than $`X`$. In fact, we may naturally embed $`\mathrm{B}^{p,i}`$ in
$$\stackrel{ห}{C}^p(\pi ^1๐,\mathrm{H}\text{om}^i(\pi ^1\mathrm{M}^{},\pi ^1\mathrm{M}^{})).$$
(48)
This complex, in turn, can be viewed as a sub-complex of $`\mathrm{C}^{p,i,0}`$. To see why, note that, from (39), a section of $`\mathrm{H}\text{om}^i(\pi ^1\mathrm{M}^{},\pi ^1\mathrm{M}^{})`$ determines a section of $`\mathrm{H}\text{om}^{i,0}(\mathrm{M}^,,\mathrm{M}^,)`$; basically it gives directly the maps among the sheaves in the upper row of (39), and, taking the identity map on $`\pi ^{}(K_{S}^{}{}_{}{}^{})\pi ^1\mathrm{M}^{}`$, it determines the maps for the sheaves in the lower row as well. Also, from (46), we see that the composition of these embeddings induces an isomorphism from the cohomology of $`\mathrm{B}^{p,i}`$ to the $`j=0`$ part of the cohomology of $`\mathrm{C}^{p,i,j}`$.
Now, in order to carry out the $`A_{\mathrm{}}`$ procedure we must choose representatives for all cohomology classes in $`\mathrm{C}^{p,i,j}`$. The upshot of the construction in the previous paragraph is that it gives us a natural choice of representatives of the $`j=0`$ part of the cohomology; in fact, it shows that the $`A_{\mathrm{}}`$ products of these $`j=0`$ cohomology classes are exactly the same as those in $`\mathrm{B}^{p,i}`$. In other words, for the $`j=0`$ generators the $`A_{\mathrm{}}`$ products are just those defined over $`S`$. This accomplishes part of our goal of reducing the computation over $`X`$ to a computation over $`S`$; to finish we have to deal with products that may contain some $`j=1`$ generators.
The $`j=1`$ cohomology generators are the ones that contribute the โextraโ $`\mathrm{Ext}^1`$s. To carry out the $`A_{\mathrm{}}`$ procedure, we must pick representatives of their cohomology classes. We choose these to be homogeneous of $`j`$ degree $`1`$, or, in other words, to lie in
$$\stackrel{ห}{C}^p(\pi ^1๐,\mathrm{H}\text{om}^i(\pi ^1\mathrm{M}^{}\pi ^1K_S,\pi ^1\mathrm{M}^{}))$$
(49)
Clearly this is the most natural choice, though it should be pointed out that we could have done something stupid and chosen the generator to have a nonzero (exact) $`j=0`$ part, for example. The advantage of having homogeneous generators is that their products are homogeneous as well, and so vanish if they have $`j>1`$.
Now we claim that any $`m_k`$ that contains more than one $`j=1`$ generator vanishes. The naive argument would invoke the $`j`$ grading and the fact that there are no elements in $`\mathrm{C}^{p,i,j}`$ with $`j>1`$. The obvious flaw is the fact that $`m_k`$ does not respect the overall grading โ it in fact has degree $`2k`$. Thus we have to be more careful. We claim that although $`m_k`$ has nonzero degree with respect to the overall grading $`p+i+j`$, through a careful choice of $`f_k`$, which we now construct, we can make $`m_k`$ respect the $`j`$ grading. The claim at the top of this paragraph then immediately follows.
We show by induction that all the $`f_k`$ and $`m_k`$ respect the $`j`$ grading. Clearly $`f_1`$ and $`m_1`$ respect the $`j`$ grading. Suppose that this is also true for all $`kn`$. We have for all $`n+1`$ (24), which can be rewritten as an equation determining $`m_{n+1}`$ in terms of the lower $`m_k`$ and $`f_k`$ (for $`n+1=3`$, for example, this is (26)). So we immediately see that itโs true for $`m_{n+1}`$. We now deal with $`f_{n+1}`$. We suppress its arguments, but everywhere below $`f_{n+1}`$ and $`df_{n+1}`$ will stand for $`f_{n+1}`$ and $`df_{n+1}`$ applied to their arguments. Now equation (24) again gives $`df_{n+1}`$ as an expression in terms of $`m_{n+1}`$ and the $`m_k`$ and $`f_k`$ for $`kn`$. We have to make a choice of $`f_{n+1}`$ that respects the $`j`$ grading, i.e., we want $`f_{n+1}`$ to have the same $`j`$ degree as $`df_{n+1}`$. Now, the case when all the arguments have $`j=0`$ has been discussed above and clearly we have already defined $`f_{n+1}`$ to have $`j=0`$. When more than one argument has $`j=1`$ then, because all the terms in the expression for $`df_{n+1}`$ are homogeneous, we have $`df_{n+1}=0`$, so that we can choose $`f_{n+1}=0`$. The nontrivial case is when exactly one argument has $`j=1`$. In that case, $`df_{n+1}`$ is homogeneous of $`j`$ degree $`1`$ and hence lies in the sub-complex
$$\mathrm{C}^{p,i,1}=\stackrel{ห}{C}^p(\pi ^1๐,\mathrm{H}\text{om}^i(\pi ^1\mathrm{M}^{}\pi ^1K_S,\pi ^1\mathrm{M}^{})).$$
(50)
The crucial point is now that the embedding $`\mathrm{C}^{p,i,1}\mathrm{C}^{p,i,j}`$ induces an injection in cohomology. This can easily be seen from the spectral sequence (46) โ the cohomology of $`\mathrm{C}^{p,i,1}`$ reproduces precisely the upper, $`j=1`$ row in the diagram. Therefore $`df_{n+1}`$ is exact not only in $`\mathrm{C}^{p,i,j}`$ but also in the sub-complex $`\mathrm{C}^{p,i,1}`$. Therefore we can choose $`f_{n+1}`$ to be in $`\mathrm{C}^{p,i,1}`$, so that it will have $`j=1`$. Thus we see that we can always choose $`f_{n+1}`$ to respect the $`j`$ grading. This completes the inductive step.
Together with the cyclicity property this determines all the $`A_{\mathrm{}}`$ products in $`\mathrm{C}^{p,i,j}`$ in terms of those over $`S`$. To restate, we have the original $`A_{\mathrm{}}`$ algebra reproduced when all the arguments are the original $`\mathrm{Ext}^1`$โs (i.e., have $`j=0`$), any product that involves more than one โextraโ $`\mathrm{Ext}^1`$ (i.e., one that has $`j=1`$) must vanish, while any product that contains exactly one โextraโ $`\mathrm{Ext}^1`$ is determined uniquely by requiring it to reproduce correlators that obey the cyclicity property (30). Having accomplished the reduction and thus shown that the superpotential is linear in the โextraโ $`\mathrm{Ext}^1`$s, we now determine the $`A_{\mathrm{}}`$ products over $`S`$ and relate them to the quiver relations.
## 5 $`A_{\mathrm{}}`$ relations and quivers
We must determine the $`A_{\mathrm{}}`$ products over $`S`$, i.e., those of $`\mathrm{B}^,`$, defined in (33). We know by Bondalโs theorem that $`๐(S)`$ is equivalent to $`๐(A\text{}\text{mod})`$, where $`A`$ is the path algebra of the associated quiver. The operational version of this equivalence that will suffice for us is as follows. First, construct a complex of quiver representations $`M^{}`$ by summing the projective resolutions of the $`L_i`$. In the usual way it gives rise to the graded dga $`\mathrm{End}(M)^{}`$. There is a natural map of this complex into $`\mathrm{B}`$ given by interpreting maps in $`\mathrm{End}(M)`$ as global sections of the $`\mathrm{Hom}(\mathrm{F}_i,\mathrm{F}_j)`$ and mapping them to ฤech $`0`$-cochains. Because these $`0`$-cochains are global sections, they are annihilated by the ฤech part of the differential, and one thus quickly sees that this map is a map of dgaโs. In fact, (the derivation of) Bondalโs theorem shows that it is a quasi-isomorphism. This allows us to apply Kadeishviliโs theorem and compute the $`A_{\mathrm{}}`$ structure of $`\mathrm{B}`$ in the quiver dga $`\mathrm{End}(M)^{}`$.
We will see that we obtain a form of the superpotential exactly as conjectured in section 4.1.
### 5.1 A simple example
We start with the Beilinson quiver, corresponding to $`S=^2`$:
(51)
with relations $`a_\alpha b_\beta =a_\beta b_\alpha `$. We recall that we have the projective resolutions:
(52)
and that
$$\begin{array}{cc}\hfill dim\mathrm{Ext}^1(L_i,L_j)& =n_{ij}\hfill \\ \hfill dim\mathrm{Ext}^2(L_i,L_j)& =r_{ij},\hfill \end{array}$$
(53)
where $`n_{ij}`$ counts arrows and $`r_{ij}`$ counts relations.
We start by choosing specific generators of the $`\mathrm{Ext}^i`$. Recalling that the $`\mathrm{Ext}^{}`$ can be represented as morphisms between resolutions of the $`L_i`$, we can choose the three generators $`๐บ_i`$ of $`\mathrm{Ext}^1(L_1,L_0)`$ to be
(54)
where $`\pi _i`$ is projection on the $`i`$โth factor. As far as the generators $`๐ป_i`$ of $`\mathrm{Ext}^1(L_2,L_1)`$ we have $`๐ป_0`$ represented as
(55)
and the other $`๐ป_i`$ represented similarly. We also have the relations $`๐_i`$ in $`\mathrm{Ext}^2(L_3,L_2)`$, represented by
(56)
Clearly, the only potentially nontrivial products are $`m_2(๐บ_j,๐ป_i)`$, and one easily sees by composing the representatives for $`๐บ_j`$ and $`๐ป_i`$ that $`m_2(๐บ_j,๐ป_i)=ฯต^{ijk}r_k`$. This gives rise to the superpotential
$$W=ฯต^{ijk}A_iB_jR_k$$
(57)
which is the correct superpotential for this quiver gauge theory on the orbifold $`^3/_3`$ . Again, the $`A_i`$ and $`B_j`$ are massless moduli corresponding to the internal structure of the shrinking cycle $`^2`$, while $`R_k`$ is the modulus that corresponds to moving the $`D3`$-brane off the singularity. We note that the superpotential is of the desired form, linear in the โextraโ $`\mathrm{Ext}^1`$s $`R_i`$, which multiply the relations. To write it out explicitly, we have
$$W=(A_0B_1B_1A_0)R_2+(A_1B_2B_2A_1)R_0+(A_2B_0B_0A_2)R_1$$
(58)
### 5.2 del Pezzo 1
Let us consider the quiver associated to $`\mathrm{dP}_1`$. It is:
(59)
subject to the relations $`r_0=b_0d_1b_1d_0`$, $`s_0=ab_0d_2cd_0=0`$, and $`s_1=ab_1d_2cd_1=0`$. Denote the corresponding generators of $`\mathrm{Ext}^2`$ by $`๐_0`$, $`๐_0`$, and $`๐_1`$. We first pick maps of projective resolutions representing these generators, which all turn out to be uniquely determined. We have the projective resolutions:
(60)
We choose representatives of $`\mathrm{Ext}^1(L_i,L_j)`$ as follows: for $`i3`$, there are no relations originating at the $`i`$โth node of the quiver and hence the maps representing $`๐ป_i`$, $`๐บ`$, and $`๐ผ`$ are uniquely determined. The choice of representative of $`๐ฝ_i`$ is uniquely determined as well. That is to say, in the diagram
(61)
the bottom horizontal map is injective, so that the left vertical map is uniquely determined by the right vertical map, which we take to be projection on the $`i`$โth factor. Finally, for the generators of $`\mathrm{Ext}^2`$ we take the obvious uniquely determined maps from the projective resolution of $`L_3`$ to the other $`L_i`$.
We want to compute all products $`m_k`$ of the various $`\mathrm{Ext}^1`$โs. Such products will all be in $`\mathrm{Ext}^2`$, and because only $`\mathrm{Ext}^2(L_3,L_0)`$ and $`\mathrm{Ext}^2(L_3,L_1)`$ are nonzero we see that the possible nonzero products are $`m_2(๐ป_i,๐ฝ_j)`$, $`m_2(๐ผ,๐ฝ_i)`$, and $`m_3(๐บ,๐ป_i,๐ฝ_j)`$. Letโs look at the first of these; the relevant composition is:
(62)
The map from the first to the second row is $`๐ฝ_j`$ and that from the second to the third row is $`๐ป_i`$. For convenience, we label the individual $`P_k`$s that occur in various parts of the diagram. The upper left entry is $`P_0^2P_1`$, where each summand corresponds to a different relation. We naturally label the two $`P_0`$s as $`S_0`$ and $`S_1`$, and we label the $`P_1`$ as $`R_0`$. The entry below this one is $`P_0P_1^2`$, and here each $`P_k`$ corresponds to an arrow emanating from $`P_2`$. Thus we label the $`P_0`$ as $`C`$ and the two $`P_1`$s as $`B_0`$ and $`B_1`$.
Let us consider the possible maps we can have. For $`k>l`$ there are no nonzero maps from $`P_k`$ to $`P_l`$. From each $`P_k`$ to itself there is the identity map, and it is the only one that will be of use to us. For $`k<l`$, however, there are several ways to map $`P_k`$ to $`P_l`$, each corresponding to a path from node $`l`$ to node $`k`$. Thus, for example, there is one map from $`P_0`$ to $`P_1`$, denoted by $`a`$.
To see how $`๐ฝ_j`$ acts note that it maps $`R_0S_0S_1`$ to $`CB_0B_1`$. We can thus represent its action on $`R_0S_0S_1`$ as a 3 by 3 matrix. From the definition of $`๐ฝ_j`$ it is easy to obtain (by slight abuse of notation we denote by $`๐ฝ_j`$ both itself and its restriction to $`R_0S_0S_1`$):
$$๐ฝ_0=\left(\begin{array}{ccc}0& 1& 0\\ 0& 0& 0\\ 1& 0& 0\end{array}\right),๐ฝ_1=\left(\begin{array}{ccc}0& 0& 1\\ 1& 0& 0\\ 0& 0& 0\end{array}\right),๐ฝ_2=\left(\begin{array}{ccc}0& 0& 0\\ 0& a& 0\\ 0& 0& a\end{array}\right).$$
(63)
Note that these matrix elements are simply obtained by โcontractingโ the relevant relation (which indexes the column) with $`d_j`$ to obtain the elements of the column. Similar reasoning shows that the $`๐ป_i`$ act as follows:
$$๐ป_0=\left(\begin{array}{ccc}0& 1& 0\end{array}\right),๐ป_1=\left(\begin{array}{ccc}0& 0& 1\end{array}\right).$$
(64)
The nonzero compositions are
$$๐ป_0๐ฝ_1=\left(\begin{array}{ccc}1& 0& 0\end{array}\right),๐ป_0๐ฝ_2=\left(\begin{array}{ccc}0& a& 0\end{array}\right),$$
(65)
$$๐ป_1๐ฝ_0=\left(\begin{array}{ccc}1& 0& 0\end{array}\right),๐ป_1๐ฝ_2=\left(\begin{array}{ccc}0& 0& a\end{array}\right),$$
(66)
Now, note that the compositions $`๐ป_0๐ฝ_2`$ and $`๐ป_1๐ฝ_2`$ can both be factored through the leftmost map $`P_0P_1`$, and hence are exact in the quiver dga. So the only nonzero products are
$$m_2(๐ป_0,๐ฝ_1)=๐_0,m_2(๐ป_1,๐ฝ_0)=๐_0.$$
(67)
A good shorthand way of expressing this result is that $`m_2(๐ป_i,๐ฝ_j)=(b_id_j,R_0)๐_0+(b_id_j,S_0)๐_0+(b_id_j,S_1)๐_1`$, where the parentheses denote the coefficient of the string represented by the left argument in the relation represented by the right argument. As one traces through the above manipulations it is clear that this is a general result that always holds when one computes the products $`m_2`$. Thus we have $`m_2(๐ผ,๐ฝ_0)=๐_0`$, $`m_2(๐ผ,๐ฝ_1)=๐_1`$, and $`m_2(๐ผ,๐ฝ_2)=0`$.
To compute $`m_3`$, we first have to define a choice of $`f_2`$, which must satisfy
$$im_2=(ii)+df_2.$$
(68)
Quick inspection shows that we may take $`f_2=0`$ everywhere except for $`f_2(๐ป_i,๐ฝ_j)`$. From the above analysis, we see that
$$df_2(๐ป_0,๐ฝ_2)=\left(\begin{array}{ccc}0& a& 0\end{array}\right),df_2(๐ป_1,๐ฝ_2)=\left(\begin{array}{ccc}0& 0& a\end{array}\right).$$
(69)
where $`a`$ denotes right multiplication and the notation indicates a map $`P_0^2P_1P_1`$. Hence we can define $`f_2(๐ป_0,๐ฝ_2)`$ and $`f_2(๐ป_1,๐ฝ_2)`$ respectively as:
(70)
(71)
Now, we have
$$im_3=f_2(\mathrm{๐}m_2)f_2(m_2\mathrm{๐})+(if_2)(f_2i)+df_3,$$
(72)
so that, recalling the sign rule (22), $`m_3(๐บ,๐ป_i,๐ฝ_j)=\left[๐บf_2(๐ป_i,๐ฝ_j)\right]`$. Composing with $`๐บ`$ we see that $`m_3(๐บ,๐ป_0,๐ฝ_2)=๐_0`$ and $`m_3(๐บ,๐ป_1,๐ฝ_2)=๐_1`$. Again, a shorthand way of expressing this result is $`m_3(๐บ,๐ป_i,๐ฝ_j)=(ab_id_j,R_0)๐_0+(ab_id_j,S_0)๐_0+(ab_id_j,S_1)๐_1`$. These are all the nonzero $`A_{\mathrm{}}`$ products in this example.
One can see by carrying out the $`A_{\mathrm{}}`$ algorithm that this formula generalizes to all the $`m_k`$ in any quiver. We sketch the argument modulo various signs, which have to be checked carefully. Let the relations by labeled by $`R_i`$ and the corresponding generators of $`\mathrm{Ext}^2`$ by $`๐_i`$, as above. One proceeds by induction on $`k_0`$. Letโs take the following inductive hypothesis: for all $`j<k_0`$, we have
$$m_j(a_1,\mathrm{},a_j)=\underset{i}{}(a_1\mathrm{}a_j,R_i)๐_i$$
(73)
as well as a statement about $`f_k`$ for which we need to introduce some notation. Each $`a_i`$ is an $`\mathrm{Ext}^1`$, so that it can be represented as a map between the projective resolution of $`L_{m(i)}`$ and $`L_{m(i1)}`$. According to this notation, $`a_i`$ is an arrow between node $`m(i)`$ and $`m(i1)`$. The projective resolution of $`L_{m(i)}`$ is
(74)
The second part of the inductive hypothesis is that for $`j<k_0`$, $`f_j(a_1,\mathrm{},a_j)`$ is represented as:
(75)
where the left vertical map takes each relation to an $`\mathrm{Ext}^1`$ determined by the contraction of the relation with $`a_1\mathrm{}a_j`$ (we extract only the linear terms in the contraction, as only these correspond to $`\mathrm{Ext}^1`$s). $`f_j`$ with any $`\mathrm{Ext}^2`$s as arguments vanish.
For the inductive step, we have to prove the analogous statements for $`m_{k_0}`$. We use (24) to write $`m_{k_0}`$ in terms of the lower $`m_j`$s and $`f_j`$s. We then note from the above form of the $`f_j`$ that all terms vanish except $`(if_{k_01})`$ โ basically because the only nonzero map in (75) takes relations to $`\mathrm{Ext}^1`$s, so the composition of two $`f_j`$s vanishes. This straightforwardly leads to (73) for $`m_{k_0}`$. Inverting $`d`$ shows that $`f_{k_0}`$ may be chosen to be of the form (75).
Thus essentially one knows these products as soon as one knows all the relations in the quiver. It follows that the term in the superpotential that multiplies the โextraโ $`\mathrm{Ext}^1`$ corresponding to a given relation is simply that relation (written as a polynomial in the $`\mathrm{Ext}^1`$s). We can now write down the superpotential:
$$W=R_0(B_0D_1B_1D_0)+S_0(AB_0D_2CD_0)+S_1(AB_1D_2CD_1)$$
(76)
## 6 Conclusions
We have given an effective method for computing superpotentials for quiver gauge theories associated with shrinking del Pezzo cycles. We showed that the superpotential is linear in the fields that correspond to $`\mathrm{Ext}^2`$s in the del Pezzo quiver, and that each such field multiplies a polynomial which is just the corresponding relation. To do this we performed a precise reduction of the problem from one involving sheaves on the Calabi-Yau to one involving sheaves on the del Pezzo. We solved the problem on the del Pezzo by switching to the algebraically more tractable category of quiver representations and explicitly evaluating the $`A_{\mathrm{}}`$ products there. We did only the cases where $`S`$ is $`^2`$ and $`\mathrm{dP}_1`$, but these examples show that the algorithm is trivial to carry out provided one has the quiver and the relations. These, of course, might not be so trivial to obtain, especially for the higher del Pezzos, which themselves have complex structure moduli. These complex structure moduli are contained in the choice of points to be blown up on $`^2`$, and will show up in the quiver relations.
## Acknowledgments
We wish to thank X. Liu, J. McGreevy, A. Saltman, A. Tomasiello for useful conversations. P.S.A. is supported in part by NSF grant DMS-0301476, Stanford University, SLAC and the Packard Foundation. L.M.K. is supported by Stanford University.
|
warning/0506/hep-ex0506042.html
|
ar5iv
|
text
|
# A Search for Supersymmetric Higgs Bosons in the Di-Tau Decay Mode in ๐โข๐ฬ Collisions at โ๐ =1.8 TeV
## I Introduction
The Higgs mechanism in the Minimal Supersymmetric Standard Model (MSSM) Nilles (1984); Haber and Kane (1985); Barbieri (1988) provides a way to assign a mass to each particle while preserving the gauge invariance of the theory just as in the Standard Model (SM). The CP-conserving MSSM contains two $`SU(2)`$ Higgs doublets yielding five physical particles - four CP-even scalars ($`h^0,H^0,H^{}`$ and $`H^+`$) and one CP-odd scalar ($`A^0`$J. Ellis (1991); Y. Okada (1991); Haber and Hempfling (1991). Here, $`h^0`$ is the lighter of the two neutral scalars. In the MSSM there are two parameters at tree level that are conventionally selected to be $`\mathrm{tan}\beta `$ and $`m_{A^0}`$. The parameter $`\mathrm{tan}\beta `$ is the ratio of the vacuum expectation values of the two Higgs doublets, and $`m_{A^0}`$ is the mass of the pseudoscalar Higgs particle. When $`\mathrm{tan}\beta `$ is large, the cross section for direct production of Higgs bosons through gluon fusion becomes enhanced, making that an appealing region for searches at the Tevatron. The coupling strength of the $`A^0`$ boson to down-type fermions with mass $`m_f`$ is proportional to $`m_f\times \mathrm{tan}\beta `$, hence couplings to tau leptons are enhanced. The couplings of the $`h_0`$ are similarly enhanced for many possible models.
No fundamental scalar particle has yet been observed in any experiment. The four experiments at LEP have each performed a search for $`h^0/A^0`$ produced in the process: $`e^+e^{}h^0Z`$ and $`e^+e^{}h^0A^0`$. The combined results of four experiments have constrained the theory, excluding $`m_{A^0}<91.9`$ GeV/$`c^2`$, $`m_{h^0}<91.0`$ GeV/$`c^2`$ and $`0.5<\mathrm{tan}\beta <2.4`$ at 95% confidence level lep . Another search for $`h^0/A^0`$ produced in association with two bottom ($`b`$) quarks and decaying to two $`b`$ quarks was performed at CDF earlierAffolder et al. (2001). This previous search was sensitive to the high-$`\mathrm{tan}\beta `$ region, excluding $`\mathrm{tan}\beta >50`$ for $`m_{A^0}=100`$ GeV/$`c^2`$.
This paper presents the results of a search for supersymmetric (SUSY) Higgs bosons directly produced in proton-antiproton collisions at a center-of-mass energy of 1.8 TeV using the $`86.3\pm 3.5`$ pb<sup>-1</sup> of data recorded by the Collider Detector at Fermilab (CDF) during the 1994$``$1995 data taking period of the Tevatron (Run 1b). Although the branching ratio to b quarks would be largest, that decay mode would be dominated by QCD background, so we search for Higgs bosons that have decayed to two tau ($`\tau `$) leptons. Events are selected inclusively by requiring an electron from $`\tau e\nu _e\nu _\tau `$ and a hadronically decaying tau ($`\tau _h`$) lepton. This semi-leptonic mode was chosen for this search as a trade-off between the distinctive electron signature in a QCD environment and the high branching ratio of hadronic tau decays.
This is the first time that a search for Higgs bosons has been carried out in the di-tau decay mode using data from a hadron collider. In Run 1, CDF published other analysis with taus in the final state Acosta et al. (2004); Abe et al. (1997a). We also demonstrate for the first time from such data the feasibility of a technique to reconstruct the full mass of a candidate di-tau system, which is only possible when the tau candidates are not back-to-back in the plane transverse to the beam.
The sample that passes the final selection cuts is dominated by $`Z\tau \tau `$ events. There is no evidence for a signal, so we report a limit on a set of MSSM models in a region of parameter space where $`\mathrm{tan}\beta `$ is large because this is where the best sensitivity is achieved. Since at the Tevatron most directly produced Higgs bosons would be back-to-back, the acceptance is small in the region where the mass reconstruction is a discriminating variable. Therefore, limits are reported based on a counting experiment using events from the full sample. Then, from a subset of the events where the tau candidates are not back-to-back in the transverse plane we extract a mass distribution and perform a binned likelihood to demonstrate the capability of that technique.
## II The CDF Detector
This section briefly describes the Run 1 CDF detector with an emphasis on the sub-detectors important to this analysis. The CDF detector is described in detail elsewhere Abe et al. (1988, 1994); Amidei et al. (1994); Azzi et al. (1995a).
CDF used a cylindrical coordinate system with the $`z`$ axis along the proton beam direction. The polar angle ($`\theta `$) was reported with respect to the $`z`$ axis. Pseudorapidity ($`\eta `$) was defined as $`\mathrm{ln}[\mathrm{tan}(\theta /2)]`$. Detector pseudorapidity ($`\eta _d`$) was the same quantity with dependence on vertex position removed. The azimuthal angle ($`\varphi `$) was measured relative to the positive $`x`$ direction.
The CDF electromagnetic and hadronic calorimeters were arranged in a projective tower geometry, as well as charged particle tracking chambers. The tracking chambers were immersed in a 1.4 T magnetic field oriented along the proton beam direction provided by a 3 m diameter, 5 m long superconducting solenoid magnet coil.
In the central region covering $`|\eta |<1.1`$, the electromagnetic (CEM) and hadron (CHA, WHA) calorimeters were made of absorber sheets interspersed with scintillator. Plastic light guides brought the light up to two phototubes per EM tower. The towers were constructed in 48 wedges, each consisting of 10 towers in $`\eta `$ by one tower in $`\varphi `$. The measured energy resolution for the CEM and CHA were $`\sigma (E)/E=13.7\%/\sqrt{E_T}2\%`$ and $`\sigma (E)/E=50\%/\sqrt{E_T}3\%`$, respectively.
The central EM strip chambers (CES) were proportional strip and wire chambers located six radiation lengths deep in the CEM (radial distance $`r=184.15`$ cm), where the lateral size of the electromagnetic shower was expected to be maximal Balka et al. (1988); Hahn et al. (1988); Yasuoka et al. (1988); Wagner et al. (1988); Devlin et al. (1988). It measured the position of electromagnetic showers in the plane perpendicular to the radial direction with a resolution of 2 mm in each dimension Balka et al. (1988); Yasuoka et al. (1988). In each half of the detector (east and west), and for each $`15^{}`$ section in $`\varphi `$, the CES was subdivided into two regions in $`z`$, with 128 cathode strips separated by $``$ 2 cm measuring the shower positions along the $`z`$ direction with a gap within 6.2 cm of the $`z=0`$ plane. In each such region, 64 anode wires (ganged in pairs) with a 1.45 cm pitch provided a measurement in $`\varphi `$.
A three-component tracking system measured charged particle trajectories, consisting of the silicon vertex detector (SVX), the time projection chamber (VTX) and the central tracking chamber (CTC). The SVX Azzi et al. (1995b) consisted of four concentric silicon layers sitting at radii between 2.36 cm and 7.87 cm and providing $`r\varphi `$ tracking information only. The VTX was positioned just beyond the SVX in radius and measured the position of the collision point along the beam for each event to a resolution of 2 mm.
Beyond the VTX (radially) was the CTC, a cylindrical drift chamber 3.2 m long in the $`z`$ direction, with its inner (outer) radius at 0.3 (1.3) m. The sense wires were arranged into 84 layers divided into 9 โsuper-layers.โ Five of the super-layers (axial) contained cells with 12 sense wires that ran parallel to the beam and provided measurements in $`r\varphi `$. The remaining four super-layers (stereo) sat between the axial layers in radius, contained 6 sense wires per cell, and were rotated in the $`rz`$ projection by 2.5 with respect to the beam to provide measurements in $`rz`$. The transverse momentum resolution of the CTC was $`\delta p_T/p_T0.002\text{ GeV/}c^1\times p_T`$ and when combined with the SVX tracking information when available, the resolution was $`\delta p_T/p_T0.001\text{ GeV/}c^1\times p_T`$.
## III Monte Carlo Simulation
The Pythia 6.203 Sjostrand et al. (2001) event generator is used to simulate signal events and backgrounds other than fakes. The Monte Carlo (MC) samples that are generated with the standard Pythia package required modifications that are discussed in this section.
A re-summed calculation of the cross sections for direct Higgs boson production at the Tevatron has been performed using the program HIGLU hig , which allows a user to estimate cross sections as a function of mass, $`\mathrm{tan}\beta `$ and other parameters. For a given Higgs boson mass HIGLU gives the *on-shell* cross section only. A SUSY Higgs boson produced at large $`\mathrm{tan}\beta `$ has a significant width and a tail at low values of the center of mass energy of the parton collision that produced the Higgs boson $`\sqrt{Q^2}`$ (see Fig. 1).
Therefore, scaling the Pythia differential distribution to the HIGLU one would underestimate the cross section because the off-shell events would not be accounted for.
To estimate the total cross section, we first retrieve the on-shell cross section as a function of mass in bins 1 GeV/$`c^2`$ wide using the HIGLU program. Then, the mass-dependent cross section is folded into a relativistic Breit-Wigner shape, with the width proportional to $`Q^2`$, from $`Q^2=(40`$ GeV$`)^2`$ to $`Q^2=(200`$ GeV$`)^2`$. For $`m_{A^0}=100`$ GeV/$`c^2`$ at $`\mathrm{tan}\beta =50`$, the MSSM cross-section for $`A^0+h^0`$ production is 122 pb.
The rate of Higgs boson production in the region of low $`\sqrt{Q^2}`$ is a source of significant uncertainty for the analysis, particularly at high mass and high $`\mathrm{tan}\beta `$. The low tail seen in Fig. 1 originates from a steeply falling cross section folded in with an increase in parton luminosities at small momentum transfer, folded in with a broad Higgs boson width. When we use the method outlined above to obtain a $`Q^2`$ dependent cross section, the size of the tail is bounded by those obtained when one generates events at the same parameter space point with Pythia and Isajet isa . We compare the result from the default Pythia output with that where we use the method above to estimate the systematic error from these low mass tails. At $`m_{A^0}=100`$ GeV/c<sup>2</sup> and $`\mathrm{tan}\beta =50`$, this uncertainty is $`2\%`$. At higher mass and higher $`\mathrm{tan}\beta `$, this systematic can become significant. At $`m_{A^0}=140`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =80`$ the systematic is $`30\%`$.
It is important to properly model the boson $`p_T`$ for Higgs boson and $`Z`$ boson events because it impacts the relative rates of back-to-back and non-back-to-back events, and the di-tau mass resolution for the latter events. Figure 2 shows the $`p_T`$
distribution of the $`A^0`$ ($`m_{A^0}=100`$ GeV/c<sup>2</sup>) after the $`\mathrm{sin}\mathrm{\Delta }\varphi >0.3`$ cut is imposed, where $`\mathrm{\Delta }\varphi `$ is the angle between the taus in the transverse plane. We see that this cut is approximately equivalent to a cut of $`p_T>15`$ GeV/c imposed on the parent Higgs bosons.
In the high-$`\mathrm{tan}\beta `$ region of parameter space probed in this analysis, the direct-production process occurs predominantly through gluon fusion via a triangular bottom quark loop as in Fig. 3 (the direct production of SM Higgs bosons proceeds predominantly through a $`t`$ quark loop). In the default Pythia, the effect of the lighter quark mass in the fermion loop on the $`p_T`$ of Higgs bosons is not taken into account. To correct for this, we use Ellis et al. (1988) which performs a perturbative calculation for the differential cross section $`d\sigma /dp_T`$ to order $`\alpha _s^3`$ with a variable quark mass in the triangular loop. The perturbative calculation is valid in the region $`p_T^{Higgs}\text{ }>15`$ GeV/c. As has been pointed out, this cut is nearly equivalent to the requirement $`\mathrm{sin}\mathrm{\Delta }\varphi >0.3`$. We force agreement between the Higgs boson $`p_T`$ distributions in the region $`p_T^{Higgs}>15`$ GeV/c and the result of the program with a 5 GeV/c<sup>2</sup> $`b`$ quark in the fermion loop so that the resulting MC sample will have the proper efficiency for the $`\mathrm{sin}\mathrm{\Delta }\varphi `$ cut. To do this, we use a reweighting method known as the acceptance rejection MC method Eidelman et al. (2004).
The $`p_T`$ distributions for the $`Z`$ bosons from $`Z/\gamma ^{}ee`$ events has been measured by CDF in the region 66 GeV/c$`{}_{}{}^{2}<M_{ee}<`$116 GeV/c<sup>2</sup> Affolder et al. (2000a). $`Z`$ bosons from Pythia tend to have a lower average $`p_T`$ than the measured value. Therefore, we reweight both the $`Z/\gamma ^{}\tau \tau `$ and $`Z/\gamma ^{}ee`$ samples to force agreement between the MC events and the measured spectrum. Only events that lie in the measured region are subject to rejection. This correction makes a significant difference in the relative rates of back-to-back and non-back-to-back events. Before the correction, 20% of the $`Z`$ events from Pythia are in the non-back-to-back region (here, defined by $`p_T^Z>15`$ GeV/c). After the correction, 26% of the events are in the non-back-to-back region.
CDF measured the cross section for $`Z/\gamma ^{}ee`$ in the same mass window as the $`p_T`$ distributions above, and reports this cross section to be 248 $`\pm `$ 11 pb Affolder et al. (2000a). Assuming universality, we scale the generated $`Z\tau \tau `$ sample to this cross section in the measured range.
Polarization effects, which impact the tau lepton kinematics, are properly taken into account with Tauola Jadach et al. (1990). Simulated Higgs boson events tend to produce one tau with hard (high-$`p_T`$) visible decay products and one with soft visible decay products. Taus produced from $`Z`$ bosons are either both hard or both soft.
## IV Event Selection and Efficiencies
The search is performed in a data sample of events collected using a high-$`E_T`$ electron trigger. In the following, we first describe the trigger system that an event must pass to enter this data sample, followed by a description of the electron and tau identification, and finally the event selection and mass reconstruction made at the analysis level.
### IV.1 Trigger
A three-level trigger system was used to select events for data storage Amidei et al. (1988) and the energy-dependent efficiency of these triggers has been measured Affolder et al. (2000b). At Level 1 (L1), the trigger requires at least one trigger tower with $`E_T>8`$ GeV in the CEM and the efficiency was measured to be 100% at the $`10^3`$ level. At Level 2 (L2), the trigger requires one calorimeter cluster with $`E_T>16`$ GeV and the ratio of hadronic energy to electromagnetic energy ($`E_{HAD}/E_{EM}`$) to be less than 0.125. This cluster must be close in azimuthal angle to a track reconstructed in the trigger with $`p_T>12`$ GeV/c. The triggered sample contains 128,761 events. The efficiency of the L2 trigger for a good quality electron with $`E_T>20`$ GeV is $`91\pm 2`$%, measured from a $`Zee`$ control sample. Signal events would pass this trigger approximately 20% of the time.
### IV.2 Electron Identification
To improve the purity of the data, further cuts (listed in Table 1) are made on the candidate electron.
These cuts are described in more detail in Abe et al. (1995a); we briefly summarize them here. We use $`Zee`$ control samples to quantify the degree of agreement between the simulation and the data.
A candidate electron is first identified as a calorimeter cluster in the CEM. An electron cluster in the calorimeter is formed by merging seed towers (required to have $`E_T>3`$ GeV) with neighboring towers in $`\eta `$ ($`E_T>0.1`$ GeV required). To be called an EM cluster, $`E_T>5`$ GeV is required. Corrections are made to the energy of an EM cluster to compensate for variable response across each tower, tower-to-tower gain variations and time-dependent effects. A global correction to the energy scale is also imposed Abe et al. (1995b). These corrections are typically at the level of a few percent.
We require an electron candidate to have a CTC track pointing to an EM cluster. The highest $`p_T`$ track pointing to the cluster is the โelectron track,โ and is required to satisfy $`p_T>13`$ GeV/c. The measured direction of the track momentum sets the direction of the electron candidate.
We require that the ratio of the energy deposited in the EM calorimeter to the momentum of the electron track is not too large ($`E/p<1.5`$). Also, the lateral profile of the EM shower left by the electron tau candidate is required to be consistent with electron shower profiles as measured in test beam data ($`Lshr<0.2`$). The ratio of energy measured in the CHA to energy measured in the CEM ($`E_{HAD}/E_{EM}`$) is expected to be small for an electron from a tau decay; we require $`E_{HAD}/E_{EM}<0.05`$. In addition, the electron track projected to the plane of the CES must be close to a CES shower position: $`|\mathrm{\Delta }x|<1.5`$ cm and $`|\mathrm{\Delta }z|<3.0`$ cm. This reduces the background from charged pions produced in the neighborhood of neutral pions. We also confirm that the profile of the pulse heights produced by the electron shower across CES strips is consistent with electron test beam data: $`\chi _{strip}^2<10.0`$. The electron track is required to be consistent with a vertex that lies within 60 cm of the nominal collision point. Standard CDF fiducial cuts are made on the electron to ensure that the particle arrived at an instrumented region of the calorimeter with good response. We reject electron candidates which are consistent with having originated from a photon that converted to an electron-positron pair in the detector by removing candidates that leave a low-occupancy track in the VTX or that have a nearby opposite-sign track. The opposite-sign track must be within $`90^{}`$ in $`\varphi `$, separated in $`r\varphi `$ from the electron track by no more than 0.3 cm measured at the point where the tracks are parallel, satisfying $`|\mathrm{\Delta }\mathrm{cot}\theta |<0.06`$ where $`\mathrm{\Delta }\mathrm{cot}\theta `$ is the difference between the values of $`\mathrm{cot}\theta `$ for the two tracks. The electron tau candidate must be isolated in the calorimeter and in the tracking system. In the calorimeter, we use the standard CDF isolation variable defined by:
$$R_{\mathrm{iso}}=\frac{E^{\mathrm{cone}}E^{\mathrm{cluster}}}{E^{\mathrm{cluster}}}$$
(1)
where $`E^{cone}`$ is the energy deposited in the calorimeter in a cone of $`\mathrm{\Delta }R=\sqrt{(\mathrm{\Delta }\varphi )^2+(\mathrm{\Delta }\eta )^2}=0.4`$ around the electron tau candidate and $`E^{\mathrm{cluster}}`$ is the energy of the EM cluster. We require $`R_{\mathrm{iso}}<0.1`$. We define $`N_{iso}`$ to be the number of tracks with $`p_T>1`$ GeV/c within $`\mathrm{\Delta }R<0.524`$ of the EM cluster. This cone size is 30, chosen to be the same as the outer radius of the isolation annulus used for identifying hadronic taus, described below. A track must originate within 5 cm of the electron track along the beam line to be counted in the isolation cone. We require that $`N_{iso}=0`$.
We require at least one EM cluster in the event with $`E_T>20`$ GeV passing the electron identification requirements just described. We refer to this as the *electron tau ($`\tau _e`$) candidate*. If there is more than one $`\tau _e`$ candidate in the event, we select the candidate with highest $`E_T`$.
We correct for inadequacies in simulation of electron identification variables by applying a scale factor of $`0.869\pm 0.016`$. The scale factor was determined using $`Z/\gamma ^{}ee`$ events and we have determined that is not $`p_T`$ dependent.
Since the vertex position affects the acceptance, we reweight the events to force agreement between the distribution of primary vertex positions from the simulation and that measured from the Run 1b data sample.
### IV.3 Hadronic Tau Identification
For taus coming from Higgs boson decays, the hadronic tau decay products will be collimated, with an angular deviation from the direction of the tau parent of no more than $`\mathrm{\Delta }\varphi \text{ }<m_\tau /E_\tau `$ which is $`10^{}`$ for a typical $`E_\tau 10`$ GeV. In nearly all cases, tau decay products will include 1 or 3 charged tracks and $`2`$ neutral pions, each decaying to two photons (all other decay modes have branching fractions of less than half of a percent).
The cuts used to select a hadronic tau are listed in Table 2. The $`\tau _h`$ identification cuts used here are based on those outlined in Abe et al. (1997b). The main differences are noted in what follows. The search for a $`\tau _h`$ begins with identifying a stiff track associated with a jet cluster. We require a track with $`p_T>10`$ GeV/c within $`\mathrm{\Delta }R<0.4`$ of a jet cluster with $`E_T>10`$ GeV. The calorimeter cluster size is $`\mathrm{\Delta }R=0.4`$. The track with the highest $`p_T`$ satisfying this requirement is called the *tau seed*. The $`E_T`$ cut is approximately 75% efficient for signal. The $`p_T`$ cut is approximately 65% efficient for signal while rejecting 80% of QCD jets (after the $`E_T`$ cut has already been imposed).
We make stringent fiducial requirements on the tau seed to ensure that the tau candidateโs energy is well measured. The track must be fully contained in the CTC and must pass additional fiducial cuts similar to those imposed on the electron candidate to ensure that it is not incident on an uninstrumented portion of the calorimeter. Also, to suppress fake track contamination, we require 0.5 GeV in the tower to which the track points. The seed track also must be within 5 cm of the same primary vertex ($`|z_{\mathrm{VTX}}z_{\tau _h}|<5.0`$ cm) as the electron track.
In the neighborhood of the tau seed, two isolation regions are considered separately with different requirements made in each region: the $`\mathrm{\Delta }R<0.175`$ cone and the $`0.175<\mathrm{\Delta }R<0.524`$ annulus. In either isolation region, a track is a *shoulder* track if it is a good quality track with $`p_T>1`$ GeV/c. The seed track is included in the track counting in the $`R<0.175`$ cone.
For a true hadronically decaying tau, the number of tracks in the $`\mathrm{\Delta }R<0.175`$ cone ($`N_{\mathrm{cone}}^{\mathrm{trks}}`$) is usually 1 or 3, so we require $`N_{\mathrm{cone}}^{\mathrm{trks}}<4`$. We additionally require the sum of the charges of the tracks in the cone to be $`\pm 1`$, and opposite to the charge of the electron. We expect the number of tracks in the annulus $`0.175<R<0.524`$ around the tau seed ($`N_{\mathrm{ann}}^{\mathrm{trks}}`$) to vanish in signal events, so we require $`N_{\mathrm{ann}}^{\mathrm{trks}}=0`$. These tracking isolation cuts retain approximately 80% of signal events and reject 70-80% of QCD jets.
The CES clustering algorithm used is the same as the one used in previous CDF analyses Abe et al. (1997b), with some modifications that improve tau purity and fake rejection. In particular, CES clusters were formed at a larger distance in $`r\varphi `$ from the seed track so that this information may be used for fake rejection. Also, a $`\chi ^2`$ requirement that was used in previous CDF analyses to ensure the consistency of the CES cluster profile with electron test beam data was removed here to improve efficiency without a significant sacrifice in purity.
The algorithm forms CES clusters in the $`\mathrm{\Delta }R<0.6`$ cone around the seed track by taking the highest energy strips (wires) in each calorimeter tower, in descending order, calling them seeds, and merging them with their nearest neighbors to form clusters 4 strips (6 wires) wide. To be a seed for a cluster, a strip (wire) must show a pulse height that surpasses 0.4 (0.5) GeV. The cluster position is defined as the position of the center strip or wire in the cluster. Pulse heights from strips and wires were corrected for $`\eta `$ and $`\varphi `$dependent effects, measured from test-beam data. The energy of a CES cluster in a tower is the CEM energy of that tower, weighted by the energy deposited in the CES by that cluster compared to the energy deposited by all CES clusters in the tower. The predicted response in the CEM for a charged pion is subtracted from the energy in the tower impacted by the seed track. Wire clusters in the $`r\varphi `$ view are matched with strip clusters in the $`z`$ view with similar energies and merged into new clusters. The cluster position must not be consistent with coming from the seed track. It is rejected if $`|\eta _d^{\mathrm{seed}}\eta _d^{\mathrm{cluster}}|<0.03`$ and either $`|\varphi ^{\mathrm{seed}}\varphi ^{\mathrm{cluster}}|<0.01`$ or $`|\varphi ^{towercenter}\varphi ^{\mathrm{cluster}}|<0.01`$ (the latter requirement is because the cluster position is assigned to the center of the tower in $`\varphi `$ when no wire information is available).
A CES cluster must satisfy $`E_T>1`$ GeV to be counted as a *shoulder cluster* in either of the isolation regions. We call the number of CES clusters found in the isolation cone (annulus) $`N_{\mathrm{cone}}^{\mathrm{ces}}`$ ($`N_{\mathrm{ann}}^{\mathrm{ces}}`$) and require $`N_{\mathrm{cone}}^{\mathrm{ces}}<3`$ and $`N_{\mathrm{ann}}^{\mathrm{ces}}=0`$. In addition, the CES cluster energies in the cone measured as 3-component vectors are combined with the measured momenta of the tracks to compute a tau mass, $`m_\tau `$. We require $`m_\tau <2.0`$ GeV/c<sup>2</sup>. The cuts on $`N_{\mathrm{cone}}^{\mathrm{ces}}`$, $`N_{\mathrm{ann}}^{\mathrm{ces}}`$ and $`m_\tau `$ give a combined efficiency for signal of approximately 95%. These three cuts additionally reject 30-50% of QCD jets.
As described above, a tau candidate is found to be isolated through a measurement of track and CES cluster multiplicities in the neighborhood of a seed track. We compare the isolation variables among $`Zee`$, $`Z\mu \mu `$ and $`Z\tau \tau `$ simulation MC samples, and in $`Zee`$ and $`Z\mu \mu `$ data control samples. In the $`Zee`$ and $`Z\mu \mu `$ samples, both lepton candidates in the event mimic the tau seed in this analysis. We find good agreement between data and simulation, and between electron and muon samples, in the efficiencies of the isolation cuts. The isolation efficiencies from the simulated $`Z\tau \tau `$ MC sample also agrees with the data samples in the annulus around the tau seed (where no particles from tau decays are expected).
To reduce the impact of $`Z/\gamma ee`$ background on the sensitivity, we require $`\xi >0.15`$ where $`\xi =E_T^{had}/\mathrm{\Sigma }p_T`$ Groer (1998). Here, $`E_T^{had}`$ is the hadronic energy of the tau jet cluster and $`\mathrm{\Sigma }p_T`$ is the sum of the $`p_T`$ of all tracks with $`p_T>1`$ GeV/c within the cone $`\mathrm{\Delta }R<0.175`$ centered on the jet direction.
Since the decay products of a hadronically decaying tau are highly collimated, a tau is expected to leave a narrow cluster of energy in the calorimeter. We cut on the $`\varphi \varphi `$ and $`\eta \eta `$ moments of the jet cluster associated with the hadronic tau candidate which are defined as follows:
$$I_{\varphi \varphi }=\frac{_iE_T^i(\varphi _i\varphi _0)^2}{E_T^i}$$
(2)
$$I_{\eta \eta }=\frac{_iE_T^i(\eta _i\eta _0)^2}{E_T^i}.$$
(3)
The sum is over calorimeter towers in the jet cluster, and $`\varphi _0`$ and $`\eta _0`$ are the $`E_T`$ weighted center of the jet in the $`\varphi `$ and $`\eta `$ directions. We require $`I_{\varphi \varphi }<0.1`$ and $`I_{\eta \eta }<0.1`$. These two cuts together reject approximately 30-45% of QCD jets.
Figure 4 shows the efficiency of the hadronic tau identification cuts from the simulation. We bin the efficiencies according to the visible transverse energy ($`E_T`$) from the tau at the generator level. The total efficiency plateaus near 55% for $`E_T\text{ }>35`$ GeV. For fiducial hadronic taus with $`E_T>10`$ GeV from $`Z`$ decays, the average efficiency of the tau identification cuts is 40%. $`A\tau \tau `$ events at $`m_A=100`$ GeV and $`\mathrm{tan}\beta =50`$ are also 40% efficient.
### IV.4 Additional Requirements
We make further cuts on the data sample to increase purity and further reject background. We require at most one recoil jet with $`E_T>15`$ GeV in the event. For further suppression of $`Zee`$ background, we reject any event with two electrons or one electron and one track that have a reconstructed invariant mass $`M_{ee}`$ between 70 and 110 GeV/c<sup>2</sup>, which removes 99% of $`Zee`$ events while retaining 90% of signal events. We require a separation of the tau candidates in the transverse plane, $`\mathrm{\Delta }\varphi >1.5`$, where $`\mathrm{\Delta }\varphi `$ is the azimuthal angle between $`\tau _e`$ and $`\tau _h`$. This is nearly 100% efficient for signal rejecting 20% of non-tau backgrounds as measured from a background-dominated data sample.
To take advantage of the mass reconstruction technique, we divide the events into back-to-back and non-back-to-back samples. The full invariant mass of the di-tau system can be estimated only when the tau candidates are not back-to-back, as explained in more detail in Section IV E. The tau candidates are called back-to-back when $`\mathrm{sin}\mathrm{\Delta }\varphi <0.3`$, where $`\mathrm{\Delta }\varphi `$ is the azimuthal angle between $`\tau _e`$ and $`\tau _h`$. Note that $`\mathrm{sin}\mathrm{\Delta }\varphi `$ is the determinant of the system of equations which determine the di-tau mass, so when $`\mathrm{sin}\mathrm{\Delta }\varphi 0`$, the solution is not unique.
The missing transverse energy in the event, denoted $`\stackrel{}{E_T}/`$, is the opposite of the vector sum of the measured transverse energies of the event. For the non-back-to-back events, we use the magnitude and direction of $`\stackrel{}{E_T}/`$ to derive the di-tau mass. First, we define corrected $`E_T/`$ in the following way:
$$\stackrel{}{E_T}/^{corr}=\underset{towers}{}\stackrel{}{E_T}^i\underset{muons}{}\stackrel{}{p_T}^j\mathrm{\Delta }\stackrel{}{E_T}^{ele}\underset{jets}{}\mathrm{\Delta }\stackrel{}{E_T}^i.$$
(4)
The first term on the right side of the equation is a sum of the the transverse component of the energy deposited the calorimeter towers. The remaining terms improve the resolution by accounting for the momentum carried away by muons, energy corrections applied to the electron candidate, and jet energy corrections.
It is only necessary for the simulation to correctly model $`\stackrel{}{E_T}/^{corr}`$ well in the region $`p_T^{A,h,Z}>15`$ GeV/c, since that is where the mass reconstruction is utilized. We confirm that the $`\stackrel{}{E_T}/^{corr}`$ variable from the data is well modeled by the simulation using a sample of $`Zee`$ events with $`p_T^Z>15`$ GeV/c.
### IV.5 Di-tau Mass Reconstruction
Signal events contain neutrinos that escape CDF undetected. At hadron colliders, the resulting energy imbalance may only be determined in the transverse plane because the $`z`$-component of the total momentum of the interaction is unknown.
Nonetheless, the energy of the neutrinos from each tau decay, and thus the full mass of the di-tau system, may be deduced if (i) the tau candidates are not back-to-back in the transverse plane and (ii) the neutrino directions are assumed to be the same as their parent tausEllis et al. (1988); cms ; atl (a, b).
The contributions to the total missing energy from the leptonic and hadronic decays, denoted $`E_l/`$ and $`E_h/`$, are the solution to a system of two equations and two unknowns:
$$E_l/\mathrm{sin}\theta _l\mathrm{cos}\varphi _l+E_h/\mathrm{sin}\theta _h\mathrm{cos}\varphi _h=(E^{meas}/)_x$$
(5)
$$E_l/\mathrm{sin}\theta _l\mathrm{sin}\varphi _l+E_h/\mathrm{sin}\theta _h\mathrm{sin}\varphi _h=(E^{meas}/)_y$$
(6)
Here, $`E^{meas}/`$ is the missing energy measured for the event and $`\theta _{l,h}`$ are the polar angles of the taus and $`\varphi _{l,h}`$ are the azimuthal angles of the taus. The tau candidate directions are measured from the visible decay products.
The reason for the first of the two criteria for the mass technique outlined above is that when the tau candidates are back-to-back in the transverse plane, the reconstructed mass is not a good separating variable because there are many high mass solutions.
Some events may give negative solutions for $`E_l/`$ and $`E_h/`$. We require $`E_{l,h}/>0`$ for the non-back-to-back events, which reduces non-di-tau backgrounds in this region. Figure 5 shows
$`E_l/`$ from a background-dominated sample of the data compared to simulated $`A^0/h^0\tau \tau `$ events. The analogous distributions for $`E_h/`$ are similar. The $`E_{l,h}/>0`$ requirement is 55-60% efficient for signal, increasing with mass, while reducing non-di-tau background by approximately a factor of 10. When the di-tau mass is reconstructed as described next, this cut also improves the mass resolution.
We calculate the total reconstructed mass of the di-tau system using
$$\begin{array}{cc}\hfill m_{\tau \tau }^2& =m_{Z/h}^2=(p_l+p_h)^2\hfill \\ & =2m_\tau ^2+2(E_l/+E_l^{vis})(E_h/+E_h^{vis})(1\mathrm{cos}\psi )\hfill \end{array}$$
(7)
where $`p_l`$ and $`p_h`$ are the 4-momenta of each tau, and $`E_l`$ and $`E_h`$ represent the total energy of each tau. $`E_l^{vis}`$ and $`E_h^{vis}`$ represent the energy left by their visible decay products. The $`\psi `$ is the 3-dimensional angle between the two taus. The missing energies $`E_l/`$ and $`E_h/`$ are the solutions to Eqs. 5 and 6.
Figure 6 shows the di-tau mass distribution reconstructed from signal events for the parameter point $`m_{A^0}=100`$ GeV/c<sup>2</sup> and $`\mathrm{tan}\beta =50`$, for which the $`A^0`$ or $`h^0`$ particle has an inherent width of 5.7 GeV/c<sup>2</sup>. The contribution from the calorimeter resolution may be qualitatively seen from Fig. 7, since $`Zee`$ events would have a small $`E_T/`$. The remaining contribution to the width of the di-tau mass distribution comes from the approximations that are needed to implement the technique, listed above.
Table III summarizes the cuts that we impose and the number of events remaining in the sample after each cut. Table IV summarizes the efficiency of each cut on the $`Z\tau \tau `$ and $`A^0+h^0\tau \tau `$ simulated events.
## V Backgrounds
The backgrounds can be classified into two categories: physics and instrumental backgrounds. The former includes $`Z\tau \tau `$ and $`Z/\gamma ^{}ee`$. The latter fall into three categories: events containing a conversion pair and a recoil jet, $`W(e\nu )`$+jets events and events containing a jet that fakes an electron. All of these contain a fake hadronic tau. We model signal and physics backgrounds using Pythia Monte-Carlo and the CDF detector simulation Sjostrand et al. (2001).
We estimate the rate of all fake tau contributions combined instead of estimating each source of fakes separately. This way, we do not rely on modeling of jets, nor on limited control samples for each separate background. We estimate the number of fakes expected to pass all of the cuts using the fake rate technique described next.
We use five different control samples from CDF data to measure the rate at which a jet fakes a hadronic tau by passing the tau identification cuts for this analysis. They are: (i) a sample dominated by events where a photon produced an electron pair, (ii) a sample dominated by $`W(e\nu )`$+jets, (iii) another dominated by $`W(\mu \nu )`$+jets and two samples of events collected using inclusive triggers with a jet satisfying (iv) $`E_T>20`$ GeV and (v) $`E_T>50`$ GeV. The first three are classified as lepton samples and the latter two are jet samples.
A fake rate is defined as the probability that a jet passes the hadronic tau identification requirements as described in Sec. IV.3. For all data control samples, we measure the fake rates in bins defined by the $`E_T`$ of the jet associated with the tau candidate. Figure 8 shows the fake
rates measured from all three lepton samples combined and from both jet samples combined. We measure fake rates less than about 1% from the jet samples, which are as good as in previous tau analyses Groer (1998). We find that the fake rates measured from the lepton samples are approximately a factor of 2 higher than from the jet samples. Two of the lepton samples are dominated by W+jets events where the recoil jet comes from a quark; quark jets are narrower, have lower multiplicity and are thus more likely to pass the tau identification cuts qua . Still, the reason for this difference is not definitively understood. Since the analysis is performed in a lepton sample, we use the fake rates measured from the lepton samples as the central value and the difference between the rates from the two different types of samples as a systematic uncertainty. The histogram in Fig. 8 shows the $`E_T`$ distribution of jets in the data sample just before tau identification cuts are applied. The fake rates are folded into this $`E_T`$ distribution to predict the rate of fake tau background.
We find that fake taus from the $`W`$+jets samples are opposite-sign from the lepton in the event (67.1 $`\pm `$ 3.0)% of the time. This is due to a correlation between the W and the recoiling quark. The isolation cut enhances this correlation due to charge conservation. In the conversion sample, the opposite-sign requirement is consistent with being 50% efficient. We take the central value of the efficiency for the opposite-sign cut to be the average of the two ($`(67.1+50)/2=58.6\%`$) and equally likely to be anywhere between 50.0% and 67.1%.
To estimate the number of fake taus expected to pass all of the cuts, we use the data sample with the following cuts removed: (i) tau ID cuts (ii) the opposite sign requirement and (iii) $`E/_{\tau _1,\tau _2}>0`$. Then, we fold in the measured fake rates with the $`E_T`$ spectrum of jets. There are 6478 events that pass these cuts and which contain at least one jet passing the fiducial cuts. We apply our measured fake rates to 6972 jets from these events. For each jet, we give it a weight equal to the fake rate measured for its $`E_T`$. We expect 21.0 $`\pm `$ 12.1 back-to-back and 29.8 $`\pm `$ 16.8 non-back-to-back for a total of 50.8 $`\pm `$ 29.0 fakes before the remaining cuts are applied.
We apply two final cuts to improve the purity of the sample. We apply the opposite sign requirement, taking the efficiency to be 58.6%. Including the systematic error from this cut, we expect 10.5 $`\pm `$ 6.0 back-to-back and 14.9 $`\pm `$ 8.5 non-back-to-back events at this stage. The final cut is the $`E/_{\tau _1,\tau _2}>0`$ cut applied to the non-back-to-back events only. This is measured from the same background-dominated sample, subtracting out $`Z\tau \tau `$ and $`Zee`$ contamination. We find that this cut removes 89.0% of the fake tau background. This brings the number of predicted non-back-to-back events with a fake tau to 1.6. Added to the back-to-back events, we expect 1.6 + 10.5 = 12.1 events containing a fake hadronic tau to pass the analysis cuts. At each stage of this estimation, we account for the 10%-level $`Z\tau \tau `$ and $`Zee`$ contamination of the background-dominated sample.
## VI Summary of Systematic Uncertainties
In Tab. V , we summarize the systematics on the backgrounds and signal. $`Zee`$ is not included in the table because the expected rate is based on a low number of background events. We expect 0.6 $`\pm `$ 0.3 $`Zee`$ events in the counting experiment. The systematics on the fake tau background are described in Sec. V.
The error on the Run 1b luminosity at CDF is 4.1% Cronin-Hennessy et al. (2000). The systematic error on the electron identification cuts is 1.8%, as noted in Sec. IV. This error comes from the limited size of the $`Zee`$ sample used for comparing the simulation with data. The systematic error due to the modeling of the trigger efficiency is obtained by moving the parameters in the energy and $`\eta `$-dependent efficiency by one standard deviation in each direction and measuring the effect on the total efficiency of all cuts. This systematic is 1.6% (1.7%) for $`A^0/h^0\tau \tau `$ ($`Z\tau \tau `$).
The one systematic uncertainty on the yield of signal events that varies significantly with the mass of the Higgs boson is the uncertainty on the cross section due to the low $`\sqrt{Q^2}`$ tail. Here are the systematic uncertainties due to this effect for each parameter space point considered in this note: $`m_{A^0}=100`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$, 0.5%; $`m_{A^0}=110`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$, 2.5%; $`m_{A^0}=120`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$, 3.5%, $`m_{A^0}=140`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$, 7.2%; $`m_{A^0}=140`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =80`$, 21.3%. The systematic uncertainty on the $`Z`$ production cross section is $`1.7\%`$ based on the CDF measurement Affolder et al. (2000c).
By studying isolated pions, we have shown that the detector simulation overestimates the width of jets from charged pions, and thus underestimates the efficiency of the cut on jet width on the hadronic tau candidates. Therefore, we take the efficiency for that cut to be the average of the simulated efficiency and 100%, which is (100.0+92.2)/2 = 96.1%. The systematic error on this cut is half the difference between the efficiency from simulation and 100%, which is (100.0-92.2)/2 = 3.9%. The electron rejection cut applied to the hadronic tau candidate is not modeled sufficiently for hadronic decays. The same study of isolated pions previously mentioned showed that the hadronic energy deposited by charged pions is underestimated by the simulation, so that the efficiency of this cut is underestimated. We take the efficiency to be the average of the efficiency from simulation for hadronic taus and 100%, which is (100.0+95.5)/2 = 97.8%. The systematic error on this cut is half the difference between the simulated efficiency and 100%, which is (100.0-95.5)/2 = 2.2%. We obtain the jet energy scale systematic by varying the energies of all of the jets (except those identified as electrons) up and down by 5%. The resulting systematic error is 1.0% (1.2%) for $`A^0/h^0\tau \tau `$ ($`Z\tau \tau `$). There is a systematic uncertainty attributed to the tau fake rates since we measured different rates in lepton and jet samples. We set this systematic uncertainty to the the difference between the fake rates measured in the two types of samples; it is the dominant systematic error at 57.1%.
## VII Results
We set limits on direct $`A^0/h^0`$ production in the MSSM based on a counting experiment using events from both the back-to-back and non-back-to-back samples. Then we show the limits achieved from a binned likelihood fit to the di-tau mass distribution from the non-back-to-back events alone. Our nominal limits come from the counting experiment utilizing both back-to-back and non-back-to-back events.
### VII.1 All Events
We plot the track multiplicity of the tau candidates after imposing all of the analysis cuts *except* the following cuts: i) $`|Q_i|=1`$, ii) $`N_{cone}^{trks}<4`$ and iii) the opposite sign requirement. Figure 9 shows the number of tracks in the 0.175 cone around the tau seed in the hadronic tau candidate in the event. We expect 78.2 events to appear in this plot and we observe 81.
Table 6 lists the number of events expected of each background type. The data and the prediction show good agreement at each stage. Of the 47 final observed events, 35 are 1-track and 12 are 3-track. The final three cuts reduce the fake background in these two bins by nearly a factor of 2 compared to that shown in Fig. 9 and leaves the $`Z\tau \tau `$ background in those bins virtually unchanged.
### VII.2 Non-Back-to-Back Events
Figure 10 shows the track multiplicity distribution for the non-back-to-back events only (since we will be performing the di-tau mass reconstruction on these events). These events have passed the $`E_{l,h}/>0`$ requirement but the $`|Q|=1`$, opposite sign, and $`N_{cone}^{trks}<4`$ requirements have not been imposed. We expect 9.2 events to appear in the plot and we observe 15. The probability for the disagreement between data and the prediction to be equal to or more than what was observed is 3.2%.
Table 7 lists the number of events expected of each background type in Fig. 10 including non-back-to-back events only. We also show the expectation compared to the observed after the subsequent cuts are imposed. The final 8 events contain 5 1-track events and 3 3-track events.
### VII.3 Limits
The numbers of observed events do not show an excess above the standard model expectation for background processes. Therefore, we set a limit on the product of the cross section and branching ratio of $`A^0/h^0\tau \tau `$ ($`\sigma `$ BR) at 95% confidence level. We take the branching ratio to be 9% for all parameter space points considered. We use a Bayesian method with a flat prior based on a likelihood that is smeared to account for systematic errors.
Table 8 shows the predicted and observed upper limits on $`\sigma `$ BR for $`A^0/h^0`$ production in the MSSM as a function of $`m_{A^0}`$ at $`\mathrm{tan}\beta =50`$. For $`m_{A^0}=100`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$ in the MSSM, the expected limit is 23.4 signal events, corresponding to 102 pb. Since we observed slightly fewer events than we expected, the observed limits are better than our expected limits. The observed limit is 77.9 pb. The limits on the $`\sigma `$ BR improve with increasing mass since the efficiency improves, but we are less sensitive to the MSSM theory at higher mass due to the steeply falling predicted cross section.
The cross section for producing $`A^0/h^0`$ in the MSSM scales with $`(\mathrm{tan}\beta )^2`$. The sensitivity does not improve by that same factor, however, because the Higgs boson width scales as $`(\mathrm{tan}\beta )^2`$, and the tail at low $`\sqrt{s}`$ also becomes more prominent with increasing $`\mathrm{tan}\beta `$, increasing the systematic error due to the uncertainty in the cross section in this region. At $`m_{A^0}=140`$ GeV/c<sup>2</sup> and $`\mathrm{tan}\beta =80`$, the uncertainty on the efficiency of the selection of Higgs boson events due to this low mass tail is 20%, compared to 7.2% at $`m_{A^0}=140`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$. Also, both effects bring down the efficiency at higher $`\mathrm{tan}\beta `$: at $`m_{A^0}=140`$ GeV/c<sup>2</sup> and $`\mathrm{tan}\beta =80`$, the efficiency is similar to the efficiency at a lower mass point: $`m_{A^0}=110`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$. Table 9 shows the limits for two different values of $`\mathrm{tan}\beta `$ for $`m_{A^0}=140`$ GeV/c<sup>2</sup>. These limits are also summarized in Figure 11.
In the non-back-to-back region, after all cuts are applied, we expect 5.9 events and observe 8. Figure 12 shows the di-tau mass distribution for these 8 events compared with the expectation. In this region, limits are obtained by fitting the mass distribution using a binned likelihood, with 14 bins between 60 GeV/c<sup>2</sup> and 200 GeV/c<sup>2</sup> in di-tau mass.
Table X shows the upper limits on the MSSM $`\sigma `$ BR obtained from the binned likelihood for four different values of $`m_{A^0}`$ for $`\mathrm{tan}\beta =50`$. At $`m_{A^0}=100`$ GeV/c<sup>2</sup>, where the Higgs boson mass is nearly on top of the $`Z`$ mass, the mass reconstruction is less effective than at higher masses, so the expected limit from the binned likelihood in the non-back-to-back region is approximately 2.4 times worse than the expected limit from the counting experiment without the non-back-to-back requirement. At $`m_{A^0}=140`$ GeV/c<sup>2</sup>, which is approximately 2 RMS away from the $`Z`$ in the di-tau mass variable, the expected limit from the binned likelihood using the non-back-to-back events is 2.1 times worse than the counting experiment limit, showing a modest improvement, but still not coming close to the expected limit from the counting experiment. With more data collected in Run 2, the power of the di-tau mass reconstruction technique will improve. Table XI shows the upper limits obtained from the binned likelihood for two different values of $`\mathrm{tan}\beta `$.
## VIII Conclusions
We have performed a search for directly produced Higgs bosons decaying to two taus where one tau decays to an electron and the other hadronically in Run 1b data at CDF. This is the first Higgs boson search based on the di-tau final state at a hadron collider. The number of events that pass all of the cuts is consistent with the background expectation. This agreement between data and background demonstrates our capability to reconstruct $`Z\tau \tau `$ final states, the irreducible background for this analysis. At a benchmark parameter space point, $`m_{A^0}=100`$ GeV/c<sup>2</sup> and $`\mathrm{tan}\beta =50`$, we are sensitive to a $`\sigma `$ BR of 102 pb compared to the 11.0 pb predicted in the MSSM. The observed limit at this parameter space point is 77.9 pb.
A di-tau mass reconstruction is performed for tau candidate pairs which are not back-to-back, for the first time with hadron collider data. The modest sensitivity that one gains from a limit binned in mass is not nearly enough to make up for the hit in efficiency taken when only non-back-to-back events are considered. At $`m_{A^0}=100`$ GeV/c<sup>2</sup>, the binned mass limit at $`\mathrm{tan}\beta =50`$ from non-back-to-back events alone is 395 pb. At $`m_{A^0}=140`$ GeV/c<sup>2</sup>, $`\mathrm{tan}\beta =50`$, the binned mass limit is 309 pb, showing a modest improvement as the Higgs boson mass is moved away from the $`Z`$ mass.
While this search does not have the sensitivity to the Standard Model Higgs boson that prior searches using decays to pairs of $`b`$ quarks, it lays the groundwork for similar analysis to be performed by future experiments. The di-tau mass reconstruction technique demonstrated here may also be useful for searches for other processes where a Higgs boson is produced with a recoil, such as $`Hb`$ or $`Hb\overline{b}`$.
## IX Acknowledgements
We thank the Fermilab staff and the technical staffs of the participating institutions for their vital contributions. This work was supported by the U.S. Department of Energy and National Science Foundation; the Italian Istituto Nazionale di Fisica Nucleare; the Ministry of Education, Science and Culture of Japan; the Natural Sciences and Engineering Research Council of Canada; the National Science Council of the Republic of China; the A.P. Sloan Foundation.
|
warning/0506/astro-ph0506532.html
|
ar5iv
|
text
|
# Gauge-invariant perturbations at second order: multiple scalar fields on large scales
## I Introduction
Cosmological perturbation theory Bardeen80 ; KS has become the standard tool to study inflation and its observational consequences for the Cosmic Microwave Background (CMB) and the formation of large scale structure LLBook . Linear perturbation theory is sufficient to study the power spectrum of the primordial perturbations generated during inflation, in particular to calculate the spectral index and its scale dependence.
One way to glean more information from the CMB is to go beyond first order perturbation theory, which allows the study of higher order statistics, such as the bispectrum, or to calculate the amount of non-gaussianity expected from ones favourite early universe model.
On super-horizon scales, i.e. scales much larger than the particle horizon, there are mainly two approaches to study higher order effects such as non-gaussianity: the first uses second order perturbation theory following Bardeen Bardeen80 ; Mukhanov ; Bruni ; Acquaviva ; Maldacena ; Nakamura ; Noh ; Bartolo:2001cw ; Rigopoulos:2002mc ; Bernardeau:2002jf ; Bernardeau:2002jy ; MW ; Bartolo:2004if ; Bartolo:2004ty ; Enqvist:2004bk ; filippo ; Tomita:2005et ; Lyth:2005du ; Seery:2005wm ; Seery:2005gb , the second uses nonlinear theory and different manifestations of the separate universe approximation, either employing a gradient expansion, as originally used by Salopek and Bond SB ; RS ; Rigopoulos:2004ba ; Rigopoulos:2005xx or using the $`\mathrm{\Delta }N`$ formalism SaTa ; LMS ; Lyth:2005fi <sup>1</sup><sup>1</sup>1Recently there has also been a covariant study LV .. In the formalism used by Salopek and Bond SB the metric and the matter variables are not expanded into a power series, instead an expansion in spatial gradients is used. The $`\mathrm{\Delta }N`$ formalism was introduced at linear order by Sasaki and Stewart in SaSt95 (see also Starobinsky1982 ; Starobinsky1986 ) and relates the comoving curvature perturbation to the perturbation in the number of e-foldings and has recently been extended to the non-linear case LMS ; Lyth:2005fi . In the Bardeen approach the metric and the matter fields are expanded in a power series in a small parameter. There has been increased interest in second order perturbation theory due to the prospect of new high precision data and following the papers by Acquaviva et al. Acquaviva and Maldacena Maldacena focusing on the study of non-gaussianity from inflation.
In this article we concentrate on the scalar field dynamics at second order on large scales in a universe dominated by multiple scalar fields, including metric perturbations. In linear theory the Klein-Gordon equation, including metric perturbations, can be written as a system of coupled evolution equations in terms of the linear field fluctuations in the flat gauge, the Sasaki-Mukhanov variables Sasaki1986 ; Mukhanov88 . We show here that the Klein-Gordon equation at second order on large scales in the multiple field case can also be written solely in terms of the Sasaki-Mukhanov variables, albeit at second and first order. The Klein-Gordon equation is linear in the second order perturbations, with source terms quadratic in the first order perturbations (the same holds for the field equations). We also give an expression for the curvature perturbation on uniform density slices at second order, $`\zeta _2`$, in terms of the field fluctuations on flat slices at first and second order. After calculating the evolution of the field fluctuations at first and second order we therefore immediately get the evolution of the curvature perturbation $`\zeta _2`$. This is similar to the $`\mathrm{\Delta }N`$ formalism, where the evolution of the comoving curvature perturbation is given by the scalar field dynamics in the background, but which necessitates the slow roll approximation at higher order.
We consider scalar perturbations including and up to second order in the metric and in the scalar fields using the Bardeen approach. We focus on scales larger than the horizon, which allows us to neglect gradient terms. We do not include first order vector and tensor perturbations since they do not couple to the field fluctuations on large scales; also scalar fields do not support vector modes and the tensor contribution from inflation is small and constant or decaying. The large scale focus also makes the equations more transparent and displays the relevant physics more succinctly, a big bonus at second order. We leave the inclusion of small scales, and vector and tensor perturbations to a future publication M2005 .
The paper is organised as follows: in the next section we give the field equations and the Klein-Gordon equation up to second order in gauge dependent form. In Section III, after reviewing the behaviour of perturbations at first and second order under gauge transformations, we construct physically transparent gauge-invariant combinations. In Section IV we show how the gauge-invariant variables defined in different gauges are related to each other. We also discuss the construction of total variables in the presence of several fields or fluids. We review the derivation of the Klein-Gordon equation at first order in Section V. In Section VI.1 we give the field equations at second order in the flat gauge and in Section VI.2 we finally derive the Klein-Gordon equation at second order. We give the second order Klein-Gordon equation in the single field case in Section VII. In Section VIII we apply the formalism to a simple two field inflation model. We conclude in Section IX.
Throughout this paper we assume a flat Friedmann-Robertson-Walker (FRW) background spacetime and work in conformal time, $`\eta `$. Derivatives with respect to conformal time are denoted by a dash. Greek indices, $`\mu ,\nu ,\lambda `$, run from $`0,\mathrm{}3`$, while lower case Latin indices, $`i,j,k`$, run from $`1,\mathrm{}3`$. Upper case Latin indices, $`I,J,K`$, denote different scalar fields.
## II Governing equations
The covariant Einstein equations are given by
$$G_{\mu \nu }=8\pi GT_{\mu \nu },$$
(1)
where $`G_{\mu \nu }`$ is the Einstein tensor, $`T_{\mu \nu }`$ the total energy-momentum tensor, and $`G`$ Newtonโs constant. Through the Bianchi identities, the field equations (1) give the local conservation of the total energy and momentum,
$$_\mu T^{\mu \nu }=0,$$
(2)
where $`_\mu `$ is the covariant derivative. The energy momentum tensor for $`N`$ scalar fields minimally coupled to gravity is
$$T_{\mu \nu }=\underset{K=1}{\overset{N}{}}\left[\phi _{K,\mu }\phi _{K,\nu }\frac{1}{2}g_{\mu \nu }g^{\alpha \beta }\phi _{K,\alpha }\phi _{K,\beta }\right]g_{\mu \nu }U(\phi _1,\mathrm{},\phi _N),$$
(3)
where $`\phi _K`$ is the $`K`$th scalar field and $`U`$ the scalar field potential and $`\phi _{K,\mu }\frac{\phi _K}{x^\mu }`$.
We split tensorial quantities into a background value and perturbations according to
$$๐=๐_0+\delta ๐_1+\frac{1}{2}\delta ๐_2+\mathrm{},$$
(4)
where the background part is a time dependent quantity only $`๐_0๐_0(\eta )`$, whereas the perturbations depend on time and space coordinates $`x^\mu =[\eta ,x^i]`$, that is $`\delta ๐_n\delta ๐_n(x^\mu )`$. The order of the perturbation is indicated by a subscript, e.g. $`\delta ๐_1=O(ฯต)`$.
The metric tensor up to second order, including only scalar perturbations, is
$`g_{00}`$ $`=`$ $`a^2\left(1+2\varphi _1+\varphi _2\right),`$ (5)
$`g_{0i}`$ $`=`$ $`a^2\left(B_1+{\displaystyle \frac{1}{2}}B_2\right)_{,i},`$ (6)
$`g_{ij}`$ $`=`$ $`a^2\left[\left(12\psi _1\psi _2\right)\delta _{ij}+2E_{1,ij}+E_{2,ij}\right],`$ (7)
where $`a=a(\eta )`$ is the scale factor, $`\eta `$ conformal time, $`\delta _{ij}`$ is the flat background metric, $`\varphi _1`$ and $`\varphi _2`$ the lapse function, and $`\psi _1`$ and $`\psi _2`$ the curvature perturbations at first and second order; $`B_1`$ and $`B_2`$ and $`E_1`$ and $`E_2`$ are scalar perturbations describing the shear. The contravariant form of the metric tensor is given in Appendix B.
### II.1 Einstein tensor
We expand the Einstein tensor in a power series according to Eq. (4) up to second order
$$G_\nu ^\mu G_{(0)\nu }^\mu +\delta G_{(1)\nu }^\mu +\frac{1}{2}\delta G_{(2)\nu }^\mu .$$
(8)
Then the components of the Einstein tensor at zeroth order are
$$G_0^0=\frac{3}{a^2}\frac{a^2}{a^2},G_j^i=\frac{1}{a^2}\left[\frac{a^2}{a^2}2\frac{a^{\prime \prime }}{a}\right],$$
(9)
at first order
$`\delta G_{(1)0}^0`$ $`=`$ $`{\displaystyle \frac{6}{a^2}}\left[{\displaystyle \frac{a^2}{a^2}}\varphi _1+{\displaystyle \frac{a^{}}{a}}\psi _1^{}\right]+O(k^2),`$ (10)
$`\delta G_{(1)i}^0`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}\left[2\psi _1^{}+2{\displaystyle \frac{a^{}}{a}}\varphi _1\right]_{,i}`$ (11)
$`\delta G_{(1)j}^i`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}\left[4\left({\displaystyle \frac{a^{\prime \prime }}{a}}\varphi _1+{\displaystyle \frac{a^{}}{a}}\psi _1^{}\right)+2\left(\psi _1^{\prime \prime }+{\displaystyle \frac{a^{}}{a}}\varphi _1^{}{\displaystyle \frac{a^2}{a^2}}\varphi _1\right)\right]\delta _j^i+O(k^2),`$ (12)
and at second order
$`\delta G_{(2)0}^0`$ $`=`$ $`{\displaystyle \frac{6}{a^2}}\left[\psi _1^2{\displaystyle \frac{a^2}{a^2}}\varphi _2{\displaystyle \frac{a^{}}{a}}\psi _2^{}+4\left({\displaystyle \frac{a^{}}{a}}\varphi _1\psi _1^{}+{\displaystyle \frac{a^2}{a^2}}\varphi _1^2{\displaystyle \frac{a^{}}{a}}\psi _1\psi _1^{}\right)\right]+O(k^2),`$ (13)
$`\delta G_{(2)i}^0`$ $`=`$ $`{\displaystyle \frac{2}{a^2}}\left[\psi _2^{}+{\displaystyle \frac{a^{}}{a}}\varphi _2+4\psi _1\psi _1^{}4{\displaystyle \frac{a^{}}{a}}\varphi _1^2\right]_{,i}{\displaystyle \frac{4}{a^2}}\left[2\varphi _1\psi _{1,i}^{}\psi _1^{}\varphi _{1,i}\right]+O(k^3),`$ (14)
$`\delta G_{(2)j}^i`$ $`=`$ $`{\displaystyle \frac{2}{a^2}}[\psi _{1}^{}{}_{}{}^{2}2\varphi _1^{}\psi _1^{}+8({\displaystyle \frac{a^{}}{a}}\psi _1\psi _1^{}{\displaystyle \frac{a^{}}{a}}\varphi _1\varphi _1^{}{\displaystyle \frac{a^{}}{a}}\varphi _1\psi _1^{}{\displaystyle \frac{a^{\prime \prime }}{a}}\varphi _1^2)+4(\psi _1\psi _1^{\prime \prime }\varphi _1\psi _1^{\prime \prime }+{\displaystyle \frac{a^2}{a^2}}\varphi _1^2)`$ (15)
$`+2({\displaystyle \frac{a^{\prime \prime }}{a}}\varphi _2+{\displaystyle \frac{a^{}}{a}}\psi _2^{})+\psi _2^{\prime \prime }+{\displaystyle \frac{a^{}}{a}}\varphi _2^{}{\displaystyle \frac{a^2}{a^2}}\varphi _2]\delta ^i_j+O(k^2),`$
where $`O(k^n)`$ denotes terms of order $`n`$ or higher in gradients, since we are mainly interested in the large results, i.e. the case $`k0`$.
### II.2 Energy-momentum tensor
We split the scalar fields $`\phi _I`$ into a background and perturbations up to and including second order according to Eq. (4),
$$\phi _I(x^\mu )=\phi _{0I}(\eta )+\delta \phi _{1I}(x^\mu )+\frac{1}{2}\delta \phi _{2I}(x^\mu ).$$
(16)
The potential $`UU(\phi _I)`$ can be split similarly according to
$$U(\phi _I)=U_0+\delta U_1+\frac{1}{2}\delta U_2,$$
(17)
where
$`\delta U_1`$ $`=`$ $`{\displaystyle \underset{K}{}}U_{,\phi _K}\delta \phi _{K1},`$ (18)
$`\delta U_2`$ $`=`$ $`{\displaystyle \underset{K,L}{}}U_{,\phi _K\phi _L}\delta \phi _{1K}\delta \phi _{1L}+{\displaystyle \underset{K}{}}U_{,\phi _K}\delta \phi _{2K}.`$ (19)
and we use the shorthand $`U_{,\phi _K}\frac{U}{\phi _K}`$.
The energy-momentum tensor for $`N`$ scalar fields with potential $`U(\phi _I)`$ is then split into background, first, and second order perturbations, using Eq. (4), as
$$T_\nu ^\mu T_{(0)\nu }^\mu +\delta T_{(1)\nu }^\mu +\frac{1}{2}\delta T_{(2)\nu }^\mu ,$$
(20)
and we get for the components, from Eq. (3), at zeroth order
$`T_{(0)0}^0`$ $`=`$ $`\left({\displaystyle \underset{K}{}}{\displaystyle \frac{1}{2a^2}}\phi _{0K}^{}{}_{}{}^{2}+U_0\right),T_{(0)j}^i=\left({\displaystyle \frac{1}{2a^2}}{\displaystyle \underset{K}{}}\phi _{0K}^{}{}_{}{}^{2}U_0\right)\delta _j^i,`$ (21)
at first order
$`\delta T_{(1)0}^0`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{1K}^{}{}_{}{}^{}\phi _{0K}^{}{}_{}{}^{2}\varphi _1\right)\delta U_1,`$ (22)
$`\delta T_{(1)i}^0`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{1K}^{}{}_{,i}{}^{}\right),`$ (23)
$`\delta T_{(1)j}^i`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}\left[{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{1K}^{}{}_{}{}^{}\phi _{0K}^{}{}_{}{}^{2}\varphi _1\right)a^2\delta U_1\right]\delta _j^i,`$ (24)
and at second order in the perturbations
$`\delta T_{(2)0}^0`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{2K}^{}{}_{}{}^{}4\phi _{0K}^{}\varphi _1\delta \phi _{1K}^{}{}_{}{}^{}\phi _{0K}^{}{}_{}{}^{2}\varphi _2+4\phi _{0K}^{}{}_{}{}^{2}\varphi _1^2+\delta \phi _{1K}^{}{}_{}{}^{2}+a^2\delta U_2\right)+O(k^2),`$ (25)
$`\delta T_{(2)i}^0`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{2K}^{}{}_{,i}{}^{}4\varphi _1\phi _{0K}^{}\delta \phi _{1K}^{}{}_{,i}{}^{}+2\delta \phi _{1K}^{}{}_{}{}^{}\delta \phi _{1K}^{}{}_{,i}{}^{}\right),`$ (26)
$`\delta T_{(2)j}^i`$ $`=`$ $`{\displaystyle \frac{1}{a^2}}\left[{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}\delta \phi _{2K}^{}{}_{}{}^{}4\phi _{0K}^{}\varphi _1\delta \phi _{1K}^{}{}_{}{}^{}\phi _{0K}^{}{}_{}{}^{2}\varphi _2+4\phi _{0K}^{}{}_{}{}^{2}\varphi _1^2+\delta \phi _{1K}^{}{}_{}{}^{2}\right)a^2\delta U_2\right]\delta _j^i+O(k^2).`$ (27)
The $`00`$ component of the energy momentum tensor gives the corresponding energy density at zeroth, first and second order for a universe filled by $`N`$ scalar fields Noh ,
$$T_{(0)0}^0=\rho _0,\delta T_{(1)0}^0=\delta \rho _1,\delta T_{(2)0}^0=\delta \rho _2.$$
(28)
### II.3 Evolution of the background
The evolution of the background is governed by the Friedmann equation given from the $`00`$ component of the Einstein equations
$$^2=\frac{8\pi G}{3}\left(\frac{1}{2}\underset{K}{}\phi _{0K}^{}{}_{}{}^{2}+a^2U(\phi _{0I})\right),$$
(29)
where $`\frac{a^{}}{a}`$. The scalar fields are governed by the Klein-Gordon equation, which is from the energy conservation equation, (2), at zeroth order for the $`I`$th field given by
$`\phi _{0I}^{\prime \prime }+2\phi _{0I}^{}+a^2U_{,\phi _I}=0.`$ (30)
### II.4 Perturbed Klein-Gordon equation
Using the energy momentum tensor given above in Section II.2 and the energy conservation equation (2) we get the Klein-Gordon equation in the multiple filed case for the field $`I`$ at first order
$$\delta \phi _{1I}^{}{}_{}{}^{\prime \prime }+2\delta \phi _{1I}^{}{}_{}{}^{}+2a^2U_{,\phi _I}\varphi _13\phi _{0I}^{}\psi _1^{}\phi _{0I}^{}\varphi _1^{}+a^2\underset{K}{}U_{,\phi _I\phi _K}\delta \phi _{1K}+O(k^2)=0.$$
(31)
We get the Klein-Gordon equation at second order from Eq. (2) and using Eq. (31)
$`\delta \phi _{2I}^{}{}_{}{}^{\prime \prime }`$ $`+`$ $`2\delta \phi _{2I}^{}{}_{}{}^{}+2a^2U_{,\phi _I}\varphi _26\psi _1^{}\delta \phi _{1I}^{}{}_{}{}^{}2\varphi _1^{}\delta \phi _{1I}^{}{}_{}{}^{}+4\phi _{0I}^{}\varphi _1\varphi _1^{}+4\varphi _1a^2{\displaystyle \underset{K}{}}U_{,\phi _I\phi _K}\delta \phi _{1K}`$ (32)
$``$ $`12\phi _{0I}^{}\psi _1\psi _1^{}3\phi _{0I}^{}\psi _2^{}\phi _{0I}^{}\varphi _2^{}+a^2{\displaystyle \underset{K}{}}U_{,\phi _I\phi _K}\delta \phi _{2K}+a^2{\displaystyle \underset{K,L}{}}U_{,\phi _I\phi _K\phi _L}\delta \phi _{1K}\delta \phi _{1L}+O(k^2)=0.`$
Note, that Eqs. (31) and (32) are in an arbitrary gauge.
## III Variables
In this section we first calculate the changes coordinate transformations induce in the metric and matter perturbations at first and second order. We then construct physically meaningful gauge-invariant combinations from these variables.
### III.1 Gauge transformations
We now briefly review how tensorial quantities change under coordinate transformations Mukhanov ; Bruni . We consider two coordinate systems, $`\stackrel{~}{x^\mu }`$ and $`x^\mu `$, which are related by the coordinate transformation
$$\stackrel{~}{x^\mu }=e^{\xi ^\lambda \frac{}{x^\lambda }}x^\mu ,$$
(33)
where $`\xi ^\lambda `$ is the vector field generating the transformation and $`\xi ^\mu \xi _1^\mu +\frac{1}{2}\xi _2^\mu +O(ฯต^3)`$. Equation (33) can then be expanded up to second-order as
$`\stackrel{~}{x^\mu }=x^\mu +\xi _1^\mu +{\displaystyle \frac{1}{2}}\left(\xi _{1,\nu }^\mu \xi _1^\nu +\xi _2^\mu \right).`$ (34)
A tensor $`๐`$ transforms under the change of coordinate system defined above as
$$\stackrel{~}{๐}=e^{\mathrm{\pounds }_{\xi ^\lambda }}๐,$$
(35)
where $`\mathrm{\pounds }_{\xi ^\lambda }`$ denotes the Lie derivative with respect to $`\xi ^\lambda `$. Splitting the tensor$`๐`$ according to Eq. (4) we get Mukhanov ; Bruni
$`\stackrel{~}{\delta ๐_1}`$ $`=`$ $`\delta ๐_1+\mathrm{\pounds }_{\xi _1}๐_0`$
$`\stackrel{~}{\delta ๐_2}`$ $`=`$ $`\delta ๐_2+\mathrm{\pounds }_{\xi _2}๐_0+\mathrm{\pounds }_{\xi _1}^2๐_0+2\mathrm{\pounds }_{\xi _1}\delta ๐_1.`$ (36)
Under a first-order transformation $`\xi _1^\mu =(\alpha _1,\beta _{1,}^i)`$, a four scalar such as the energy density, $`\rho `$, or the scalar field $`\phi `$, transforms therefore as
$$\stackrel{~}{\delta \rho _1}=\delta \rho _1+\rho _0^{}\alpha _1,$$
(37)
and or the first order metric perturbations we have
$`\stackrel{~}{\psi }_1`$ $`=`$ $`\psi _1\alpha _1,`$ (38)
$`\stackrel{~}{\varphi }_1`$ $`=`$ $`\varphi _1+\alpha _1+\alpha _1^{}.`$ (39)
At second order, writing $`\xi _2^\mu =(\alpha _2,\beta _{2,}^i)`$, we find from Eq. (III.1) that a four scalar transforms as
$`\stackrel{~}{\delta \rho _2}`$ $`=`$ $`\delta \rho _2+\rho _0^{}\alpha _2+\alpha _1\left[\rho _0^{\prime \prime }\alpha _1+\rho _0^{}\alpha _{1}^{}{}_{}{}^{}+2\delta \rho _{1}^{}{}_{}{}^{}\right]`$ (40)
$`+\left(2\delta \rho _1+\rho _0^{}\alpha _1\right)_{,i}\beta _{1,}^i,`$
and the second order metric perturbations transform as
$`\stackrel{~}{\psi }_2`$ $`=`$ $`\psi _2\alpha _2\alpha _1\left[\alpha _{1}^{}{}_{}{}^{}+\left(^{}+2^2\right)\alpha _12\psi _1^{}4\psi _1\right]`$ (41)
$`\left(\alpha _12\psi _1\right)_{,k}\beta _{1,}^k,`$
$`\stackrel{~}{\varphi }_2`$ $`=`$ $`\varphi _2+\alpha _2+\alpha _{2}^{}{}_{}{}^{}+\alpha _1\left[\alpha _{1}^{}{}_{}{}^{\prime \prime }+5\alpha _{1}^{}{}_{}{}^{}+\left(^{}+2^2\right)\alpha _1+4\varphi _1+2\varphi _1^{}\right]`$ (42)
$`+\alpha _{1}^{}{}_{}{}^{}\left(2\alpha _{1}^{}{}_{}{}^{}+4\varphi _1\right)+\beta _{1,k}\left(\alpha _{1}^{}{}_{}{}^{}+\alpha _1+2\varphi _1\right)_,^k+\beta _{1,k}^{}\left(\alpha _12B_1\beta _1^{}\right)_,^k.`$
From Eqs. (40), (41), and (42) we see that on super horizon scales, where gradient terms can be neglected, the definition of the second order perturbations in the โnewโ coordinate is independent of the spatial coordinate choice (the โthreadingโ) at second order in the gradients. It is therefore sufficient on large scales (at $`O(k^2)`$) to specify the time slicing by prescribing $`\alpha _1`$ and $`\alpha _2`$, in order to define gauge-invariant variables MW ; LMS . The procedure to neglect the gradient terms, is explained in detail in Ref. LMS . For the approximation to hold one assumes that each quantity can be treated as smooth on some sufficiently large scale. Formally one multiplies each spatial gradient $`_i`$ by a fictitious parameter $`k`$, and expands the exact equations as a power series in $`k`$, keeping only the zero- and first-order terms, finally setting $`k=1`$.
*Note*, in the following sections we shall omit the symbol $`O(k^n)`$ denoting the order of the gradient terms neglected, assuming that if not stated otherwise the equations are valid to $`O(k^2)`$.
### III.2 Gauge-invariant combinations
Using the transformation behaviour of the perturbations derived in the last section we can now construct combinations of these variables that do not change under gauge transformations, i.e. gauge-invariant variables. The combinations given below have definite physical meaning, such as the curvature perturbation on a hypersurface of uniform density.
We define the various time slicings and then substitute the results into the transformation equations to arrive at the gauge-invariant variables. Whereas the first order results are valid on all scales, we only consider the large scale case at second order.
#### III.2.1 First order
Hypersurfaces of uniform $`I`$-field, i.e. $`\stackrel{~}{\delta \phi _{1I}}=0`$, are given from Eq. (37) by the temporal slicing
$$\alpha _1=\frac{\delta \phi _{1I}}{\phi _{0I}^{}}.$$
(43)
The curvature perturbation on uniform field hypersurfaces Lukash ; Lyth1985 , or comoving curvature perturbation, can then be defined for each field using Eq. (38) as
$$_{1I}=\psi _1+\frac{\delta \phi _{1}^{}{}_{I}{}^{}}{\phi _{0I}^{}}.$$
(44)
Flat slices are defined as $`\stackrel{~}{\psi _1}=0`$ and we therefore get from Eq. (38) the first order time shift
$$\alpha _1=\frac{\psi _1}{}.$$
(45)
The Sasaki-Mukhanov variable Sasaki1986 ; Mukhanov88 , or the field fluctuation on uniform curvature hypersurfaces, is then given by
$$๐ฌ_{1I}\stackrel{~}{\delta \phi _{1I}}=\delta \phi _{1I}+\frac{\phi _{0I}^{}}{}\psi _1.$$
(46)
The density perturbation on uniform curvature hypersurfaces is defined as
$$\stackrel{~}{\delta \rho _{1\alpha }}=\delta \rho _{1\alpha }+\frac{\rho _{0\alpha }^{}}{}\psi _1,$$
(47)
where $`\rho _{0\alpha }`$ and $`\delta \rho _{1\alpha }`$ are the energy density of the $`\alpha `$-fluid in the background and at first order <sup>2</sup><sup>2</sup>2 Greek indices from the beginning of the alphabet, $`\alpha ,\beta ,\gamma `$ will be used to denote different fluids.. The curvature perturbation on uniform $`\alpha `$-fluid density hypersurfaces is at first order defined as
$$\zeta _{1\alpha }=\psi _1\frac{\delta \rho _{1\alpha }}{\rho _{0\alpha }^{}}.$$
(48)
The lapse function on flat slices is from the definition (39) and Eq. (45) given as
$$\stackrel{~}{\varphi _1}=\varphi _1+\frac{1}{}\psi _1^{}+\left(1\frac{^{}}{^2}\right)\psi _1.$$
(49)
#### III.2.2 Second order
At second order uniform field slices are, from Eq. (40), defined by the temporal shift
$$\alpha _2=\frac{1}{\phi _{0I}^{}}\left(\delta \phi _{2I}\frac{1}{\phi _{0I}^{}}\delta \phi _{1I}^{}{}_{}{}^{}\delta \phi _{1I}\right),$$
(50)
on large scales. The curvature perturbation on uniform $`I`$-field slices is therefore on large scales at second order, from Eq. (41) and using (43) and (50) MW
$$_{2I}=\psi _2+\frac{}{\phi _{0I}^{}}\delta \phi _{2I}2\frac{}{\phi _{0I}^{}{}_{}{}^{2}}\delta \phi _{1I}^{}{}_{}{}^{}\delta \phi _{1I}2\frac{\delta \phi _{1I}}{\phi _{0I}^{}}\left(\psi _1^{}+2\psi _1\right)+\frac{\delta \phi _{1I}^{}{}_{}{}^{2}}{\phi _{0I}^{}{}_{}{}^{2}}\left(\frac{\phi _{0I}^{\prime \prime }}{\phi _{0I}}^{}2^2\right).$$
(51)
The time shift that defines the flat slicing is at second order, from Eq. (41), on large scales given by
$$\alpha _2=\frac{1}{}\left(\psi _2+2\psi _1^2+\frac{1}{}\psi _1^{}\psi _1\right).$$
(52)
Then the field fluctuation on uniform curvature hypersurfaces, or Sasaki-Mukhanov variable, at second order for the $`I`$th field, from Eq. (40) and definitions of the time shifts (45) and (52) is MW
$$๐ฌ_{2I}\stackrel{~}{\delta \phi _{2I}}=\delta \phi _{2I}+\frac{\phi _{0I}^{}}{}\psi _2+\left(\frac{\psi _1}{}\right)^2\left[2\phi _{0I}^{}+\phi _{0I}^{\prime \prime }\frac{^{}}{}\phi _{0I}^{}\right]+2\frac{\phi _{0I}^{}}{^2}\psi _1^{}\psi _1+\frac{2}{}\psi _1\delta \phi _{1I}^{}{}_{}{}^{}.$$
(53)
The second order lapse function in the flat gauge is given from Eqs. (42), (45), and (52) as
$`\stackrel{~}{\varphi _2}`$ $`=`$ $`\varphi _2+\left(1{\displaystyle \frac{^{}}{^2}}\right)\psi _2+{\displaystyle \frac{1}{}}\psi _2^{}+{\displaystyle \frac{2}{}}\left(54{\displaystyle \frac{^{}}{^2}}\right)\psi _1^{}\psi _1+\left(46{\displaystyle \frac{^{}}{^2}}{\displaystyle \frac{^{\prime \prime }}{^3}}+4{\displaystyle \frac{^2}{^4}}\right)\psi _1^2`$ (54)
$`+{\displaystyle \frac{3}{^2}}\psi _1^2+{\displaystyle \frac{2}{}}\psi _1^{\prime \prime }\psi _1+4\varphi _1\left[{\displaystyle \frac{1}{}}\psi _1^{}+\left(1{\displaystyle \frac{^{}}{^2}}\right)\psi _1\right]+{\displaystyle \frac{2}{}}\varphi _1^{}\psi _1.`$
The curvature perturbation on uniform $`\alpha `$-density hypersurfaces is at second order on large scales from Eq. (41) and using (43) and (50) given by MW
$`\zeta _{2\alpha }`$ $`=`$ $`\psi _2+{\displaystyle \frac{}{\rho _{0\alpha }^{}}}\delta \rho _{2\alpha }2{\displaystyle \frac{}{\rho _{0\alpha }^{}{}_{}{}^{2}}}\delta \rho _{1\alpha }^{}\delta \rho _{1\alpha }2{\displaystyle \frac{\delta \rho _{1\alpha }}{\rho _{0\alpha }^{}}}\left(\psi _1^{}+2\psi _1\right)+{\displaystyle \frac{\delta \rho _{1\alpha }^{}{}_{}{}^{2}}{\rho _{0\alpha }^{}{}_{}{}^{2}}}\left({\displaystyle \frac{\rho _{0\alpha }^{\prime \prime }}{\rho _{0\alpha }^{}}}^{}2^2\right),`$ (55)
and from Eq. (40) and the definitions of the time shifts (45) and (52) we get the second order density perturbation on uniform curvature hypersurfaces MW
$`\stackrel{~}{\delta \rho _{2\alpha }}`$ $``$ $`\delta \rho _{2\alpha }+{\displaystyle \frac{\rho _{0\alpha }^{}}{}}\psi _2+{\displaystyle \frac{1}{}}\left(2\rho _{0\alpha }^{}+{\displaystyle \frac{\rho _{0\alpha }^{\prime \prime }}{}}{\displaystyle \frac{\rho _{0\alpha }^{}^{}}{^2}}\right)\psi _1^2+2{\displaystyle \frac{\rho _{0\alpha }^{}}{^2}}\psi _1^{}\psi _1+{\displaystyle \frac{2}{}}\psi _1\delta \rho _{1\alpha }^{}.`$ (56)
## IV Relating variables
In this section we relate the gauge-invariant variables defined above on different hypersurfaces to each other. Using the field equations we then define total perturbations from the individual field and fluid ones, where possible.
### IV.1 First order
From Eqs. (61) and (46) the Sasaki-Mukhanov variable of the field $`I`$ is related to the curvature perturbation on uniform field hypersurfaces at linear order simply by
$$๐ฌ_{1I}=\frac{\phi _{0I}^{}}{}_{1I}.$$
(57)
From Eq. (47) and (48) we find that the curvature perturbation on uniform density hypersurfaces is related to the density perturbation on uniform curvature hypersurfaces simply as
$$\zeta _{1\alpha }=\frac{\stackrel{~}{\delta \rho _{1\alpha }}}{\rho _{0\alpha }^{}}.$$
(58)
The total density perturbation at first order in the flat gauge is given in terms of the density perturbations of individual fluids simply by
$$\stackrel{~}{\delta \rho _1}=\underset{\alpha }{}\stackrel{~}{\delta \rho _{1\alpha }},$$
(59)
which allows us to relate the total curvature perturbation to the individual fluid curvature perturbations by
$$\zeta _1=\underset{\alpha }{}\frac{\rho _{0\alpha }^{}}{\rho _0^{}}\zeta _{1\alpha }.$$
(60)
We can define the total comoving curvature perturbation (i.e. relative to the average fluid velocity or an โaverage fieldโ) using the $`0i`$ component of the field equations as
$$_{1\phi }\frac{1}{_K\phi _{0K}^{}{}_{}{}^{2}}\underset{I}{}\phi _{0I}^{}{}_{}{}^{2}_{1I}.$$
(61)
The total curvature perturbation on uniform density hypersurfaces in terms of the field fluctuations on flat slices is given by
$$\zeta _1=\frac{}{_L\phi _{0L}^{}{}_{}{}^{2}}\underset{K}{}\phi _{0K}^{}\stackrel{~}{๐ฌ_{K1}}.$$
(62)
### IV.2 Second order
The Sasaki-Mukhanov variable at second order defined above in Eq. (53) can be expressed in terms of the curvature perturbations on uniform field hypersurfaces, $`_{1I}`$ and $`_{2I}`$,
$$๐ฌ_{2I}=\frac{\phi _{0I}^{}}{}_{2I}+\left(\frac{_{1I}}{}\right)^2\left[2\phi _{0I}^{}+\phi _{0I}^{\prime \prime }\frac{^{}}{}\phi _{0I}^{}\right]+2\frac{\phi _{0I}^{}}{^2}_{1I}^{}_{1I}.$$
(63)
Similarly we can express the curvature perturbations on uniform field hypersurfaces in terms of the Sasaki-Mukhanov variables at first and second order and get,
$$_{2I}=\frac{}{\phi _{0I}^{}}๐ฌ_{2I}2\frac{}{\phi _{0I}^{}{}_{}{}^{2}}๐ฌ_{1I}^{}๐ฌ_{1I}+\frac{๐ฌ_{1I}^2}{\phi _{0I}^{}{}_{}{}^{2}}\left(\frac{\phi _{0I}^{\prime \prime }}{\phi _{0I}}^{}2^2\right).$$
(64)
To define the total comoving curvature perturbation at second order $`_{2\phi }`$ in terms of the $`_{2I}`$ we would need the $`0i`$ Einstein equation at this order in an appropriate form, that is without gradients. Since it is not clear how to arrive at this form of the $`0i`$ Einstein equation (without imposing slow roll, see Section VI.1 below) we shall leave the definition of $`_{2I}`$ open for the moment and shall return to this issue in a future publication M2005 . We note however, that the definition of the total comoving curvature perturbation at second order $`_{2\phi }`$ is not a problem in itself, since it was already shown in Ref. LMS that on large scales $`_{2\phi }`$ and $`\zeta _2`$ coincide. The definition of the total curvature perturbation $`\zeta _2`$ in terms of the field fluctuations is given below in Eq. (72), and we can therefore get $`_{2\phi }`$ from $`\zeta _2`$ if we should need it.
The definition of โtotalโ quantities from quantities defined for a specific field or fluid is not more problematic at second order than at first order, it is a mere question of having the โrightโ equations. As an example we shall now digress slightly from the main theme of this paper, multiple scalar fields, and analyse the definition of the total curvature perturbation at second order in the case of multiple fluids, which is unlike the definition of the total comoving curvature perturbation, straight forward <sup>3</sup><sup>3</sup>3The relation between scalar fields and fluids at first order, in particular how fields can be treated as fluids, has been studied in detail in Ref. MW2005 ..
We can relate the curvature perturbation on uniform $`\alpha `$-fluid hypersurfaces, Eq. (55), to the density perturbation on flat slices, Eq. (56),
$$\zeta _{2\alpha }=\frac{}{\rho _{0\alpha }^{}}\stackrel{~}{\delta \rho _{2\alpha }}2\frac{}{\rho _{0\alpha }^{}{}_{}{}^{2}}\stackrel{~}{\delta \rho _{1\alpha }}^{}\stackrel{~}{\delta \rho _{1\alpha }}+\frac{\stackrel{~}{\delta \rho _{1\alpha }}^2}{\rho _{0\alpha }^{}{}_{}{}^{2}}\left(\frac{\rho _{0\alpha }^{\prime \prime }}{\rho _{0\alpha }^{}}^{}2^2\right),$$
(65)
and similarly the density perturbation of the $`\alpha `$-fluid at second order on flat slices in terms of $`\zeta _{1\alpha }`$ and $`\zeta _{2\alpha }`$,
$$\stackrel{~}{\delta \rho _{2\alpha }}=\frac{\rho _{0\alpha }^{}}{}\zeta _{2\alpha }+2\frac{\rho _{0\alpha }^{}}{^2}\zeta _{1\alpha }^{}\zeta _{1\alpha }+\zeta _{1\alpha }^2\left(2\frac{\rho _{0\alpha }^{}}{}+\frac{\rho _{0\alpha }^{\prime \prime }}{^2}\frac{\rho _{0\alpha }^{}^{}}{^3}\right).$$
(66)
The total density perturbation at second order on flat slices is given in terms of the density perturbations of individual fluids on large scales simply by
$$\stackrel{~}{\delta \rho _2}=\underset{\alpha }{}\stackrel{~}{\delta \rho _{2\alpha }},$$
(67)
which allows us to write the total curvature perturbation at second order to the individual fluid curvature perturbations, using Eq. (66), as
$$\zeta _2=\underset{\alpha }{}\frac{\rho _{0\alpha }^{}}{\rho _0^{}}\zeta _{2\alpha }\left(1+3c_\mathrm{s}^2\right)\zeta _1^2+\underset{\alpha }{}\left[\left(1+3c_\alpha ^2\right)\frac{\rho _{0\alpha }^{}}{\rho _0^{}}+\frac{1}{\rho _0^{}}\left(Q_\alpha \frac{^{}}{}Q_\alpha ^{}\right)\right]\zeta _{1\alpha }^2\frac{2}{}\underset{\alpha }{}\zeta _{1\alpha }\left(\zeta _{1\alpha }^{}\zeta _1^{}\right),$$
(68)
where we used the background energy conservation equation for the $`\alpha `$-fluid, Eq. (111), given in Appendix A, $`c_\alpha ^2`$ is the adiabatic sound speed of the $`\alpha `$-fluid, and the total adiabatic sound speed is denoted $`c_\mathrm{s}^2`$. <sup>4</sup><sup>4</sup>4The relation between the total $`\zeta _2`$ and the curvature perturbations of the individual fluids, $`\zeta _{2\alpha }`$, has been given in the two field case in Ref. Bartolo:JCAP .
The relative entropy (or isocurvature) perturbation at first order is defined as WMLL ; MWU
$$๐ฎ_{1\alpha \beta }3\left(\zeta _{1\alpha }\zeta _{1\beta }\right),$$
(69)
which can be used to rewrite Eq. (68) in terms of adiabatic and non-adiabatic quadratic first order terms as
$`\zeta _2`$ $`=`$ $`{\displaystyle \underset{\alpha }{}}{\displaystyle \frac{\rho _{0\alpha }^{}}{\rho _0^{}}}\zeta _{2\alpha }{\displaystyle \frac{2}{3}}{\displaystyle \underset{\alpha }{}}{\displaystyle \frac{\rho _{0\alpha }^{}}{\rho _0^{}}}\left(\zeta _1+{\displaystyle \frac{1}{3}}{\displaystyle \underset{\gamma }{}}{\displaystyle \frac{\rho _{0\gamma }^{}}{\rho _0^{}}}๐ฎ_{1\alpha \gamma }\right){\displaystyle \underset{\gamma }{}}\left({\displaystyle \frac{\rho _{0\gamma }^{}}{\rho _0^{}}}๐ฎ_{1\alpha \gamma }\right)^{}`$ (70)
$`+{\displaystyle \frac{1}{3}}{\displaystyle \underset{\alpha }{}}\left[\left(1+3c_\alpha ^2\right){\displaystyle \frac{\rho _{0\alpha }^{}}{\rho _0^{}}}+{\displaystyle \frac{1}{\rho _0^{}}}\left(Q_\alpha {\displaystyle \frac{^{}}{}}Q_\alpha ^{}\right)\right]{\displaystyle \underset{\gamma }{}}{\displaystyle \frac{\rho _{0\gamma }^{}}{\rho _0^{}}}๐ฎ_{1\alpha \gamma }\left(2\zeta _1+{\displaystyle \frac{1}{3}}{\displaystyle \underset{\gamma }{}}{\displaystyle \frac{\rho _{0\gamma }^{}}{\rho _0^{}}}๐ฎ_{1\alpha \gamma }\right).`$
The evolution equation for the relative entropy perturbation at first order, $`๐ฎ_{1\alpha \gamma }`$, was given in MW2005 . We see that there is a purely non-adiabatic contribution to $`\zeta _2`$, quadratic in $`๐ฎ_{1\alpha \gamma }`$. Note that for purely adiabatic first order perturbations Eq. (70) simplifies to
$$\zeta _2=\underset{\alpha }{}\frac{\rho _{0\alpha }^{}}{\rho _0^{}}\zeta _{2\alpha }.$$
(71)
Turning back to scalar fields, the curvature perturbation $`\zeta _2`$ in terms of field fluctuations on flat slices at first and second order is, using Eq. (55) evaluated for a single fluid, the energy densities in terms of scalar fields Eq. (28), and the definitions for the Sasaki-Mukhanov variables Eqs. (46) and (53), given by
$`\zeta _2`$ $`=`$ $`{\displaystyle \frac{\rho _0}{3U_0_L\phi _{0L}^{}{}_{}{}^{2}}}{\displaystyle \underset{K}{}}\left(\phi _{0K}^{}๐ฌ_{2K}^{}+๐ฌ_{1K}^{}{}_{}{}^{2}+a^2\stackrel{~}{\delta U_2}\right){\displaystyle \frac{2}{_L\phi _{0L}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{2a^2U_0_L\phi _{0L}^{}{}_{}{}^{2}}{U_0_K\phi _{0K}^{}{}_{}{}^{2}}}\right){\displaystyle \underset{K,L}{}}U_{,\phi _K}\phi _{0L}^{}๐ฌ_{1K}๐ฌ_{1L}`$ (72)
$`^2\left({\displaystyle \frac{_K\phi _{0K}^{}๐ฌ_{1K}}{_L\phi _{0L}^{}{}_{}{}^{2}}}\right)^2\left[73c_\mathrm{s}^2{\displaystyle \frac{6_L\phi _{0L}^{}{}_{}{}^{2}}{a^2\rho _0}}\right],`$
where $`\stackrel{~}{\delta U_2}`$ is given from Eq. (19) above evaluated on flat slices,
$$\stackrel{~}{\delta U_2}=\underset{K,L}{}U_{,\phi _K\phi _L}๐ฌ_{1K}๐ฌ_{1L}+\underset{K}{}U_{,\phi _K}๐ฌ_{2K},$$
(73)
and the adiabatic sound speed for a universe filled by $`N`$ scalar fields is given by
$$c_\mathrm{s}^2=1+\frac{2}{3}\frac{_KU_{,\phi _K}\phi _{0K}^{}}{\frac{1}{a^2}_K\phi _{0K}^{}{}_{}{}^{2}},$$
(74)
and $`\rho _0`$ is the background energy density defined in Eq. (28) above.
Equation (72) can be readily evaluated either analytically or numerically given the Sasaki-Mukhanov variables at first and second order, $`๐ฌ_{1I}`$ and $`๐ฌ_{2I}`$. However, in Section VII below we use $`0i`$ Einstein equation at second order in the single field case and and in Section VIII we use the slow roll approximation in a particular model to simplify Eq. (72) further, without the time derivatives of the Sasaki-Mukhanov variables.
## V Governing equations in the uniform curvature gauge at first order
We begin this section by giving the field equations at first order in the uniform curvature gauge and then review the derivation of the Klein-Gordon equation for multiple scalar fields at first order on large scales.
### V.1 Field equations
The components of the Einstein tensor and the energy-momentum tensor are given in Sections II.1 and II.2. Substitution into Eq. (1) then gives for the $`00`$ component of the Einstein equations gives at first order
$$2a^2U_0\stackrel{~}{\varphi _1}+\underset{K}{}\phi _{0K}^{}๐ฌ_{1K}^{}{}_{}{}^{}+a^2\stackrel{~}{\delta U_1}=0,$$
(75)
and from the $`ij`$ component of the Einstein equation we get
$$\frac{a^{}}{a}\stackrel{~}{\varphi _1}^{}8\pi G\underset{K}{}\phi _{0K}^{}๐ฌ_{1K}^{}{}_{}{}^{}=0.$$
(76)
At first order the $`0i`$ components of the Einstein tensor, Eq. (11), and energy momentum tensor, Eq. (23), are linear in the spatial gradients (by definition the background quantities are only time dependent), which allows us to write the $`0i`$ Einstein equation without gradients as
$$\frac{a^{}}{a}\stackrel{~}{\varphi _1}4\pi G\underset{K}{}\phi _{0K}^{}๐ฌ_{1K}=0.$$
(77)
Combining Eqs. (75) and (77) we get
$$\underset{K}{}\phi _{0K}^{}๐ฌ_{1K}^{}{}_{}{}^{}=a^2\underset{K}{}\left(U_{,\phi _K}+\frac{8\pi G}{}U_0\phi _{0K}^{}\right)๐ฌ_{1K},$$
(78)
relating the time derivative of the field fluctuations to the field fluctuations themselves.
### V.2 Klein Gordon equation
At first order the Klein-Gordon equation for the field $`I`$, on flat slices, is from Eq. (31) given by
$$๐ฌ_{1I}^{}{}_{}{}^{\prime \prime }+2๐ฌ_{1I}^{}{}_{}{}^{}+2a^2U_{,\phi _I}\stackrel{~}{\varphi _1}\phi _{0I}^{}\stackrel{~}{\varphi _1}^{}+a^2\underset{K}{}U_{,\phi _I\phi _K}๐ฌ_{1K}=0,$$
(79)
which can be simplified using Eqs. (75) and (76), to give
$$๐ฌ_{1I}^{}{}_{}{}^{\prime \prime }+2๐ฌ_{1I}^{}{}_{}{}^{}\left(\frac{U_{,\phi _I}}{U_0}+\phi _{0I}^{}\frac{8\pi G}{}\right)\underset{K}{}\phi _{0K}^{}๐ฌ_{1K}^{}{}_{}{}^{}+a^2\underset{K}{}\left(U_{,\phi _I\phi _K}\frac{1}{U_0}U_{,\phi _I}U_{,\phi _K}\right)๐ฌ_{1K}=0.$$
(80)
Equation (80) can be further rewritten using Eq. (78) above, to give Hwang ; Taruya ; Chris
$$๐ฌ_{1I}^{}{}_{}{}^{\prime \prime }+2๐ฌ_{1I}^{}{}_{}{}^{}+\underset{K}{}\left[a^2U_{,\phi _I\phi _K}\frac{8\pi G}{a^2}\left(a^2\phi _{0I}^{}\left(\frac{\phi _{0K}^{}}{}\right)\right)^{}\right]๐ฌ_{1K}=0,$$
(81)
which now displays the โcanonicalโ time derivatives (in conformal time) $`^2/\eta ^2+2/\eta `$, but has a rather involved mass term.
## VI Governing equations in the uniform curvature gauge at second order
In this section we first give the field equation in the uniform curvature gauge at second order, highlighting problems arising from the $`0i`$ equation at second order and possible ways to solve them. We then give the Klein-Gordon equation in the multiple field case on large scales in terms of the Sasaki-Mukhanov variables.
### VI.1 Field equations
In this section we present the field equations in the uniform curvature gauge at second order on large scales. The components of the Einstein tensor and the energy-momentum tensor are given in Sections II.1 and II.2. Substitution into Eq. (1) then gives for the $`00`$ field equation in the flat gauge at second order is
$$2a^2U_0\left(4\stackrel{~}{\varphi _1}^2\stackrel{~}{\varphi _2}\right)=\underset{K}{}\left(\phi _{0K}^{}๐ฌ_{2K}^{}4\phi _{0K}^{}\stackrel{~}{\varphi _1}๐ฌ_{1K}^{}+๐ฌ_{1K}^{}{}_{}{}^{2}\right)+a^2\stackrel{~}{\delta U_2}.$$
(82)
and, using Eq. (77) and Eq. (82) above, the $`ij`$ Einstein equation can be rewritten as
$$\frac{a^{}}{a}\left(\stackrel{~}{\varphi _2}^{}4\stackrel{~}{\varphi _1}^{}\stackrel{~}{\varphi _1}\right)=8\pi G\left[\underset{K}{}\left(\phi _{0K}^{}๐ฌ_{2K}^{}+๐ฌ_{1K}^{}{}_{}{}^{2}\right)\right].$$
(83)
The $`0i`$ Einstein equation at second order is given by
$$\left(\stackrel{~}{\varphi _2}_{,i}4\stackrel{~}{\varphi _1}\stackrel{~}{\varphi _1}_{,i}\right)4\pi G\underset{K}{}\left(\phi _{0K}^{}๐ฌ_{2K,i}+2๐ฌ_{1K}^{}{}_{}{}^{}๐ฌ_{1K,i}\right)+O(k^3)=0,$$
(84)
where we used Eq. (77). The left hand side of Eq. (84) can be rewritten as $`\left(\stackrel{~}{\varphi _2}2\stackrel{~}{\varphi _1}^2\right)_{,i}`$, with the spatial derivative outside the brackets. However the right hand side of Eq. (84) canโt be written, in the general case, as an overall gradient, as can be seen above: whereas the $`๐ฌ_{2K,i}`$ term is multiplied by a background quantity, similar to the first order case, the $`๐ฌ_{1K}^{}{}_{}{}^{}๐ฌ_{1K,i}`$ term canโt be written as an overall gradient and the $`0i`$ Einstein equation at second order cannot be brought into scalar form immediately.
In order to recast Eq. (84) in a more useful form, without the gradients, we have several possibilities:
* The easiest solution is to require $`๐ฌ_{1K}=const`$ or $`๐ฌ_{1K}=0`$. Unfortunately this case is not particularly interesting, since one of the most interesting applications of second order theory is the study of non-gaussianity, which necessitates the inclusion and evolution of the first order and in particular the first order squared terms. We shall therefore not pursue this option further.
* The next case is to assume we only have one field to consider and we shall study this case in detail in Section VII. Actually, we need only one field at first order, but to assume a single field at first order and multiple field perturbations at second order seems to be rather contrived.
* Another possibility is to use the slow roll approximation such that Eq. (81) can be rewritten as $`๐ฌ_{1I}^{}f(\phi _{0J})๐ฌ_{1I}`$, where $`f(\phi _{0J})`$ is a function of the background fields, which allows as in the single field case to replace $`๐ฌ_{1I}^{}๐ฌ_{1I}`$ by $`๐ฌ_{1I}^2`$ in Eq. (84) (for a recent detailed exposition of the slow roll approximation in the multi-field case see e.g. Seery:2005gb ). We shall illustrate this particular case, i.e. using the slow roll approximation, by studying a simple two-field inflation model in Section VIII.
* Finally, we can take the divergence of Eq. (84), which we shall do next.
Following Ref. Acquaviva we take the divergence of Eq. (84) and get
$`^2\stackrel{~}{\varphi _2}`$ $``$ $`4\left(\stackrel{~}{\varphi _1}^2\stackrel{~}{\varphi _1}+\stackrel{~}{\varphi _1}_{,k}\stackrel{~}{\varphi _{1,}}^k\right)`$ (85)
$``$ $`4\pi G{\displaystyle \underset{K}{}}\left[\phi _{0K}^{}^2๐ฌ_{2K}+2๐ฌ_{1K}^{}^2๐ฌ_{1K}+2๐ฌ_{1K,j}^{}๐ฌ_{1K,}^j\right]+O(k^3)=0.`$
Using now the inverse Laplacian, defined as $`^2^2X=X`$, Eq. (85) can be rewritten as
$$\left(\stackrel{~}{\varphi _2}4\stackrel{~}{\varphi _1}^2\right)4\pi G\underset{K}{}\left[\phi _{0K}^{}๐ฌ_{2K}+2๐ฌ_{1K}^{}๐ฌ_{1K}\right]R(\stackrel{~}{\varphi _1},๐ฌ_{1I})+O(k^3)=0,$$
(86)
where we define $`R(\stackrel{~}{\varphi _1},๐ฌ_{1I})`$ as
$$R(\stackrel{~}{\varphi _1},๐ฌ_{1I})4\left[^2\stackrel{~}{\varphi _1}^2\stackrel{~}{\varphi _1}+^2\left(\stackrel{~}{\varphi _1}_{,k}\stackrel{~}{\varphi _{1,}}^k\right)\right]+8\pi G\underset{K}{}\left[^2๐ฌ_{1K}^{}^2๐ฌ_{1K}+^2\left(๐ฌ_{1K,j}^{}๐ฌ_{1K,}^j\right)\right].$$
(87)
But since neither the effect of $`R(\stackrel{~}{\varphi _1},๐ฌ_{1I})`$ on the evolution of the field fluctuations nor its large scale limit is clear, we shall not use this form of the $`0i`$ Einstein equation below.
Not being able to make use of the $`0i`$ Einstein equation is inconvenient, but since it is only a constraint equation and therefore redundant, we are able to get a closed system of equations without using it, as shown in Section VI.2 below, which is sufficient to get the evolution of $`\zeta _2`$ from Eq. (72). However, the $`0i`$ Einstein equation does allow us to rewrite the Klein Gordon equation in a more compact form with canonical time derivatives (see above the first order case, Section V, and below the single field second order case, Section VII).
### VI.2 Klein Gordon equation
In this section we derive the Klein-Gordon equation in the flat gauge for $`N`$ scalar fields on large scales.
At second order the Klein Gordon equations on flat slices is for the field $`I`$, from Eq. (32), on large scales given by
$`๐ฌ_{2I}^{}{}_{}{}^{\prime \prime }`$ $`+`$ $`2๐ฌ_{2I}^{}{}_{}{}^{}+2a^2U_{,\phi _I}\stackrel{~}{\varphi _2}2\stackrel{~}{\varphi _1}^{}๐ฌ_{1I}^{}{}_{}{}^{}+4\phi _{0I}^{}\stackrel{~}{\varphi _1}\stackrel{~}{\varphi _1}^{}+4\stackrel{~}{\varphi _1}a^2{\displaystyle \underset{K}{}}U_{,\phi _I\phi _K}๐ฌ_{1K}`$ (88)
$``$ $`\phi _{0I}^{}\stackrel{~}{\varphi _2}^{}+a^2{\displaystyle \underset{K}{}}U_{,\phi _I\phi _K}๐ฌ_{2K}+a^2{\displaystyle \underset{K,L}{}}U_{,\phi _I\phi _K\phi _L}๐ฌ_{1K}๐ฌ_{1L}=0.`$
We can rewrite Eq. (88) above, using the field equations at first and second order given above in Sections V.1 and VI.1, to substitute for the lapse functions at first and second order, $`\stackrel{~}{\varphi _1}`$ and $`\stackrel{~}{\varphi _2}`$, and get
$`๐ฌ_{2I}^{}{}_{}{}^{\prime \prime }`$ $`+`$ $`2๐ฌ_{2I}^{}{}_{}{}^{}+a^2{\displaystyle \underset{K}{}}\left[U_{,\phi _I\phi _K}{\displaystyle \frac{1}{U_0}}U_{,\phi _I}U_{,\phi _K}\right]๐ฌ_{2K}\left({\displaystyle \frac{U_{,\phi _I}}{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _{0I}^{}\right){\displaystyle \underset{K}{}}\left(\phi _{0K}^{}๐ฌ_{2K}^{}{}_{}{}^{}+๐ฌ_{1K}^{}{}_{}{}^{2}\right)`$ (89)
$`+`$ $`a^2{\displaystyle \underset{K,L}{}}\left\{U_{,\phi _I\phi _K\phi _L}{\displaystyle \frac{1}{U_0}}U_{,\phi _I}U_{,\phi _K\phi _L}+{\displaystyle \frac{16\pi G}{}}{\displaystyle \underset{K,L}{}}\left(U_{,\phi _I\phi _K}{\displaystyle \frac{1}{U_0}}U_{,\phi _I}U_{,\phi _K}\right)\phi _{0L}^{}\right\}๐ฌ_{1L}๐ฌ_{1K}`$
$`+`$ $`a^2{\displaystyle \frac{16\pi G}{}}{\displaystyle \underset{K}{}}\left({\displaystyle \frac{U_{,\phi _K}}{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _{0K}^{}\right)๐ฌ_{1K}๐ฌ_{1I}^{}{}_{}{}^{}=0.`$
This is the gauge-invariant Klein-Gordon equation in the uniform curvature gauge for $`N`$ minimally coupled scalar fields on large scales in terms of the field fluctuations in the flat gauge at first and second order, $`๐ฌ_{1I}`$ and $`๐ฌ_{2I}`$. It is linear in the second order variables, but quadratic in the first order ones.
## VII Klein Gordon equation and curvature perturbation: single field case
Having dealt with the multi-field case in the previous section, we now turn to the single field case. The general form of the Klein Gordon equation at second order, Eq. (89), reduces in this case to
$`๐ฌ_{2}^{}{}_{}{}^{\prime \prime }`$ $`+`$ $`2๐ฌ_{2}^{}{}_{}{}^{}+a^2\left[U_{\phi \phi }{\displaystyle \frac{1}{U_0}}U_\phi U_\phi \right]๐ฌ_2\left({\displaystyle \frac{U_\phi }{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _0^{}\right)\left(\phi _0^{}๐ฌ_{2}^{}{}_{}{}^{}+๐ฌ_{1}^{}{}_{}{}^{2}\right)`$ (90)
$`+`$ $`a^2\left\{U_{,\phi \phi \phi }{\displaystyle \frac{1}{U_0}}U_\phi U_{\phi \phi }+{\displaystyle \frac{16\pi G}{}}\left(U_{\phi \phi }{\displaystyle \frac{1}{U_0}}U_\phi U_\phi \right)\phi _0^{}\right\}๐ฌ_{1}^{}{}_{}{}^{2}`$
$`+`$ $`a^2{\displaystyle \frac{16\pi G}{}}\left({\displaystyle \frac{U_\phi }{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _0^{}\right)๐ฌ_{I}^{}{}_{}{}^{}๐ฌ_K=0.`$
In the single field case we can use Eq. (78) to rewrite the $`0i`$ Einstein equation at second order, Eq. (84), as
$$\left(\stackrel{~}{\varphi _2}2\stackrel{~}{\varphi _1}^2\right)_{,i}=4\pi G\left[\phi _0^{}๐ฌ_2a^2\left(\frac{U_{,\phi }}{\phi _0^{}}+8\pi G\frac{U_0}{}\right)๐ฌ_1^2\right]_{,i}+O(k^3)=0,$$
(91)
which allows us to immediately get rid of the spatial gradient in a similar fashion as in the first order case. From Eqs. (82) and (91) we then get the useful relation, similar to Eq. (78) at first order,
$`\phi _0^{}๐ฌ_2^{}+๐ฌ_{1}^{}{}_{}{}^{2}`$ $`=`$ $`a^2U_0\left({\displaystyle \frac{U_{,\phi }}{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _0^{}\right)๐ฌ_2a^2U_{,\phi \phi }๐ฌ_1^2`$ (92)
$`a^2U_0{\displaystyle \frac{8\pi G}{}}\left[2\phi _0^{}{\displaystyle \frac{U_{,\phi }}{U_0}}+{\displaystyle \frac{8\pi G}{}}\phi _{0}^{}{}_{}{}^{2}a^2\left({\displaystyle \frac{U_{,\phi }}{\phi _0^{}}}+{\displaystyle \frac{8\pi G}{}}U_0\right)\right]๐ฌ_1^2.`$
Substituting Eq. (92) into Eq. (90) we arrive at the single field Klein Gordon equation at second order,
$`๐ฌ_{2}^{}{}_{}{}^{\prime \prime }`$ $`+`$ $`2๐ฌ_{2}^{}{}_{}{}^{}+a^2\left[U_{,\phi \phi }+{\displaystyle \frac{16\pi G}{}}U_0\phi _0^{}\left({\displaystyle \frac{U_{,\phi }}{U_0}}+{\displaystyle \frac{4\pi G}{}}\phi _0^{}\right)\right]๐ฌ_2`$ (93)
$`+`$ $`a^2\{U_{,\phi \phi \phi }+{\displaystyle \frac{24\pi G}{}}\phi _0^{}U_{,\phi \phi }+{\displaystyle \frac{8\pi G}{}}[3U_{,\phi }({\displaystyle \frac{8\pi G}{}}(\phi _{0}^{}{}_{}{}^{2}2a^2U_0)a^2{\displaystyle \frac{U_{,\phi }}{\phi _0^{}}})`$
$`+\left({\displaystyle \frac{8\pi G}{}}\right)^2U_0\phi _0^{}(\phi _{0}^{}{}_{}{}^{2}3a^2U_0)]\}๐ฌ_1^2=0.`$
now with the time derivatives in canonical form, which is nice.
In the single field case the curvature perturbation on uniform density hypersurfaces at first order in terms of the field fluctuation on the flat slice is given, from Eq. (62), as
$$\zeta _1=\frac{}{\phi _0^{}}๐ฌ_1,$$
(94)
and at second order we get from Eq. (72), using Eq. (92),
$`\zeta _2={\displaystyle \frac{}{\phi _0^{}}}๐ฌ_2\left[73c_\mathrm{s}^23{\displaystyle \frac{\phi _{0}^{}{}_{}{}^{2}a^2U_0}{a^2\rho _0}}+3{\displaystyle \frac{a^2U_{,\phi }}{\phi _0^{}}}\right]\left({\displaystyle \frac{}{\phi _0^{}}}\right)^2๐ฌ_1^2.`$ (95)
Using Eqs. (93) and (95) should be sufficient to show the conservation of $`\zeta _2`$ on large scales in the single field case, which was shown in Refs. SB ; filippo , and we will return to this issue in a future work M2005 .
## VIII A simple application using slow roll
We can now apply the formalism derived in the previous sections to the simple two field model of Ref. Enqvist:2004bk which has subsequently been studied in Refs. Lyth:2005du ; Lyth:2005fi and most recently in Ref. antti (for earlier work on two-field inflation at linear order see e.g. Polarski:1992dq ). Using the slow-roll approximation we calculate the curvature perturbation at second order, $`\zeta _2`$, in terms of the field fluctuations.
The potential is given as
$$U=\overline{U}_0+\frac{1}{2}m_1^2\phi _1^2+\frac{1}{2}m_2^2\phi _2^2,$$
(96)
with the first term, $`\overline{U}_0=const`$, dominating and where $`m_I`$ denotes the mass of the $`I`$th field. Furthermore we assume slow roll and
$$\phi _{02}=0,\phi _{02}^{}=0.$$
(97)
The background Friedmann constraint, Eq. (29), then simplifies to
$$^2=\frac{8\pi G}{3}\overline{U}_0,$$
(98)
and the background Klein-Gordon equation, Eq. (30), gives
$$\phi _{01}^{}+\frac{a^2m_1^2}{3}\phi _{01}=0.$$
(99)
From Eq. (62) we immediately get the curvature perturbation at first order in terms of the field fluctuations on flat slices in this model,
$$\zeta _1=8\pi G\frac{\overline{U}_0}{m_1^2\phi _{01}}๐ฌ_{11}.$$
(100)
The calculation of the second order curvature perturbation in terms of the field fluctuations is slightly more involved. As pointed out in Section IV.2, in order to get $`\zeta _2`$ solely in terms of the field fluctuations without their time derivatives we have to use the $`0i`$ Einstein equation at second order, and, as discussed in Section VI.1, to get the latter in a useful form we need the first order Klein-Gordon equations.
The perturbed Klein-Gordon equation at first order on flat slices, Eq. (79), reduces in the slow roll limit for the model specified by Eqs. (96) and (97) to
$$๐ฌ_{11}^{}+\frac{a^2m_1^2}{3}๐ฌ_{11}=0,๐ฌ_{12}^{}+\frac{a^2m_2^2}{3}๐ฌ_{12}=0,$$
(101)
where $`๐ฌ_{11}`$ and $`๐ฌ_{12}`$ are the field fluctuations in the flat gauge, the Sasaki-Mukhanov variables, at first order for the two fields.
The $`0i`$ equation at second order on flat slices, Eq. (84), simplifies then using Eqs. (101) to
$$\left(\stackrel{~}{\varphi }_22\stackrel{~}{\varphi }_1^2\right)=4\pi G\left[\phi _{01}^{}๐ฌ_{21}\frac{a^2m_1^2}{3}๐ฌ_{11}^2\frac{a^2m_2^2}{3}๐ฌ_{12}^2\right].$$
(102)
Rewriting Eq. (82), the $`00`$ Einstein equation at second order, in a similar fashion we arrive at the useful relation
$$\underset{K}{}\left(\phi _{0K}๐ฌ_{2K}^{}+๐ฌ_{1K}^{}{}_{}{}^{2}\right)+a^2\delta U_2=a^2\left(m_1^2\phi _{01}๐ฌ_{21}+m_1^2๐ฌ_{11}^2+m_2^2๐ฌ_{12}^2\right),$$
(103)
which upon substitution in Eq. (72) yields
$$\zeta _2=8\pi G\frac{\overline{U}_0}{m_1^2\phi _{01}^2}\left[\phi _{01}๐ฌ_{21}+\frac{m_2^2}{m_1^2}๐ฌ_{12}^2+\left(1+2\frac{8\pi G\overline{U}_0}{m_1^2}\right)๐ฌ_{11}^2\right],$$
(104)
the curvature perturbation on uniform density hypersurfaces at second order in terms of the field fluctuations on flat slices for the model specified above.
This expression for $`\zeta _2`$, Eq. (104) above, agrees with the one found by Lyth and Rodriguez in Ref. Lyth:2005fi using the $`\mathrm{\Delta }N`$ formalism, if we take into account that $`\zeta _2=\zeta _{2\mathrm{L}\mathrm{R}}+2\zeta _{1}^{}{}_{}{}^{2}`$ LMS . In particular we find that $`\zeta _2`$ does not contain any non-local terms.
However, the curvature perturbation $`\zeta _2`$ found here disagrees with the expression for the second order curvature perturbation found by Enqvist and Vaihkonen in Ref. Enqvist:2004bk , although the order of magnitude estimate published subsequently by Vaihkonen antti seems to agree with the result found here.
## IX Discussion and conclusions
In this paper we have presented the Klein-Gordon equation for multiple minimally coupled scalar fields at second order in the perturbations in a perturbed FRW background on large scales. We have shown that using suitable gauge-invariant variables, namely the field fluctuations in the flat gauge or Sasaki-Mukhanov variables, the Klein-Gordon equation at second order, Eq. (89), can be written solely in terms of these variables, as in the first order case, and is linear in the second order variables, but has source terms quadratic in the first order field fluctuations.
We have also given the relation between gauge-invariant quantities in different gauges, and hence on different hypersurfaces, which at second order is non-trivial. In particular we give the curvature perturbation on uniform density hypersurfaces, $`\zeta _2`$, in terms of the Sasaki-Mukhanov variables of the individual fields, $`๐ฌ_{1I}`$ and $`๐ฌ_{2I}`$ in Eq. (72). We calculated $`\zeta _2`$ for a particular two-field model during slow-roll inflation, Eq. (104), and found excellent agreement with the expression derived using the $`\mathrm{\Delta }N`$ formalism Lyth:2005fi . In particular we find using second order cosmological perturbation theory that there are no non-local terms in the expression for $`\zeta _2`$ in this model.
Having an expression for $`\zeta _2`$ solely in terms of the field fluctuations allows us to get the evolution of $`\zeta _2`$ directly from solving the Klein-Gordon equations at zeroth, first, and second order, Eqs. (30), (81), and (89), respectively, together with the Friedmann constraint, Eq. (29). Alternatively we could have calculated the non-adiabatic pressure due to the multiple scalar fields first and then solve an evolution equation for $`\zeta _2`$ WMLL ; MW ; MW2005 , but using Eq. (72) eliminates the integration of the $`\zeta _2`$ evolution equation and is therefore simpler, since the Klein-Gordon equations have to be solved in both cases.
Given suitable initial conditions for the Klein-Gordon equations governing the dynamics of the fields at zeroth, first, and second order the equations can be integrated numerically. The equations can be solved order by order, the background and first order fields acting as source terms for the second order equations, since we do not consider back-reaction of the higher order perturbations on the lower order ones. Solving the second order equations numerically will therefore be very similar to solving the first order system.
Alternatively, employing a slow roll approximation at all orders, the equations can be approximated analytically. It will be particularly interesting to compare the evolution of $`\zeta _2`$ from a slow roll version of the Klein-Gordon equation, Eq. (89), to the $`\mathrm{\Delta }N`$-formalism Lyth:2005fi , which also uses slow roll and also gives the evolution of the curvature perturbation directly from the solution of the Klein-Gordon equations, without having to integrate an evolution equation for $`\zeta _2`$.
The next step will be the extension of the formalism presented here to small scales, i.e. deriving a formalism valid on all scales. This will also be an advantage compared to other approaches based on the separate universe paradigm, such as the $`\mathrm{\Delta }N`$-formalism, which can not be extended to small scales since it is by assumption only valid on scales of order or larger than the horizon. First steps towards extending the formalism are under way at present, based on the work presented here M2005 .
###### Acknowledgements.
The author is grateful to David Lyth, David Matravers, and David Wands for useful discussions and comments. KAM is supported by PPARC grant PPA/G/S/2002/00098. Algebraic computations of tensor components were performed using the GRTensorII package for Maple.
## Appendix A Background field equations
The Friedmann equation is given from the $`00`$ component of the Einstein equations
$$^2=\frac{8\pi G}{3}a^2\rho _0,$$
(105)
where $`\frac{a^{}}{a}`$. From the $`ij`$ component we find
$$\left(\frac{a^{}}{a}\right)^22\frac{a^{\prime \prime }}{a}=8\pi Ga^2P_0.$$
(106)
The two previous equations can be rewritten as
$$^{}=\frac{4\pi G}{3}a^2\left(\rho _0+3P_0\right),$$
(107)
or alternatively as
$$\frac{a^{\prime \prime }}{a}=\frac{4\pi G}{3}a^2\left(\rho _03P_0\right).$$
(108)
Also useful are
$$2\left(\frac{a^{}}{a}\right)^2\frac{a^{\prime \prime }}{a}=4\pi Ga^2\left(P_0+\rho _0\right),$$
(109)
where the right hand side in the case of scalar fields simplifies to $`a^2\left(P_0+\rho _0\right)=_K\phi _{0K}^{}{}_{}{}^{2}`$, and
$$^{}+2^2=4\pi Ga^2\left(\rho _0P_0\right),$$
(110)
where the right hand side in the case of scalar fields simplifies to $`\left(\rho _0P_0\right)=2U_0`$.
The background energy conservation for the $`\alpha `$-fluid is given by KS ; MW2005
$$\rho _{0\alpha }^{}=3\left(\rho _{0\alpha }+P_{0\alpha }\right)+Q_\alpha ,$$
(111)
where $`\rho _{0\alpha }`$, $`P_{0\alpha }`$, and $`Q_\alpha `$ are the energy density, the pressure and the energy transfer to the $`\alpha `$-fluid. Note that the energy transfer defined here is related to the one define in MW2005 , $`\widehat{Q}_\alpha `$, by $`aQ_\alpha =\widehat{Q}_\alpha `$. The adiabatic sound speed of the $`\alpha `$-fluid is
$$c_\alpha ^2\frac{P_{0\alpha }^{}}{\rho _{0\alpha }^{}},$$
(112)
related to the total adiabatic sound speed by
$$c_\mathrm{s}^2=\underset{\alpha }{}\frac{\rho _{0\alpha }^{}}{\rho _0^{}}c_\alpha ^2.$$
(113)
For more details on the multi-fluid formalism see MW2005 .
## Appendix B The metric tensor
The metric tensor up to second order, including only scalar perturbations, is
$`g_{00}`$ $`=`$ $`a^2\left(1+2\varphi _1+\varphi _2\right),`$ (114)
$`g_{0i}`$ $`=`$ $`a^2\left(B_1+{\displaystyle \frac{1}{2}}B_2\right)_{,i},`$ (115)
$`g_{ij}`$ $`=`$ $`a^2\left[\left(12\psi _1\psi _2\right)\delta _{ij}+2E_{1,ij}+E_{2,ij}\right].`$ (116)
and its contravariant form is
$`g^{00}`$ $`=`$ $`a^2\left[12\varphi _1\varphi _2+4\varphi _1^2B_{1,k}B_{1,}^k\right],`$ (117)
$`g^{0i}`$ $`=`$ $`a^2\left[B_{1,}^i+{\displaystyle \frac{1}{2}}B_{2,}^i2B_{1,k}E_{1,}^{ki}+2\left(\psi _1\varphi _1\right)B_{1,}^i\right],`$ (118)
$`g^{ij}`$ $`=`$ $`a^2\left[\left(1+2\psi _1+\psi _2+4\psi _1^2\right)\delta ^{ij}\left(2E_{1,}^{ij}+E_{2,}^{ij}4E_{1,}^{ik}E_{1,k}^j+8\psi _1E_{1,}^{ij}+B_{1,}^iB_{1,}^j\right)\right],`$ (119)
|
warning/0506/hep-lat0506005.html
|
ar5iv
|
text
|
# Precision Lattice Calculation of SU(2) โt Hooft loops
## 1 Introduction
Because of asymptotic freedom, the running coupling $`g(T)`$ in an $`SU(N)`$ Yang-Mills theory becomes small at high temperature. However, a perturbative calculation of $`SU(N)`$ thermodynamic properties faces two obstacles. $`g(T)`$ only runs logarithmically, so that subleading orders of the perturbative expansion contribute significantly as the temperature is lowered towards $`T_c`$, the confinement/deconfinement transition temperature. And infrared divergences prevent an analytic, perturbative treatment altogether beyond some order. Heroic efforts have been devoted to the calculation of the pressure to this maximum order $`g^6`$ . The expansion converges poorly. One may fear that this is a general feature, and that an accurate perturbative calculation of any thermodynamic property is only possible at astronomically high temperatures. The purpose of this paper is to show a counter-example to this pessimistic view: the spatial โt Hooft loop.
The tension of a spatial โt Hooft loop has been calculated in perturbation theory to the first three non-trivial orders, and the expansion appears to converge fast . We show here that this perturbative calculation agrees with non-perturbative Monte Carlo measurements in the $`SU(2)`$ theory to a high precision, down to temperatures $`๐ช(10)T_c`$.
The โt Hooft loop has not received the same attention as the Wilson loop in numerical simulations of $`SU(N)`$ lattice gauge theories. There are several causes for this relative neglect.
$``$ First, as โt Hooft has shown , area law or perimeter law for Wilson and โt Hooft loops together is forbidden, in the absence of massless modes. As a result, in $`SU(N)`$ Yang-Mills theory, an area law is only observed for the spatial โt Hooft loop (dual to the temporal Wilson loop) above the deconfinement temperature $`T_c`$. The associated dual string tension $`\stackrel{~}{\sigma }`$ serves as an order parameter for deconfinement . The deconfined phase has traditionally received less attention than the confined phase, although the situation has been changing with the experimental search for the quark-gluon plasma.
$``$ Second, the intuitive, โphysicalโ meaning of a spatial โt Hooft loop has not been widely appreciated, and it is often considered as an โexoticโ observable of marginal interest. The fact is that a โt Hooft loop enforces a $`Z_N`$ interface in the Euclidean system, and that the dual string tension $`\stackrel{~}{\sigma }`$ is nothing else but the interface tension (up to a conventional factor $`T`$).
This is easiest to see on the lattice. Starting from an Euclidean lattice of size $`L_x\times L_y\times L_z\times L_t`$, with lattice spacing $`a`$ ($`L_i=N_ia`$) and periodic boundary conditions in all directions, let us construct a rectangular โt Hooft loop $`\stackrel{~}{W}(\stackrel{~}{\mathrm{\Sigma }})`$ in the $`(x,y)`$ plane. The contour $`\stackrel{~}{\mathrm{\Sigma }}`$, which we take rectangular for simplicity, is the boundary of a surface $`\stackrel{~}{\mathrm{\Sigma }}`$, both on the dual lattice. Let us take for $`\stackrel{~}{\mathrm{\Sigma }}`$ the minimal, planar surface, with coordinates $`(z_0,t_0)`$. Each plaquette of $`\stackrel{~}{\mathrm{\Sigma }}`$ is dual to a $`(z,t)`$ plaquette. These $`(z,t)`$ plaquettes with fixed coordinates $`(z_0,t_0)`$ form a โstackโ $`๐ซ(\stackrel{~}{\mathrm{\Sigma }})`$. The โt Hooft loop expectation value is then
$$\stackrel{~}{W}(\stackrel{~}{\mathrm{\Sigma }})\frac{๐U\mathrm{exp}(\beta _{U_P๐ซ(\stackrel{~}{\mathrm{\Sigma }})}(1\frac{1}{N}\mathrm{ReTr}\zeta U_P)\beta _{U_P๐ซ(\stackrel{~}{\mathrm{\Sigma }})}(1\frac{1}{N}\mathrm{ReTr}U_P))}{๐U\mathrm{exp}(\beta _{U_P}(1\frac{1}{N}\mathrm{ReTr}U_P))}$$
(1.1)
where $`\zeta Z_N`$ (for $`SU(2)`$, it amounts to flipping the coupling $`\beta \beta `$). Another choice of surface $`\stackrel{~}{\mathrm{\Sigma }}`$ leaves $`\stackrel{~}{W}(\stackrel{~}{\mathrm{\Sigma }})`$ invariant, as can be shown by a change of variables in (1.1): only the contour $`\stackrel{~}{\mathrm{\Sigma }}`$ matters. The dual string tension is then defined by
$$\stackrel{~}{\sigma }\underset{A(\stackrel{~}{\mathrm{\Sigma }})\mathrm{}}{lim}\frac{1}{A}\mathrm{log}\stackrel{~}{W}(\stackrel{~}{\mathrm{\Sigma }})$$
(1.2)
where $`A`$ is the minimal area bounded by $`\stackrel{~}{\mathrm{\Sigma }}`$. We can then consider a โt Hooft loop of maximum size $`L_x\times L_y`$, equal to that of the lattice in the $`x`$ and $`y`$ directions. In that case, the stack $`๐ซ(\stackrel{~}{\mathrm{\Sigma }})`$ contains one plaquette in every $`(z,t)`$ plane. Following Fig. 1 for each $`(z,t)`$ plane, we can move the special plaquette to a corner, then absorb the center element $`\zeta `$ into the boundary link $`U_t`$ <sup>1</sup><sup>1</sup>1The center element $`\zeta `$ could be absorbed in $`U_z`$ instead, leading to the same free energy for a system with twisted boundary conditions for the $`z`$-like โPolyakovโ loop.. As a result, the time-like links satisfy $`U_t(x,y,z+L_z,t=1)=\zeta U_t(x,y,z,t=1)`$. The same happens for the Polyakov loop $`P(x,y,z)_tU_t(x,y,z,t)`$ (the product of time-like links at position $`(x,y,z)`$), as
$$P(x,y,z+L_z)=\zeta P(x,y,z).$$
(1.3)
Therefore, a โt Hooft loop of maximal size in the $`(x,y)`$ plane is equivalent to twisted boundary conditions for the Polyakov loop, and
$$\stackrel{~}{W}(L_x,L_y)=\frac{Z_{tbc}}{Z_{pbc}}$$
(1.4)
where the numerator and denominator are the partition functions of a system with ordinary action, but with boundary conditions respectively twisted (by $`\zeta `$) and periodic in the $`z`$-direction for the Polyakov loop. Taking the logarithm on both sides, dividing by $`L_x\times L_y`$, and taking the thermodynamic limit, one recovers, on the left-hand side, the dual string tension $`\stackrel{~}{\sigma }`$, and on the right-hand side, the interface free energy per unit 2-dimensional area, i.e. the reduced interface tension, conventionally written as $`\sigma /T`$ in a rather confusing notation. The dual string tension is identical to the reduced interface tension. Therefore, all the old studies of $`SU(N)`$ interface tensions, both numerical and perturbative , can be re-labeled as โt Hooft loop studies. It also becomes intuitively clear why $`\stackrel{~}{\sigma }`$ vanishes below $`T_c`$: the Polyakov loop becomes disordered as the $`Z_N`$ center symmetry is restored, and the interface tension, i.e. $`\stackrel{~}{\sigma }`$, vanishes.
$``$ Third, the numerical study of the โt Hooft loop has been considered more difficult than that of the Wilson loop. The reason is an โoverlapโ problem. In the โt Hooft loop expectation value eq.(1.4), the two partition functions $`Z_{tbc}`$ and $`Z_{pbc}`$ are physically different. Gauge configurations which dominate the integral in the numerator $`Z_{tbc}`$ contain an interface; configurations which dominate in the denominator $`Z_{pbc}`$ do not. These two ensembles have little overlap, and importance sampling with respect to the denominator fails. This technical problem can be approached in various ways, all of which entail multiple simulations and increased computer cost. Decisive progress was achieved with the โsnakeโ algorithm which factorizes the ratio $`Z_{tbc}/Z_{pbc}`$ into $`N_x\times N_y`$ factors, each of $`๐ช(1)`$, which can each be estimated by an independent Monte Carlo simulation. In each factor, the area of the interface is increased by one elementary plaquette $`a^2`$. Finally, it was realized in that a single factor converges to $`\mathrm{exp}(\stackrel{~}{\sigma }a^2)`$ in the thermodynamic limit. This brings the computing cost of $`\stackrel{~}{\sigma }`$ on a par with that of the string tension $`\sigma `$ of the Wilson loop (actually, we will see that the force $`\stackrel{~}{\sigma }`$ between two dual charges can be computed with a constant accuracy independent of their separation, in sharp contrast with the ordinary string tension). Contact with perturbation theory for $`\stackrel{~}{\sigma }(T)`$ becomes feasible at high temperature, as we will show.
One surprise comes on the way. We expect our lattice measurement of $`\stackrel{~}{\sigma }(T)`$ to be affected by discretization errors $`๐ช(a^2)`$, as for any bosonic theory, i.e.
$$\stackrel{~}{\sigma }(T,a)=\stackrel{~}{\sigma }(T,a=0)(1+c_1a^2+\mathrm{})$$
(1.5)
where $`c_1`$ starts with $`๐ช(g^2)`$. Here, $`c_1`$ starts with $`๐ช(g^0)`$, which can be traced to the non-perturbative nature of $`\stackrel{~}{\sigma }`$. In this paper, we calculate the complete correction $`(\stackrel{~}{\sigma }(T,a)/\stackrel{~}{\sigma }(T,a=0)1)`$ for $`g=0`$. It turns out to be large and quite different from $`๐ช(a^2)`$ for the lattice sizes $`N_t`$ accessible to current numerical simulations. In fact, it is not even monotonic as a function of $`N_t`$. The expected behaviour $`c/N_t^2,c1`$, is only recovered for $`N_t>8`$. These large lattice corrections obscure the analysis of lattice data, and mar the continuum extrapolations of old $`SU(3)`$ studies of the critical interface tension . Our 1-loop calculation of these lattice correction factors, shown in Table I and Fig. 4, should be of general use.
Finally, we are in a position to compare our lattice Monte Carlo results for $`SU(2)`$ with continuum perturbation theory . Remarkable agreement is seen (cf. Fig. 7), at the 2% level, down to temperatures $`๐ช(10)T_c`$, with perturbation theory at $`๐ช(g^2)`$.
Our paper is organized as follows. In Sec. II, we recall for completeness the perturbative calculation of the interface tension of Ref. , then show how this calculation is modified on the lattice, and extract the lattice correction factors Table I. In Sec. III, we compare our Monte Carlo results with the perturbative ones. Conclusions follow.
## 2 1-Loop Calculation of the Interface Tension
### 2.1 Continuum Derivation
Following , we calculate the $`Z_2`$ interface tension, which is derived from the effective action of a kink interpolating between the two $`SU(2)`$ vacua. The effective action is the sum of the classical action plus a quantum term, obtained by integrating out fluctuations at 1-loop order.
The calculation is done in euclidean space-time at temperature $`T`$. The euclidean time $`x_0`$ runs from $`0`$ to $`1/T`$. The system has spatial size $`L_x\times L_y\times L_z`$, which is taken to infinity in the end.
A Z(2) interface along the $`z`$-direction is constructed by assigning one vacuum to $`z=0`$ and the other to $`z=L_z`$, and minimizing the effective action subject to these boundary conditions. By definition, the interface tension $`\stackrel{~}{\sigma }`$ is equal to the action of this interface divided by the area of the plane $`L_xL_y`$ in the thermodynamic limit ($`\stackrel{~}{\sigma }`$ is often called the โreducedโ interface tension; one can equivalently consider the (full) interface tension $`\stackrel{~}{\sigma }_F`$, which is the interface action divided by the transverse volume $`L_xL_y/T`$. Of course, $`\stackrel{~}{\sigma }=\stackrel{~}{\sigma }_F/T`$).
We consider a gauge field $`A`$ which is non-trivial only in the $`x_0`$-direction:
$$A_\mu ^{cl}=C_\mu \tau _3;C_0=\frac{2\pi T}{g}q,C_i=0i=1,2,3$$
(2.1)
where $`\tau _3`$ is the diagonal generator of SU(2) ($`\tau _3=\frac{\sigma _3}{2}`$). Then the Polyakov loop is
$$P(\text{x})=\frac{1}{2}Tr\left[๐ซ\mathrm{exp}\left(ig_0^{\frac{1}{T}}๐x_0A_0^{cl}(x)\right)\right]=\mathrm{cos}(\pi q)$$
(2.2)
The trivial vacuum is at $`A_\mu ^{cl}=q=0`$ with $`P(\text{x})=1`$. The $`Z(2)`$ transform of the trivial vacuum occurs for $`q=1`$, with $`P(\text{x})=\mathrm{exp}(i\pi )=1`$. We can therefore introduce an interface by choosing $`q`$ in (2.1) to be a function of $`z`$ and fixing $`q(0)=0`$ and $`q(L_z)=1`$. Then $`P(\text{x})=1`$ at $`z=0`$ and $`P(\text{x})=1`$ at $`z=L`$.
We need to justify why the path of minimal action between the two vacua can be brought to the form (2.1), with $`q=q(z)`$. That is because a constant background field can be brought to the diagonal, Cartan subalgebra by a global gauge rotation. But for SU(2), there is only one diagonal generator.
The *classical action* is
$$S^{cl}(A)=_0^{\frac{1}{T}}๐x_0d^3x\frac{1}{2}Tr\left[G_{\mu \nu }^2\right]$$
(2.3)
with the field-strength tensor $`G_{\mu \nu }=_\mu A_\nu _\nu A_\mu ig[A_\mu ,A_\nu ]`$. For the field in (2.1) the classical action is
$$S^{cl}=\frac{2}{g^2}TL_xL_y\pi ^2_0^{L_z}๐z\left[\frac{dq}{dz}(z)\right]^2$$
(2.4)
Minimizing the action subject to the boundary conditions $`q(0)=0,q(L_z)=1`$, yields $`q(z)=z/L_z`$. There is no true interface since the action vanishes like $`1/L_z`$ when $`L\mathrm{}`$ is taken. But this is to be expected: classically there is no energy barrier, since the action is constant for any value of $`q`$.
This degeneracy of the action with respect to $`q`$ is broken by quantum effects. To show this, we calculate the action in the presence of the (classical) background field defined in (2.1). To do this, we first assume $`q`$ to be constant in space-time. The neglect of gradient terms will be justified a posteriori.
With $`A_\mu =A_\mu ^{cl}+A_\mu ^{qu}`$ the lagrangian consists of the following Yang-Mills term plus gauge fixing term (background field gauge) and Faddeev-Popov ghost term
$``$ $`=`$ $`_{YM}+_{GF}+_{FPG}`$
$`_{YM}`$ $`=`$ $`{\displaystyle \frac{1}{2}}Tr\left[G_{\mu \nu }^2\right]`$
$`_{GF}`$ $`=`$ $`{\displaystyle \frac{1}{\xi }}Tr\left[F^2\right]`$
$`_{FPG}`$ $`=`$ $`Tr\left[\overline{\eta }P\eta \right]`$ (2.5)
where the gauge fixing term is $`F=D_\mu ^{cl}A_\mu ^{qu}`$. The operator $`P`$ in the ghost part is defined as $`P2g\frac{\delta _\mathrm{\Lambda }F}{\delta \mathrm{\Lambda }}`$, with a gauge transformation $`\mathrm{\Lambda }`$. Also we define the covariant derivative in the adjoint representation for the background field:
$$D_\mu ^{cl}=_\mu ig[A_\mu ^{cl},]$$
(2.6)
The whole lagrangian *quadratic* in the fields is:
$`_{YM}+_{GF}`$ $`=`$ $`Tr\left[A_\mu ^{qu}(x)\left(\delta _{\mu \nu }D_{cl}^2+(1{\displaystyle \frac{1}{\xi }})D_\mu ^{cl}D_\nu ^{cl}\right)A_\nu ^{qu}(y)\right]\delta ^4(xy)`$ (2.7)
$`_{FPG}`$ $`=`$ $`Tr\left[\overline{\eta }(x)\left(2D_{cl}^2\right)\eta (y)\right]\delta ^4(xy)`$ (2.8)
Using Feynman gauge ($`\xi =1`$) it follows:
$``$ $`=`$ $`_{YM}+_{GF}+_{FPG}`$ (2.9)
$`=`$ $`{\displaystyle \frac{1}{2}}A_{\mu a}^{qu}(x)_{\mu \nu }^{ab}(x,y)A_{\nu b}^{qu}(y)+\overline{\eta }_a(x)_{ab}(x,y)\eta _b(y)`$
$`_{ab}(x,y)`$ $`=`$ $`\delta ^4(xy)\left[D_{cl}^2\right]_{ab}`$ (2.10)
$`_{ab}^{\mu \nu }(x,y)`$ $`=`$ $`\delta _{\mu \nu }\delta ^4(xy)\left[D_{cl}^2\right]_{ab}`$ (2.11)
To obtain the quantum effective action to one-loop $`S_1^{qu}`$ we have to integrate out the quantum fields $`A^{qu},\overline{\eta },\eta `$
$`e^{S_1^{qu}}`$ $`=`$ $`{\displaystyle ๐A^{qu}๐\overline{\eta }๐\eta e^S}`$
$`=`$ $`{\displaystyle }๐A^{qu}๐\overline{\eta }๐\eta \mathrm{exp}[{\displaystyle \frac{1}{2}}{\displaystyle }d^4xd^4yA_{\mu a}^{qu}(x)_{\mu \nu }^{ab}(x,y)A_{\nu b}^{qu}(y)`$
$`{\displaystyle }d^4xd^4y\overline{\eta }_a(x)_{ab}(x,y)\eta _b(y)]`$
$`S_1^{qu}`$ $`=`$ $`{\displaystyle \frac{1}{2}}Tr\left[\mathrm{log}\left(\right)\right]Tr\left[\mathrm{log}\left(\right)\right]`$ (2.12)
To calculate such traces it is helpful to diagonalize the โmatricesโ $``$ and $``$ by passing to momentum space. We then need the eigenvalues of the matrix
$$\stackrel{~}{D}_{cl}^2(k)=\left(\begin{array}{ccc}๐ค^2+k_0^2+(2\pi Tq)^2& 2i(2\pi Tq)k_0& 0\\ 2i(2\pi Tq)k_0& ๐ค^2+k_0^2+(2\pi Tq)^2& 0\\ 0& 0& ๐ค^2+k_0^2\end{array}\right)$$
(2.13)
At finite temperature, $`k_0=2\pi Tn`$ with $`n`$ integer, and, using the definition
$$k_\pm 2\pi T(n\pm q)$$
(2.14)
it follows
$$\begin{array}{ccccc}\lambda _1\hfill & =\hfill & ๐ค^2+[2\pi T(n+q)]^2\hfill & =\hfill & ๐ค^2+k_+^2\hfill \\ \lambda _2\hfill & =\hfill & ๐ค^2+[2\pi T(nq)]^2\hfill & =\hfill & ๐ค^2+k_{}^2\hfill \\ \lambda _3\hfill & =\hfill & ๐ค^2+k_0^2\hfill & & (\mathrm{independent}\mathrm{of}\mathrm{q})\hfill \end{array}$$
(2.15)
This shift in momenta is a typical effect of a constant background field.
The sum over $`k_+`$ is equal to the sum over $`k_{}`$, therefore the quantum action reads, after substitution of (2.10) and (2.11) into (2.12):
$$S_1^{qu}=V\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\frac{d^3๐ค}{(2\pi )^3}2\mathrm{log}(๐ค^2+k_+^2)$$
(2.16)
where terms independent of $`q`$ (eigenvalue $`\lambda _3`$) have been dropped.
Now we need to perform the sums. To do this, we first differentiate with respect to $`q`$.
$$\frac{S_1^{qu}}{q}=\frac{TV}{\pi ^2}\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}d^3๐ค\frac{k_+}{๐ค^2+k_+^2}=8T^3V\pi ^2\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}(n+q)\left|n+q\right|$$
(2.17)
The divergent sum is interpreted using zeta-function regularization, with $`\zeta (s,q)_{n=0}^+\mathrm{}(n+q)^s`$. Noting that $`_{n=\mathrm{}}^+\mathrm{}(n+q)\left|n+q\right|=\zeta (2,q)\zeta (2,1q)`$, and using the identity $`\zeta (s,q)=\frac{B_{s+1}(q)}{s1}`$ for $`s0`$, where $`B`$ is the Bernouilli polynomial, one gets $`\frac{dS_1^{qu}}{dq}=\frac{4}{3}T^3V\pi ^2\frac{d}{dq}\left[q^2(1q)^2\right]`$. Integrating back with respect to $`q`$, one obtains:
$$S_1^{qu}=\frac{4}{3}T^3V\pi ^2\left[q^2(1q)^2\right]$$
(2.18)
So far we treated $`q`$ as if constant, but actually we want it to be a function $`q=q(z)`$ of $`z`$. The reason why we can still make use of the action $`S_1^{qu}`$ for constant $`q`$ is, that $`A^{cl}`$ varies slowly with $`z`$, and at leading order this variation can be neglected (gradient expansion), as we will see in the final kink solution (2.21). Now we reintroduce the $`z`$-dependence of $`qq(z)`$ and replace the size of the system in the $`z`$-direction $`L_z`$ ($`V=L_xL_yL_z`$) by an integral $`L_z_0^{L_z}๐z`$.
$$S_1^{qu}=\frac{4}{3}T^3L_xL_y\pi ^2_0^{L_z}๐z\left[q(z)^2(1q(z))^2\right]$$
(2.19)
Now the classical degeneracy in $`q`$ has been lifted by quantum effects and the minima (vacua) are at $`q=0`$ and $`q=1`$ (see Figure 2, left). Since the sum (2.17) is periodic in $`q`$ and hence invariant under shifts $`qq+k`$ for any integer $`k`$, this is also true for the quantum action. Thus $`q`$ in $`S_1^{qu}`$ (2.19) is defined modulo one.
From (2.4) the classical action is
$$S^{cl}=\frac{2}{g^2}TL_xL_y\pi ^2_0^{L_z}๐z\left[\frac{dq}{dz}(z)\right]^2$$
Now introduce the dimensionless coordinate $`z^{}=cz`$ with $`c=\sqrt{\frac{2}{3}}gT`$ in both $`S^{cl}`$ and $`S_1^{qu}`$. The complete effective kink action then becomes:
$`S^{kink}=S^{cl}+S_1^{qu}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{3}}}{\displaystyle \frac{2}{g}}T^2L_xL_y\pi ^2{\displaystyle _0^{L_z^{}}}๐z^{}\left[\left({\displaystyle \frac{dq}{dz^{}}}\right)^2+q^2(1q)^2\right]`$ (2.20)
Defining the Lagrangian $`L=\left(\frac{dq}{dz^{}}\right)^2+V(q)`$, $`V(q)q^2(1q)^2`$, the Euler-Lagrange equation of motion gives $`\frac{dq}{dz^{}}=\sqrt{V(q)}`$. The solution which satisfies the boundary conditions $`q(\mathrm{})=0`$ and $`q(+\mathrm{})=1`$ is
$$q(z=z^{}/c)=\frac{e^z^{}}{1+e^z^{}}$$
(2.21)
This solution is called a *kink* (see Figure 2, right). The width of the kink is $`c^1\frac{1}{gT}`$, which is much larger than the scale $`\frac{1}{T}`$ of our system since $`g1`$. Hence it was indeed justified to neglect gradient terms $`\frac{dq}{dz}`$: they only contribute at the next order in $`g^2`$.
Thus, we finally have for the effective action of the kink between the two vacua
$$S^{kink}=\frac{4\pi ^2}{3\sqrt{6}}\frac{T^2}{g}L_xL_y$$
(2.22)
Our interface tension $`\stackrel{~}{\sigma }`$ is the kink action per unit planar area, i.e. $`S^{kink}/L_xL_y`$:
$$\overline{)\stackrel{~}{\sigma }(T)=\frac{4\pi ^2}{3\sqrt{6}}\frac{T^2}{g}}$$
(2.23)
This formula superficially differs from the corresponding one in by a factor $`T`$, but this is due to our different definition of the interface tension ($`S^{kink}=\frac{L_xL_y}{T}\stackrel{~}{\sigma }_F`$ in ).
### 2.2 Corrections on the Lattice
When measuring the interface tension on the lattice, one may expect that the measurements of $`\frac{\stackrel{~}{\sigma }}{T^2}(N_t)`$ ($`N_t`$ number of temporal sites) approach the perturbative continuum form (2.23) for $`g0`$. Somewhat surprisingly, this is not the case. Multiplicative correction factors are needed, which are large for all $`N_t`$ accessible to numerical simulations. Only for very large $`N_t`$ do these factors go to $`1`$. In , Nathan Weiss gave the main expression leading to these correction factors, but without giving details about the calculation. This is done here for completeness. Moreover, we show how well these results describe the measurements we obtain from Monte Carlo simulations.
Our derivation on the lattice is very similar to the one in the continuum and in the end, we compare the obtained effective potentials from both calculations. Actually the only thing to do is to find a lattice expression for the inverse propagator $`D_{cl}^2`$, which took the form (2.13) in the continuum.
A four-dimensional lattice of size $`aN_s`$ in the spatial directions and $`aN_t`$ in the temporal direction is used.
We calculate the classical action first. The Wilson action is
$$S_{cl}^L=\beta \underset{P}{}(1\frac{1}{2}\mathrm{ReTr}U_P)$$
(2.24)
Only plaquettes in the $`zt`$-plane are different from $`\mathrm{๐}`$. Such a plaquette at position $`z`$ takes the value
$$U_P=e^{ia\frac{\pi }{N_t}\frac{1}{a}(q(z+a)q(z))\sigma _3}=e^{ia\frac{\pi }{N_t}\mathrm{\Delta }_z^fq(z)\sigma _3}$$
(2.25)
with the forward derivative $`\mathrm{\Delta }_z^f`$. The trace of the plaquette is
$`2\mathrm{cos}\left(a{\displaystyle \frac{\pi }{N_t}}\mathrm{\Delta }_z^fq(z)\right)`$ $``$ $`2\left(1{\displaystyle \frac{1}{2}}\left(a{\displaystyle \frac{\pi }{N_t}}\mathrm{\Delta }_z^fq(z)\right)^2\right)`$ (2.26)
Placing this into the action we have
$$S_{cl}^L=\beta \underset{P_{zt}}{}\frac{1}{2}\left(a\frac{\pi }{N_t}\mathrm{\Delta }_z^fq(z)\right)^2=\frac{2}{g^2}\frac{N_s^2}{N_t}\pi ^2a^2\underset{z}{}\left(\mathrm{\Delta }_z^fq(z)\right)^2$$
(2.27)
where $`_{P_{zt}}`$ means sum over plaquettes in the $`zt`$-plane and $`\beta =\frac{4}{g^2}`$.
Mimicking the steps from the continuum derivation, we add quantum fluctuations to this action. The calculation on the lattice yields for the momentum space propagator (2.13) ($`k\frac{k^{}}{a}`$ where $`k^{}`$ is now a dimensionless quantity)
$$\left[\stackrel{~}{D}_{cl}^{2L}\right]=\left(\begin{array}{ccc}_\mu 4\mathrm{sin}^2\left(\frac{k_\mu ^{}}{2}\right)+4\mathrm{sin}^2\left(\frac{\pi q}{N_t}\right)\mathrm{cos}(k_0^{})& 2i\mathrm{sin}\left(\frac{2\pi q}{N_t}\right)\mathrm{sin}(k_0^{})& 0\\ 2i\mathrm{sin}\left(\frac{2\pi q}{N_t}\right)\mathrm{sin}(k_0^{})& _\mu 4\mathrm{sin}^2\left(\frac{k_\mu ^{}}{2}\right)+4\mathrm{sin}^2\left(\frac{\pi q}{N_t}\right)\mathrm{cos}(k_0^{})& 0\\ 0& 0& _\mu 4\mathrm{sin}\left(\frac{k_\mu ^{}}{2}\right)\end{array}\right)$$
The eigenvalues are
$`\lambda _1`$ $`=`$ $`{\displaystyle \frac{4}{a^2}}\left({\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{sin}^2\left({\displaystyle \frac{k_i}{2}}\right)+\mathrm{sin}^2\left(\pi {\displaystyle \frac{(n+q)}{N_t}}\right)\right)`$
$`\lambda _2`$ $`=`$ $`{\displaystyle \frac{4}{a^2}}\left({\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{sin}^2\left({\displaystyle \frac{k_i}{2}}\right)+\mathrm{sin}^2\left(\pi {\displaystyle \frac{(nq)}{N_t}}\right)\right)`$
$`\lambda _3`$ $`=`$ $`{\displaystyle \frac{4}{a^2}}\left({\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{sin}^2\left({\displaystyle \frac{k_i}{2}}\right)+\mathrm{sin}^2\left(\pi {\displaystyle \frac{n}{N_t}}\right)\right)`$ (2.28)
which can be directly obtained from (2.15) by replacing momenta with their lattice counterparts $`k\widehat{k}=\frac{2}{a}\mathrm{sin}\left(\frac{a}{2}k\right)`$.
The lattice quantum action is (dropping again the eigenvalue $`\lambda _3`$ since it is independent of $`q`$):
$`S_1^{quL}=`$ $`V`$ $`{\displaystyle \underset{n=0}{\overset{N_t1}{}}}{\displaystyle _0^{\frac{2\pi }{a}}}{\displaystyle \frac{d^3๐ค^{}}{(2\pi )^3}}\{\mathrm{log}[{\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{sin}^2\left({\displaystyle \frac{k_i^{}}{2}}\right)+\mathrm{sin}^2\left(\pi {\displaystyle \frac{(n+q)}{N_t}}\right)]`$ (2.29)
$`+`$ $`\mathrm{log}[{\displaystyle \underset{i=1}{\overset{3}{}}}\mathrm{sin}^2\left({\displaystyle \frac{k_i^{}}{2}}\right)+\mathrm{sin}^2\left(\pi {\displaystyle \frac{(nq)}{N_t}}\right)]\}`$
After performing the sum, the lattice quantum action is
$`S_1^{quL}`$ $`=`$ $`2N_s^3{\displaystyle _0^{2\pi }}{\displaystyle \frac{d^3๐ค^{}}{(2\pi )^3}}\mathrm{log}[12\mathrm{cos}(2\pi q)e^{2N_th}+e^{4N_th}]`$ (2.30)
$`h`$ $`=`$ $`\mathrm{log}(๐+\sqrt{๐^2+1})`$ (2.31)
$`๐^2`$ $`=`$ $`\mathrm{sin}^2\left({\displaystyle \frac{k_x^{}}{2}}\right)+\mathrm{sin}^2\left({\displaystyle \frac{k_y^{}}{2}}\right)+\mathrm{sin}^2\left({\displaystyle \frac{k_z^{}}{2}}\right)`$ (2.32)
which is the same as stated by Weiss in .
We want to compare this to the continuum expression to make a quantitative prediction for corrections to the interface tension measured on the lattice. From (2.18) we have in the continuum ($`V=L_xL_yL_z`$)
$$S_1^{qu}=\frac{4}{3}VT^3\pi ^2\left[q^2(1q)^2\right]$$
(2.33)
Defining
$`V_L(q,N_t)`$ $`=`$ $`\left({\displaystyle \frac{3}{2\pi ^2}}N_t^3\right){\displaystyle _0^{2\pi }}{\displaystyle \frac{d^3๐ค^{}}{(2\pi )^3}}\mathrm{log}[12\mathrm{cos}(2\pi q)e^{2N_th}+e^{4N_th}]`$
and subtracting $`V_L(q=0)`$ from the action in order to make it zero when the system is in one of the vacua we get
$`S_1^{quL}`$ $`=`$ $`{\displaystyle \frac{4}{3}}VT^3\pi ^2(V_L(q,N_t)V_L(0,N_t))`$ (2.35)
Comparing this to (2.33), we see that $`V(q)=q^2(1q)^2`$ is replaced by $`V_L(q,N_t)V_L(0,N_t)`$ on the lattice. Both are shown in Figure 3.
Taking the continuum limit $`N_t\mathrm{}`$ we have $`h\frac{|k|}{2}`$ and therefore
$`V_L(q)`$ $`=`$ $`\left({\displaystyle \frac{3}{2\pi ^2}}N_t^3\right){\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2dk}{(2\pi )^3}}\mathrm{log}[12\mathrm{cos}(2\pi q)e^{N_tk}+e^{2N_tk}]`$
$`=`$ $`\left({\displaystyle \frac{3}{4\pi ^4}}\right){\displaystyle _0^{\mathrm{}}}x^2๐x\mathrm{log}[12\mathrm{cos}(2\pi q)e^x+e^{2x}]`$
which can be shown to be identical to $`q^2(1q)^2+\mathrm{constant}`$, by differentiating with respect to $`q`$.
This gives for the lattice correction factors $`C_{lat}(N_t)\stackrel{~}{\sigma }(N_t)/\stackrel{~}{\sigma }(N_t=\mathrm{})`$:
$$\overline{)C_{lat}(N_t)=6_0^1๐q\sqrt{V_L(q,N_t)V_L(0,N_t)}}$$
(2.36)
A numerical evaluation yields the factors in Table 1.
As is to be expected in a bosonic theory, the corrections are of order $`O(a^2)`$, i.e. the factors are of the form $`\left(1+O(\frac{1}{N_t^2})\right)`$. This is shown in Figure 4. However, this asymptotic behaviour sets in only for large values of $`N_t`$. For small values of $`N_t`$, the ones accessible to numerical simulations, the correction factors show a quite different, not even monotonic ($`N_t=1`$), behaviour. This knowledge is crucial when trying to extrapolate to the continuum limit with only small-$`N_t`$ Monte Carlo data.
To leading order the lattice correction factors $`C_{lat}`$ do not depend on the number of colors $`N`$, which only occurs as a prefactor in the 1-loop quantum action and cancels out when comparing the continuum to the lattice effective potential. Thus Table I applies to $`SU(N)`$ as well.
There is one more thing to say, concerning finite size effects. So far, we chose the spatial dimensions $`N_s`$ to be $`\mathrm{}`$. But we can also calculate the impact of a finite size in the spatial dimensions on the lattice correction factors. Let us consider a cubic geometry for simplicity. For finite $`N_s`$ the integral over the momenta $`๐ค`$ becomes a sum and we obtain
$`V_L(q,N_t,N_s)`$ $`=`$ $`{\displaystyle \frac{3}{2\pi ^2}}\left({\displaystyle \frac{N_t}{N_s}}\right)^3{\displaystyle \underset{n_x,n_y,n_z=0}{\overset{N_s1}{}}}\mathrm{log}[12\mathrm{cos}(2\pi q)e^{2N_th}+e^{4N_th}]`$
$`๐^2`$ $`=`$ $`\mathrm{sin}^2\left({\displaystyle \frac{\pi }{N_s}}n_x\right)+\mathrm{sin}^2\left({\displaystyle \frac{\pi }{N_s}}n_y\right)+\mathrm{sin}^2\left({\displaystyle \frac{\pi }{N_s}}n_z\right)`$ (2.37)
One finds for example at $`N_t=2`$ and aspect ratio $`\rho =2`$ ($`\rho \frac{N_s}{N_t}`$) that $`C_{lat}`$ in Table 1 should be multiplied by 0.885, for $`\rho =4`$: 0.966 and for $`\rho =8`$: 0.992. This shows, that with increasing aspect ratio, the impact of a finite $`N_s`$ becomes quickly negligible. If finite size effects are taken into account, $`C_{lat}`$ decreases slightly, and therefore the corrected interface tension $`\stackrel{~}{\sigma }(T)/C_{lat}`$ increases by a tiny amount.
## 3 Comparing with Simulations
### 3.1 Measuring the Interface Tension
A direct Monte Carlo measurement of $`\stackrel{~}{\sigma }`$ from the free energy definition (1.4,1.2) is impractical. First, there is an overlap problem: configurations contributing to the numerator and the denominator of (1.4) are physically, macroscopically different, since they respectively contain an interface or donโt. Thus, importance sampling with respect to the denominator fails for all but the smallest volumes. Second, the interface is translationally invariant, so that the corresponding entropy must be carefully subtracted to avoid $`\mathrm{log}L`$ corrections. Both problems can be solved by a clever choice of algorithm. Historically, this choice has been arrived at in two steps.
An elegant approach to the overlap problem is provided by the โsnakeโ algorithm . It consists of building the interface in $`N_x\times N_y`$ steps, one plaquette at a time, and factorizing the ratio eq.(1.4):
$$\frac{Z_{tbc}}{Z_{pbc}}=\frac{Z_N}{Z_{N1}}\frac{Z_{N1}}{Z_{N2}}\mathrm{}\frac{Z_1}{Z_0}$$
(3.1)
where $`N=N_x\times N_y`$, and $`Z_k`$, $`k=0,\mathrm{},N`$ ($`Z_NZ_{tbc}`$ and $`Z_0Z_{pbc}`$) is the partition function for a โt Hooft loop of area $`ka^2`$, with a stack $`๐ซ_k`$ of $`k`$ corresponding plaquettes โtwistedโ, i.e. having the coupling $`\beta `$ changed to $`\beta `$:
$$Z_k=๐U\mathrm{exp}(+\beta \underset{๐ซ_k}{}(1\frac{1}{2}\mathrm{ReTr}U_P)\beta \underset{\overline{๐ซ_k}}{}(1\frac{1}{2}\mathrm{ReTr}U_P))$$
(3.2)
where $`\overline{๐ซ}_k`$ is the complement of $`๐ซ_k`$. When $`mod(k,N_y)=0`$, the โt Hooft loop is rectangular with perimeter $`2N_y`$, otherwise it has perimeter $`(2N_y+2)`$, and one side contains 2 kinks (the bold $`x`$-links in Fig. 5). The progressive flipping of the plaquettes is described by a snake-like motion, hence the name attached to the algorithm . Each ratio in (3.1) is of $`๐ช(1)`$ and can be efficiently estimated by a separate Monte Carlo simulation.
In fact, this algorithm can be simplified , and the $`N_x\times N_y`$ Monte Carlo simulations reduced to a single one, by the following observation. $`\frac{Z_{k+1}}{Z_k}=\frac{Z_{k+1}/Z_0}{Z_k/Z_0}`$ is a ratio of two โt Hooft loops differing by $`a^2`$ in area. If $`mod(k,N_y)\{0,1\}`$, they have the same perimeter and number of corners. They form the dual analogue of a Creutz ratio, commonly used to measure the ordinary string tension. Therefore,
$$\underset{N_x,N_y\mathrm{}}{lim}\mathrm{log}\frac{Z_{k+1}}{Z_k}=\stackrel{~}{\sigma }a^2.$$
(3.3)
A crucial advantage over the Creutz ratio is that the integrand in $`Z_k`$ is of exponential form: it is positive and can be used as a sampling probability. Then, a given statistical precision on $`Z_{k+1}/Z_k`$ requires the same computer effort, independently of the โt Hooft loop size. As stressed in , this is in sharp contrast to the ordinary string tension, where the statistical accuracy degrades exponentially with the Wilson loop area. Therefore, (3.3) can be rewritten and measured as an expectation value with respect to $`Z_k`$:
$$\stackrel{~}{\sigma }a^2=\underset{N_x,N_y\mathrm{}}{lim}\mathrm{log}\mathrm{exp}(2\beta \frac{1}{2}\mathrm{ReTr}U_{P_{k+1}})$$
(3.4)
In our simulations, we use the equivalent $`\mathrm{exp}(\beta \frac{1}{2}\mathrm{ReTr}U_{P_{k+1}})_0/\mathrm{exp}(+\beta \frac{1}{2}\mathrm{ReTr}U_{P_{k+1}})_0`$, where the expectation value $`.._0`$ is taken in the ensemble which interpolates between $`Z_k`$ and $`Z_{k+1}`$, where the coupling attached to $`U_{P_{k+1}}`$ is set to zero. This observable has a smaller variance, which can be further reduced by performing multi-hit (analytically) over the 4 links making up $`U_{P_{k+1}}`$, since they are now decoupled from each other.
Note that translation invariance is broken. Our โt Hooft loops $`Z_k`$ and $`Z_{k+1}`$ have fixed boundaries, and the observable is a function of a single plaquette in the system. This solves the second problem mentioned at the beginning: there is no entropy to subtract from the free energy of a partial interface.<sup>2</sup><sup>2</sup>2In the original โsnakeโ algorithm, translation invariance is restored in the last ratio $`Z_N/Z_{N1}`$. Its Monte Carlo evaluation is plagued by very long autocorrelation times, corresponding to translations of the full interface. It also suggests to organize the Monte Carlo updates hierarchically in shells centered around $`P_{k+1}`$ . The gauge links of the innermost shells, upon which the observable depends most sensitively, are integrated more thoroughly with more frequent Monte Carlo updates.
Since taking a large โt Hooft loop in (3.4) requires no more work than taking a small one, we choose the size as large as practical, as per Fig. 5, with length $`N_y`$ and width $`rN_x/2`$. $`r`$ can be increased even more, but taking it too close to $`N_x`$ brings about the possibility of reducing the total free energy by trading the single interface of width $`r`$ for two of them: one interface of width $`(N_xr)`$, and one full interface of width $`N_x`$. Translation invariance of the latter causes the free energy reduction. The choice $`rN_x/2`$ completely eliminates this finite-size effect.
For large โt Hooft loops of size $`N_y,r\xi =1/\sqrt{\stackrel{~}{\sigma }a^2}`$, the leading deviation of (3.4) from its asymptotic value comes from two contributions: Gaussian fluctuations of the interface away from its minimal area (known as capillary waves, or Lรผscher correction), and interaction between the two kinks in the โt Hooft loop perimeter Fig. 5<sup>3</sup><sup>3</sup>3Excited states of the interface also bring a correction. Assuming an energy gap $`\pi /r`$ between the ground- and first excited state propagating along the $`y`$ direction, one can check that this correction is smaller than our statistical errors, for all lattice sizes and couplings considered here.. $`(i)`$ The Lรผscher correction can be evaluated analytically. For a vibrating $`L_y\times r`$ interface with $`\xi rL_y`$, it takes the familiar form $`\stackrel{~}{\sigma }\stackrel{~}{\sigma }+\frac{\pi }{12}\frac{1}{r^2}`$. In our case where $`r\frac{1}{2}L_y`$, it is obtained from the partition function appropriate to our boundary conditions
$$Z_{\mathrm{surface}}=\eta ^2\left[i\frac{L_y}{2r}\right],\eta (\tau )=e^{i\frac{\pi }{12}\tau }\underset{n=1}{\overset{\mathrm{}}{}}(1e^{i2\pi n\tau }),$$
(3.5)
We subtract this analytic correction from our Monte Carlo results. $`(ii)`$ A kink of size $`a`$ in the โt Hooft loop perimeter creates UV excited string states of energy $`๐ช(a^1)`$, which propagate along the $`y`$ direction over a distance $`saN_y/2`$. Therefore, we expect a correction of order $`\mathrm{exp}(N_y/2)`$ due to the two kinks. Indeed, Monte Carlo data obtained on lattices of increasing size $`L^3\times 2`$ at fixed $`\beta `$ are well described by the ansatz $`(\stackrel{~}{\sigma }a^2+`$ Lรผscher correction $`+c_1\mathrm{exp}(c_2N_y/2))`$ with $`c_1,c_21`$ . This finite-size correction is negligible compared to our statistical errors provided $`N_y2\mathrm{log}(\stackrel{~}{\sigma }a^2)`$. Numerical tests at several values of $`\beta `$ and several lattice extents $`N_t`$ indicated that finite-size effects cause an underestimate of the true interface tension, and led us to aspect ratios $`\{N_x,N_y,N_z\}/N_t`$ as large as 16.
To increase statistics, we typically estimated 2 ratios $`Z_{k+1}/Z_k`$, corresponding to plaquettes $`U_{P_{k+1}}`$ at the โcenterโ of the lattice (i.e. with coordinates $`(\frac{N_x}{2},\frac{N_y}{2}+\{0,1\})`$). For each run, an average of 20k multi-hit, multi-shell measurements were taken.
### 3.2 Results versus 2-Loop Perturbation Theory
In Section 2, we derived the interface tension at 1-loop, in the continuum and on the lattice. The continuum 2-loop result has been obtained in and in , where the running coupling $`\stackrel{~}{g}(T)`$ is chosen such that the 1-loop effects disappear in the renormalization of the coupling in the dimensionally reduced theory. For SU(N):
$$\stackrel{~}{\sigma }(T)=\frac{4\pi ^2(N1)}{3\sqrt{3N}}\frac{T^2}{\stackrel{~}{g}(T)}(1(15.27853..\frac{11}{3}(\gamma _E+\frac{1}{22}))\frac{\stackrel{~}{g}^2(T)N}{(4\pi )^2})$$
(3.6)
with Eulerโs constant $`\gamma _E=0.577215\mathrm{}`$. The coupling $`\stackrel{~}{g}(T)`$ is given in as
$$\stackrel{~}{g}^2(T)g_{\overline{MS}}^2(\mu )|_{\mu =4\pi Te^{(\gamma _E+1/22)}}$$
(3.7)
which defines the $`\stackrel{~}{\mathrm{\Lambda }}`$-parameter
$$\stackrel{~}{\mathrm{\Lambda }}=\frac{e^{(\gamma _E+1/22)}}{4\pi }\mathrm{\Lambda }_{\overline{MS}}$$
(3.8)
With this result it is possible to express the interface tension in (3.6) to subleading order in terms of the lattice bare coupling $`g_{LAT}(T)`$. To do that, we use the $`SU(2)`$ relation
$$\mathrm{\Lambda }_{\overline{MS}}=19.8228\mathrm{\Lambda }_{LAT}$$
(3.9)
to obtain the 1-loop result
$$\stackrel{~}{g}^2=g_{LAT}^2(1+0.10016g^2)$$
(3.10)
which is substituted into (3.6) to get
$$\stackrel{~}{\sigma }(T)=\frac{4\pi ^2}{3\sqrt{6}}\frac{T^2}{g_{LAT}(T)}\left(10.21467g_{LAT}^2(T)\right)$$
(3.11)
But we want the lattice bare coupling to run with the scale $`a^1`$ instead of $`T=(aN_t)^1`$ because what is used in a simulation is $`\beta =2N/g_{LAT}^2(a^1)`$. This can be easily translated using the $`\beta `$-function (3.16). Also the lattice correction factors $`C_{lat}(N_t)`$, calculated in Section 2.2 and given in Table 1, need to be included. Altogether, one gets for SU(2)
$$\overline{)\frac{\stackrel{~}{\sigma }}{T^2}(\beta ,N_t)=C_{lat}(N_t)\frac{4\pi ^2}{3\sqrt{6}}\frac{\sqrt{\beta }}{2}\left(1(0.21467+\beta _0\mathrm{log}(N_t))\frac{4}{\beta }\right)}$$
(3.12)
to subleading order, where $`\beta _0`$ is given in (3.15). Note that we have neglected $`๐ช(g^2)`$ corrections in the lattice correction factors $`C_{lat}(N_t)`$. How well this (parameter free!) formula describes the measurements of the interface tension at different $`N_t`$โs for high $`\beta `$ can be seen in Figure 6. The $`N_t=1`$ comparison is shown separately because the data do not lie neatly above the other curves but cross right through.
We have not performed a similar comparison for SU(3). But analogous steps predict:
$$\frac{\stackrel{~}{\sigma }}{T^2}(\beta ,N_t)=C_{lat}(N_t)\frac{8\pi ^2}{9}\sqrt{\frac{\beta }{6}}\left(1(0.34805+\beta _0\mathrm{log}(N_t))\frac{6}{\beta }\right)$$
(3.13)
### 3.3 Results versus Temperature
Let us collect our measurements of the interface tension for different values of $`\beta `$ and $`N_t`$, and correct them with the appropriate lattice correction factor $`C_{lat}(N_t)`$, forming $`\frac{\stackrel{~}{\sigma }}{T^2}(T)_{corrected}=\frac{\stackrel{~}{\sigma }}{T^2}(\beta ,N_t)_{measured}/C_{lat}(N_t)`$. The bulk of the cutoff effects has been removed with the division by $`C_{lat}(N_t)`$, so that the left-hand side can be compared with the corresponding continuum quantity evaluated in perturbation theory, as a function of temperature. Since we lack the possibility to set a physical temperature scale (to do that, we would need the non-perturbative $`\beta `$-function obtained, e.g., from the step scaling function ), we calculate the temperature $`T=1/(a(\beta )N_t)`$, associated with each value of $`\frac{\stackrel{~}{\sigma }}{T^2}`$, using the perturbative 2-loop formula for the lattice spacing
$$a(\beta ,\mathrm{\Lambda }_{LAT})=\frac{1}{\mathrm{\Lambda }_{LAT}}\mathrm{exp}\left(\frac{\beta }{8\beta _0}\right)\left(\frac{4\beta _0}{\beta }\right)^{\beta _1/(2\beta _0^2)}$$
(3.14)
with ($`N=2`$ here):
$$\beta _0=\frac{11}{3}\frac{N}{16\pi ^2},\beta _1=\frac{34}{3}\left(\frac{N}{16\pi ^2}\right)^2$$
(3.15)
which is justified as long as $`T\mathrm{\Lambda }_{\overline{MS}}`$. We want to express the temperature in terms of the parameter $`\mathrm{\Lambda }_{\overline{MS}}`$. One obtains
$$a(\beta ,\mathrm{\Lambda }_{\overline{MS}})=\frac{19.8228}{\mathrm{\Lambda }_{\overline{MS}}}\mathrm{exp}\left(\frac{\beta }{8\beta _0}\right)\left(\frac{4\beta _0}{\beta }\right)^{\beta _1/(2\beta _0^2)}$$
(3.16)
From equation (3.6), we know the relation between $`\frac{\stackrel{~}{\sigma }}{T^2}`$ and the coupling $`\stackrel{~}{g}`$ at 2-loop order. In , Giovannangeli and Korthals Altes show that higher order corrections to equation (3.6) are very small. Thus one may hope, that the evaluation of (3.6) for $`SU(2)`$:
$$\frac{\stackrel{~}{\sigma }}{T^2}(T)_{GKA}=\frac{4\pi ^2}{3\sqrt{6}}\frac{1}{\stackrel{~}{g}(T)}\left(10.16459\stackrel{~}{g}^2(T)\right)$$
(3.17)
is sufficient to describe $`\frac{\stackrel{~}{\sigma }}{T^2}`$ over a large range of temperatures. For $`\stackrel{~}{g}^2(T)`$ we use the solution to the renormalization group equation to 2-loop:
$$\stackrel{~}{g}^2(T)=\frac{1}{\beta _0\mathrm{log}(T^2/\stackrel{~}{\mathrm{\Lambda }}^2)+(\beta _1/\beta _0)\mathrm{log}\mathrm{log}(T^2/\stackrel{~}{\mathrm{\Lambda }}^2)}$$
(3.18)
where $`\stackrel{~}{\mathrm{\Lambda }}`$ is related to $`\mathrm{\Lambda }_{\overline{MS}}`$ through equation (3.8). The perturbative approximation $`\frac{\stackrel{~}{\sigma }}{T^2}(T)_{GKA}`$ (3.17) is compared to our measurements $`\frac{\stackrel{~}{\sigma }}{T^2}(T)_{corrected}`$ in Figure 7, which shows the ratio $`\stackrel{~}{\sigma }(T)_{corrected}/\stackrel{~}{\sigma }(T)_{GKA}`$ versus $`T/\mathrm{\Lambda }_{\overline{MS}}`$. As can be seen, the deviations from $`1`$ are remarkably small and remain at the 2% level down to temperatures $`๐ช(10)T_c`$. This figure is a beautiful confirmation of the work of Giovannangeli and Korthals Altes.
## 4 Discussion
We have shown that cutoff effects on lattice measurements of the Yang-Mills interface tension (or dual string tension) are large for practical lattice sizes used in Monte Carlo simulations. They do not disappear as $`\beta \mathrm{}`$ as could be naively expected. Table I, which lists these corrections calculated at 1-loop order as a function of the temporal lattice size $`N_t`$ for any $`SU(N)`$ group, should be of practical use.
Moreover, we have shown precise agreement between our $`SU(2)`$ Monte Carlo data and 2-loop continuum perturbation theory, once the correction factors above have been folded in. The ratio of the measured over the perturbative interface tension remains 1 within $`2\%`$, from very high temperatures down to $`๐ช(10)T_c`$. Our results put the perturbative calculation of the interface tension on firm ground, and show that the perturbative expansion converges well for this quantity, in contrast to the notorious expansion of the pressure . The reason for this remarkable contrast lies presumably in the topological nature of the โt Hooft loop. It naturally encodes the $`Z_N`$ center symmetry, which is absent in the perturbative expansion of the pressure.
The small remaining discrepancy between the measured data and perturbation theory can be assigned various causes. We list them in a subjective order of decreasing importance: $`(i)`$ $`๐ช(g^3)`$ correction in $`\stackrel{~}{\sigma }`$ ; $`(ii)`$ $`๐ช(g^2)`$ corrections in the lattice correction factors $`C_{lat}`$; $`(iii)`$ 3-loop terms in the lattice $`\beta `$-function (3.14), which will change the $`x`$-axis $`T/\mathrm{\Lambda }_{\overline{MS}}`$ in Figure 7; $`(iv)`$ 3-loop terms in the running of $`\stackrel{~}{g}^2(T)`$ (3.18); $`(v)`$ remaining finite size effects in the Monte Carlo simulations.
Of course, the low-temperature drop in Figure 7 is natural. The interface tension is a (dual) order parameter for the phase transition between the cold confining and the hot deconfining phases, which for $`SU(2)`$ is second-order. The corresponding correlation length therefore diverges near the critical temperature $`T_c`$ like $`(T/T_c1)^\nu `$, with the critical exponent $`\nu `$ of the 3d Ising model as expected from universality . This singularity is not seen by the perturbative calculation. Hence, the ratio $`\stackrel{~}{\sigma }(T)_{corrected}/\stackrel{~}{\sigma }(T)_{GKA}`$ in Fig. 7 goes to zero like $`(T/T_c1)^{2\nu }`$.
It would be useful to extend our calculation of the lattice correction factors $`C_{lat}`$ to order $`g^2`$, for arbitrary $`SU(N)`$ group. At this order, one may observe a dependence on the relative $`Z_N`$ separation of the two vacua on either side of the interface. This dependence would enter in the comparison of dual $`k`$-string tensions $`(\stackrel{~}{\sigma }_k/\stackrel{~}{\sigma }_1)(T)`$ between lattice measurements and continuum perturbation theory .
Finally, we became aware of Ref. as this work was being written up. Ref. deals with the same issue of connecting lattice measurements and continuum calculations of dual tensions, for the $`3d`$ theory (see Table VIII there). Nevertheless, it seems that some of their analytic results generalize to the $`4d`$ case we studied.
## Note added
The generalization mentioned above has just been presented in Ref. . Table I there contains similar lattice correction factors as our Table I. The small numerical differences can presumably be assigned to truncation errors in the series used in Ref. . The numerical simulations in the paper focus on ratios of $`k`$-interface tensions, or rather of their derivatives, in $`SU(N),N>3`$.
## Acknowledgments
We have benefited from many useful discussions with our colleagues, including Jรผrg Frรถhlich, Pierre Giovannangeli, Chris Korthals Altes, Slavo Kratochvila, Biagio Lucini, Kari Rummukainen and Michele Vettorazzo. The results presented are part of the work done by D.N. for his diploma thesis in 2004 at ETH Zรผrich.
|
warning/0506/quant-ph0506136.html
|
ar5iv
|
text
|
# Concurrence of Arbitrary Dimensional Bipartite Quantum States
## Abstract
We derive an analytical lower bound for the concurrence of a bipartite quantum state in arbitrary dimension. A functional relation is established relating concurrence, the Peres-Horodecki criterion and the realignment criterion. We demonstrate that our bound is exact for some mixed quantum states. The significance of our method is illustrated by giving a quantitative evaluation of entanglement for many bound entangled states, some of which fail to be identified by the usual concurrence estimation method.
Entanglement is a striking feature of quantum systems and is the key physical resource to realize quantum information tasks such as quantum cryptography, quantum teleportation and quantum computation nielsen , which cannot be accounted for by classical physics. This has provided a strong motivation for the study of detection and quantification of entanglement in an operational way. Despite of a great deal of effort in past years Peres96 ; HorodeckiPLA96 ; Wootters98 ; Audenaert01 ; Rudolph02 ; ChenQIC03 ; Horodecki02 ; realignmentcriteria ; chenp02-Gerjuoy03 ; Lozinski03 ; sepcriteria ; Mintert04-MintertPhD for the moment only partial solutions are known for generic mixed states. As for quantitative measures of entanglement, there is an elegant formula for 2 qubits in terms of *concurrence*, which is derived analytically by Wootters in Ref. Wootters98 . This quantity has recently been shown to play an essential role in describing quantum phase transition in various interacting quantum many-body systems Osterloh02-Wu04 and may affect macroscopic properties of solids significantly Ghosh2003 . Furthermore, value of concurrence will provide an estimation Mintert04-MintertPhD for the entanglement of formation (EOF) BDSW , which quantifies the required minimally physical resources to prepare a quantum state. It is thus very important to have a precise quantitative picture of entanglement in order to get a better insight into the corresponding physical systems.
However, calculation of the concurrence is a formidable task as the Hilbert space dimension is increasing, like in the case of two parts in a real solid-state system considered for quantum computation. Good algorithms and progresses have been obtained concerning lower bounds for qubit-qudit system chenp02-Gerjuoy03 ; Lozinski03 and for bipartite systems in arbitrary dimension Audenaert01 ; Mintert04-MintertPhD . Considerable progress is made in Mintert04-MintertPhD to give a purely algebraic lower bound. Nevertheless, an optimized bound generally involves numerical optimization over a large number of free parameters in a level (at least $`m(m1)n(n1)/4`$ for a $`mn`$ bipartite system, where $`m,n`$ are Hilbert space dimension for two subsystems respectively Audenaert01 ; Lozinski03 ; Mintert04-MintertPhD ). This leads to a computationally untractable problem for realistic system with a higher dimension. In addition, these methods for evaluating concurrence can not detect reliably *arbitrary* entangled states even if one applies all known optimization methods Mintert04-MintertPhD .
Our aim in this work is to improve this situation dramatically by giving an analytical lower bound for concurrence of any mixed bipartite quantum state. We find an essential quantitative relation among this measure and available strong separability criteria. A functional relation is explicitly derived to give a tightly lower bound for the concurrence. It is shown to be exact for some special class of states. Our method is further demonstrated to be better than the regular method for concurrence optimization, in the sense that it can detect and give an evaluation of entanglement for many bound entangled states (BES) which cannot be identified by the latter. This also complements a number of existing methods involving numerical optimization and provides a computational method to estimate manifestly the actual value of concurrence for any bipartite quantum state.
We start with a generalized definition Rungta01-AlbeverioFei01 of concurrence for a pure state $`|\psi `$ in the tensor product $`_A_B`$ of two (finite dimensional) Hilbert spaces $`_A,_B`$ for 2 systems $`A,B`$. The concurrence is defined by $`C(|\psi )=\sqrt{2(1\text{Tr}\rho _A^2)}`$, where the reduced density matrix $`\rho _A`$ is obtained by tracing over the subsystem $`B`$. The concurrence is then extended to mixed states $`\rho `$ by the convex roof,
$$C(\rho )\underset{\{p_i,|\psi _i\}}{\mathrm{min}}\underset{i}{}p_iC(|\psi _i),$$
(1)
for all possible ensemble realizations $`\rho =_ip_i|\psi _i\psi _i|`$, where $`p_i0`$ and $`_ip_i=1`$. For any pure product state $`|\psi `$, $`C(|\psi )`$ vanishes according to the definition. Consequently, a state $`\rho `$ is *separable* if and only if $`C(\rho )=0`$ and hence can be represented as a convex combination of product states as $`\rho =_ip_i\rho _i^A\rho _i^B`$ where $`\rho _i^A`$ and $`\rho _i^B`$ are pure state density matrices of the subsystems $`A`$ and $`B`$, respectively werner89 .
The key point of our idea is to relate directly the concurrence and the Peres-Horodecki criterion of positivity under partial transpose (PPT criterion) Peres96 ; HorodeckiPLA96 and the realignment criterion Rudolph02 ; ChenQIC03 by means of Schmidt coefficients of a pure state. Let us firstly consider the concurrence for a pure state. $`C(|\psi )`$ is invariant under a local unitary transformation (LU) Wootters98 ; Rungta01-AlbeverioFei01 . Without loss of generality, we suppose that a pure $`mn`$ $`(mn)`$ quantum state has the standard Schmidt form
$$|\psi =\underset{i}{}\sqrt{\mu _i}|a_ib_i,$$
(2)
where $`\sqrt{\mu _i}`$ $`(i=1,\mathrm{}m)`$ are the Schmidt coefficients, $`|a_i`$ and $`|b_i`$ are orthonormal basis in $`_A`$ and $`_B`$, respectively. It is evident that the reduced density matrices $`\rho _A`$ and $`\rho _B`$ have the same eigenvalues of $`\mu _i`$. It follows
$$C^2(|\psi )=2\left(1\underset{i}{}\mu _i^2\right)=4\underset{i<j}{}\mu _i\mu _j,$$
(3)
which varies smoothly from $`0`$, for pure product states, to $`2(m1)/m`$ for maximally entangled pure states.
In order to derive a quantitative connection with the PPT criterion and the realignment criterion, we recall some details of the two criteria. Peres made firstly an important step forward for separability criterion in Peres96 by showing that $`\rho ^{T_A}0`$ should be satisfied for a separable state, where $`\rho ^{T_A}`$ stands for a partial transpose with respect to the subsystem $`A`$. $`\rho ^{T_A}0`$ is further shown by Horodecki et al. HorodeckiPLA96 to be sufficient for $`2\times 2`$ and $`2\times 3`$ bipartite systems. $`\rho ^{T_A}`$ is LU invariant as shown in Refs.Peres96 ; Vidal02 where $`||||`$ stands for the trace norm defined by $`G=Tr(GG^{})^{1/2}`$. Thus it is sufficient to consider only the pure states with standard Schmidt form given by Eq. (2). It is easy to see that $`\rho =|\psi \psi |=_{i,j}\sqrt{\mu _i\mu _j}|a_ib_ia_jb_j|`$ and $`\rho ^{T_A}=_{i,j}\sqrt{\mu _i\mu _j}|a_j^{}b_ia_i^{}b_j|.`$ Then we arrive at
$`\rho ^{T_A}`$ $`=`$ $`{\displaystyle \underset{i,j}{}}\sqrt{\mu _i\mu _j}|a_j^{}b_ib_ja_i^{}|`$ (4)
$`=`$ $`{\displaystyle \underset{j}{}}\sqrt{\mu _j}|a_j^{}b_j|{\displaystyle \underset{i}{}}\sqrt{\mu _i}|b_ia_i^{}|`$
$`=`$ $`GG^{}=G^2=({\displaystyle \underset{i}{}}\sqrt{\mu _i})^2.`$
where $`G=_j\sqrt{\mu _j}|a_j^{}b_j|`$. In this derivation we have used the unitarily invariant property of the trace norm when applying the elementary column transformation: $`a_i^{}b_j|b_ja_i^{}|`$ in the derivation of the first formula. The last formula is obtained from the observation that $`GG^{}=_{i,j}\sqrt{\mu _i\mu _j}|a_i^{}b_i||b_ja_j^{}|=_i\mu _i|a_i^{}a_i^{}|,`$ the property of the trace norm $`PQ=PQ`$ and the fact that $`G`$ is the sum of the square root of eigenvalues $`\mu _i`$ of $`GG^{}`$.
Another complementary operational criterion for separability called the *realignment* criterion is very strong in detecting many of BES Rudolph02 ; ChenQIC03 and even genuinely tripartite entanglement Horodecki02 . Recently there has been considerable progress in the further analysis, and in finding stronger variants and multipartite generalizations for this criterion realignmentcriteria . We recall that this criterion states that a realigned version $`(\rho )`$ of $`\rho `$ should satisfy $`(\rho )1`$ for any separable state $`\rho `$. $`(\rho )`$ is simply $`(\rho )_{ij,kl}=\rho _{ik,jl}`$ where $`i`$ and $`j`$ are the row and column indices for the subsystem $`A`$ respectively, while $`k`$ and $`l`$ are such indices for the subsystem $`B`$ Rudolph02 ; ChenQIC03 ; Horodecki02 . $`(\rho )`$ is also shown to be LU invariant in ChenQIC03 . One has $`(\rho )=_{i,j}\sqrt{\mu _i\mu _j}|a_ia_j^{}b_i^{}b_j|`$ for the state Eq. (2), as follows easily from the definition. Similar to (4) one has
$`(\rho )`$ $`=`$ $`{\displaystyle \underset{i}{}}\sqrt{\mu _i}|a_ib_i^{}|{\displaystyle \underset{j}{}}\sqrt{\mu _j}|a_j^{}b_j|`$ (5)
$`=`$ $`GG^{}=G^2=({\displaystyle \underset{i}{}}\sqrt{\mu _i})^2.`$
where $`G=_i\sqrt{\mu _i}|a_ib_i^{}|`$. The last formula follows from the observation $`GG^{}=_{i,j}\sqrt{\mu _i\mu _j}|a_ib_i^{}||b_j^{}a_j|=_i\mu _i|a_ia_i|`$.
We now derive the main result of this Letter.
Theorem: *For any $`mn`$ $`(mn)`$ mixed quantum state $`\rho `$, the concurrence $`C(\rho )`$ satisfies*
$$C(\rho )\sqrt{\frac{2}{m(m1)}}\left(\mathrm{max}(\rho ^{T_A},(\rho ))1\right).$$
(6)
*Proof.โ* To obtain the desired lower bound, let us assume that one has already found an optimal decomposition $`_ip_i\rho ^i`$ for $`\rho `$ to achieve the infimum of $`C(\rho )`$, where $`\rho ^i`$ are pure state density matrices. Then $`C(\rho )=_ip_iC(\rho ^i)`$ by definition. Noticing that $`\rho ^{T_A}_ip_i(\rho ^i)^{T_A}`$ and $`(\rho )_ip_i(\rho ^i)`$ due to the convex property of the trace norm, one needs to show $`C(\rho ^i)\sqrt{2/\left(m(m1)\right)}((\rho ^i)^{T_A}1)`$ and $`C(\rho ^i)\sqrt{2/\left(m(m1)\right)}((\rho ^i)1).`$ For a pure state $`\rho ^i`$ one has $`(\rho ^i)=(\rho ^i)^{T_A}=(_k\sqrt{\mu _k})^2`$ from Eqs. (4) and (5), where $`\sqrt{\mu _k}`$ are the Schmidt coefficients for the pure state $`\rho ^i`$. From the expression of Eq. (3) it remains to prove that
$`4{\displaystyle \underset{i<j}{}}\mu _i\mu _j`$ $``$ $`{\displaystyle \frac{2}{m(m1)}}(({\displaystyle \underset{k}{}}\sqrt{\mu _k})^21)^2`$ (7)
$`=`$ $`{\displaystyle \frac{8}{m(m1)}}({\displaystyle \underset{i<j}{}}\sqrt{\mu _i\mu _j})^2,`$
where we have used $`_i\mu _i=1`$.
The verification of the inequality Eq. (7) is straightforward: by summing over all of arithmetic mean inequalities $`\mu _i\mu _j+\mu _k\mu _l2\sqrt{\mu _i\mu _j\mu _k\mu _l}`$ for $`i<j`$ and $`k<l`$, one gets
$`{\displaystyle \underset{i<j}{}}{\displaystyle \underset{k<l}{}}(\mu _i\mu _j+\mu _k\mu _l)`$ $``$ $`2{\displaystyle \underset{i<j}{}}{\displaystyle \underset{k<l}{}}\sqrt{\mu _i\mu _j\mu _k\mu _l}`$ (8)
$`=`$ $`2({\displaystyle \underset{i<j}{}}\sqrt{\mu _i\mu _j})^2.`$
It is seen that the number of appearance times is $`m(m1)`$ for the term $`\mu _i\mu _j`$ on the lhs of Eq. (8). Therefore Eq. (7) is confirmed and the conclusion Eq. (6) is proved.
The most prominent feature of the Theorem is that it allows to obtain an analytical lower bound for the concurrence without any numerical optimization procedure. The bound has the same range as $`C(\rho )`$ and goes from $`0`$ to $`\sqrt{2(m1)/m}`$ for pure product states and maximally entangled pure states, respectively. One can of course renormalizes the maximum value $`C(\rho )`$ to be 1 with a change of the corresponding constant factor.
We highlight some of the benefits of this new bound. Firstly, it serves to detect and gives a lower bound of concurrence for *all* entangled states of two qubits and qubit-qutrit system. This is so because the PPT criterion is necessary and sufficient for separability in the two cases HorodeckiPLA96 . Secondly, it generalizes to bipartite systems of arbitrary dimension a relation given in Eisert99-Zyczkowski99-Verstraete01 which is only valid for two qubits case, that the concurrence is lower bounded by the negativity Vidal02 (defined to be $`\rho ^{T_A}1`$). Thirdly, for any qubit-qudit system our bound can contribute an analytical lower bound for EOF which is a convex function of the concurrence, $`E(|\psi )=H_2\left((1+\sqrt{1C^2(|\psi )})/2\right)`$ where $`H_2(.)`$ is the binary entropy function Lozinski03 . In fact our bound can furnish a lower bound of EOF $`E(\rho )`$ for arbitrary bipartite state $`\rho `$ Mintert04-MintertPhD . Given any monotonously increasing, convex function $``$ satisfying $`(C(|\psi ))\mathrm{\Sigma }_r\mu _r\mathrm{log}_2\mu _r`$, one has $`E(\rho )(C(\rho ))`$, with the rhs bounded from below by our bound Eq. (6). Next we consider some examples to illustrate further the tightness and significance of our bound.
*Example 1:* Isotropic states
Isotropic states Horodecki1999 ; Vollbrecht01 are a class of $`UU^{}`$ invariant mixed states in $`d\times d`$ systems
$$\rho _F=\frac{1F}{d^21}\left(I|\mathrm{\Psi }^+\mathrm{\Psi }^+|\right)+F|\mathrm{\Psi }^+\mathrm{\Psi }^+|,$$
(9)
where $`|\mathrm{\Psi }^+\sqrt{1/d}_{i=1}^d|ii`$ and $`F=\mathrm{\Psi }^+|\rho _F|\mathrm{\Psi }^+`$, satisfying $`0F1`$, is the *fidelity* of $`\rho _F`$ and $`|\mathrm{\Psi }^+`$. These states were shown to be separable for $`F1/d`$ Horodecki1999 . It is shown in Vidal02 ; Rudolph02 that $`\rho _F^{T_A}=(\rho _F)=dF`$ for $`F>1/d`$. The concurrence $`C(\rho )`$ for this class of states is recently derived in Rungta03 to be $`\sqrt{2d/(d1)}(F1/d)`$ by an extremization procedure. An application of our Theorem gives $`C(\rho )\sqrt{2/\left(d(d1)\right)}(dF1)=C(\rho )`$. Thus the bound gives surprisingly exact value of the concurrence for this sort of states.
*Remark:* One can see that the equality of Eq. (8) holds when $`|\psi `$ are product states or maximally entangled states (MES) (all $`\mu _i`$ are equal). Thus our bound will be tight if an optimal decomposition for achieving concurrence only involves product states and MES, and also attains the value of $`\rho ^{T_A}`$ or $`(\rho )`$. Roughly speaking, the difference between our lower bound and the exact value of concurrence will be small if there are few deviations from these two types of states in the optimal ensemble decomposition. Exact estimation of this difference would be an interesting subject for future study. In the case of isotropic states, it is shown in Rungta03 that the optimal decomposition falls exactly into this class and the concurrence is just our bound.
*Example 2:* $`3\times 3`$ BES constructed from unextendible product bases (UPB)
In UPB , Bennett et al. introduced a $`3\times 3`$ BES from the following bases:
$`|\psi _0`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}|0(|0|1),|\psi _1={\displaystyle \frac{1}{\sqrt{2}}}(|0|1)|2,`$
$`|\psi _2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}|2(|1|2),\text{ }|\psi _3={\displaystyle \frac{1}{\sqrt{2}}}(|1|2)|0,`$
$`|\psi _4`$ $`=`$ $`{\displaystyle \frac{1}{3}}(|0+|1+|2)(|0+|1+|2),`$
from which the density matrix could be expressed as
$$\rho =\frac{1}{4}(Id\underset{i=0}{\overset{4}{}}|\psi _i\psi _i|).$$
(10)
A simple calculation gives $`\rho ^{T_A}=1`$ and $`(\rho )=1.087`$ ChenQIC03 , therefore $`C(\rho )0.05`$ according to the Theorem. This shows that the state is entangled.
When BES are constructed from the UPB UPB given by $`|\psi _j=|\stackrel{}{v}_j|\stackrel{}{v}_{2jmod5},(j=0,\mathrm{},4)`$ with $`\stackrel{}{v}_j=N(\mathrm{cos}(2\pi j/5),\mathrm{sin}(2\pi j/5),h)`$, with $`j=0,\mathrm{},4`$, $`h=\sqrt{1+\sqrt{5}}/2`$ and $`N=2/\sqrt{5+\sqrt{5}}`$, then the PPT state of Eq. (10) gives $`\rho ^{T_A}=1`$ and $`(\rho )=1.098`$ ChenQIC03 , therefore $`C(\rho )0.056`$ according to the Theorem, which identifies this BES.
It is conjectured by Audenaert et al. that the optimization method for concurrence is a necessary and sufficient for separability when one considers all possible complex linear combination of the concurrence-vectors Audenaert01 . Our numerical verification suggests a disproval for this conjecture, because of the failure to identify entanglement by applying their optimization method for the above two UPB states. Thus the direct estimation method for concurrence in Audenaert01 may not be able to detect *all* entangled states through numerical optimizations. Here our Theorem complements other existing approaches to make a quite good estimate of entanglement for BES.
*Example 3:* Horodeckiโs $`3\times 3`$ entangled state
A mixed two qutrits is introduced in HorodeckiPRL99 :
$$\sigma _\alpha =\frac{2}{7}|\mathrm{\Psi }^+\mathrm{\Psi }^+|+\frac{\alpha }{7}\sigma _++\frac{5\alpha }{7}\sigma _{},$$
(11)
where
$`\sigma _+`$ $`=`$ $`{\displaystyle \frac{1}{3}}(|0|10|1|+|1|21|2|+|2|02|0|),`$
$`\sigma _{}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(|1|01|0|+|2|12|1|+|0|20|2|),`$
$`|\mathrm{\Psi }^+`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{3}}}\left(|0|0+|1|1+|2|2\right).`$ (12)
In HorodeckiPRL99 Horodecki et al. demonstrate that the states Eq. (11) admit a simple characterization with respect to the parameter $`2\alpha 5`$: separable for $`2\alpha 3`$; bound entangled for $`3<\alpha 4`$; free entangled for $`4<\alpha 5`$. It is computed by using the realignment criterion in Rudolph02 that $`(\sigma _\alpha )=(19+2\sqrt{3\alpha ^215\alpha +19})/21`$ and one can recognize all the entangled states for $`3<\alpha 5.`$ One can obtain further that $`\sigma _\alpha ^{T_A}=1`$ for $`2\alpha 4`$ and $`\sigma _\alpha ^{T_A}=(2+\sqrt{4\alpha ^220\alpha +41})/7`$ for $`4<\alpha 5.`$ Therefore one has $`C(\sigma _\alpha )1/\sqrt{3}((\sigma _\alpha )1)=2\sqrt{3}(\sqrt{3\alpha ^215\alpha +19}1)/63`$ due to the observation that $`(\sigma _\alpha )`$ is always greater than $`\sigma _\alpha ^{T_A}`$ in the entangled region $`3<\alpha 5`$.
However, the concurrence optimization procedure proposed in Mintert04-MintertPhD can only identify the entangled states for $`3.52\alpha 5`$ Mintert04-MintertPhD . This suggests that Mintert et al.โs methods may not be necessary and sufficient for detecting entanglement. A rough comparison with the result of Mintert04-MintertPhD shows that our lower bound is much better than their optimized bound in the entangled region of $`3<\alpha 4.75`$, though a little bit weaker than theirs in the region $`4.75\alpha 5`$.
We remark that, like any other known approaches, there are also some drawbacks for our estimation. Our lower bound cannot detect all the entangled states due to limitation of the PPT criterion and the realignment criterion. For example, it can neither recognize the $`2\times 4`$ Horodecki BES HorodeckiPLA97 , which instead can be detected by the methods of Mintert04-MintertPhD , nor give the exact value of concurrence for 2 qubits known from Wootters98 ; Mintert04-MintertPhD .
In summary, we have provided an entirely analytical formula for lower bound of concurrence, by making a novel connection with the known strong separability criteria. The bound leads to actual values of concurrence for some special class of quantum states. One only needs to calculate the trace norm of certain matrices, which avoids complicated optimization procedure over a large number of free parameters in numerical approaches. The formula also permits to furnish lower bounds of EOF for arbitrary bipartite quantum state. This complements the nice result of Wootters for 2 qubits, as well as a number of existing optimization methods for concurrence. Profiting from the strong realignment criterion, our bound can give easy entanglement evaluation for many BES, which fail to be recognized by the regular optimization methods. This shows that our method can serve as a powerful tool for investigating both static and dynamical entanglement properties in realistic quantum computing devices. As applications the method could be used in indicating a possible quantum phase transition for condensed matter system, and in analyzing finite size or scaling behavior of entanglement in various interacting quantum many-body systems.
K.C. gratefully acknowledges support from the Alexander von Humboldt Foundation. This work has been supported the Deutsche Forschungsgemeinschaft SFB611 and German(DFG)-Chinese(NSFC) Exchange Programme 446CHV113/231. We thank Zhi-Xi Wang for valuable discussions.
|
warning/0506/math0506343.html
|
ar5iv
|
text
|
# Limiting search cost distribution for the move-to-front rule with random request probabilities
## 1 Introduction and model
Consider a list of $`n`$ files which is updated as follows: at each unit of discrete time, a file is requested independently of the previous requests and is moved to the front of the list. This heuristic is called the move-to-front rule and was first introduced by and to sort files. Such strategy is used when the request probabilities are unknown, otherwise we would list the files in order to have decreasing request probabilities. The move-to-front rule induces a Markov chain over the permutations of $`n`$ elements which has a unique stationary distribution, (see and reference to the work of Hendricks, Dies and Letac therein). This distribution turns out to be the size-biased permutation of the request probabilities.
Here, we consider that these request probabilities are themselves random, as in a Bayesian analysis. Let $`\omega =(\omega _i)_{iIN^{}}`$ be a sequence of iid positive random variables. The Laplace transform of a weight will be denoted by $`\varphi `$ and its expectation by $`\mu `$. For any $`i\text{IN}^{}`$, $`\omega _i`$ represents the weight of the file $`i`$. We can construct request probabilities $`๐ฉ=(p_1,..,p_n)`$ as follows:
$$i\{1,\mathrm{},n\},p_i=\frac{\omega _i}{W_n}\text{where}W_n=\underset{i=1}{\overset{n}{}}\omega _i.$$
Such random vector $`๐ฉ`$ is called a random discrete distribution .
Let us denote by $`S_n`$ the search cost of an item (i.e. the position in the list of the requested item) when the underlying Markov chain is in steady state (the first position will be 0). For this model, obtained exact and asymptotic formulae for the Laplace transform of $`S_n`$ (some results were also extended to the case of independent random weights). In particular, they found the limit of the expectation and the variance of $`S_n`$. Moreover, in the case of i.i.d. gamma weights, , obtained the exact and asymptotic distribution of $`S_n`$, using an exact representation of the size-biased permutation arising from Dirichlet partitions. Note that found the limiting distribution of $`S_n`$ when weights are deterministic but non-identical, in some cases (uniform, Zipfโs law, generalized Zipfโs law, power law and geometric).
In section 2, we shall give a general formula for the density of the limiting search cost distribution $`S`$, provided that the expected weight is finite. Then we derive the moment function and the cumulative distribution function of $`S`$. We also discuss the relationship between the move-to-front rule and the least-recently-used strategy. In section 3 we study some examples for which computations can be done explicitly: both continuous and discrete distributions are considered.
## 2 Limiting search cost distribution
The early analysis of the heuristic move-to-front focused on the expected search cost, see , and , for instance. Later, researchers paid much attention to the (transient and stationary) distribution of the search cost (). Some of them investigated the limiting behavior as the number $`n`$ of items tends to infinity (see ). In a more recent article, obtained an integral representation of the Laplace transform of $`S_n`$ in the Bayesian model described in the introduction. Their main theorem is the following:
###### Theorem 2.1
For a sequence $`\omega `$ of iid positive random variables,
$$s0,\varphi _{S_n}(s)=n_0^{\mathrm{}}_t^{\mathrm{}}\varphi ^{\prime \prime }(r)\left[\varphi (r)+e^s\left(\varphi (rt)\varphi (r)\right)\right]^{n1}๐r๐t.$$
In the same article the integral representation for the two first moments of $`S_n`$ were derived. Moreover, they obtained a point-wise asymptotic equivalent for the Laplace transform of $`S_n`$ and the limit of the first two moments of $`S_n/n`$ when the number $`n`$ of items tends to infinity. From theorem 2.1, we can obtain the following closed-form expression for the density function of the limiting distribution of $`Sn/n`$:
###### Theorem 2.2
For a sequence $`\omega `$ of iid positive random weights with finite expectation $`\mu `$,
$$\frac{S_n}{n}\underset{n\mathrm{}}{\overset{๐}{}}S,$$
where $`S`$ is a continuous random variable with the following density function $`f_S`$:
$$f_S(x)=\frac{1}{\mu }\frac{\varphi ^{\prime \prime }\left(\varphi ^1(1x)\right)}{\varphi ^{}\left(\varphi ^1(1x)\right)}\text{11}_{[0,1p_0]}(x),$$
(1)
where $`p_0=\text{IP}(\omega _i=0)`$ and $`\varphi ^1`$ is the inverse function of $`\varphi `$.
###### Remark 2.1
The quantity $`p_0`$ can be interpreted as follows: $`p_0`$ is the probability that an item is never requested. At stationarity, one expects that any such item will be at the bottom of the list: $`np_0`$ is the mean number of unrequested items. So it is not surprising that the support of $`S`$ is not the entire unit interval. Note that if the distribution of the weight is continuous, then $`p_0=0`$.
###### Proof.
We have to prove that $`S_n/n`$ converges in distribution, as $`n`$ tends to infinity, to a certain random variable that will be denote by $`S`$. First, observe that:
$$s0,\varphi _{S_n/n}(s)=\varphi _{S_n}\left(\frac{s}{n}\right).$$
So we are now interested in the limit of $`\varphi _{S_n}(s/n)`$.
For any reals $`a`$ and $`b`$ such that $`0ab\mathrm{}`$, let:
$$I_n(a,b)=_a^b\varphi ^{\prime \prime }(r)\left[\varphi (r)+e^{s/n}(\varphi (rt)\varphi (r))\right]^{n1}๐r.$$
If $`b=\mathrm{}`$, then we will omit this parameter, i.e. $`I_n(a)=I_n(a,\mathrm{})`$. Using these notations, theorem 2.1 gives:
$$\varphi _{S_n}\left(\frac{s}{n}\right)=n_0^{\mathrm{}}I_n(t)๐t.$$
(2)
We now decompose $`I_n(t)`$ into two parts: $`I_n(t)=I_n(t,t+\epsilon )+I_n(t+\epsilon )`$. We will prove that $`nI_n(t+\epsilon ,\mathrm{})`$ tends to $`0`$ when $`n`$ tends to infinity:
$`nI_n(t+\epsilon ,\mathrm{})`$ $`=`$ $`n{\displaystyle _{t+\epsilon }^{\mathrm{}}}\varphi ^{\prime \prime }(r)\left[e^{s/n}(\varphi (rt)+(1e^{s/n})\varphi (r))\right]^{n1}๐r,`$
$``$ $`n{\displaystyle _{t+\epsilon }^{\mathrm{}}}\varphi ^{\prime \prime }(r)\varphi (rt)^{n1}๐r,`$
$``$ $`n\varphi (\epsilon )^{n1}\varphi ^{}(t+\epsilon ),`$
since $`\varphi `$ is decreasing. Then $`lim_n\mathrm{}nI_n(t+\epsilon ,\mathrm{})=0`$, for all $`\epsilon >0`$.
Now we will estimate $`I_n(t,t+\epsilon )`$. Let $`h_n(r,t)=\varphi (r)+e^{s/n}(\varphi (rt)\varphi (r))`$. For a fixed value of $`t`$, the function $`h_n(,t)`$ behaves as $`\varphi `$. In particular $`\frac{h_n}{r}`$ is an increasing function for $`r[t,t+\epsilon ]`$. Then we obtain the following bounds:
$$\frac{h_n}{r}(t,t)\frac{h_n}{r}(r,t)\frac{h_n}{r}(t+\epsilon ,t),$$
and:
$$\varphi ^{\prime \prime }(t+\epsilon )\varphi ^{\prime \prime }(r)\varphi ^{\prime \prime }(t).$$
Hence, we can bound $`I_n(t,t+\epsilon )`$ by:
$`I_n(t,t+\epsilon )`$ $`=`$ $`{\displaystyle _t^{t+\epsilon }}\varphi ^{\prime \prime }(r)\left(h_n(r,t)\right)^{n1}{\displaystyle \frac{h_n}{r}}(r,t){\displaystyle \frac{h_n}{r}}(r,t)^1๐r`$
$``$ $`\varphi ^{\prime \prime }(t){\displaystyle \frac{h_n}{r}}(t,t)^1{\displaystyle _t^{t+\epsilon }}\left(h_n(r,t)\right)^{n1}{\displaystyle \frac{h_n}{r}}(r,t)๐r`$
$``$ $`\varphi ^{\prime \prime }(t){\displaystyle \frac{h_n}{r}}(t,t)^1{\displaystyle \frac{1}{n}}\left[\left(h_n(t+\epsilon ,t)\right)^n\left(h_n(t,t)\right)^n\right].`$
Proceeding similarly, we can find a lower bound:
$$I_n(t,t+\epsilon )\varphi ^{\prime \prime }(t+\epsilon )\frac{h_n}{r}(t+\epsilon ,t)^1\frac{1}{n}\left[\left(h_n(t+\epsilon ,t)\right)^n\left(h_n(t,t)\right)^n\right].$$
Then, for any $`\epsilon >0`$, one can prove the following limits hold:
$`\underset{n\mathrm{}}{lim}\left(h_n(t+\epsilon ,t)\right)^n`$ $`=`$ $`0,`$
$`\underset{n\mathrm{}}{lim}\left(h_n(t,t)\right)^n`$ $`=`$ $`\mathrm{exp}\left[s(1\varphi (t))\right],`$
$`\underset{n\mathrm{}}{lim}{\displaystyle \frac{h_n}{r}}(t,t)`$ $`=`$ $`\varphi ^{}(\epsilon ),`$
$`\underset{n\mathrm{}}{lim}{\displaystyle \frac{h_n}{r}}(t+\epsilon ,t)`$ $`=`$ $`\varphi ^{}(0).`$
Replacing these limits in the equations above, we have computed upper and lower bounds of $`I_n(t,t+\epsilon )`$. In other words, if the limit of $`nI_n(t,t+\epsilon )`$ exists, then it is bounded by:
$$\frac{\varphi ^{\prime \prime }(t+\epsilon )}{\varphi ^{}(0)}\mathrm{exp}\left((1\varphi (t))s\right)\underset{n\mathrm{}}{lim}nI_n(t,t+\epsilon )\frac{\varphi ^{\prime \prime }(t)}{\varphi ^{}(\epsilon )}\mathrm{exp}\left((1\varphi (t))s\right).$$
This is true for any $`\epsilon >0`$; then letting $`\epsilon `$ tends to $`0`$, we have:
$$\underset{n\mathrm{}}{lim}nI_n(t)=\frac{\varphi ^{\prime \prime }(t)}{\mu }\mathrm{exp}\left((1\varphi (t))s\right).$$
Replacing this limit in equation (2) we obtain
$$\underset{n\mathrm{}}{lim}\varphi _{S_n/n}(s)=\frac{1}{\mu }_0^{\mathrm{}}\varphi ^{\prime \prime }(t)e^{(1\varphi (t))s}๐t,$$
(3)
which will be denoted by $`\varphi _S(s)`$. Although this limit a priori is not necessarily the Laplace transform of a random variable, according to the Continuity theorem (page 431 Ch. XIII in ), one has to check that $`lim_{s0}\varphi _S(s)=1`$, which can be proved by using the dominated convergence theorem.
A suitable change of variable $`y=1\varphi (r)`$ in equation (3) gives:
$$\varphi _S(s)=\frac{1}{\mu }_0^{1p_0}\frac{\varphi ^{\prime \prime }\left(\varphi ^1(1y)\right)}{\varphi ^{}\left(\varphi ^1(1y)\right)}e^{ys}๐r,$$
where for the integral limits we used the property that $`\varphi (\mathrm{})=p_0`$ (see remark in theorem 1(a) page 439 Ch. XIII). Therefore, we have that:
$$f_S(y)=\frac{1}{\mu }\frac{\varphi ^{\prime \prime }\left(\varphi ^1(1y)\right)}{\varphi ^{}\left(\varphi ^1(1y)\right)}\text{11}_{[0,1p_0]}(y)$$
is the probability density of $`S`$. โ
As a corollary to this theorem, we can compute the $`q`$-th moment and the cumulative distribution function (c.d.f.) of $`S`$:
###### Corollary 2.1
For any $`q\text{IR}`$
$$E[S^q]=\frac{1}{\mu }_0^{\mathrm{}}(1\varphi (t))^q\varphi ^{\prime \prime }(t)๐t,$$
and, for any $`x[0,1]`$,
$$\text{IP}(Sx)=\left(\frac{1}{\mu }_0^{\varphi ^1(1x)}\varphi ^{\prime \prime }(t)๐t\right)\text{11}_{[0,1p_0]}(x)+\text{11}_{(1p_0,1]}(x).$$
One could be interested in the cumulative distribution function of $`S`$ (or more precisely in the survival function), since the move-to-front rule is related to the least-recently-used strategy (see for instance). Indeed, many operating systems or softwares use a memory (also called cache) that could be quickly addressed (think of a web browser, for instance). Hence, one needs to define a strategy to organize it. Let us consider that the cache is made of $`k`$ files. The least-recently-used strategy is the following: at each unit of discrete time, a file is requested and is moved in front of the cache; if the file was not just previously in the cache, then the last file is deleted from the cache and all other files are shifted by one position to the right; if the file was just previously in the cache, then the file is moved exactly as in the move-to-front rule. So, the move-to-front rule can be viewed as a special case of the least-recently-used strategy for which the length of the cache is equal to the number of files ($`k=n`$). An important question arises: what is the probability that the requested file is not in the cache? The probability of this event is called the page default; we will denote it by $`\pi _k`$ in the sequel. Because of the link between the move-to-front rule and the least-recently-used strategy (as underlined above), we clearly have that $`\pi _k=\text{IP}(S_nk)`$. So, if we assume that the cache length is proportional to the number of files, say $`k=\alpha n`$ with $`\alpha [0,1]`$ fixed, for a large collection of files, the following approximation holds:
$$\pi _{\alpha n}\frac{1}{\mu }_0^{\varphi ^1(1\alpha )}\varphi ^{\prime \prime }(t)๐t$$
if $`\alpha <p_0`$ and $`\pi _{\alpha n}1`$ otherwise.
## 3 Examples
In this section, we study some examples for which we are able to do explicitly all computations. We will consider both continuous and discrete distribution for the random weights.
###### Example 3.1
Suppose that the weights have the Dirac distribution at point mass $`1`$ (in other words, weights are deterministic and are equally requested). Then $`\varphi (r)=e^r`$, the expectation $`\mu =1`$ and $`p_0=0`$, we deduce that:
$$f_{S_1}(x)=\text{11}_{[0,1]}(x).$$
Thus, $`S_1`$ has the uniform distribution over $`[0,1]`$: this result was already proved in (theorem 4.2, p. 198 of ). The $`k`$-th moment (with $`k\text{IR}_+`$) and the c.d.f. of $`S_1`$ is:
$$\text{IE}[S_1^k]=\frac{1}{k+1}\text{and}x[0,1],F_{S_1}(x)=\text{IP}(S_1x)=x.$$
###### Example 3.2
Suppose that the weights have the Gamma distribution with parameter $`\alpha >0`$. In this example, the random vector $`(p_1,\mathrm{},p_n)`$ has the symmetric Dirichlet distribution $`D_n(\alpha )`$ (see or ). In such a case, $`p_0=0`$, $`\mu =\alpha `$ and $`\varphi (r)=(1+r)^\alpha `$. Computations give:
$$f_{S_2}(x)=\left(1+\frac{1}{\alpha }\right)(1x)^{1/\alpha }\text{11}_{[0,1]}(x),$$
which is the density function of the Beta distribution with parameters $`(1,1+1/\alpha )`$. Note that this result has already been proved by with a specific technique using properties of Dirichlet distribution (in this case we were able not only to find the limiting search cost distribution but also the transient search cost distribution for any finite $`n`$). The $`k`$-th moment (with $`k\text{IR}_+`$) of $`S_2`$ is:
$$\text{IE}[S_2^k]=\frac{\mathrm{\Gamma }(k+1)\mathrm{\Gamma }(2+\frac{1}{\alpha })}{\mathrm{\Gamma }(2+k+\frac{1}{\alpha })}.$$
In particular, we have $`\text{IE}[S_2]=\frac{\alpha }{2\alpha +1}`$ and $`\text{Var}[S_2]=\frac{(\alpha +1)\alpha ^2}{(3\alpha +1)(2\alpha +1)^2}`$. One can also compute the c.d.f of $`S_2`$ and, for any $`x[0,1]`$, we get:
$$F_{S_2}(x)=\text{IP}(S_2x)=1(1x)^{1+1/\alpha }.$$
We can easily deduce that, for any $`x[0,1]`$, $`\overline{F}_{S_2}(x)\overline{F}_{S_1}(x)`$, where $`\overline{F}_{S_1}()=1F_{S_1}()`$. So we have $`S_2_{st}S_1`$ (where $`_{st}`$ denotes the usual stochastic ordering; see or , for instance).
###### Example 3.3
Suppose that the weights have the Geometric distribution on IN with parameter $`p(0,1)`$. In such case, $`p_0=p`$, $`\mu =(1p)/p`$ and $`\varphi (r)=p/(1(1p)e^r)`$. Elementary computations give:
$$f_{S_3}(x)=\frac{2(1x)p}{1p}\text{11}_{[0,1p]}(x).$$
The $`k`$-th moment (with $`k\text{IR}_+`$) of $`S_3`$ is:
$$\text{IE}[S_3^k]=\frac{(2+pk)(1p)^k}{(k+1)(k+2)}.$$
In particular, we have $`\text{IE}[S_3]=\frac{(2+p)(1p)}{6}`$ and $`\text{Var}[S_3]=\frac{(1p)^2(2+2pp^2)}{36}`$. One can also compute the c.d.f of $`S_3`$ and, for any $`x[0,1]`$, get:
$$F_{S_3}(x)=\text{IP}(S_3x)=\frac{x(2px)}{1p}\text{11}_{[0,1p]}(x)+\text{11}_{(1p,1]}(x).$$
Hence, from the above expression, one can check that $`S_4_{st}S_1`$.
###### Example 3.4
Suppose that the weights have the Poisson distribution with parameter $`\lambda `$. In such case, $`p_0=e^\lambda `$, $`\mu =\lambda `$ and $`\varphi (r)=\mathrm{exp}\left(\lambda e^r1\right)`$. Simple computations give:
$$f_{S_4}(x)=\frac{\mathrm{ln}(1x)+\lambda +1}{\lambda }\text{11}_{[0,1e^\lambda ]}(x).$$
Using formula 1.6.5.3 of (page 244), one can compute the $`k`$-th moment (with $`k\text{IN}`$) of $`S_4`$:
$$\text{IE}[S_4^k]=\frac{1}{\lambda (k+1)}\left[\lambda +(1e^\lambda )^{k+1}\underset{i=1}{\overset{k+1}{}}\frac{(1e^\lambda )^i}{i}\right].$$
In particular, we have $`\text{IE}[S_4]=\frac{1}{2}\frac{1e^{2\lambda }}{4\lambda }`$. One can also compute the c.d.f of $`S_4`$ and, for any $`x[0,1]`$, we get:
$$F_{S_4}(x)=\text{IP}(S_4x)=(x\frac{1}{\lambda }(1x)\mathrm{ln}(1x))\text{11}_{[0,1e^\lambda ]}(x)+\text{11}_{(1e^\lambda ,1]}(x).$$
Thus, from the expression above, one can deduce that $`S_4_{st}S_1`$.
From the study of these four examples, one can observe that both $`S_2`$, $`S_3`$ and $`S_4`$ are stochastically smaller than $`S_1`$. Hence, the following conjecture looks appealing:
###### Conjecture 3.1
Let $`S`$ be the limiting distribution of the search cost associated to a sequence $`\omega `$ of iid positive random variables. Then, $`S_{st}S_1`$ where $`S_1`$ is a random distribution having the uniform distribution on the unit interval.
This conjecture is compatible with some remarks in , more precisely with proposition 3.1 therein. Indeed, if the conjecture is right, then as a consequence we have $`\text{IE}[S]\text{IE}[S_1]=\frac{1}{2}`$. And this is precisely what is stated in proposition 3.1. This conjecture can be interpreted as follows: the case with Dirac weights corresponds to the worst case. Despite our conjecture seems to be true, its proof seems to be difficult.
### Acknowledgment
The authors thank support from FONDAP-CONICYT in Applied Mathematics and Millenium Nucleus in Information and Randomness ICM P01-005. JB wishes to thank CONICYT National Postgraduate Fellowship Program which supports her PhD.
|
warning/0506/math0506146.html
|
ar5iv
|
text
|
# Line-Bundle-Valued Ternary Quadratic Bundles Over Schemes Dedicated to the Memory of Professor Martin Kneser
#### Keywords:
Azumaya algebra, Clifford algebra, orthogonal group, quaternion algebra, semiregular form, Witt-invariant.
#### MSC:
11E, 14A25, 14F05, 14L15, 14M, 14Q, 15A63, 15A66, 15A75, 15A78, 16H05, 16S60, 16W20, 20G05, 20G35
## 1 Introduction
To explain the results of this work, in this introduction we shall restrict ourselves to the case of affine schemes for the sake of simplicity, leaving the general formulation to ยง3. So let $`R`$ be a commutative ring with $`1.`$ The objects of our study are ternary quadratic forms $`q:VR`$ on projective $`R`$-modules $`V`$ of rank 3. Zariski-locally these are quadratic forms in three variables. We shall denote such forms by pairs $`(V,q).`$ A similarity from $`(V,q)`$ to $`(V^{},q^{})`$ is a pair $`(g,m)`$, where $`g:VV^{}`$ and $`m:RR`$ are linear isomorphisms such that $`q^{}g=mq.`$ When $`m`$ is the identity, we call the similarity an isometry.
Our broad aim is to study the sets of similarities between ternary forms (when they exist) in terms of their even-Clifford algebras and indicate applications. We are especially interested in bad quadratic forms, and would like to develop a theory for such forms. We obtain results extending (though not deduced from, but rather motivated by) what is known for the good ones.
What are good forms? These are the same as the usual regular (or nonsingular) forms, provided none of the residue fields of $`R`$ is of characteristic two. The correct technical notion is that of semiregular form, due to Martin Kneser, on which we will elaborate in ยง2.3, page 2.3. At this point we only remark that this notion works universally, so includes the case of regular forms as well.
For a semiregular form $`q:VR`$, its even-Clifford algebra $`A=C_0(V,q)`$ is an Azumaya $`R`$-algebra with underlying module of rank 4. Azumaya means that the natural $`R`$-algebra homomorphism
$$A_RA^{\text{op}}\text{End}_{R\text{Mod}}(A):ab^{\text{op}}(xaxb^{\text{op}})$$
is an isomorphism, where the superscript refers to the opposite algebra. Matrix algebras, for example algebras of $`(2\times 2)`$ matrices in our case, are simple examples of Azumaya algebras, and they are all one gets if say $`R`$ were an algebraically closed field. Any similarity $`(g,m):(V,q)(V^{},q^{})`$ induces an isomorphism of $`R`$-algebras $`C_0(g,m):C_0(V,q)C_0(V^{},q^{}).`$ So if we denote the set of isometry classes of semiregular ternary quadratic forms by $`๐ฌ_3^{sr}(R)`$, and the set of $`R`$-algebra isomorphism classes of rank 4 Azumaya $`R`$-algebras by $`๐๐ต๐ฐ_4(R)`$, then we get a natural map, which is in fact functorial in $`R`$:
$$()๐ฌ_3^{sr}(R)๐๐ต๐ฐ_4(R)\text{ via }(V,q)C_0(V,q).$$
It is well-known that the above map is surjective but may not be injective. In fact, given an Azumaya algebra $`A`$ of rank 4, there is a well-defined unique standard involution $`\sigma _A`$ on $`A`$, which defines a regular quadratic form: the norm $`n_A:AR`$ and a linear form: the trace $`tr_A:AR.`$ Then the restriction of $`n_A`$ to the kernel $`A^{}`$ of $`tr_A`$, which is a rank 3 $`R`$-module (due to the surjectivity of the trace), remains good i.e., is a semiregular quadratic form, and one has an isomorphism of $`R`$-algebras
$$C_0(A^{},n_A|A^{})A.$$
Thus the map $`()`$ is surjective. Let us explain why it is not injective. Let $`I`$ be a rank 1 projective $`R`$-module (i.e., an invertible module) equipped with an $`R`$-linear isomorphism
$$h:I^2:=IIR.$$
The pair $`(I,h)`$ is called a discriminant module and can be thought of as a good symmetric bilinear form on $`I`$ as well as a prescribed square root of $`R.`$ Now given a ternary quadratic module $`(V,q)`$, we may define a new quadratic module:
$$(V,q)(I,h):=(VI,qh):(qh)(vl):=q(v)h(ll).$$
Then it can be verified that we have an isomorphism of $`R`$-algebras
$$C_0(V,q)C_0(VI,qh).$$
Thus the map in $`()`$ may not be injective. Now the set of isomorphism classes of $`(I,h)`$ forms a commutative group of exponent 2, the group operation induced by the usual tensor product. It is denoted by $`\text{Disc}(R)`$ and acts on the left side of $`().`$ It is a Theorem of Max-Albert Knus that we have a bijection
$$()๐ฌ_3^{sr}(R)/\text{Disc(R)}๐๐ต๐ฐ_4(R),$$
which may also be thought of as a statement in cohomology (cf. discussion following Theorem 2.10, page 2.10). A natural question is to ask if there is a statement analogous to $`()`$ if we consider non-semiregular forms on the left side as well. Our central result, Theorem 3.1 (page 3.1), shows that the answer is yes, thus providing a limiting version of the cohomological statement $`().`$ However, when considering non-semiregular forms, we would need to make some changes to our present formulation; these we set out to explain next.
Firstly, since we wanted to include degenerate forms in the left side of $`()`$, and since such forms are limits, locally in the Zariski-topology, of semiregular forms, it is natural to consider similar limits on the right side. In other words, we consider rank 4 $`R`$-algebras which are schematic limits of rank 4 Azumaya algebras. They were introduced in , and are recalled in ยง2.9 (page 2.9). Instead of going now into the definition of such limiting algebras, we note that it follows from op. cit., that they are precisely those rank 4 $`R`$-algebras which are Zariski-locally isomorphic to even-Clifford algebras of ternary quadratic forms. Let us denote the set of isomorphism classes of such algebras by $`๐ฎ๐ซ๐๐ต๐ฐ_4(R).`$ This will be the replacement for the right side of $`()`$.
Given an algebra $`A`$ representing an element of $`๐ฎ๐ซ๐๐ต๐ฐ_4(R)`$, all we know by definition is that there are elements $`f_i`$ that generate $`R`$, and ternary quadratic $`R_{f_i}`$-forms $`(V_i,q_i)`$ such that there are isomorphisms $`A_RR_{f_i}C_0(V_i,q_i).`$ It is not clear a priori that there is a choice for which the $`V_i`$ would glue to give a rank 3 $`R`$-module $`V`$, and even if this were so, that the $`q_i`$ would glue to give a quadratic form on $`V.`$ In fact, it may not. We show that there is a choice for which the $`V_i`$ glue to give a $`V`$, but that the $`q_i`$ glue to give a quadratic form $`q`$ with values in the invertible module $`I:=\text{det}^1(A)`$, such that
$$C_0(V,q,I)A,$$
(cf. Theorem 3.6, part (a), page 3.6) and further that that we could get a quadratic form with values in $`R`$ iff $`\text{det}(A)2.\text{Pic}(R)`$ (cf. Theorem 3.6, part (b)). At this point we note that the even Clifford algebra $`C_0(V,q,I)`$ is the one defined by Bichsel-Knus in , briefly recalled in ยง2.4, page 2.4, which reduces to the familiar even-Clifford algebra when the invertible module is $`R`$ itself. Thus, going back to our idea of replacing the left side of $`()`$ with degenerate forms, we see that we should work with ternary quadratic forms with values in invertible $`R`$-modules. Let us denote the set of isometry classes of such quadratic modules by $`๐ฌ_3(R).`$ Further, since we had to divide out by the action of discriminant modules to get the bijection $`()`$, and since such discriminant modules are square roots of $`R`$, it is natural that we should consider square roots of invertible $`R`$-modules as well. Such objects could be called twisted discriminant modules, and they do form a group under the tensor product which we denote by $`\text{T-Disc}(R).`$ Any such object is given by a triple $`(L,h,J)`$ where $`h:LLJ`$ is an $`R`$-linear isomorphism with $`L,J`$ being invertible $`R`$-modules ($`h`$ makes $`L`$ a specified square-root of $`J`$). Given a quadratic form $`q:VI`$ with values in the invertible module $`I`$, one defines in the obvious way:
$$(V,q,I)(L,h,J):=(VL,qh,IJ):vlq(v)h(ll),$$
and there is a canonical isomorphism of $`R`$-algebras
$$C_0(VI,qh,LJ)C_0(V,q,I).$$
It happens that this generalisation to values in invertible modules does not disturb the bijection $`()`$ i.e., if we considered semiregular quadratic modules with values in invertible modules on the left side modulo the action of $`\text{T-Disc}(R)`$, we still obtain a bijection to the right side. Theorem 3.1 (page 3.1) shows that we do have a bijection
$$()๐ฌ_3(R)/\text{T-Disc}(R)๐ฎ๐ซ๐๐ต๐ฐ_4(R)$$
which is a limiting version of $`().`$ We next indicate the ingredients that go into the proof of $`()`$, which are of interest by themselves, since they involve the study of the sets of similarities between ternary quadratic modules.
Given ternary quadratic forms $`q,q^{}:VI`$, we may consider the set $`\text{Sim}(q,q^{})`$ of similarities from $`q`$ to $`q^{}`$: these are pairs $`(g,l)`$ where $`g\text{GL}(V)`$ and $`lR^{}`$ that satisfy $`q^{}g=\lambda _lq`$ where $`\lambda _l`$ is the element of $`\text{Aut}(I)`$ given by multiplication by $`l.`$ The second member $`l`$ of $`(g,l)`$ is called the multiplier of the similarity. The set $`\text{Sim}(q,q^{})`$ is a subset of $`\text{GL}(V)\times R^{}.`$ The subset $`\text{Iso}(q,q^{})`$ of orthogonal transformations or isometries are those similarities which have trivial multipliers. A smaller subset, $`\text{S-Iso}(q,q^{})`$, consists of special orthogonal transformations, and corresponds to those for which $`\text{det}(g)=1.`$ The latter two subsets may be considered as subsets of $`\text{GL}(V)`$ and $`\text{SL}(V)`$ respectively.
Since the even-Clifford algebra is functorial in $`q`$, any similarity $`qq^{}`$ gives rise to an isomorphism of $`R`$-algebras $`C_0(V,q,I)C_0(V,q^{},I).`$ Theorem 3.4, page 3.4, identifies the images of the subsets of similarities (defined in the previous paragraph) in the set of isomorphisms $`\text{Iso}(C_0(q),C_0(q^{}))`$ from $`C_0(q)`$ to $`C_0(q^{}).`$ It shows firstly, that each isomorphism of even-Clifford algebras lifts to a similarity; further, given such an isomorphism $`\varphi `$, Prop. 3.2, page 3.2, shows how to define an $`R`$-linear automorphism $`\varphi _{\mathrm{\Lambda }^2}`$ of $`\mathrm{\Lambda }^2(V)I^1.`$ We may thus consider those isomorphisms $`\varphi `$ for which $`\text{det}(\varphi _{\mathrm{\Lambda }^2})R^{}`$ is a square, (respectively, is 1), and denote it by $`\text{Iso}^{}(C_0(q),C_0(q^{}))`$ (respectively, by $`\text{S-Iso}(C_0(q),C_0(q^{})).`$) Then Theorem 3.4 shows that the image of $`\text{Iso}(q,q^{})`$ is preciesly $`\text{Iso}^{}(C_0(q),C_0(q^{}))`$ and that of $`\text{S-Iso}(q,q^{})`$ is precisely $`\text{S-Iso}(C_0(q),C_0(q^{})).`$
These surjectivities are accomplished by explicitly constructing families of sections indexed by the integers. The proof of the existence of $`\varphi _{\mathrm{\Lambda }^2}`$ involves the use of Bourbakiโs Tensor Operations for forms with values in invertible modules (cf. ยง2.5). In order to construct the sections, we make computations analysing the correspondence from between locally-even-Clifford algebras and specialisations of Azumaya algebras with the same underlying module. The further uses of Theorem 3.4 are principally threefold: it is used to prove the injectivity part of $`()`$ (Theorem 3.1) and later on in the proof of the surjectivity part (Theorem 3.6, (a)). It is also used in Theorem 3.5, page 3.5, in the description of the general, usual and special orthogonal groups $`\text{GO}(q):=\text{Sim}(q,q)`$, $`\text{O}(q):=\text{Iso}(q,q)`$ and $`\text{SO}(q):=\text{S-Iso}(q,q)`$ respectively, as fitting into a neat commutative diagram. Further, $`\text{GO}(q)`$ is shown to be a semi-direct product, and when $`R`$ is an integral domain, and $`q`$ is semiregular even at one point of $`\text{Spec}(R)`$, then all the subgroups of automorphisms of $`C_0(q)`$ described above are seen to coincide, so that $`\text{O}(q)`$ is the semi-direct product of the multiplicative subgroup $`\mu _2(R)`$ consisting of the square roots of 1 and of $`\text{SO}(q).`$
The injectivity of Theorem 3.1 is the statement that two ternary quadratic modules with isomorphic even-Clifford algebras differ by an isometry after one of them is tensored with a suitable twisted discriminant module. The existence of such a module is reduced to the lifting of a given isomorphism of even-Cliffords to a similarity in the free case (cf.ยง4).
A suitable member of the family of sections constructed to show the surjectivities in the proof of Theorem 3.4 allows one to get a cocycle defining a ternary quadratic module from a suitably chosen cocycle for a given specialised algebra (cf.ยง6).
Since Bourbakiโs Tensor Operations form a key tool in our computations, we explain how they are of use to us. Given two ternary quadratic forms $`q,q^{}:VI`$, one may ask how one would compare, if at all, the underlying modules of their even-Clifford algebras. Now a bilinear form $`b:V\times VI`$ defines a quadratic form $`q_b:VI`$ by $`q_b(v):=b(v,v).`$ If there exists a $`b`$ so that $`q_b=qq^{}`$, then the Tensor Operations provide us with an explicitly computable $`R`$-linear isomorphism
$$\psi _b:C_0(V,q,I)C_0(V,q^{},I).$$
Since we can always find a $`b`$ such that $`q_b=q`$, by taking $`q^{}=0`$ we see that we could, via $`\psi _b`$, transport the algebra structure on $`C_0(V,q,I)`$ to one on the underlying module $`W`$ of
$$C_0(V,0,I)=R\mathrm{\Lambda }^2(V)I^1$$
with multiplicative identity $`w=1_R.`$ In this way, we may map the space of $`I`$-valued bilinear forms on $`V`$ into the space of $`R`$-algebra structures on the same underlying module $`W`$ with identity $`w.`$ And we could do this functorially in $`R`$-algebras $`S.`$ Theorem 3.8, page 3.8, shows that in this way we obtain bijectively for any $`S`$, all the locally-even-Clifford $`S`$-algebra structures on $`WS`$ with identity $`wS.`$ In other words, bilinear forms correspond to such algebras. This generalizes a similar result due to S. Ramanan for the case $`R=k`$ an algebraically closed field (cf. ยง2.3 of ). When $`\text{char.}(k)2`$, it was given another interpretation by C. S. Seshadri in (cf. Theorem 3.10, page 3.10), which will also be our main key to computations.
Not only in the proof of Theorem 3.8, but throughout this work, we use the description from of the locally-even-Clifford algebras as schematic specialisations of Azumaya algebras since it is this description that allows us to compute without losing track of global information.
For the rest of this introduction, we depart to the language of schemes. Suppose $`W`$ is a vector bundle of rank 4 over a scheme $`X.`$ If there exists an Azumaya algebra structure on $`W`$, then it can be seen that $`W`$ has to be self-dual. One may ask for a converse: if $`W`$ is self dual, does there exist a global Azumaya algebra structure on $`W`$? We answer this question more generally, and show in Theorem 3.17, page 3.17, that this is indeed the case if we on the one hand let $`X`$ to be a proper scheme of finite type over a base scheme with connected fibres, while on the other hand we assume that $`W`$ has square rank $`n^2`$ with $`n2`$, and moreover that there exists some associative unital algebra structure $`A`$ on $`W`$ which is Azumaya at some point of each fiber over the base. We then show in fact that $`A`$ has to be Azumaya, giving a โpunctual to globalโ result.
By an application of Theorem 3.1 to this result, we deduce that the hypothesis of self-duality on the underlying bundle of the even-Clifford algebra of a ternary quadratic bundle implies that the quadratic bundle is semiregular everywhere if it is semiregular even at a single point of each fiber over the base scheme.
Such results can be used to give examples of rank 4 vector bundles $`W`$ over which there do not exist any Azumaya algebra structures but which nevertheless admit algebra structures that are Azumaya on a nonempty open subscheme. They can also be used to give examples of pairs $`(V,I)`$, consisting of a rank 3 vector bundle $`V`$ and a line bundle $`I`$, such that there exists no semiregular quadratic form on $`V`$ with values in $`I`$, though there do exist forms that are semiregular on a nonempty open subscheme. For any given irreducible smooth projective curve $`C`$ of genus $`g2`$ over an algebraically closed field, such examples naturally occur on certain smooth projective varieties $`๐ฉ_C(4,0)`$ of dimension $`4g3`$, where $`g`$ is the genus of $`C`$. In fact these $`๐ฉ_C(4,0)`$ appear as fine moduli spaces of certain parabolic stable vector bundles of rank 4 and degree zero over $`C`$ and arise as Seshadri desingularisations of the moduli space of semistable vector bundles of rank 2 and degree 0 over $`C.`$ The whole situation can be generalised to the case of a curve relative to a base scheme and the results are given in Theorem 3.18, page 3.18. A detailed account with proofs will appear in .
Suppose $`X`$ is a scheme and $`W`$ a rank 4 vector bundle on $`X`$ with a nowhere-vanishing global section $`w.`$ In Theorem 3.19, page 3.19, we delve a little into the topological and geometric properties of the scheme of specialised algebra structures on $`W`$ with identity $`w`$ (functorially in $`X`$-schemes). If $`X`$ is locally-factorial, we show that the natural homomorphism from the Picard group of $`X`$ to that of the scheme of specialised algebra structures on $`W`$ with unit $`w`$ is an isomorphism. We also study the specialised algebras that are nowhere Azumaya. When $`X`$ is locally-factorial and $`W`$ is self-dual, these algebras define a Weil divisor $`D_X`$ such that some positive integer multiple $`n.D_X`$ is principal. Therefore the natural surjective homomorphism from the Picard group of the scheme of specialisations to that of the open subscheme of Azumaya algebra structures is an isomorphism iff $`n=1`$ and in this situation, every specialised algebra structure on $`WT`$ arises from a quadratic form with values in the trivial line bundle, for any $`X`$-scheme $`T.`$ For example, this happens if $`X`$ is locally-factorial and if there exists an Azumaya algebra structure on $`W.`$ The proof uses the stratification of the variety of specialisations of $`(2\times 2)`$-matrix algebras over an algebraically closed field (Theorem 3.20, page 3.20).
#### Plan of the Paper.
Background material is to be found in ยง2, which also fixes some notation, explains definitions and recalls results for future use. The formulation of the statements of the main results follow in ยง3. The proofs will occupy ยง4 through ยง8.
#### Note on Numbering.
Various items such as theorems, propositions, definitions etc are all consecutively numbered within a section, irrespective of the subsection that they may appear in.
## 2 Notations and Preliminaries
This section collects together definitions and results necessary for future use. We omit proofs for statements involving quadratic and bilinear forms, since these can be reduced to the corresponding results for the affine case which are treated in Knusโ book . For the systematic treatment of the generalised Clifford algebra and its properties, we refer the reader to the paper of Bichsel-Knus . We also collect some preliminaries on the notion of schematic image and recall some background material from Part A of .
### 2.1 Quadratic and Bilinear Forms with Values in a Line Bundle
Let $`V`$ be a vector bundle (of constant positive rank) and $`I`$ a line bundle on a scheme $`X.`$ Let $`๐ฑ`$ and $``$ respectively denote the coherent locally-free sheaves corresponding to $`V`$ and $`I.`$ We define the coherent locally-free sheaves of bilinear and alternating forms on $`V`$ with values in $`I`$ respectively as follows:
$$\text{Bil}_{(๐ฑ,)}:=(T_{๐ช_X}^2(๐ฑ))^{}\text{ and }\text{Alt}_{(๐ฑ,)}^2:=(\mathrm{\Lambda }_{๐ช_X}^2(๐ฑ))^{}.$$
We let $`\text{Bil}_{(V,I)}`$ and $`\text{Alt}_{(V,I)}^2`$ denote the corresponding vector bundles. Now define the (coherent, locally-free) sheaf of $`I`$-valued quadratic forms on $`V`$ by the exactness of the following sequence:
$$0\text{Alt}_{(๐ฑ,)}^2\text{Bil}_{(๐ฑ,)}\text{Quad}_{(๐ฑ,)}0.$$
Let the corresponding bundle of $`I`$-valued quadratic forms on $`V`$ be denoted by $`\text{Quad}_{(V,I)}.`$ Thus a bilinear form (resp. alternating form, resp. quadratic form) with values in $`I`$ on $`V`$ over an open set $`UX`$ is by definition a section over $`U`$ of the vector bundle
$$\text{Bil}_{(V,I)}\text{ (resp. of }\text{Alt}_{(V,I)}^2,\text{ resp. of }\text{Quad}_{(V,I)}),$$
or equivalently, an element of
$$\mathrm{\Gamma }(U,\text{Bil}_{(๐ฑ,)})\text{ (resp. of }\mathrm{\Gamma }(U,\text{Alt}_{(๐ฑ,)}^2),\text{ resp. of }\mathrm{\Gamma }(U,\text{Quad}_{(๐ฑ,)})).$$
In terms of the corresponding (geometric) vector bundles over $`X`$, the exact sequence above translates into the following sequence of morphisms of vector bundles, with the first one a closed immersion and the second one a Zariski locally-trivial principal $`\text{Alt}_{(V,I)}^2`$-bundle:
$$\text{Alt}_{(V,I)}^2\text{Bil}_{(V,I)}\text{Quad}_{(V,I)}.$$
Given a quadratic form $`q\mathrm{\Gamma }(U,\text{Quad}_{(๐ฑ,)})`$, recall that the usual โassociatedโ symmetric bilinear form $`b_q\mathrm{\Gamma }(U,\text{Bil}_{(๐ฑ,)})`$ is given on sections (over open subsets of $`U`$) by
$$vv^{}q(v+v^{})q(v)q(v^{}).$$
Given a (not-necessarily symmetric!) bilinear form $`b`$, we also have the induced quadratic form $`q_b`$ given on sections by $`vb(vv).`$ A global quadratic form may not be induced from a global bilinear form, unless we assume something more, for e.g., that the scheme is affine, or more generally that the sheaf cohomology group $`\text{H}^1(X,\text{Alt}_{(๐ฑ,)}^2)=0`$.
### 2.2 Sets of Similarities of Quadratic Bundles
Let $`(V,q,I)`$ and $`(V^{},q^{},I^{})`$ be quadratic bundles on the scheme $`X.`$ We denote by
$$\text{Sim}[(V,q,I),(V^{},q^{},I^{})]$$
the set of generalised similarities from $`(V,q,I)`$ to $`(V^{},q^{},I^{}).`$ These consist of pairs $`(g,m)`$ such that $`g:VV^{}`$ and $`m:II^{}`$ are linear isomorphisms and the following diagram commutes (where $`q`$ and $`q^{}`$ are considered as morphisms of sheaves of sets):
$$\begin{array}{ccc}V& \underset{}{\overset{g}{}}& V^{}\\ q& & q^{}& & \\ I& \underset{m}{\overset{}{}}& I^{}\end{array}$$
When $`I=I^{}`$, since an $`m\text{Aut}(I)`$ may be thought of as multiplication by a scalar $`l\mathrm{\Gamma }(X,๐ช_X^{})\text{Aut}(I)`$, we may call the isomorphism $`(g,m)`$ as an $`I`$-similarity with multiplier $`l.`$ In such a case we may as well denote $`(g,m)`$ by the pair $`(g,l)`$ and we often write
$$g:(V,q,I)_l(V^{},q^{},I).$$
Let
$$\text{Iso}[(V,q,I),(V^{},q^{},I)]$$
be the subset of isometries (i.e., those pairs $`(g,m)`$ with $`m=\text{Identity}`$ or $`I`$-similarities with trivial multipliers). When $`V=V^{}`$, the subset of isometries with trivial determinant is denoted
$$\text{S-Iso}[(V,q,I),(V,q^{},I)].$$
On taking $`q=q^{}`$ these sets naturally become subgroups of
$$\text{Aut}(V)\times \mathrm{\Gamma }(X,๐ช_X^{})=\text{GL}(V)\times \mathrm{\Gamma }(X,๐ช_X^{})$$
and we define
| $`\text{Sim}[(V,q,I),(V,q,I)]=:\text{GO}(V,q,I)\text{Iso}[(V,q,I),(V,q,I)]=:\text{O}(V,q,I)`$ |
| --- |
| $`\text{S-Iso}[(V,q,I),(V,q,I)]=:\text{SO}(V,q,I).`$ |
Of course, $`\text{O}(V,q,I)`$ and $`\text{SO}(V,q,I)`$ may be thought of as subgroups of $`\text{GL}(V)\text{GL}(V)\times \{1\}`$ and $`\text{SL}(V)\text{SL}(V)\times \{1\}`$ respectively.
### 2.3 Semiregular Bilinear forms
Fundamental problems in dealing with quadratic forms over arbitrary base schemes arise essentially from two abnormalities in characteristic two: firstly, the mapping that associates a quadratic form to its symmetric bilinear form is not bijective and secondly, there do not exist regular quadratic forms on any odd-rank bundle. The remedy for this is to consider semiregular quadratic forms, a concept due to M.Kneser and elaborated upon by Knus in , which in fact works over an arbitrary base scheme (and hence in a characteristic-free way) and further reduces to the usual notion of regular form in characteristics $`2.`$
Let $`\text{Spec}(R)=UX`$ be an open affine subscheme of $`X`$ such that $`V|U`$ is trivial. Let $`\text{Quad}_V:=\text{Quad}_{(V,๐ช_X)}.`$ Consider a quadratic form $`q\mathrm{\Gamma }(U,\text{Quad}_V)`$ on $`V|U`$ and its associated symmetric bilinear form $`b_q`$. The matrix of this bilinear form relative to any fixed basis is a symmetric matrix of odd rank and in particular, if $`U`$ is of pure characteristic two (i.e., the residue field of any point of $`U`$ is of characteristic two), then this matrix is also alternating and is hence singular, immediately implying that $`q`$ cannot be regular. However, computing the the determinant of such a matrix in formal variables $`\{\zeta _i,\zeta _{ij}\}`$ shows that it is twice the following polynomial
$$P_3(\zeta _i,\zeta _{ij})=4\zeta _1\zeta _2\zeta _3+\zeta _{12}\zeta _{13}\zeta _{23}(\zeta _1\zeta _{23}^2+\zeta _2\zeta _{13}^2+\zeta _3\zeta _{12}^2).$$
The value $`P_3(q(e_i),b_q(e_i,e_j))`$ corresponding to a basis $`\{e_1,e_2,e_3\}`$ is called the half-discriminant of $`q`$ relative to that basis, and $`q`$ is said to be semiregular if its half-discriminant is a unit. It turns out that this definition is independent of the basis chosen (ยง3, Chap.IV, ).
Even if $`V|U`$ were only projective (i.e., locally-free but not free), the semiregularity of $`q`$ may be defined as the semiregularity of $`q_RR_๐ช`$ for each maximal ideal $`๐ชR`$, and it turns out that with this definition, the notion of a quadratic form being semiregular is local and is well-behaved under base-change (Prop.3.1.5, Chap.IV, ).
We may thus define the subfunctor of $`\text{Quad}_V:=\text{Quad}_{(V,๐ช_X)}`$ of semiregular quadratic forms. This subfunctor is represented by a $`\text{GL}_V`$-invariant open subscheme
$$i:\text{Quad}_V^{sr}\text{Quad}_V$$
because, over each affine open subscheme $`UX`$ which trivialises $`V`$, it corresponds to localisation by the non-zerodivisor $`P_3`$. Note that this canonical open immersion is affine and schematically dominant as well. We next turn to semiregular bilinear forms. Recall that we had defined a bilinear form $`b`$ to be semiregular iff its induced quadratic form $`q_b`$ is semiregular. Thus by definition, $`\text{Bil}_V^{sr}`$ is the fiber product:
$$\begin{array}{ccc}\text{Bil}_V& \stackrel{p}{}& \text{Quad}_V\\ i^{}& & i& & \\ \text{Bil}_V^{sr}& \underset{p^{}}{}& \text{Quad}_V^{sr}\end{array}$$
Since $`p`$ is a Zariski-locally-trivial principal $`\text{Alt}_V^2`$-bundle, it is smooth and surjective (in particular faithfully flat). It therefore follows that the affineness and schematic dominance of $`i`$ imply those of $`i^{}.`$ We record these facts below.
###### Proposition 2.1
The open immersion
$$\text{Bil}_V^{sr}\text{Bil}_V$$
is a $`\text{GL}_V`$-equivariant schematically dominant affine morphism. Further this open immersion behaves well under base-change (relative to $`X`$).
### 2.4 The Generalised Clifford Algebra of Bichsel-Knus
Let $`R`$ be a commutative ring (with 1), $`I`$ an invertible $`R`$-module and $`V`$ a projective $`R`$-module. Consider the Laurent-Rees algebra of $`I`$ defined by
$$L[I]:=R\left(\underset{n>0}{}(T^n(I)T^n(I^1))\right)$$
and define the $``$-gradation on the tensor product of algebras $`TVL[I]`$ by requiring elements of $`V`$ (resp. of $`I`$) to be of degree one (resp. of degree two). Let $`q:VI`$ be an $`I`$-valued quadratic form on $`V.`$ Following the definition of Bichsel-Knus , let $`J(q,I)`$ be the two-sided ideal of $`TVL[I]`$ generated by the set
$$\{(x_{TV}x)1_{L[I]}1_{TV}q(x)|xV\}$$
and let the generalised Clifford algebra of $`q`$ be defined by
$$\stackrel{~}{C}(V,q,I):=TVL[I]/J(q,I).$$
This is an $``$-graded algebra by definition. Let $`C_n`$ be the submodule of elements of degree $`n.`$ Then $`C_0`$ is a subalgebra, playing the role of the even Clifford algebra in the classical situation (i.e., $`I=R`$) and $`C_1`$ is a $`C_0`$-bimodule. Bichsel and Knus baptize $`C_0`$ and $`C_1`$ respectively as the even Clifford algebra and the Clifford module associated to the triple $`(V,q,I).`$ The generalised Clifford algebra satisfies an appropriate universal property which ensures it behaves well functorially. Since $`V`$ is projective, the canonical maps
$$V\stackrel{~}{C}(V,q,I)\text{ and }L[I]\stackrel{~}{C}(V,q,I)$$
are injective. For proofs of these facts, see Sec.3 of . If $`(V,q,I)`$ is an $`I`$-valued quadratic form on the vector bundle $`V`$ over a scheme $`X`$, with $`I`$ a line bundle, then the above construction may be carried out to define the generalised Clifford algebra bundle $`\stackrel{~}{C}(V,q,I)`$ which is an $``$-graded algebra bundle on $`X.`$ Its degree zero subalgebra bundle is denoted $`C_0(V,q,I)`$ and is called the even Clifford algebra bundle of $`(V,q,I).`$
### 2.5 Bourbakiโs Tensor Operations with Values in a Line Bundle
Let $`R`$ be a commutative ring and $`L[I]:=R\left(_{n>0}(T^n(I)T^n(I^1))\right)`$ as above. We denote by $`_T`$ (resp. by $`_L`$) the tensor product and by $`1_T`$ (resp. $`1_L`$) the unit element in the algebra $`TV`$ (resp. in $`L[I]`$).
###### Theorem 2.2 (with the above notations)
Let $`q:VI`$ be an $`I`$-valued quadratic form on $`V`$ and $`f\text{Hom}_R(V,I).`$ Then there exists an $`R`$-linear endomorphism $`t_f`$ of the algebra $`TVL[I]`$ which is unique with respect to the first three of the following properties it satisfies:
for each $`\lambda L[I]`$ we have $`t_f(1_T\lambda )=0;`$
for any $`xV`$, $`yTV`$, and $`\lambda L[I]`$ we have
$$t_f((x_Ty)\lambda )=y(f(x)_L\lambda )(x_T1_T)t_f(y\lambda );$$
if $`J(q,I)`$ is the two-sided ideal of $`TVL[I]`$ as defined in ยง2.4 above, then we have
$$t_f(J(q,I))J(q,I);$$
$`t_f`$ is homogeneous of degree $`+1`$ (except for elements which it does not annihilate);
if $`\stackrel{~}{C}(V,q,I)`$ is the generalised Clifford algebra of $`(V,q,I)`$ as defined in ยง2.4 above, then due to assertion (c) above, $`t_f`$ induces a $``$-graded endomorphism of degree $`+1`$ denoted by
$$d_f^q:\stackrel{~}{C}(V,q,I)\stackrel{~}{C}(V,q,I);$$
$`t_ft_f=0;`$
if $`g\text{Hom}_R(V,I)`$, then $`t_ft_g+t_gt_f=0;`$
if $`\alpha \text{End}_R(V)`$ and $`\alpha ^{}f\text{Hom}_R(V,I)`$ is defined by $`xf(\alpha (x))`$ then
$$t_f(T(\alpha )\text{Id}_{L[I]})=(T(\alpha )\text{Id}_{L[I]})t_{\alpha ^{}f};$$
$`t_f0`$ on the $`R`$-subalgebra of $`TVL[I]`$ generated by
$$\text{kernel}(f)L[I].$$
In fact, atleast when $`V`$ is projective, the smallest $`R`$-subalgebra of $`TVL[I]`$ containing
$$\text{kernel}(f)R.1_L$$
is given by
$$\text{Image}(T(\text{kernel}(f))R.1_L$$
and $`t_f`$ vanishes on this $`R`$-subalgebra.
Let $`q,q^{}:VI`$ be two $`I`$-valued quadratic forms whose difference is the quadratic form $`q_b`$ induced by an $`I`$-valued bilinear form
$$b\text{Bil}_R(V,I):=\text{Hom}_R(V_RV,I).$$
This means that
$$q^{}(x)q(x)=q_b(x):=b(x,x)xV.$$
Further, for any $`xV`$ denote by $`b_x`$ the element of $`\text{Hom}_R(V,I)`$ given by $`yb(x,y).`$ Then there exists an $`R`$-linear automorphism $`\mathrm{\Psi }_b`$ of $`TVL[I]`$ which is unique with respect to the first three of the following properties it satisfies:
for any $`\lambda L[I]`$ we have $`\mathrm{\Psi }_b(1_T\lambda )=(1_T\lambda );`$
for any $`xV`$, $`yTV`$ and $`\lambda L[I]`$ we have
$$\mathrm{\Psi }_b((x_Ty)\lambda )=(x1_L).\mathrm{\Psi }_b(y\lambda )+t_{b_x}(\mathrm{\Psi }_b(y\lambda ));$$
$`\mathrm{\Psi }_b(J(q^{},I))J(q,I);`$
by the previous property, $`\mathrm{\Psi }_b`$ induces an isomorphism of $``$-graded $`R`$-modules
$$\psi _b:\stackrel{~}{C}(V,q^{},I)\stackrel{~}{C}(V,q,I);$$
in particular, given a quadratic form $`q_1:VI`$, since there always exists an $`I`$-valued bilinear form $`b_1`$ that induces $`q_1`$ (i.e., such that $`q_1=q_{b_1}`$), setting $`q^{}=q_1`$, $`q=0`$ and $`b=b_1`$ in the above gives an $``$-graded linear isomorphism
$$\psi _{b_1}:\stackrel{~}{C}(V,q_1,I)\stackrel{~}{C}(V,0,I)=\mathrm{\Lambda }(V)L[I];$$
$`\mathrm{\Psi }_b(T^{2n}VL[I])_{(in)}(T^{2i}VL[I]);`$
$`\mathrm{\Psi }_b(T^{2n+1}VL[I])_{(\text{odd }i2n+1)}(T^iVL[I]);`$
$`\mathrm{\Psi }_b(T^{2n}VI^n)_{(in)}(T^{2i}VI^i);`$
$`\mathrm{\Psi }_b(T^{2n+1}VI^n)_{(\text{odd }i2n+1)}(T^iVI^{\frac{1i}{2}});`$
in particular, for $`x,x^{}V`$,
$$\mathrm{\Psi }_b((x_Tx^{})1_L)=(x_Tx^{})1_L+1_Tb(x,x^{})$$
so that for
$$\psi _b:C_0(V,q_b,I)C_0(V,0,I)=_{n0}(\mathrm{\Lambda }^{2n}(V)I^n)$$
we have
$$\psi _b(((x_Tx^{})\zeta )\text{ mod }J(q_b,I))=(xx^{})\zeta +\zeta (b(x,x^{})).1$$
for any $`x,x^{}V`$ and $`\zeta I^1\text{Hom}_R(I,R);`$
if $`f\text{Hom}_R(V,I)`$, and $`t_f`$ is given by (1) above, then $`\mathrm{\Psi }_bt_f=t_f\mathrm{\Psi }_b;`$
for $`I`$-valued bilinear forms $`b_i`$ on $`V`$,
$$\mathrm{\Psi }_{b_1+b_2}=\mathrm{\Psi }_{b_1}\mathrm{\Psi }_{b_2}\text{ and }\mathrm{\Psi }_0=\text{Identity on }TVL[I];$$
for any $`\alpha \text{End}_R(V)`$, we have
$$\mathrm{\Psi }_b(T(\alpha )\text{Id}_{L[I]})=(T(\alpha )\text{Id}_{L[I]})\mathrm{\Psi }_{(b.\alpha )}$$
where $`(b.\alpha )(x,x^{}):=b(\alpha (x),\alpha (x^{}))x,x^{}V;`$
by property (h), one has a homomorphism of groups
$$(\text{Bil}_R(V,I),+)(\text{Aut}_R(TVL[I]),):b\mathrm{\Psi }_b;$$
the associative unital monoid $`(\text{End}_R(V),)`$ acts on $`\text{Bil}_R(V,I)`$ on the right by $`b^{}b^{}.\alpha `$ and acts on the left (resp. on the right) of $`\text{End}_R(TVL[I])`$ by
$$\alpha .\mathrm{\Phi }:=(T(\alpha )\text{Id}_{L[I]})\mathrm{\Phi }\text{ (resp. by }\mathrm{\Phi }.\alpha :=\mathrm{\Phi }(T(\alpha )\text{Id}_{L[I]})\text{ )},$$
and the homomorphism $`b\mathrm{\Psi }_b`$ satisfies
$$\alpha .\mathrm{\Psi }_{(b.\alpha )}=\mathrm{\Psi }_b.\alpha ;$$
the group $`\text{Aut}_R(V)=\text{GL}_R(V)`$ acts on the left of $`\text{Bil}_R(V,I)`$ by
$$g.b:(x,x^{})b(g^1(x),g^1(x^{}))$$
and on the left of $`\text{Aut}_R(TVL[I])`$ by conjugation via the natural group homomorphism
$$\text{GL}_R(V)\text{Aut}_R(TVL[I]):gT(g)\text{Id}_{L[I]}$$
namely we have
$$g.\mathrm{\Phi }:=(T(g)\text{Id}_{L[I]})\mathrm{\Phi }(T(g^1)\text{Id}_{L[I]}),$$
and the homomorphism $`b\mathrm{\Psi }_b`$ is $`\text{GL}_R(V)`$-equivariant: $`\mathrm{\Psi }_{g.b}=g.\mathrm{\Psi }_b.`$
For a commutative $`R`$-algebra $`S`$ (with 1), let
$$(q_RS),(q^{}_RS):(V_RS=:V_S)(I_RS=:I_S)$$
be the $`I_S`$-valued quadratic forms induced from the quadratic forms $`q,q^{}`$ of (2) above and
$$(b_RS)\text{Bil}_S(V_S,I_S)$$
the $`I_S`$-valued $`S`$-bilinear form induced from the bilinear form $`b`$ of (2) above. Then as a result of the uniqueness properties (2a)โ(2c) satisfied by $`\mathrm{\Psi }_b`$ and $`\mathrm{\Psi }_{(b_RS)}`$, the $`S`$-linear automorphisms $`(\mathrm{\Psi }_b_RS)`$ and $`\mathrm{\Psi }_{(b_RS)}`$ may be canonically identified. In particular, the $``$-graded $`S`$-linear isomorphism
$$(\psi _b_RS):\stackrel{~}{C}(V_S,(q^{}_RS),I_S)\stackrel{~}{C}(V_S,(q_RS),I_S)$$
induced from $`\psi _b`$ of (2d) above may be canonically identified with $`\psi _{(b_RS)}.`$
###### Remark 2.3
As mentioned in , F. van Oystaeyen has observed that $`L[I]`$ is a faithfully-flat splitting for $`I`$, and the generalised Clifford algebra $`\stackrel{~}{C}(V,q,I)`$ is nothing but the โclassicalโ Clifford algebra of the triple
$$(V_RL[I],q_RL[I],I_RL[I]L[I])$$
over $`L[I].`$ In the same vein, $`I`$-valued forms (both multilinear and quadratic) on an $`R`$-module $`V`$ can be treated as the usual ($`L[I]`$-valued) forms on $`V_RL[I].`$ With this in mind, the proof of Theorem 2.2 follows from the usual Bourbaki tensor operations with respect to $`V_RL[I]`$ on $`L[I].`$ (See ยง9, Chap.9, or para.1.7, Chap.IV, for the โclassicalโ Bourbaki operations). However one needs to remember that the $``$-gradation on $`TV_RL[I]`$ as defined above is different from the usual $`_0`$-gradation on $`T(V_RL[I]).`$
### 2.6 Tensoring by Bilinear- and Twisted Discriminant Bundles
Let $`V,M`$ be vector bundles on a scheme $`X`$ and let $`I,J`$ be line bundles on $`X.`$ Let $`q`$ be a quadratic form on $`V`$ with values in $`I`$ and let $`b`$ be a symmetric bilinear form on $`M`$ with values in $`J.`$ By abuse of notation, we also use $`b`$ to denote the corresponding $`J`$-valued linear form on $`MM.`$
###### Proposition 2.4 (with the above notations)
The tensor product of $`(V,q,I)`$ with $`(M,b,J)`$ gives a unique quadratic bundle
$$(VM,qb,IJ).$$
The quadratic form $`qb`$ on $`VM`$ is given on sections by
$$vmq(v)b(mm)$$
and has associated bilinear form $`b_{qb}=b_qb.`$
When $`M`$ is a line bundle, $`(M,b,J)`$ is regular (=nonsingular) iff $`(M,q_b,J)`$ is semiregular iff
$$b:MMJ$$
is an isomorphism.
Let $`V`$ be of odd rank and $`M`$ a line bundle such that $`(M,b,J)`$ is regular. Then $`(V,q,I)`$ is semiregular iff
$$(V,q,I)(M,b,J)=(VM,qb,IJ)$$
is semiregular.
###### Definition 2.5
A triple $`(L,h,J)`$, consisting of a linear isomorphism $`h:LLJ`$, with $`L`$ and $`J`$ both line bundles, is called a twisted discriminant bundle on $`X.`$
A twisted discriminant bundle $`(L,h,J)`$ specifies $`L`$ as a square root of the line bundle $`J`$ via $`h`$. The terminology is motivated by the following: when $`J`$ is the trivial line bundle, such a datum is referred to as a discriminant bundle in ยง3, Chap.III, of Knusโ book . By part (2) of the preceding Proposition, $`h`$ is a regular bilinear form on $`L`$ (necessarily symmetric) with values in $`J`$, so that we may speak of an isometry between two twisted discriminant bundles $`(L,h,J)`$ and $`(L^{},h^{},J^{})`$: it is a pair $`(\zeta ,\eta )`$ consisting of linear isomorphisms $`\zeta :LL^{}`$ and $`\eta :JJ^{}`$ such that $`\eta h=h^{}(\zeta \zeta ).`$
###### Lemma 2.6
On the set $`\text{T-Disc}(X)`$ of isometry classes of twisted discriminant bundles on $`X`$, we have a natural group structure induced by the tensor product. $`\text{T-Disc}(X)`$ is functorial in $`X.`$ If we consider only isometry classes of discriminant bundles, i.e., of triples $`(L,h,๐ช_X)`$, then we obtain a subgroup $`\text{Disc}(X)\text{T-Disc}(X)`$ which is of exponent 2.
###### Proposition 2.7
Let $`V`$ and $`V^{}`$ be vector bundles of the same rank on the scheme $`X`$, $`(L,h,J)`$ a twisted discriminant line bundle on $`X`$ and
$$\alpha :V^{}VL$$
an isomorphism of bundles.
Over any open subset $`UX`$, given a bilinear form
$$b^{}\mathrm{\Gamma }(U,\text{Bil}_{(V^{},I)}),$$
we can define a bilinear form
$$b\mathrm{\Gamma }(U,\text{Bil}_{(V,IJ^1)})$$
using $`\alpha `$ and $`h`$ as follows: we let
$$b:=(b^{}J^1)(\zeta _{(\alpha ,h)})^1$$
where
$$\zeta _{(\alpha ,h)}:V^{}V^{}J^1VV$$
is the linear isomorphism given by the composition of the following natural morphisms:
| $`V^{}V^{}J^1\stackrel{\alpha \alpha \text{Id}()}{}VLVLJ^1\stackrel{\text{SWAP}(2,3)()}{}`$ |
| --- |
| $`VVL^2J^1\stackrel{\text{Id}h\text{Id}()}{}`$ |
| $`\stackrel{\text{Id}h\text{Id}()}{}VVJJ^1\stackrel{\text{CANON}()}{}VV.`$ |
Then the association $`b^{}b`$ induces linear isomorphisms shown by vertical upward arrows in the following diagram of associated locally-free sheaves (with exact rows) making it commutative:
$$\begin{array}{ccccccccc}0& & \text{Alt}_{(๐ฑ,๐ฅ^1)}^2& & \text{Bil}_{(๐ฑ,๐ฅ^1)}& & \text{Quad}_{(๐ฑ,๐ฅ^1)}& & 0\\ & & & & & & & & & & \\ 0& & \text{Alt}_{(๐ฑ^{},)}^2& & \text{Bil}_{(๐ฑ^{},)}& & \text{Quad}_{(๐ฑ^{},)}& & 0.\end{array}$$
Therefore one also has the following commutative diagram of vector bundle morphisms with the vertical upward arrows being isomorphisms:
$$\begin{array}{ccccc}\text{Alt}_{(V,IJ^1)}^2& \underset{\text{immersion}}{\overset{\text{closed}}{}}& \text{Bil}_{(V,IJ^1)}& \underset{\text{trivial}}{\overset{\text{locally}}{}}& \text{Quad}_{(V,IJ^1)}\\ & & & & & & \\ \text{Alt}_{(V^{},I)}^2& \underset{\text{immersion }}{\overset{\text{closed}}{}}& \text{Bil}_{(V^{},I)}& \underset{\text{trivial}}{\overset{\text{locally}}{}}& \text{Quad}_{(V^{},I)}\end{array}$$
Let $`b^{}\mathrm{\Gamma }(X,\text{Bil}_{(V^{},I)})`$ be a global bilinear form and let it induce
$$b\mathrm{\Gamma }(X,\text{Bil}_{(V,IJ^1)})$$
via $`\alpha `$ and $`h`$ as defined in (1) above. Let
$$\mathrm{\Psi }_b^{}\text{Aut}_{๐ช_X}(TV^{}L[I])\text{ (resp. }\mathrm{\Psi }_b\text{Aut}_{๐ช_X}(TVL[IJ^1]))$$
be the $``$-graded linear isomorphism induced by $`b^{}`$ (resp. by $`b`$) defined locally (and hence globally) as in (2), Theorem 2.2 above. Let
$$Z_{(\alpha ,h)}:_{n0}(T_{๐ช_X}^{2n}(V^{})I^n)_{n0}(T_{๐ช_X}^{2n}(V)I^nJ^n)$$
be the $`๐ช_X`$-algebra isomorphism induced via the isomorphism $`\zeta _{(\alpha ,h)}`$ defined in (1) above. Then, taking into account (2e), Theorem 2.2, the following diagram commutes:
$$\begin{array}{ccc}_{n0}(T_{๐ช_X}^{2n}(V^{})I^n)& \underset{}{\overset{Z_{(\alpha ,h)}}{}}& _{n0}(T_{๐ช_X}^{2n}(V)I^nJ^n)\\ \mathrm{\Psi }_b^{}& & \mathrm{\Psi }_b& & \\ _{n0}(T_{๐ช_X}^{2n}(V^{})I^n)& \underset{Z_{(\alpha ,h)}}{\overset{}{}}& _{n0}(T_{๐ช_X}^{2n}(V)I^nJ^n)\end{array}$$
thereby inducing by (2d), Theorem 2.2 the following commutative diagram of $`๐ช_X`$-linear isomorphisms
$$\begin{array}{ccc}C_0(V^{},q_b^{},I)& \underset{}{\overset{\text{via }Z_{(\alpha ,h)}}{}}& C_0(V,q_b,IJ^1)\\ \psi _b^{}& & \psi _b& & \\ _{n0}(\mathrm{\Lambda }_{๐ช_X}^{2n}(V^{})I^n)& \underset{\text{via }Z_{(\alpha ,h)}}{\overset{}{}}& _{n0}(\mathrm{\Lambda }_{๐ช_X}^{2n}(V)I^nJ^n)\end{array}$$
Let $`b`$ and $`b^{}`$ be as in (2) above. Then $`\alpha :V^{}VL`$ induces an isometry of bilinear form bundles
$$\alpha :(V^{},b^{},I)(V,b,IJ^1)(L,h,J)$$
and also an isometry of the induced quadratic bundles
$$\alpha :(V^{},q_b^{},I)(V,q_b,IJ^1)(L,h,J).$$
Moreover, if we are just given a global $`IJ^1`$-valued quadratic form $`q`$ on $`V`$ (resp. an $`I`$-valued $`q^{}`$ on $`V^{}`$), then we may define the global $`I`$-valued quadratic form $`q^{}`$ on $`V^{}`$ (resp. $`IJ^1`$-valued $`q`$ on $`V`$) via
$$q^{}:=(qh)\alpha \text{ (resp. via }q:=(q^{}\alpha ^1)(h^{})^1)$$
and again we have an isometry of quadratic bundles
$$\alpha :(V^{},q^{},I)(V,q,IJ^1)(L,h,J).$$
###### Proposition 2.8
Let $`g:(V,q,I)_l(V^{},q^{},I)`$ be an $`I`$-similarity with multiplier $`l\mathrm{\Gamma }(X,๐ช_X^{}).`$
There exists a unique isomorphism of $`๐ช_X`$-algebra bundles
$$C_0(g,l,I):C_0(V,q,I)C_0(V^{},q^{},I)$$
such that for sections $`v,v^{}`$ of $`V`$ and $`s`$ of $`I^1`$ we have
$$C_0(g,l,I)(v.v^{}.s)=g(v).g(v^{}).l^1s.$$
There exists a unique vector bundle isomorphism
$$C_1(g,l,I):C_1(V,q,I)C_1(V^{},q^{},I)$$
such that the following hold for any section $`v`$ of $`V`$ and any section $`c`$ of $`C_0(V,q)`$:
$`C_1(g,l,I)(v.c)=g(v).C_0(g,l,I)(c)`$ and
$`C_1(g,l,I)(c.v)=C_0(g,l,I)(c).g(v).`$
Thus $`C_1(g,l,I)`$ is $`C_0(g,l,I)`$-semilinear.
If $`g_1:(V^{},q^{},I)_{l_1}(V^{\prime \prime },q^{\prime \prime },I)`$ is another similarity with multiplier $`l_1`$, then the composition
$$g_1g:(V,q,I)_{ll_1}(V^{\prime \prime },q^{\prime \prime },I)$$
is also a similarity with multiplier given by the product of the multipliers. Further
$$C_i(g_1g,ll_1,I)=C_i(g_1,l_1,I)C_i(g,l,I)\text{ for }i=0,1.$$
A local computation shows that tensoring by a twisted discriminant bundle is locally the same as applying a similarity. In this case also one gets a global isomorphism of even Clifford algebras:
###### Proposition 2.9
Let $`(V,q,I)`$ be a quadratic bundle on a scheme $`X`$ and $`(L,h,J)`$ be a twisted discriminant bundle. There exists a unique isomorphism of algebra bundles
$$\gamma _{(L,h,J)}:C_0\left((V,q,I)(L,h,J)\right)C_0(V,q,I)$$
such that for any sections $`v,v^{}`$ of $`V`$, $`\lambda ,\lambda ^{}`$ of $`L`$, $`s`$ of $`I^1I^{}`$ and $`t`$ of $`J^1J^{}`$ we have
$$\gamma _{(L,h,J)}((v\lambda ).(v^{}\lambda ^{}).(st))=t(h(\lambda \lambda ^{}))v.v^{}.s.$$
### 2.7 The Theorem of Max-Albert Knus
For any scheme $`X,`$ denote by $`๐ฌ_3^{sr}(X;๐ช_X)`$ the set of isomorphism classes of semiregular ternary quadratic bundles with values in the trivial line bundle on $`X.`$ Here isomorphism stands for isometry as defined in ยง2.2. Recall the group $`\text{Disc}(X)`$ of discriminant bundles on $`X`$ (Lemma 2.6). By assertions (2) and (3) of Prop. 2.4, this group acts on $`๐ฌ_3^{sr}(X;๐ช_X).`$
For a ternary quadratic form $`q:V๐ช_X,`$ recall from ยง2.4 the Bichsel-Knus even-Clifford algebra $`C_0(V,q,๐ช_X)`$, namely the degree zero subalgebra of the generalised Clifford algebra $`\stackrel{~}{C}(V,q,๐ช_X).`$ Since the values of the form are in the trivial line bundle, this even-Clifford algebra is the same as the the classically-defined even-Clifford algebra.
Now the even-Clifford algebra of a semiregular form is Azumaya (see for instance, Prop.3.2.4, ยง3, Chap.IV, ). So if we let $`๐๐ต๐ฐ_4(X)`$ to denote the set of algebra-isomorphism classes of rank 4 Azumaya bundles over $`X.`$ then by Prop. 2.9, the association $`(V,q,๐ช_X)C_0(V,q,๐ช_X)`$ induces a map
$$\text{Witt-Invariant}_{(X;๐ช_X)}^{sr}:๐ฌ_3^{sr}(X;๐ช_X)/\text{Disc}(X)๐๐ต๐ฐ_4(X),$$
where the left side represents the set of orbits.
###### Theorem 2.10 (Max-Albert Knus, ยง 3, Chap.V, )
For any scheme $`X`$,
the map $`\text{Witt-Invariant}_{(X;๐ช_X)}^{sr}`$ defined above is a bijection.
By Prop.3.2.2, ยง3, Chap.III, , the group $`\text{Disc}(X)`$ is naturally isomorphic to the cohomology (abelian group) $`\text{ศ}_{\text{fppf}}^1(X,\mu _2).`$ Further, by Lemma 3.2.1, ยง3, Chap.IV, , the cohomology $`\text{ศ}_{\text{fppf}}^1(X,\text{O}_3)`$ classifies the set of isomorphism classes of semiregular rank 3 quadratic bundles with values in the trivial line bundle, so it is the same as $`๐ฌ_3(X;๐ช_X).`$ On the other hand, the set of isomorphism classes of rank 4 Azumaya algebras on $`X`$ may be interpreted as the cohomology $`\text{ศ}_{\text{รฉtale}}^1(X,\text{PGL}_2)`$ (see page 145, ยง5, Chap.III, ). Thus the bijection of Theorem 2.10 can be thought of as a statement in cohomology:
$$\text{ศ}_{\text{fppf}}^1(X,\text{O}_3)/\text{ศ}_{\text{fppf}}^1(X,\mu _2)\text{ศ}_{\text{รฉtale}}^1(X,\text{PGL}_2).$$
### 2.8 Results on the notion of Schematic Image
###### Definition 2.11 (Defs.6.10.1-2, Chap.I, EGA I )
Let $`f:XY`$ be a morphism of schemes. If there exists a smallest closed subscheme $`Y^{}Y`$ such that the inverse image scheme
$$f^1(Y^{}):=Y^{}\times _Y(_fX)$$
is equal to $`X`$, one calls $`Y^{}`$ the schematic image of $`f`$ (or of $`X`$ in $`Y`$ under $`f`$). If $`X`$ were a subscheme of $`Y`$ and $`f`$ the canonical immersion, and if $`f`$ has a schematic image $`Y^{}`$, then $`Y^{}`$ is called the schematic limit or the limiting scheme of the subscheme $`X\stackrel{f}{}Y.`$
###### Proposition 2.12 (Prop.6.10.5, Chap.I, EGA I)
The schematic image $`Y^{}`$ of $`X`$ by a morphism $`f:XY`$ exists in the following two cases:
$`f_{}(๐ช_X)`$ is a quasi-coherent $`๐ช_Y`$-module, which is for example the case when $`f`$ is quasi-compact and quasi-separated;
$`X`$ is reduced.
###### Proposition 2.13
In each of the following statements whenever a schematic image is mentioned, we assume that one of the two hypotheses of the above Prop. 2.12 is true so that the schematic image does exist.
Let $`Y^{}`$ be the schematic image of $`X`$ under a morphism $`f:XY`$ and let $`f`$ factor as
$$X\stackrel{g}{}Y^{}\stackrel{j}{}Y.$$
Then $`Y^{}`$ is topologically the closure of $`f(X)`$ in $`Y`$, the morphism $`g`$ is schematically dominant i.e.,
$$g^\mathrm{\#}:๐ช_Y^{}g_{}(๐ช_X)$$
is injective and the schematic image of $`X`$ in $`Y^{}`$ (under $`g`$) is $`Y^{}`$ itself. If $`X`$ is reduced (respectively integral) then the same is true of $`Y^{}.`$
The schematic image of a closed subscheme under its canonical closed immersion is itself.
(Transitivity of Schematic Image.) Let there be given morphisms
$$X\stackrel{f}{}Y\stackrel{g}{}Z,$$
such that the schematic image $`Y^{}`$ of $`X`$ under $`f`$ exists, and further such that if $`g^{}`$ is the restriction of $`g`$ to $`Y^{}`$, the schematic image $`Z^{}`$ of $`Y^{}`$ by $`g^{}`$ exists. Then the schematic image of $`X`$ under $`gf`$ exists and equals $`Z^{}.`$
Let $`f:XY`$ be a morphism which factors through a closed subscheme $`Y_1`$ of $`Y`$ by a morphism $`f_1:XY_1.`$ Then the scheme-theoretic image $`Y^{}`$ of $`X`$ in $`Y`$ is the same as the scheme-theoretic image $`Y_1^{}`$ of $`X`$ in $`Y_1`$ considered canonically as closed subscheme of $`Y.`$
If $`f:XY`$ has a schematic image $`Y^{}`$ then $`f`$ is schematically dominant iff $`Y^{}=Y.`$
The formation of schematic image commutes with flat base change: if $`f:XY`$ is a morphism of $`S`$-schemes which has a schematic image $`Y^{}`$ then for a flat morphism $`S^{}S`$, one has that the induced $`S^{}`$-morphism
$$f\times _SS^{}:X\times _SS^{}Y\times _SS^{}$$
has a schematic image and it may be canonically identified with $`Y^{}\times _SS^{}.`$ In particular this means that the formation of schematic image is local over the base.
Assertions (1) and (3) are respectively Prop.6.10.5 and Prop.6.10.3 in EGA I. The defining property of schematic image gives (2), while (4) can be deduced from the first three. As for (5), from (1) it follows that $`Y^{}=Y`$ implies $`f=g`$ is schematically dominant. For the other way around, one uses the following characterisation of a schematically dominant morphism in Prop.5.4.1 of EGA I: if $`f:XY`$ is a morphism of schemes, then $`f`$ is schematically dominant iff for every open subscheme $`U`$ of $`Y`$ and every closed subscheme $`Y_1`$ of $`U`$ such that there exists a factorisation
$$f^1(U)\stackrel{g_1}{}Y_1\stackrel{j_1}{}U,$$
of the restriction $`f^1(U)U`$ of $`f`$ (where $`j_1`$ is the canonical closed immersion), one has $`Y_1=U.`$ Given $`f`$ is schematically dominant, one just has to take $`U=Y`$, $`Y_1=Y^{}`$ and $`g_1=g.`$ Assertion (6) follows from statement (ii) (a) of Theorem 11.10.5 of EGA IV .
### 2.9 Specialisations of Rank 4 Azumaya Algebras: Recap
Until further notice we assume that $`W`$ is a vector bundle of fixed positive rank on the scheme $`X`$ with associated coherent locally-free sheaf $`๐ฒ.`$ Given any $`X`$-scheme $`T`$, by a $`T`$-algebra structure on $`W_T:=W\times _XT`$ (also referred to as $`T`$-algebra bundle), we mean a morphism
$$W_T\times _TW_TW_T$$
of vector bundles on $`T`$ arising from a morphism of the associated locally-free sheaves. Given such a $`T`$-algebra structure and $`T^{}T`$ an $`X`$-morphism, it is clear that one gets by pullback (i.e., by base-change) a canonical $`T^{}`$-algebra structure on $`W_T^{}.`$ Thus one has a contravariant โfunctor of algebra structures on $`W`$โ from $`\{XSchemes\}`$ to $`\{Sets\}`$ denoted $`\text{Alg}_W`$ with
$$\text{Alg}_W(T)=\{T\text{algebra structures on }W_T\}=\text{Hom}_{๐ช_T}(๐ฒ_T๐ฒ_T,๐ฒ_T).$$
It follows from Prop.9.6.1, Chap.I of EGA I that the functor $`\text{Alg}_W`$ is represented by the $`X`$-scheme
$$\text{Alg}_W:=\text{Spec }\left(\text{Sym}_X\left[\left(๐ฒ_{X}^{}{}_{}{}^{}_X๐ฒ_{X}^{}{}_{}{}^{}_X๐ฒ_X\right)^{}\right]\right).$$
Hence $`\text{Alg}_W`$ is affine (therefore separated), of finite presentation over $`X`$ and in fact smooth of relative dimension $`\text{rank}_X(W)^3`$. If $`X^{}X`$ is an extension of base, then the construction $`\text{Alg}_W`$ base-changes well i.e., one may canonically identify $`\text{Alg}_W\times _XX^{}`$ with $`\text{Alg}_W^{}`$ where $`W^{}=W\times _XX^{}`$ (cf. Prop.9.4.11, Chap.I, EGA I ).
The general linear groupscheme associated to $`W`$ viz $`\text{GL}_W`$ naturally acts on $`\text{Alg}_W`$ on the left, so that for each $`X`$-scheme $`T`$, $`\text{Alg}_W(T)`$ mod $`\text{GL}_W(T)`$ is the set of isomorphism classes of $`T`$-algebra structures on $`W\times _XT`$.
We remark that an algebra structure may fail to be associative and may fail to have a (two-sided) identity element for multiplication. However, a multiplicative identity for an associative algebra structure must be a nowhere vanishing section (Lemma 2.3, and (2)$``$(4) of Lemma 2.4, Part A, ).
Let $`w\mathrm{\Gamma }(X,๐ฒ)`$ be a nowhere vanishing section. For any $`X`$-scheme $`T`$, let $`\text{Assoc}_{W,w}(T)`$ denote the subset of $`\text{Alg}_W(T)`$ consisting of associative algebra structures with multiplicative identity the nowhere vanishing section $`w_T`$ over $`T`$ induced from $`w`$. Thus we obtain a contravariant subfunctor $`\text{Assoc}_{W,w}`$ of $`\text{Alg}_W.`$
Let $`\text{Stab}_w(T)\text{GL}_W(T)`$ denote the stabiliser subgroup of $`w_T`$, so that one gets a subfunctor in subgroups $`\text{Stab}_w\text{GL}_W.`$ It is in fact represented by a closed subgroupscheme (also denoted by) $`\text{Stab}_w`$ and further behaves well under base change relative to $`X`$ i.e., $`\text{Stab}_w\times _XT`$ can be canonically identified with $`\text{Stab}_{w_T}`$ for any $`X`$-scheme $`T.`$ These follow from para 9.6.6 of Chap.I, EGA I .
It is clear that the natural action of $`\text{GL}_W`$ on $`\text{Alg}_W`$ induces one of $`\text{Stab}_w`$ on $`\text{Assoc}_{W,w}.`$ It is easy to check (p.489, Part A, ) that the functor $`\text{Assoc}_{W,w}`$ is a sheaf in the big Zariski site over $`X`$ and further that this functor is represented by a natural closed subscheme
$$\text{Assoc}_{W,w}\text{Alg}_W$$
which is $`\text{Stab}_w`$-invariant. Further the construction $`\text{Assoc}_{W,w}`$ behaves well with respect to base-change (relative to $`X`$). Consider the subfunctor
$$\text{Azu}_{W,w}\text{Assoc}_{W,w}$$
corresponding to Azumaya algebras.
###### Theorem 2.14 (Theorem 3.4, Part A, )
$`\text{Azu}_{W,w}`$ is represented by a $`\text{Stab}_w`$-stable open subscheme
$$\text{Azu}_{W,w}\text{Assoc}_{W,w}$$
and the canonical open immersion is an affine morphism.
$`\text{Azu}_{W,w}`$ is affine (hence separated) and of finite presentation over $`X`$, and $`\text{Azu}_{W,w}`$ behaves well with respect to base-change (relative to $`X`$).
Further, $`\text{Azu}_{W,w}`$ is smooth of relative dimension $`(m^21)^2`$ and geometrically irreducible over $`X`$, where $`m^2:=\text{rank}_X(๐ฒ)`$.
###### Theorem 2.15 (Theorem 3.8, Part A, )
The open immersion $`\text{Azu}_{W,w}\text{Assoc}_{W,w}`$ has a schematic image denoted
$$\text{SpAzu}_{W,w}\text{Assoc}_{W,w}$$
which is affine (hence separated) and of finite type over $`X`$ and is naturally a $`\text{Stab}_w`$-stable closed subscheme of $`\text{Assoc}_{W,w}`$, the action extending the natural one on the open subscheme $`\text{Azu}_{W,w}`$.
When the rank of $`W`$ over $`X`$ is 4, $`\text{SpAzu}_{W,w}`$ is locally (over $`X`$) isomorphic to relative 9-dimensional affine space; in fact over every open affine subscheme $`U`$ of $`X`$ where $`W`$ becomes trivial and $`w`$ becomes part of a global basis we have
$$\text{SpAzu}_{W,w}|_U๐ธ_U^9.$$
For the explicit isomorphism, see Theorem 3.10, page 3.10. Thus $`\text{SpAzu}_{W,w}`$ is smooth of relative dimension 9 and geometrically irreducible over $`X.`$ In particular, it is of finite presentation over $`X.`$
When $`\text{rank}_X(W)=4`$, the construction $`\text{SpAzu}_{W,w}X`$ behaves well with respect to base change (relative to $`X`$).
## 3 Statements of the Main Results
The theory of semiregular/regular quadratic forms of low rank is well-known, as in Chap.V of Knusโ book . This theory is satisfactory for such forms, since it classifies them in terms of various invariants and also gives information on the groups of generalised similarities, isometries and special isometries of such forms. It is natural to look for a corresponding theory for limiting or degenerate forms. We formulate in the following such a theory for the case of ternary quadratic forms.
### 3.1 A Limiting Version of a Theorem in Cohomology
For any scheme $`X`$, denote by $`๐ฌ_3(X)`$ (respectively, by $`๐ฌ_3^{sr}(X)`$) the set of isomorphism classes of line-bundle-valued ternary quadratic bundles (respectively line-bundle-valued semiregular ternary quadratic bundles) on $`X.`$ Here isomorphism stands for isometry as defined in ยง2.2. Consider the group T-Disc(X) of twisted discriminant bundles on $`X`$ (cf. Lemma 2.6). By Prop. 2.4, this group acts on $`๐ฌ_3(X)`$ and on the subset $`๐ฌ_3^{sr}(X).`$
For a line-bundle-valued quadratic form $`(V,q,I)`$, recall from ยง2.4 the Bichsel-Knus even-Clifford algebra $`C_0(V,q,I),`$ which is the degree zero subalgebra of the generalised Clifford algebra $`\stackrel{~}{C}(V,q,I)`$ and reduces to the usual even Clifford algebra for a quadratic form with values in the structure sheaf.
Let $`๐ฎ๐ซ๐๐ต๐ฐ_4(X)`$ (respectively, $`๐๐ต๐ฐ_4(X)`$) denote the set of isomorphism classes of associative unital algebra structures on vector bundles of rank 4 over $`X`$ that are Zariski-locally isomorphic to even-Clifford algebras of rank 3 quadratic bundles (respectively, that are Azumaya). Recall that the Theorem 2.10 (page 2.10) of Max-Albert Knus gives a bijection
$$\text{Witt-Invariant}_{(X;๐ช_X)}^{sr}:๐ฌ_3^{sr}(X;๐ช_X)/\text{Disc(X)}\stackrel{}{}๐๐ต๐ฐ_4(X).$$
It follows that $`๐๐ต๐ฐ_4(X)๐ฎ๐ซ๐๐ต๐ฐ_4(X).`$ By Prop. 2.9, the association $`(V,q,I)C_0(V,q,I)`$ induces a map
$$\text{Witt-Invariant}_X:๐ฌ_3(X)/\text{T-Disc}(X)๐ฎ๐ซ๐๐ต๐ฐ_4(X),$$
where the left side represents the set of orbits. Since the even-Clifford algebra of a semiregular quadratic module is Azumaya, it follows that the above map restricts to a map:
$$\text{Witt-Invariant}_X^{sr}:๐ฌ_3^{sr}(X)/\text{T-Disc}(X)๐๐ต๐ฐ_4(X).$$
###### Theorem 3.1
For any scheme $`X`$, both the map $`\text{Witt-Invariant}_X`$ as well as its restriction $`\text{Witt-Invariant}_X^{sr}`$ are bijections.
Thus the above Theorem 3.1 may be viewed as a โlimiting versionโ of the Theorem 2.10 of Knus which as noted in page 2.10 may be interpreted as a statement in cohomology.
We shall see later ((b1), Theorem 3.6) that it is necessary to consider line-bundle-valued quadratic forms to obtain the surjectivity of Theorem 3.1 in those cases for which a given $`A`$ representing an element in $`๐ฎ๐ซ๐๐ต๐ฐ_4(X)`$ is such that $`\text{det}(A)2.\text{Pic}(X).`$
It was shown in Part A, that algebra bundles belonging to $`๐ฎ๐ซ๐๐ต๐ฐ_4(X)`$ are precisely the scheme-theoretic specialisations (or limits) of rank 4 Azumaya algebra bundles on $`X`$ (Theorem 2.15).
Thus one may also restate the surjectivity as schematic specialisations of rank 4 Azumaya bundles arise as even-Clifford algebras of ternary quadratic bundles and the injectivity as follows: if the even-Clifford algebras of two ternary quadratic bundles are isomorphic, then the quadratic bundles are isometric upto tensoring by a twisted discriminant bundle.
The main result of was the smoothness of the schematic closure of Azumaya algebra structures on a fixed vector bundle of rank 4 over any scheme. Part B of had applied this result to obtain the generalised Seshadri desingularisation of the moduli space of semistable rank 2 degree zero vector bundles on a smooth proper curve relative to a locally universally-japanese (Nagata) base scheme and also to obtain the generalised Nori desingularisation of the Artin moduli space of invariants of several matrices in rank 2 over such a base scheme together with good specialisation properties over $`.`$ The present work is concerned with applications to ternary quadratic forms and the results obtained follow from an analysis of the computations that lead to the smoothness.
The good algebraic properties of Azumaya algebras are reflected as good geometric properties of the scheme of Azumaya algebra structures on a fixed vector bundle: this scheme is separated, of finite type and smooth relative to the base scheme (over which the vector bundle is fixed) and also base-changes well relative to the base scheme (Theorem 2.14, page 2.14). When the vector bundle is of rank 4, the nice thing that happens is that all these good properties also pass over to the limit i.e., to the scheme of specialisations, defined to be the schematic image of the scheme of Azumaya algebra structures (Theorem 2.15, page 2.15.) In the same vein, the present work shows that the theories of rank 3 semiregular quadratic forms and of rank 4 Azumaya algebras and their inter-relationships extend to the limit.
### 3.2 Study of Groups of Similitudes
A bilinear form $`b`$, with values in a line bundle $`I`$, defined on a vector bundle $`V`$ over the scheme $`X`$ induces an $`I`$-valued quadratic form $`q_b`$ given on sections by $`xb(x,x).`$ Let
$$L[I]:=๐ช_X\left(\underset{n>0}{}(T^n(I)T^n(I^1))\right)$$
be the Laurent-Rees algebra of $`I`$, where sections of $`V`$ (resp. of $`I`$) are declared to be of degree one (resp. of degree two). Then, as we saw in (2d), Theorem 2.2, $`b`$ naturally defines an $``$-graded linear isomorphism
$$\psi _b:\stackrel{~}{C}(V,q_b,I)\mathrm{\Lambda }(V)L[I].$$
In fact we have
$$\psi _0:\stackrel{~}{C}(V,0,I)=\mathrm{\Lambda }(V)L[I].$$
Since in general a quadratic bundle $`(V,q,I)`$ on a non-affine scheme $`X`$ may not be induced from a global $`I`$-valued bilinear form, one is unable to identify the $``$โgraded vector bundle underlying its generalised Clifford algebra bundle with $`\mathrm{\Lambda }(V)L[I].`$ The following result overcomes this problem.
###### Proposition 3.2
To every isomorphism of algebra-bundles
$$\varphi :C_0(V,q,I)C_0(V^{},q^{},I^{}),$$
one may naturally associate an isomorphism of bundles
$$\varphi _{\mathrm{\Lambda }^2}:\mathrm{\Lambda }^2(V)I^1\mathrm{\Lambda }^2(V^{})(I^{})^1$$
inducing a map
$$\zeta _{\mathrm{\Lambda }^2}:\text{Iso}[C_0(V,q,I),C_0(V^{},q^{},I^{})]\text{Iso}[\mathrm{\Lambda }^2(V)I^1,\mathrm{\Lambda }^2(V^{})(I^{})^1]:\varphi \varphi _{\mathrm{\Lambda }^2}$$
where
$$\text{Iso}[C_0(V,q,I),C_0(V^{},q^{},I^{})]$$
is the set of algebra bundle isomorphisms.
###### Definition 3.3
When $`V=V^{}`$ and $`I=I^{}`$, we may thus denote the subset of those $`\varphi `$ for which
$$\text{det}(\varphi _{\mathrm{\Lambda }^2})\text{Aut}[\text{det}(\mathrm{\Lambda }^2(V)I^1)]\mathrm{\Gamma }(X,๐ช_X^{})$$
is a square by
$$\text{Iso}^{}[C_0(V,q,I),C_0(V,q^{},I)]$$
and those for which $`\text{det}(\varphi _{\mathrm{\Lambda }^2})=1`$ by the smaller subset
$$\text{S-Iso}[C_0(V,q,I),C_0(V,q^{},I)].$$
Taking $`q=q^{}`$ in these sets and replacing โ Isoโ by โAutโ in their notations respectively defines the groups
$$\text{Aut}(C_0(V,q,I))\text{Aut}^{}(C_0(V,q,I))\text{S-Aut}(C_0(V,q,I)).$$
###### Theorem 3.4
For $`I`$-valued quadratic forms $`q`$ and $`q^{}`$ on a rank 3 vector bundle $`V`$ over a scheme $`X`$, we have the following commuting diagram of natural maps of sets with the downward arrows being the canonical inclusions, the horizontal arrows being surjective and the top horizontal arrow being bijective:
$$\begin{array}{ccc}\text{S-Iso}[(V,q,I),(V,q^{},I)]& \stackrel{}{}& \text{S-Iso}[C_0(V,q,I),C_0(V,q^{},I)]\\ \text{inj}& & \text{inj}& & \\ \text{Iso}[(V,q,I),(V,q^{},I)]& \stackrel{\text{onto}}{}& \text{Iso}^{}[C_0(V,q,I),C_0(V,q^{},I)]\\ \text{inj}& & \text{inj}& & \\ \text{Sim}[(V,q,I),(V,q^{},I)]& \stackrel{\text{onto}}{}& \text{Iso}[C_0(V,q,I),C_0(V,q^{},I)]\end{array}$$
With respect to the surjections of the horizontal arrows in the diagram above, we further have the following (where $`l`$ is the function that associates a similarity to its multiplier, $`\text{det}(g,l):=\text{det}(g)`$ for an $`I`$-similarity $`g`$ with multiplier $`l`$ and $`\zeta _{\mathrm{\Lambda }^2}`$ is the map of Prop.3.2 above):
there is a family of sections
$$s_{2k+1}:\text{Iso}[C_0(V,q,I),C_0(V,q^{},I)]\text{Sim}[(V,q,I),(V,q^{},I)]$$
indexed by the integers such that
$$ls_{2k+1}=\text{det}^{2k+1}\zeta _{\mathrm{\Lambda }^2}\text{ and }(\text{det}^2s_{2k+1})\times (l^3s_{2k+1})=\text{det}\zeta _{\mathrm{\Lambda }^2};$$
there is also a section
$$s^{}:\text{Iso}^{}[C_0(V,q,I),C_0(V,q^{},I)]\text{Iso}[(V,q,I),(V,q^{},I)]$$
such that $`\text{det}^2s^{}=\text{det}\zeta _{\mathrm{\Lambda }^2};`$
there is a family of sections
$$s_{2k+1}^+:\text{Iso}[C_0(V,q,I),C_0(V,q^{},I)]\text{Sim}[(V,q,I),(V,q^{},I)]$$
indexed by the integers which is multiplicative when followed by the natural inclusions into $`\text{GL}(V)\times \mathrm{\Gamma }(X,๐ช_X^{})`$, i.e., if
$$\varphi _i\text{Iso}[C_0(V,q_i,I),C_0(V,q_{i+1},I)]$$
then
$$s_{2k+1}^+(\varphi _2\varphi _1)=s_{2k+1}^+(\varphi _2)s_{2k+1}^+(\varphi _1)\text{GL}(V)\times \mathrm{\Gamma }(X,๐ช_X^{}).$$
Further,
$$ls_{2k+1}^+=\text{det}^{2k+1}\zeta _{\mathrm{\Lambda }^2}\text{ and }(\text{det}^2s_{2k+1}^+)\times (l^3s_{2k+1}^+)=\text{det}\zeta _{\mathrm{\Lambda }^2}.$$
The maps $`s_{2k+1}`$ and $`s^{}`$ above may not be multiplicative but are mutliplicative upto $`\mu _2(\mathrm{\Gamma }(X,๐ช_X))`$ i.e., these maps followed by the canonical quotient map, on taking the quotient of $`\text{GL}(V)\times \mathrm{\Gamma }(X,๐ช_X^{})`$ by $`\mu _2(\mathrm{\Gamma }(X,๐ช_X)).\text{Id}_V\times \{1\}`$, become multiplicative.
###### Theorem 3.5
For a rank 3 quadratic bundle $`(V,q,I)`$ on a scheme $`X`$, one has the following natural commutative diagram of groups with exact rows, where the downward arrows are the canonical inclusions and where $`l`$ is the function that associates to any $`I`$-(self)similarity its multiplier:
Further, we have:
There are splitting homomorphisms
$$s_{2k+1}^+:\text{Aut}(C_0(V,q,I))\text{GO}(V,q,I)$$
such that
$$ls_{2k+1}^+=\text{det}^{2k+1}\text{ and }(\text{det}^2s_{2k+1}^+)\times (l^3s_{2k+1}^+)=\text{det}.$$
In particular, $`\text{GO}(V,q,I)`$ is a semidirect product.
The restriction of $`s_{2k+1}^+`$ to $`\text{Aut}^{}(C_0(V,q,I))`$ does not necessarily take values in $`\text{O}(V,q,I)`$, but the further restriction to $`\text{S-Aut}(C_0(V,q,I))`$ does take values in $`\text{SO}(V,q,I).`$
The maps $`s_{2k+1}`$ and $`s^{}`$ of Theorem 3.4 above (under the current hypotheses) may not be homomorphisms but are homomorphisms upto $`\mu _2(\mathrm{\Gamma }(X,๐ช_X)).`$
Suppose $`X`$ is integral and $`q\kappa (x)`$ is semiregular at some point $`x`$ of $`X`$ with residue field $`\kappa (x).`$ Then any automorphism of $`C_0(V,q,I)`$ has determinant 1. Hence
$$\text{Aut}(C_0(V,q,I))=\text{Aut}^{}(C_0(V,q,I))=\text{S-Aut}(C_0(V,q,I))$$
and $`\text{O}(V,q,I)`$ is the semidirect product of $`\mu _2(\mathrm{\Gamma }(X,๐ช_X))`$ and $`\text{SO}(V,q,I).`$
The proofs of the above results, and of the injectivity part of Theorem 3.1, will be given in ยง4 and ยง5.
### 3.3 Study of Bilinear Forms and Interpretation as Specialisations
As for the proof of the surjectivity part of Theorem 3.1, we have the following which will be proved in ยง6:
###### Theorem 3.6
Let $`X`$ be a scheme and $`A`$ a specialisation of rank 4 Azumaya algebra bundles on $`X.`$ Let $`๐ช_X.1_AA`$ be the line sub-bundle generated by the nowhere-vanishing global section of $`A`$ corresponding to the unit for algebra multiplication.
There exist a rank 3 vector bundle $`V`$ on $`X`$, a quadratic form $`q`$ on $`V`$ with values in the line bundle $`I:=\text{det}^1(A)`$, and an isomorphism of algebra bundles $`AC_0(V,q,I).`$ This gives the surjectivity in the statement of Theorem 3.1. Further, the following linear isomorphisms may be deduced:
$`\text{det}(A)\mathrm{\Lambda }^2(V)A/๐ช_X.1_A,`$ from which follow:
$`\text{det}(\mathrm{\Lambda }^2(V))(\text{det}(A))^2;`$
$`V(A/๐ช_X.1_A)^{}\text{det}(V)\text{det}(A);`$
$`\text{det}(A^{})(\text{det}(A))^3(\text{det}(V))^2`$ which implies that $`\text{det}(A)\text{det}(A^{})2.\text{Pic}(X).`$
There exists a quadratic bundle $`(V^{},q^{},I^{})`$ such that $`C_0(V^{},q^{},I^{})A`$ and with
$`I^{}=๐ช_X`$ iff $`\text{det}(A)2.\text{Pic}(X)`$;
with $`q^{}`$ induced from a global $`I^{}`$-valued bilinear form iff $`๐ช_X.1_A`$ is an $`๐ช_X`$-direct summand of $`A`$;
with both $`I^{}=๐ช_X`$ and with $`q^{}`$ induced from a global bilinear form (with values in $`I^{}`$) iff $`๐ช_X.1_A`$ is an $`๐ช_X`$-direct summand of $`A`$ and $`\text{det}(A)2.\text{Pic}(X).`$
If $`2\mathrm{\Gamma }(X,๐ช_X^{})`$, then any quadratic form is induced from a symmetric bilinear form. Therefore from assertion (b2) of the above, we have the following:
###### Corollary 3.7
Suppose $`2\mathrm{\Gamma }(X,๐ช_X^{}).`$ Then for any specialisation $`A`$ of rank 4 Azumaya algebras on $`X`$, $`๐ช_X.1_A`$ is an $`๐ช_X`$-direct summand of $`A.`$
There are two ingredients in the proof of part (a) of Theorem 3.6. The first is Theorem 3.4. The second is the following theorem which describes specialisations as bilinear forms under certain conditions. As a preparation towards its statement, we briefly remind the reader of a few results from from Part A, (cf. ยง2.9).
For a rank $`n^2`$ vector bundle $`W`$ on a scheme $`X`$ and $`w\mathrm{\Gamma }(X,W)`$ a nowhere-vanishing global section, recall that if $`\text{Azu}_{W,w}`$ is the open $`X`$โsubscheme of Azumaya algebra structures on $`W`$ with identity $`w`$ then its schematic image (or the scheme of specialisations or the limiting scheme) in the bigger $`X`$โscheme $`\text{Assoc}_{W,w}`$ of associative $`w`$-unital algebra structures on $`W`$ is the $`X`$โscheme $`\text{SpAzu}_{W,w}.`$ By definition, the set of distinct specialised $`w`$-unital algebra structures on $`W`$ corresponds precisely to the set of global sections of this last scheme over $`X`$.
If $`\text{Stab}_w\text{GL}_W`$ is the stabiliser subgroupscheme of $`w`$, recall from Theorems 3.4 and 3.8, Part A, , that there exists a canonical action of $`\text{Stab}_w`$ on $`\text{SpAzu}_{W,w}`$ such that the natural inclusions
$$(\mathrm{})\text{Azu}_{W,w}\text{SpAzu}_{W,w}\text{Assoc}_{W,w}$$
are all $`\text{Stab}_w`$-equivariant. Now let $`V`$ be a rank 3 vector bundle on the scheme $`X`$ and $`\text{Bil}_{(V,I)}`$ be the associated rank 9 vector bundle of bilinear forms on $`V`$ with values in the line bundle $`I.`$ We say that a bilinear form $`b`$ over an open subset $`UX`$ is semiregular if there is a trivialisation $`\{U_i\}`$ of $`I|U`$, such that over each open subscheme $`U_i`$, the quadratic form $`q_i`$ with values in the trivial line bundle induced from $`q_b|U_i`$ is semiregular (it may turn out that a semiregular bilinear form may be degenerate). This definition is independent of the choice of a trivialisation, since $`q_i`$ is semiregular iff $`\lambda q_i`$ is semiregular for every $`\lambda \mathrm{\Gamma }(U_i,๐ช_X^{})`$ (for further details see ยง2.3). In this way we obtain the open subscheme
$$\text{Bil}_{(V,I)}^{sr}\text{Bil}_{(V,I)}$$
of semiregular bilinear forms on $`V`$ with values in $`I.`$ We next take for $`W`$ the following special choice:
$$W:=\mathrm{\Lambda }^{even}(V,I):=\underset{n0}{}\mathrm{\Lambda }^{2n}(V)I^n$$
and we let $`w\mathrm{\Gamma }(X,W)`$ be the nowhere-vanishing global section corresponding to the unit for the natural multiplication in the twisted even-exterior algebra bundle $`W.`$ There is an obvious natural action of $`\text{GL}_V`$ on $`\text{Bil}_{(V,I)}.`$ There is also a natural morphism of groupschemes $`\text{GL}_V\text{Stab}_w`$ given on valued points by
$$g\underset{n0}{}\mathrm{\Lambda }^{2n}(g)\text{Id}$$
and therefore the natural inclusions marked by $`(\mathrm{})`$ above are $`\text{GL}_V`$-equivariant. Finally, note that there is an obvious involution $`\mathrm{\Sigma }`$ on $`\text{Assoc}_{W,w}`$ given by $`A\text{opposite}(A)`$ which leaves the open subscheme $`\text{Azu}_{W,w}`$ invariant.
###### Theorem 3.8
Let $`V`$ be a rank 3 vector bundle on the scheme $`X`$, $`W:=\mathrm{\Lambda }^{even}(V,I)`$ and $`w\mathrm{\Gamma }(X,W)`$ correspond to 1 in the twisted even-exterior algebra bundle. There is a natural $`\text{GL}_V`$-equivariant morphism of $`X`$โschemes
$$\mathrm{{\rm Y}}^{}=\mathrm{{\rm Y}}_X^{}:\text{Bil}_{(V,I)}\text{Assoc}_{W,w}$$
whose schematic image is precisely the scheme of specialisations
$$\text{SpAzu}_{W,w}.$$
Further if $`\mathrm{{\rm Y}}^{}`$ factors canonically through
$$\mathrm{{\rm Y}}=\mathrm{{\rm Y}}_X:\text{Bil}_{(V,I)}\text{SpAzu}_{W,w},$$
then $`\mathrm{{\rm Y}}`$ is a $`\text{GL}_V`$-equivariant isomorphism and it maps the $`\text{GL}_V`$-stable open subscheme $`\text{Bil}_{(V,I)}^{sr}`$ isomorphically onto the $`\text{GL}_V`$-stable open subscheme
$`\text{Azu}_{W,w}.`$
The involution $`\mathrm{\Sigma }`$ of $`\text{Assoc}_{W,w}`$ defines a unique involution (also denoted by $`\mathrm{\Sigma }`$) on the scheme of specialisations $`\text{SpAzu}_{W,w}`$ leaving the open subscheme $`\text{Azu}_{W,w}`$ invariant, and therefore via the isomorphism $`\mathrm{{\rm Y}}`$, it defines an involution on $`\text{Bil}_{(V,I)}.`$ This involution is none other than the one on valued points given by $`B\text{transpose}(B).`$
For an $`X`$-scheme $`T`$, let $`V_T`$ (resp.$`W_T`$, resp.$`I_T`$) denote the pullback of $`V`$ (resp.$`W`$, resp.$`I`$) to $`T`$, and let $`w_T`$ be the global section of $`W_T`$ induced by $`w.`$ Then the base-changes of $`\mathrm{{\rm Y}}_X^{}`$ and $`\mathrm{{\rm Y}}_X`$ to $`T`$, namely
$$\mathrm{{\rm Y}}_X^{}\times _XT:\text{Bil}_{(V,I)}\times _XT\text{Assoc}_{W,w}\times _XT$$
and
$$\mathrm{{\rm Y}}_X\times _XT:\text{Bil}_{(V,I)}\times _XT\text{SpAzu}_{W,w}\times _XT$$
may be canonically identified with the corresponding ones over $`T`$ namely with
$$\mathrm{{\rm Y}}_T^{}:\text{Bil}_{(V_T,I_T)}\text{Assoc}_{W_T,w_T}$$
and
$$\mathrm{{\rm Y}}_T:\text{Bil}_{(V_T,I_T)}\text{SpAzu}_{W_T,w_T}\text{ respectively.}$$
The explicit computation of the morphism $`\mathrm{{\rm Y}}`$ locally over $`X`$ is an important step in proving the above theorem. To describe this, suppose that $`I`$ is trivial and $`V`$ is free of rank 3 over $`X`$, so that we may fix a basis $`\{e_1,e_2,e_3\}`$ for $`V`$, which naturally gives rise to a basis of $`\text{Bil}_V.`$
For any $`X`$-scheme $`T`$, a $`T`$-valued point $`B`$ of $`\text{Bil}_V`$ is just a global bilinear form with values in $`๐ช_T`$ on the pull-back $`V_XT`$ of $`V`$ to $`T.`$ Such a $`B`$ is given uniquely by a $`(3\times 3)`$-matrix $`(b_{ij})`$ with the $`b_{ij}`$ being global sections of the trivial line bundle $`๐ธ_T^1`$ (or equivalently, elements of $`\mathrm{\Gamma }(T,๐ช_T)`$). The chosen basis for $`V`$ also gives rise to the basis
$$\{ฯต_0:=w=1;ฯต_1:=e_1e_2,ฯต_2:=e_2e_3,ฯต_3:=e_3e_1\}$$
of $`W=\mathrm{\Lambda }^{even}(V).`$ A $`T`$-valued point $`A`$ of $`\text{Assoc}_{W,w}`$ is just a $`w_T:=(w_XT)`$-unital associative algebra structure on the bundle $`W_T:=W_XT.`$ Let $`_A`$ denote the multiplication in the algebra bundle $`A`$, and for ease of notation, let $`s^{}`$ denote the section $`s_XT`$ induced from a section $`s`$ (for example, $`(w_XT)=w^{},ฯต_i_XT=ฯต_i^{}`$ etc).
###### Theorem 3.9
In addition to the hypothesis of Theorem 3.8, assume that $`V`$ is free of rank 3 and that $`I=๐ช_X.`$ Then fixing a basis for $`V`$ and adopting the notations above, the map $`\mathrm{{\rm Y}}(T)`$ takes $`B=(b_{ij})`$ to $`(A,1_A,)=(W_T,w_T=w^{},_A)`$ with multiplication given as follows, where $`M_{ij}(B)`$ is the determinant of the minor of the element $`b_{ij}`$ in $`B:`$
1. $`ฯต_1^{}_Aฯต_1^{}=M_{33}(B)w^{}+(b_{21}b_{12})ฯต_1^{}`$
2. $`ฯต_2^{}_Aฯต_2^{}=M_{11}(B)w^{}+(b_{32}b_{23})ฯต_2^{}`$
3. $`ฯต_3^{}_Aฯต_3^{}=M_{22}(B)w^{}+(b_{13}b_{31})ฯต_3^{}`$
4. $`ฯต_1^{}_Aฯต_2^{}=M_{31}(B)w^{}b_{23}ฯต_1^{}b_{12}ฯต_2^{}b_{22}ฯต_3^{}`$
5. $`ฯต_2^{}_Aฯต_3^{}=+M_{12}(B)w^{}b_{33}ฯต_1^{}b_{31}ฯต_2^{}b_{23}ฯต_3^{}`$
6. $`ฯต_3^{}_Aฯต_1^{}=+M_{23}(B)w^{}b_{31}ฯต_1^{}b_{11}ฯต_2^{}b_{12}ฯต_3^{}`$
7. $`ฯต_1^{}_Aฯต_3^{}=+M_{32}(B)w^{}+b_{13}ฯต_1^{}+b_{11}ฯต_2^{}+b_{21}ฯต_3^{}`$
8. $`ฯต_2^{}_Aฯต_1^{}=M_{13}(B)w^{}+b_{32}ฯต_1^{}+b_{21}ฯต_2^{}+b_{22}ฯต_3^{}`$
9. $`ฯต_3^{}_Aฯต_2^{}=M_{21}(B)w^{}+b_{33}ฯต_1^{}+b_{13}ฯต_2^{}+b_{32}ฯต_3^{}`$
The key to the proofs of Theorems 3.1, 3.4 and 3.5 lies in an analysis of a different identification of the scheme of specialisations, namely one related to the scheme of $`๐ช_X`$-valued quadratic forms on a trivial rank 3 bundle in the special situation when $`W`$ is free and $`w`$ part of a global basis. Without loss of generality we may in this situation therefore take $`V`$ to be a free rank 3 vector bundle on $`X`$ and
$$(W,w)=(\mathrm{\Lambda }^{even}(V),1),$$
so that we are in the situation of Theorem 3.9 above. This relationship with quadratic forms was shown in Theorem 5.3, Part A, , which we briefly recall next. Let $`\text{Quad}_V`$ denote the bundle of quadratic forms on $`V`$ (with values in $`๐ช_X`$) and $`\text{Quad}_V^{sr}`$ the open subscheme of semiregular quadratic forms. Let $`A_0`$ denote the algebra bundle structure (with unit $`w=1`$) on $`W=\mathrm{\Lambda }^{even}(V)`$ given by $`\mathrm{\Lambda }^{even}(V)`$ itself. Fix a basis for $`V`$ and adopt the notations preceding Theorem 3.9 above. Then $`\text{Stab}_w`$ is the semidirect product of a commutative 3-dimensional subgroupscheme
$$\text{L}_w(๐ธ_X^3,+)$$
with the stabiliser subgroupscheme $`\text{Stab}_{A_0}`$ of $`A_0`$ in $`\text{Stab}_w`$ (Lemma 5.1, Part A, ).
###### Theorem 3.10 (Definition 5.2 & Theorem 5.3, Part A, )
There is a
natural isomorphism
$$\mathrm{\Theta }:\text{Quad}_V\times _X\text{L}_w\text{SpAzu}_{W,w}$$
which maps the open subscheme $`\text{Quad}_V^{sr}\times _X\text{L}_w`$ isomorphically onto the open subscheme $`\text{Azu}_{W,w}.`$
The isomorphism $`\mathrm{\Theta }`$ was first defined by C. S. Seshadri in for the case $`X=\text{Spec}(k)`$, $`k`$ an algebraically closed field of characteristic $`2.`$ Section 5 is essentially devoted to studying $`\mathrm{\Theta }.`$ There we compute $`\mathrm{\Theta }`$ explicitly and in Theorem 5.1 we write out the multiplication table of every specialised algebra structure on any fixed free rank 4 vector bundle with fixed unit that is part of a global basis. It turns out that $`\mathrm{\Theta }`$ is not equivariant with respect to $`\text{GL}_V`$, but nevertheless satisfies a โtwistedโ form of equivariance (Theorem 5.4). A $`T`$-valued point $`q`$ of
$$\text{Quad}_V๐ธ_X^6$$
may be identified with a 6-tuple $`(\lambda _1,\lambda _2,\lambda _3,\lambda _{12},\lambda _{13},\lambda _{23})`$ corresponding to the quadratic form
$$(x_1,x_2,x_3)\mathrm{\Sigma }_i\lambda _ix_i^2+\mathrm{\Sigma }_{i<j}\lambda _{ij}x_ix_j.$$
A $`T`$-valued point $`\underset{ยฏ}{t}`$ of $`\text{L}_w(๐ธ_X^3,+)`$ may be identified with a 3-tuple $`(t_1,t_2,t_3)`$ which corresponds to the valued point of $`\text{Stab}_w`$ given by the $`(4\times 4)`$-matrix
$$\left(\begin{array}{cccc}1& t_1& t_2& t_3\\ & & & \\ 0& & I_3& \end{array}\right)$$
where $`I_3`$ is the $`(3\times 3)`$-identity matrix. With these notations, the identification of Theorems 3.8 and 3.9 may be compared with that of the above Theorem 3.10 as follows.
###### Theorem 3.11
The isomorphism $`\mathrm{{\rm Y}}^1\mathrm{\Theta }:\text{Quad}_V\times _X\text{L}_w\text{Bil}_V`$ takes the valued point
$$(q,\underset{ยฏ}{t})=((\lambda _1,\lambda _2,\lambda _3,\lambda _{12},\lambda _{13},\lambda _{23}),(t_1,t_2,t_3))$$
to the valued point $`B=(b_{ij})`$ given by
$$B=\left(\begin{array}{ccc}\lambda _1& t_1& \lambda _{13}t_3\\ \lambda _{12}t_1& \lambda _2& t_2\\ t_3& \lambda _{23}t_2& \lambda _3\end{array}\right).$$
Moreover, under this identification, the involution $`B(B)^t`$ on $`\text{Bil}_V`$ (induced from the isomorphism $`\mathrm{{\rm Y}}`$ of Theorem 3.8) translates into the involution on
$$\text{Quad}_V\times _X\text{L}_w$$
given by
$$(q,(t_1,t_2,t_3))(q,(t_1\lambda _{12},t_2\lambda _{23},t_3\lambda _{13})).$$
### 3.4 Degenerations of Azumaya Bundles as Quaternion Algebra Bundles
We next make some comments on specialised algebras. Let $`R`$ be a unital commutative ring and $`A`$ a unital associative $`R`$-algebra. Given any involution $`\sigma `$ on $`A`$, we may define the trace and norm associated to this involution by
$$tr_\sigma :xx+\sigma (x)\text{ and }n_\sigma :xx.\sigma (x).$$
In para.1.3, Chap.I, , Knus calls $`\sigma `$ standard if $`\sigma `$ fixes $`R.1_A`$ and both $`tr_\sigma `$ and $`n_\sigma `$ take values in $`R.1_A.`$ In Prop.1.3.4 of the same chapter, he proves that a standard involution is unique if it exists, provided the $`R`$-module underlying $`A`$ is finitely generated projective and faithful.
In para.1.3.7, op.cit., Knus defines $`A`$ to be a quaternion algebra if $`A`$ is a projective $`R`$-module of rank 4 and $`A`$ has a standard involution. Thus we may define a rank 4 algebra bundle on a scheme $`X`$ to be a quaternion algebra bundle if it is locally (in the Zariski topology) a quaternion algebra in Knusโ sense, and it would follow that the local standard involutions glue to define a unique global standard involution on the bundle.
###### Proposition 3.12
Any specialised algebra bundle is a quaternion algebra bundle.
This result can be deduced from the following two facts:
Any specialised algebra is locally (in the Zariski topology) the even Clifford algebra of a rank 3 quadratic bundle (assertion (3), Theorem 2.15 and Theorem 3.10).
The even Clifford algebra of a quadratic module of rank 3 over a commutative ring has a standard involution which is none other than the restriction of the โstandardโ involution on the full Clifford algebra (Prop.3.1.1, Chap.V, ). Q.E.D.
We hasten to remark that even over an algebraically closed field there are quaternion algebras that are not even-Clifford algebras of quadratic forms.
The proof of the bijection stated in Theorem 2.10 follows from Prop.3.2.3 and Prop.3.2.4, Chap.V, generalised to the scheme-theoretic setting. We recall how the surjectivity is established.
Let $`A`$ be a specialised algebra bundle on the scheme $`X`$. By the results just quoted if $`A`$ is Azumaya, or more generally by Prop.3.12, we have the existence of a unique standard involution $`\sigma _A`$ on $`A`$, to which are associated the norm
$$n_{\sigma _A}:A๐ธ_X^1\text{ given on sections by }xx.\sigma _A(x)$$
and the trace
$$tr_{\sigma _A}:A๐ธ_X^1\text{ given on sections by }xx+\sigma _A(x).$$
Let $`A^{}:=\text{kernel}(tr_{\sigma _A})A`$ be the subsheaf of trace zero elements. As the calculations in para.3.2, Chap.V, show, the trace map is surjective if $`A`$ is itself an Azumaya algebra; if this is the case, then it is further shown there that the rank 3 quadratic bundle
$$(V,q,I):=(A^{},n_{\sigma _A}|A^{},๐ช_X.1_A)$$
is semiregular and its even Clifford algebra $`C_0(V,q,I)A.`$ However the above method of retrieving a canonical rank 3 quadratic bundle fails badly for specialised non-Azumaya algebras. Consider even the case of $`X=\text{Spec}(k)`$ where $`k`$ is a field of characteristic two and the Clifford algebra $`A=C_0(V,q)`$ of a quadratic form $`q`$ on $`V=k^3`$ which is a perfect square (i.e., a square of a linear form or equivalently a sum of squares). In this case an easy computation shows that the subspace $`A^{}`$ of trace zero elements is the full space $`A.`$
###### Proposition 3.13
Let $`S`$ be a commutative semilocal ring that is 2-perfect i.e., such that the square map
$$SS:ss^2$$
is surjective, and $`V`$ a free rank 3 $`S`$-module. Then the set of semiregular quadratic $`S`$-forms on $`V`$ forms a single $`\text{GL}(V)`$โorbit; in other words, upto isometry, $``$ only one semiregular quadratic $`S`$-module structure on $`V.`$
###### Corollary 3.14
Let $`S`$ be a commutative local ring that is 2-perfect. Then any two rank 4 Azumaya $`S`$-algebras are isomorphic. If $`S`$ were only semilocal, the conclusion still holds provided the identity elements for multiplication for each of the two Azumaya $`S`$-algebras can be completed to an $`S`$-basis.
The proof of the above Proposition will be given in ยง8. In view of Theorem 2.10, taking $`X=\text{Spec}(S)`$ with $`S`$ as in Prop.3.13 proves the first assertion of the above corollary. The second may be deduced by an application of Theorem 3.10 alongwith Prop.3.13.
### 3.5 Non-existence of Azumaya Structures and Semiregular Forms
In this and the next subsections, we study what happens when we impose the condition on self-duality on the bundle underlying an algebra (or on one on which a quadratic form is defined). In this subsection, our aim is to indicate examples of rank 4 vector bundles on which there do not exist any global Azumaya structures; for these examples it also turns out that there do not exist global regular quadratic forms with values in the trivial line bundle. However, in these examples there do exist algebra structures which are Azumaya on a nontrivial dense open subscheme and there do exist quadratic forms with values in the trivial bundle which are semiregular on a dense open subscheme. We use Theorem 3.1 to obtain examples of rank three bundles on which there do not exist any global semiregular quadratic forms with values in certain line bundles; however, there do exist forms which are semiregular on a dense open subscheme. We shall only outline the results and the proofs will be given in .
In the following we let $`X`$ be a scheme and $`W`$ a rank $`n^2`$ vector bundle over $`X.`$ The following result describes the behaviour of the locus where an algebra structure is Azumaya.
###### Proposition 3.15 (Prop.3.3, Part A, )
Let $`T`$ be an $`X`$-scheme and $`A`$ an associative unital algebra structure on $`W_T:=W_XT.`$ Then the subset
$$U(T,A):=\{tT|๐_t\text{ is an Azumaya }๐ช_{T,t}\text{algebra}\}$$
is an open (possibly empty) subset. When $`U(T,A)`$ is nonempty, denote by the same symbol the canonical open subscheme structure. Then if $`f:T^{}T`$ is an $`X`$-morphism such that the topological image intersects $`U(T,A)`$, then
$$U(T^{},f^{}(A)=A_TT^{})U(T,A)\times _TT^{}$$
as open subschemes of $`T^{}.`$ Further $`U(T,A)T`$ is an affine morphism.
$`U(T,A)`$ is the maximal open subset restricted to which $`A`$ is Azumaya.
Further let $`f:T^{}T`$ be a morphism of $`X`$-schemes such that $`f^{}(A)`$ is Azumaya. Then $`f`$ factors through the open subscheme $`U(T,A)`$ defined above.
Next we let $`V`$ denote a vector bundle and $`I`$ a line-bundle over $`X.`$ The following result describes the behaviour of the locus where a quadratic form is good.
###### Proposition 3.16
Let $`T`$ be an $`X`$-scheme and $`q:V_TI_T`$ a quadratic bundle with $`V_T:=V_XT`$ and $`I_T:=I_XT.`$ Then the subset
$$U(T,q):=\{tT|q\kappa (t):V_T\kappa (t)I_T\kappa (t)\text{ is โgoodโ}\},$$
where โgoodโ means semiregular if $`V`$ is of odd rank and regular if $`V`$ is of even rank, is an open (possibly empty) subset. When $`U(T,q)`$ is nonempty, denote by the same symbol the canonical open subscheme structure. Then if $`f:T^{}T`$ is an $`X`$-morphism such that the topological image intersects $`U(T,q)`$, and if $`f^{}(q):=q_T^{}:V_T^{}I_T^{}`$ is the induced quadratic bundle on $`T^{}`$, then
$$U(T^{},q_T^{})U(T,q)\times _TT^{}$$
as open subschemes of $`T^{}.`$ Further $`U(T,q)T`$ is an affine morphism.
$`U(T,q)`$ is the maximal open subset restricted to which $`q`$ is good.
Further let $`f:T^{}T`$ be a morphism of $`X`$-schemes such that $`q_T^{}`$ is good. Then $`f`$ factors through the open subscheme $`U(T,q)`$ defined above.
We now let $`S`$ be a scheme, and let $`XS`$ be an $`S`$-scheme which is proper, of finite-type and has connected fibers relative to $`S.`$ Again, $`W`$ denotes a vector bundle on $`X`$, $`A`$ an algebra structure on $`W`$, and $`q:VI`$ a quadratic form on the vector bundle $`V`$ with values in the line bundle $`I.`$
###### Theorem 3.17
(With the above notations)
Suppose $`W`$ is self-dual and $`U(X,A)S`$ is surjective. Then $`U(X,A)=X`$ i.e., $`A`$ is Azumaya.
Suppose that the rank of $`W`$ is 4, that there exists an algebra structure $`A^{}`$ on $`W`$ such that $`U(X,A^{})S`$ is surjective and that $`U(X,A^{})X.`$ Then there does not exist any global Azumaya algebra structure on $`W.`$ Neither does there exist any global regular quadratic form on $`W`$ with values in the trivial line bundle.
Let $`V`$ be of rank 3 and suppose that the underlying bundle of $`C_0(V,q,I)`$ is self-dual. If $`U(X,q)S`$ is surjective, then $`q`$ is semiregular.
Let $`V`$ be of rank 3 such that $`U(X,q)S`$ is surjective but $`U(X,q)X.`$ Then there does not exist any global semiregular quadratic form on $`V`$ with values in $`I.`$ If $`(L,h,J)`$ is a twisted discriminant bundle, then there does not exist any global semiregular quadratic form on $`VL`$ with values in $`IJ.`$
Assertion (3) follows from (1) and Theorem 3.1. The proofs may be reduced to the case $`B=\text{Spec}(k)`$ where $`k`$ is a field. In this case, assertions (1) and (2) may be restated as follows. Let $`X`$ be a connected proper scheme of finite type over a field and let $`W`$ be a vector bundle on $`X`$ of rank $`n^2`$ for some $`n2.`$
Let $`W`$ be self-dual and $`A`$ an associative unital algebra structure on $`W`$. If a section to $`\text{Assoc}_{W,w}`$ over $`X`$ meets $`\text{Azu}_{W,w}`$ topologically, then it factors as a morphism through the open subscheme $`\text{Azu}_{W,w}`$, where $`w:=1_A`$ and $`A`$ corresponds to the given section;
Let the rank of $`W`$ be 4. if there is a section over $`X`$ of $`\text{Assoc}_{W,w}`$ that topologically meets both the open subscheme $`\text{Azu}_{W,w}`$ and its complement (with $`w=1_A`$ where $`A`$ corresponds to the given section), then the $`X`$-schemes
$`\text{Assoc}_{W,w^{}}`$ (with $`w^{}`$ global nowhere-vanishing) cannot have sections that land topologically inside $`\text{Azu}_{W,w^{}}`$ and hence in particular the $`X`$-schemes $`\text{Azu}_{W,w^{}}`$ have no sections over $`X.`$
We leave it to the reader to formulate similar statements corresponding to the other assertions of Theorem 3.17. We next indicate situations where the above results apply to give examples of rank 4 bundles that do not admit any Azumaya algebra structures and of rank 3 vector bundles which do not admit any semiregular form with values in certain line bundles. The examples occur naturally on certain moduli spaces of vector bundles over a relative curve. As remarked earlier, we quickly mention the relevant objects, but we shall not get into definitions, proofs etc, which will appear in .
We first recall some facts about Nagata Rings. The standard reference is Chap.12 of Matsumuraโs book . An integral domain $`A`$ is said to satisfy condition N-1 if its integral closure $`A_K`$ in its quotient field $`K`$ is a finite $`A`$-module. It is said to satisfy condition N-2 if for every finite extension field $`L/K`$, the integral closure $`A_L`$ of $`A`$ in $`L`$ is a finite $`A`$-module. The properties N-1 and N-2 are preserved under localisation and N-2 $``$ N-1 whereas noetherianness with N-1 $``$ N-2 only in char.0; there exists an example of Y. Akizuki of a noetherian domain of positive char. which is not N-1. A commutative ring $`B`$ is called a Nagata ring (pseudo-geometric ring in Nagataโs own terminology and universally japanese ring in Grothendieckโs) if it is noetherian and $`B/๐ญ`$ is N-2 for each prime $`๐ญ`$ of $`B.`$ Every localisation of $`B`$ and every finitely generated (commutative) $`B`$-algebra are then also Nagata, and complete noetherian local rings are Nagata as well. Dedekind domains of characteristic zero such as $``$ are Nagata.
Let $`S`$ be a normal integral scheme, which we shall assume is locally-universally Japanese (Nagata). This means that there is a covering of $`S`$ by affine open subschemes that are isomorphic to spectra of Nagata rings. Such an assumption on $`S`$ is necessary to be able to obtain finite-type quotients by reductive groupschemes following Seshadriโs Geometric Invariant Theory over a General Base as in .
Let $`CS`$ be smooth projective of relative dimension 1 (i.e., a curve relative to $`S`$) with geometric fibres irreducible and of constant genus $`g2.`$ Let $`๐ฐ_C^{ss}(2,0)`$ denote the Seshadri moduli space of semistable vector bundles on $`C`$ of rank 2 and degree zero, generalising the construction in to the base $`S`$ using the methods of . It is a coarse moduli scheme, obtained as a Geometric Invariant Theory quotient, and is a normal geometrically irreducible projective scheme of relative dimension $`4g3`$ over $`S.`$ Let $`๐ฐ_C^s(2,0)๐ฐ_C^{ss}(2,0)`$ denote the open subscheme corresponding to stable vector bundles. Then $`๐ฐ_C^s(2,0)`$ is precisely the locus where $`๐ฐ_C^{ss}(2,0)S`$ is smooth, unless $`g=2`$ in which case $`๐ฐ_C^{ss}(2,0)S`$ is itself smooth. The Seshadri-desingularisation
$$\pi _2:๐ฉ_C(4,0)๐ฐ_C^{ss}(2,0)$$
may be constructed, extending the case $`S=\text{Spec}(k)`$, $`k`$ a field, carried out in for $`\text{Char.}(k)2`$ and in a characteristic-free way in ยง 6, Part B of . $`\pi _2`$ is an isomorphism over the open subscheme $`๐ฐ_C^s(2,0).`$ The scheme $`๐ฉ_C(4,0)`$ appears as a fine moduli space for certain parabolic stable vector bundles of rank 4 and degree 0 on $`C`$ and it turns out to be a geometrically irreducible smooth projective scheme relative to $`S.`$ By the very construction of $`๐ฉ_C(4,0)`$, there is a naturally defined rank 4 vector bundle $`W`$ on $`๐ฉ_C(4,0)`$ which is equipped with an associative unital algebra structure $`A.`$ Further, the locus where $`A`$ is Azumaya is precisely the dense open subscheme $`๐ฉ_C^s(4,0):=\pi _2^1(๐ฐ_C^s(2,0)).`$ Let $`(V,q,I)`$ be any ternary quadratic bundle on $`๐ฉ_C(4,0)`$ such that $`C_0(V,q,I)A`$ (such a choice exists by the surjectivity part of Theorem 3.1). Applying Theorem 3.17 to the present situation gives the following.
###### Theorem 3.18
(With the above notations:)
$`W`$ does not admit any (global) Azumaya algebra structure.
There does not exist any global regular quadratic form on $`W`$ with values in the trivial line bundle.
There does not exist any global semiregular quadratic form on $`V`$ with values in $`I.`$ If $`(L,h,J)`$ is a twisted discriminant bundle, there does not exist any global semiregular quadratic form on $`VL`$ with values in $`IJ.`$
### 3.6 Algebra Structures on Self-Dual Bundles
In this subsection, we investigate again the effect of the condition of self-duality on the vector bundle underlying an algebra. Our aim is to study the Picard group of the scheme of specialisations of Azumaya structures on a fixed rank 4 vector bundle over suitable schemes. Recall that an integral separated noetherian scheme is said to be locally-factorial if each of its local rings is a unique factorisation domain (=UFD=factorial ring). The proofs of the following results will be given in ยง7.
###### Theorem 3.19
Let $`X`$ be a scheme and $`W`$ a rank 4 vector bundle on $`X`$ with a global nowhere-vanishing section $`w.`$ Let $`D_X`$ denote the closed subset
$$\text{SpAzu}_{W,w}\backslash \text{Azu}_{W,w}.$$
$`X`$ is irreducible iff $`\text{SpAzu}_{W,w}`$ is irreducible iff $`\text{Azu}_{W,w}`$ is irreducible iff $`D_X`$ is irreducible.
The set of irreducible components of $`X`$ is locally finiteโfor example this happens when $`X`$ is locally noetherianโiff the same is true of the corresponding set for $`\text{SpAzu}_{W,w}`$ or for $`\text{Azu}_{W,w}.`$
If $`X`$ is noetherian and finite-dimensional then the same are true for
$`\text{SpAzu}_{W,w}`$ and $`\text{Azu}_{W,w}.`$
If $`X^{}X`$ is a morphism of schemes, and if $`(W^{},w^{})`$ denotes the pullback of $`(W,w)`$, then we have a canonical isomorphism
$$\text{Azu}_{W^{},w^{}}\text{Azu}_{W,w}\times _{\text{SpAzu}_{W,w}}\text{SpAzu}_{W^{},w^{}}$$
In particular, the topological image of $`D_X^{}`$ is $`D_X.`$ Moreover, when $`X^{}X`$ is a homeomorphism onto its topological imageโwhich is for example the case when it is a closed or an open immersion, then
$$D_X\text{SpAzu}_{W^{},w^{}}$$
can be identified with $`D_X^{}.`$
If $`X`$ is a scheme which is finite dimensional and whose set of irreducible components is locally finite, then the closed subset $`D_X`$ is a divisor i.e., it has codimension 1 in $`\text{SpAzu}_{W,w}.`$
$`X`$ is affine iff $`\text{SpAzu}_{W,w}`$ is affine iff $`\text{Azu}_{W,w}`$ is affine. If $`X`$ is regular in codimension 1 (respectively locally-factorial) then so are $`\text{SpAzu}_{W,w}`$ and $`\text{Azu}_{W,w}.`$
Assume that $`X`$ is locally-factorial. Then the canonical homomorphism
$$\text{Pic}(X)\text{Pic}(\text{SpAzu}_{W,w})$$
is an isomorphism.
Assume that $`X`$ is locally-factorial and $`W`$ is self-dual (i.e., $`WW^{}`$). Then the (Weil) divisor $`n.(D_X)`$ is principal for some positive integer $`n`$, so that the natural homomorphism given by restriction of line bundles
$$\text{Pic}(\text{SpAzu}_{W,w})\text{Pic}(\text{Azu}_{W,w})$$
is an isomorphism iff $`n=1.`$ In this case, for any $`X`$-scheme $`T`$, any specialised algebra structure on $`W_XT`$ arises from a quadratic form with values in the trivial line bundle.
Assume that $`X`$ is locally-factorial and that there exists an Azumaya algebra structure on $`W`$ with unit $`w.`$ Then the canonical homomorphisms
$$\text{Pic}(X)\text{Pic}(\text{Azu}_{W,w})\text{ and }\text{Pic}(\text{SpAzu}_{W,w})\text{Pic}(\text{Azu}_{W,w})$$
are isomorphisms and the divisor $`D_X`$ is principal.
### 3.7 Stratification of the Variety of Specialisations
Let $`W`$ be a rank 4 vector bundle on a scheme $`X`$, $`w\mathrm{\Gamma }(X,W)`$ a nowhere-vanishing global section and $`\text{Stab}_w\text{GL}_W`$ the stabiliser subgroupscheme of $`w.`$ Recall from page 3.7 that the natural inclusion
$$\text{Azu}_{W,w}\text{SpAzu}_{W,w}$$
is $`\text{Stab}_w`$-equivariant. When $`X=\text{Spec}(k)`$ where $`k`$ is an algebraically closed field, there is a canonical $`\text{Stab}_w`$-stratification of the $`k`$-variety underlying $`\text{SpAzu}_{W,w}`$ as follows (the proof will be given in ยง8).
###### Theorem 3.20
Let $`k`$ be a quadratically closed field and $`X=\text{Spec}(k).`$ Then the set of ternary quadratic modules upto similarity has 4 elements which correspond to
semiregular quadratic modules;
rank 2 quadratic modules i.e., those that are not semiregular but which are regular on a two-dimensional subspace;
nonzero perfect squares (= squares of linear forms) and
the zero form.
If $`V`$ is a 3-dimensional vector space over $`k`$ and $`\{e_1,e_2,e_3\}`$ a $`k`$-basis for $`V`$, then representatives for these 4 $`\text{GL}_V`$-orbits in the space $`\text{Quad}_V`$ of quadratic forms on $`V`$ can respectively be taken to be:
$`q^{(1)}(\mathrm{\Sigma }_{i=1}^3x_ie_i)=x_1x_2+x_3^2;`$
$`q^{(2)}(\mathrm{\Sigma }_{i=1}^3x_ie_i)=x_1x_2;`$
$`q^{(3)}(\mathrm{\Sigma }_{i=1}^3x_ie_i)=x_3^2\text{ and }`$
$`q^{(4)}=0.`$
In addition to the hypotheses and notations of (1) above, assume that $`k`$ is an algebraically closed field. Then the four orbits
$$\text{Quad}_V^{(i)}:=\text{GL}_Vq^{(i)}\text{ for }1i4$$
form a stratification of the $`k`$-variety $`\text{Quad}_V`$ in the sense that we have
$$\overline{\text{Quad}_V^{(1)}}=\text{Quad}_Vand\overline{\text{Quad}_V^{(i+1)}}=\overline{\text{Quad}_V^{(i)}}\backslash \text{Quad}_V^{(i)},1i3$$
and further we also have
$$\text{Sing}(\overline{\text{Quad}_V^{(i+1)}})=\overline{\text{Quad}_V^{(i+1)}}\backslash \text{Quad}_V^{(i+1)}for1i2$$
unless the characteristic of $`k`$ is 2 in which case $`\overline{\text{Quad}_V^{(3)}}`$ is itself smooth (the notation $`\overline{T}`$ denotes the orbit closure and $`\text{Sing}(T)`$ denotes the subset of singular (non-smooth) points of $`T`$, each given the canonical reduced induced closed subscheme structure).
Continuing with the notations and hypotheses of (2) above, set
$$(W,w):=(\mathrm{\Lambda }^{even}(V),1).$$
For ease of notation denote $`\text{SpAzu}_{W,w}`$ by SpAzu and $`\text{Stab}_w`$ by $`H.`$ Then the four orbits
$$\text{SpAzu}^{(i)}:=H\mathrm{\Theta }(q^{(i)},I_4)\text{ for }1i4$$
form a stratification of the $`k`$-variety SpAzu in the sense that we have
$$\overline{\text{SpAzu}^{(1)}}=\text{SpAzu}and\overline{\text{SpAzu}^{(i+1)}}=\overline{\text{SpAzu}^{(i)}}\backslash \text{SpAzu}^{(i)},1i3$$
and further we also have
$$\text{Sing}(\overline{\text{SpAzu}^{(i+1)}})=\overline{\text{SpAzu}^{(i+1)}}\backslash \text{SpAzu}^{(i+1)}for1i2$$
unless the characteristic of $`k`$ is 2 in which case $`\overline{\text{SpAzu}^{(3)}}`$ is itself smooth.
## 4 Injectivity: Reduction to Lifting to Similarities in the Free Case
#### Proof of Prop.3.2.
Start with an isomorphism of algebra-bundles
$$\varphi :C_0(V,q,I)C_0(V^{},q^{},I^{}).$$
Let $`\{U_i\}_i`$ be an affine open covering of $`X`$ (which may also be chosen so as to trivialise some or any of the involved bundles if needed). Choose bilinear forms
$$b_i\mathrm{\Gamma }(U_i,\text{Bil}_{(V,I)})\text{ and }b_i^{}\mathrm{\Gamma }(U_i,\text{Bil}_{(V^{},I^{})})$$
such that
$$q|U_i=q_{b_i}\text{ and }q^{}|U_i=q_{b_i^{}}\text{ for each }i.$$
By (2d), Theorem 2.2, we have isomorphisms of vector bundles $`\psi _{b_i}`$ and $`\psi _{b_i^{}}`$, which preserve 1 by (2a) of the same Theorem, and we define the isomorphism of vector bundles $`(\varphi _{\mathrm{\Lambda }^{ev}})_i`$ so as to make the following diagram commute:
$$\begin{array}{ccc}C_0(V,q,I)|U_i& \underset{}{\overset{\varphi |U_i}{}}& C_0(V^{},q^{},I^{})|U_i\\ \psi _{b_i}& & \psi _{b_i^{}}& & \\ (๐ช_X\mathrm{\Lambda }^2(V)I^1)|U_i& \underset{(\varphi _{\mathrm{\Lambda }^{ev}})_i}{\overset{}{}}& (๐ช_X\mathrm{\Lambda }^2(V^{})(I^{})^1)|U_i\end{array}$$
The linear isomorphism $`(\varphi _{\mathrm{\Lambda }^{ev}})_i`$ preserves 1 and therefore it induces a linear isomorphism
$$(\varphi _{\mathrm{\Lambda }^2})_i:(\mathrm{\Lambda }^2(V)I^1)|U_i\stackrel{}{}(\mathrm{\Lambda }^2(V^{})(I^{})^1)|U_i.$$
Observe that $`(\varphi _{\mathrm{\Lambda }^2})_i`$ is independent of the choice of the bilinear forms $`b_i`$ and $`b_i^{}.`$ For, replacing these respectively by $`\widehat{b_i}`$ and $`\widehat{b_i^{}}`$, it follows from (2f), Theorem 2.2, that
$$\psi _{b_i}(\psi _{\widehat{b_i}})^1\text{ (resp. }\psi _{b_i^{}}(\psi _{\widehat{b_i^{}}})^1)$$
followed by the canonical projection onto
$$(\mathrm{\Lambda }^2(V)I^1)|U_i\text{ (resp. onto }(\mathrm{\Lambda }^2(V^{})(I^{})^1)|U_i)$$
is the same as the projection itself. By this observation, it is also clear that the isomorphisms $`\{(\varphi _{\mathrm{\Lambda }^2})_i\}_i`$ agree on (any open affine subscheme of, and hence on all of) any intersection $`U_iU_j.`$ Therefore they glue to give a global isomorphism of vector bundles
$$\varphi _{\mathrm{\Lambda }^2}:\mathrm{\Lambda }^2(V)I^1\mathrm{\Lambda }^2(V^{})(I^{})^1$$
as required. Q.E.D, Prop.3.2.
#### Reduction of Proof of Injectivity of Theorem 3.1 to Theorem 3.4.
We start with an isomorphism of algebra-bundles
$$\varphi :C_0(V,q,I)C_0(V^{},q^{},I^{}),$$
construct the isomorphism of vector bundles
$$\varphi _{\mathrm{\Lambda }^2}:\mathrm{\Lambda }^2(V)I^1\mathrm{\Lambda }^2(V^{})(I^{})^1$$
and keep the notations introduced in the proof of Prop.3.2. Firstly we deduce a linear isomorphism
$$\text{det}((\varphi _{\mathrm{\Lambda }^2})^{})^1:\text{det}((\mathrm{\Lambda }^2(V)I^1)^{})\text{det}((\mathrm{\Lambda }^2(V^{})(I^{})^1)^{}).$$
Since $`V`$ and $`V^{}`$ are of rank 3, there are canonical isomorphisms
$$\eta :\mathrm{\Lambda }^2(V)V^{}\text{det}(V)\text{ and }\eta ^{}:\mathrm{\Lambda }^2(V^{})(V^{})^{}\text{det}(V^{}).$$
It follows therefore that if we set
$$L:=\text{det}(V^{})(\text{det}(V))^1\text{ and }J:=I^{}I^1$$
then we get a twisted discriminant line bundle $`(LJ^1,h,J)`$ and a vector bundle isomorphism
$$\alpha :V^{}V(LJ^1).$$
Now for each $`i`$, the bilinear form $`b_i\mathrm{\Gamma }(U_i,\text{Bil}_{(V,I)})`$ induces, via $`\alpha |U_i`$ and $`(LJ^1,h,J)|U_i`$ and (1), Prop.2.7, a bilinear form
$$b_i^{\prime \prime }\mathrm{\Gamma }(U_i,\text{Bil}_{(V^{},IJ)}).$$
By (3) of the same Proposition, over each $`U_i`$ we get an isometry of bilinear form bundles
$$\alpha |U_i:(V^{}|U_i,b_i^{\prime \prime },IJ|U_i)(V|U_i,b_i,I|U_i)(LJ^1,h,J)|U_i$$
and also an isometry of quadratic bundles
$$\alpha |U_i:(V^{}|U_i,q_{b_i^{\prime \prime }},IJ|U_i)(V|U_i,q_{b_i}=q|U_i,I|U_i)(LJ^1,h,J)|U_i.$$
On the other hand, by an assertion in (3), Prop.2.7, we could also define the global quadratic bundle $`(V^{},q^{\prime \prime },IJ)`$ using $`(V,q,I)`$, $`\alpha `$ and $`(LJ^1,h,J)`$, so that we have an isometry of quadratic bundles
$$\alpha :(V^{},q^{\prime \prime },IJ)(V,q,I)(LJ^1,h,J).$$
It follows therefore that the $`q_{b_i^{\prime \prime }}`$ glue to give $`q^{\prime \prime }.`$ Notice that in general the $`b_i^{\prime \prime }`$ (resp. the $`b_i`$) need not glue to give a global bilinear form $`b^{\prime \prime }`$ (resp. $`b`$) such that $`q_{b^{\prime \prime }}=q^{\prime \prime }`$ (resp. $`q_b=q`$). By (1), Prop.2.8, there exists a unique isomorphism of algebra bundles
$$C_0(\alpha ,1,IJ):C_0(V^{},q^{\prime \prime },IJ)C_0\left((V,q,I)(LJ^1,h,J)\right)$$
and by Prop.2.9 we have a unique isomorphism of algebra bundles
$$\gamma _{(LJ^1,h,J)}:C_0\left((V,q,I)(LJ^1,h,J)\right)C_0(V,q,I).$$
Therefore the composition of the following sequence of isomorphisms of algebra bundles on X
| $`C_0(V^{},q^{\prime \prime },I^{})\stackrel{\text{using }I^{}IJ}{}C_0(V^{},q^{\prime \prime },IJ)\stackrel{C_0(\alpha ,1,IJ)()}{}C_0\left((V,q,I)(LJ^1,h,J)\right)`$ |
| --- |
| $`\stackrel{\gamma _{(LJ^1,h,J)}()}{}C_0(V,q,I)\stackrel{\varphi ()}{}C_0(V^{},q^{},I^{})`$ |
is an element of
$$\text{Iso}[C_0(V^{},q^{\prime \prime },I^{}),C_0(V^{},q^{},I^{})],$$
which, granting Theorem 3.4, is induced by a similarity
$$(g,l)\text{Sim}[(V^{},q^{\prime \prime },I^{}),(V^{},q^{},I^{})].$$
Hence we would have
$$g:(V^{},q^{\prime \prime },I^{})(V^{},q^{},I^{})(๐ช_X,(ss^{}s.s^{}.l^1),๐ช_X)$$
where $`l\mathrm{\Gamma }(X,๐ช_X^{}).`$ This combined with the fact that $`(V,q,I)`$ and
$$(V^{},q^{\prime \prime },I^{})=(V^{},q^{\prime \prime },IJ)$$
are isomorphic upto the twisted discriminant bundle $`(LJ^1,h,J)`$ by the construction above, would imply that $`(V,q,I)`$ and $`(V^{},q^{},I^{})`$ also differ by a twisted discriminant bundle. Therefore the proof of the injectivity asserted in Theorem 3.1 reduces to the proof of Theorem 3.4.
#### Reduction of Theorem 3.4 to the Case when $`I`$ is free.
For a similarity $`g`$ with multiplier $`l`$, we have $`C_0(g,l,I)`$ given by (1), Prop.2.8, so that we may define the map
$$\text{Sim}[(V,q,I),(V,q^{},I)]\text{Iso}[C_0(V,q,I),C_0(V,q^{},I)]:gC_0(g,l,I).$$
The equality
$$\text{det}(C_0(g,l,I))=\text{det}\left((C_0(g,l,I))_{\mathrm{\Lambda }^2}\right)=l^3\text{det}^2(g)$$
will be shown to hold (locally, hence globally) in (1), Lemma 5.11, page 5.11. Thus
$$\text{Iso}[(V,q,I),(V,q^{},I)]\text{ and }\text{S-Iso}[(V,q,I),(V,q^{},I)]$$
are respectively mapped into
$$\text{Iso}^{}[C_0(V,q,I),C_0(V,q^{},I)]\text{ and }\text{S-Iso}[C_0(V,q,I),C_0(V,q^{},I)]$$
as claimed. We start with an isomorphism of algebra-bundles
$$\varphi :C_0(V,q,I)C_0(V,q^{},I),$$
which by Prop.3.2 leads to the automorphism of vector bundles
$$\varphi _{\mathrm{\Lambda }^2}\text{Aut}(\mathrm{\Lambda }^2(V)I^1).$$
Firstly, define the global bundle automorphism
$$g^{}\text{GL}\left(V(\text{det}(V))^1I\right)$$
so that the following diagram commutes
$$\begin{array}{ccc}(\mathrm{\Lambda }^2(V))^{}I& \underset{}{\overset{((\varphi _{\mathrm{\Lambda }^2})^{})^1}{}}& (\mathrm{\Lambda }^2(V))^{}I\\ \left(\eta ^{}\right)^1I& & \left(\eta ^{}\right)^1I& & \\ V(\text{det}(V))^1I& \underset{g^{}}{\overset{}{}}& V(\text{det}(V))^1I\end{array}$$
where $`\eta :\mathrm{\Lambda }^2(V)V^{}\text{det}(V)`$ is the canonical isomorphism (since $`V`$ is of rank 3). Now let
$$g\text{GL}(V)\stackrel{}{}\text{GL}(V(\text{det}(V))^1I)$$
be the image of $`g^{}`$ i.e., the image of $`g^{}\text{det}(V)I^1`$ under the canonical identification
$$\text{GL}(V(\text{det}(V))^1I\text{det}(V)I^1)\text{GL}(V).$$
Next, let $`l\mathrm{\Gamma }(X,๐ช_X^{})`$ be a global section such that
$$\gamma (l):=(l^3).\text{det}(\varphi _{\mathrm{\Lambda }^2})$$
has a square root in $`\mathrm{\Gamma }(X,๐ช_X^{}).`$ For example, we have the following independent special cases when this is true:
If $`\varphi \text{S-Iso}[C_0(V,q,I),C_0(V,q^{},I)]`$ i.e., if $`\text{det}(\varphi _{\mathrm{\Lambda }^2})=1,`$ then set $`l=1`$ and $`\sqrt{\gamma (l)}=1.`$
If $`\varphi \text{Iso}^{}[C_0(V,q,I),C_0(V,q^{},I)]`$ i.e., $`\text{det}(\varphi _{\mathrm{\Lambda }^2})`$ is a square, then set $`l=1`$ and let $`\sqrt{\gamma (l)}`$ denote any fixed square root of $`\text{det}(\varphi _{\mathrm{\Lambda }^2}).`$
Given an integer $`k`$, take $`l=(\text{det}(\varphi _{\mathrm{\Lambda }^2}))^{2k+1}`$ and let $`\sqrt{\gamma (l)}`$ denote any fixed square root of $`(\text{det}(\varphi _{\mathrm{\Lambda }^2}))^{6k+4}.`$
For each integer $`k`$, we now associate to $`\varphi `$ the element
$$g_l^\varphi :=(l^1\sqrt{\gamma (l)})g$$
with $`g`$ as defined above. We shall show the following locally for the Zariski topology on $`X`$ (more precisely, for each open subscheme of $`X`$ over which $`V`$ and $`I`$ are free):
that $`g_l^\varphi `$ is an $`I`$-similarity from $`(V,q,I)`$ to $`(V,q^{},I)`$ with multiplier $`l`$ (Lemma 5.9, page 5.9);
that $`g_l^\varphi `$ induces $`\varphi `$ i.e., with the notations of (1), Prop.2.8, that $`C_0(g_l^\varphi ,l,I)=\varphi `$ (Lemma 5.10, page 5.10);
that $`\text{det}(g_l^\varphi )=\sqrt{\gamma (l)}`$ so that $`\text{det}^2(g_l^\varphi )=\text{det}(\varphi _{\mathrm{\Lambda }^2})`$ in cases 1 and 2 (Lemma 5.8, page 5.8) and
that the map
$$\text{S-Iso}[(V,q,I),(V,q^{},I)]\text{S-Iso}[C_0(V,q,I),C_0(V,q^{},I)]$$
is injective (Lemma 5.12, page 5.12).
It would follow then that these statements are also true globally. The maps
$$s_{2k+1}:\varphi g_l^\varphi \text{ with }l\text{ as in Case 3}$$
and
$$s^{}:\varphi g_l^\varphi \text{ with }l\text{ as in Case 2}$$
will then give the sections to the maps (which would imply their surjectivities) as mentioned in Theorem 3.4. But these maps are not necessarily multiplicative since a computation reveals that if
$$\varphi _i\text{Iso}[C_0(V,q_i,I),C_0(V,q_{i+1},I)]$$
is associated to
$$g_{l_i}^{\varphi _i}\text{Sim}[(V,q_i,I),(V,q_{i+1},I)],$$
and $`\varphi _2\varphi _1`$ to $`g_{l_{21}}^{\varphi _2\varphi _1}`$, then
$$g_{l_{21}}^{\varphi _2\varphi _1}=\delta g_{l_2}^{\varphi _2}g_{l_1}^{\varphi _1}\text{ for }\delta \mu _2(\mathrm{\Gamma }(X,๐ช_X))$$
because of the ambiguity in the initial global choices of square roots for $`\gamma (l_i)`$ and $`\gamma (l_{21}).`$ However this can be remedied as follows. For any given
$$\varphi \text{Iso}[C_0(V,q,I),C_0(V,q^{},I)],$$
irrespective of whether or not $`\text{det}(\varphi _{\mathrm{\Lambda }^2})`$ is a square, take
$$l=(\text{det}(\varphi _{\mathrm{\Lambda }^2}))^{2k+1},\gamma (l)=l^3\text{det}(\varphi _{\mathrm{\Lambda }^2}),\sqrt{\gamma (l)}:=(\text{det}(\varphi _{\mathrm{\Lambda }^2}))^{3k+2}$$
and
$$s_{2k+1}^+(\varphi ):=g_l^\varphi =\left(l^1\sqrt{\gamma (l)}\right)g.$$
Then it is clear that each $`s_{2k+1}^+`$ is multiplicative with the properties as claimed in the statement. We thus reduce the proof of Theorem 3.4 to the case when the rank 3 vector bundle $`V`$ and the line bundle $`I`$ are free. This will be treated in the next section.
## 5 The Free Case: Investigation of the Isomorphism Theta
Throughout this section, we work with $`I=๐ช_X`$ and shorten our earlier notations $`(V,q,I)`$, $`C_0(V,q,I)`$, $`C_0(g,l,I)`$ etc respectively to $`(V,q)`$, $`C_0(V,q)`$, $`C_0(g,l)`$ etc. We conclude the proofs of the injectivity of Theorem 3.1 and Theorem 3.4 which were begun in ยง4 and also prove Theorem 3.5.
As means to these ends, we carry out two explicit computations. Firstly we compute the isomorphism $`\mathrm{\Theta }`$ of Theorem 3.10. This provides us with the multiplication table of every specialised algebra structure on any fixed free rank 4 vector bundle with fixed unit which is part of a global basis (Theorem 5.1 below). This result will also be used in ยง6 in the proof of Theorem 3.11. It turns out that $`\mathrm{\Theta }`$ is not equivariant with respect to $`\text{GL}_V`$, but nevertheless satisfies a โtwistedโ form of equivariance (Theorem 5.4). Secondly, we explicitly compute the algebra bundle isomorphism
$$C_0(g,l):C_0(V,q)C_0(V,q^{})$$
of (1), Prop.2.8 induced by a similarity $`g:(V,q)_l(V,q^{})`$ with multiplier $`l\mathrm{\Gamma }(X,๐ช_X^{})`$ in the case when $`V`$ is free of rank 3 (Theorem 5.5).
### 5.1 The Action of GL on Forms
Let $`V`$ be a vector bundle over a scheme $`X`$ with associated locally-free sheaf $`๐ฑ.`$ The $`X`$-smooth $`X`$-groupscheme $`\text{GL}_V`$ acts naturally on the left on the sheaves $`\text{Alt}_๐ฑ^2`$, $`\text{Bil}_๐ฑ`$ and $`\text{Quad}_๐ฑ`$ of alternating, bilinear and quadratic forms on $`๐ฑ`$ (with values in $`๐ช_X`$). Namely, for $`UX`$ an open subscheme, and for
$$b\mathrm{\Gamma }(U,\text{Bil}_๐ฑ)\text{ (resp. }a\mathrm{\Gamma }(U,\text{Alt}_๐ฑ^2),\text{ resp. }q\mathrm{\Gamma }(U,\text{Quad}_๐ฑ)),$$
and for $`g\mathrm{\Gamma }(U,\text{GL}_V)=\text{GL}(V|U)`$, the corresponding form of the same type $`g.b`$ (resp. $`g.a`$, resp. $`g.q`$) is defined on sections (over open subsets of $`U`$) by
$$\begin{array}{cc}\hfill (g.b)(v,v^{}):=b(g^1(v),g^1(v^{}))& \text{ (resp. }(g.a)(v,v^{}):=a(g^1(v),g^1(v^{})),\hfill \\ & \text{ resp. }(g.q)(v):=q(g^1(v))).\hfill \end{array}$$
It is immediate that the following short-exact-sequence of sheaves, indicated in ยง2.1, is equivariant with respect to this action:
$$(\mathrm{})0\text{Alt}_๐ฑ^2\text{Bil}_๐ฑ\text{Quad}_๐ฑ0.$$
Equivalently, the $`X`$-groupscheme $`\text{GL}_V`$ acts on the corresponding geometric vector bundles such that both of the $`X`$-morphisms of $`X`$-vector bundles in the following sequence are $`\text{GL}_V`$-equivariant:
$$\text{Alt}_V^2\text{Bil}_V\text{Quad}_V.$$
Notice that it is one and the same thing to require that
$$\text{GL}(V|U)g:(V|U,q)_l(V|U,q^{})$$
be a similitude with multiplier $`l\mathrm{\Gamma }(U,๐ช_X)`$, and to require that $`g.q=l^1q^{}.`$
### 5.2 Computation of the Isomorphism Theta
We briefly recall the definition of $`\mathrm{\Theta }`$ from Part A of . We keep the notations introduced just before Theorem 3.10; for ease of notation, the pullback of a section $`s`$ (of a vector bundle or its associated sheaf) is denoted by $`s^{}.`$ Since $`V`$ is free of rank 3 on $`X`$, we choose an identification
$$๐ฑ๐ช_X.e_1๐ช_X.e_2๐ช_X.e_3.$$
This gives the identification of the dual bundle as
$$๐ฑ^{}๐ช_X.f_1๐ช_X.f_2๐ช_X.f_3$$
(defined uniquely by $`f_i(e_j)=\delta _{ij}`$, the Kronecker delta). Therefore the dual of the sheaf of quadratic forms on $`V`$, which is
$$(\text{Quad}_๐ฑ)^{}:=(\text{Bil}_๐ฑ/\text{Alt}_๐ฑ^2)^{}=((T^2๐ฑ)^{}/(\mathrm{\Lambda }^2๐ฑ)^{})^{},$$
has global $`๐ช_X`$-basis given by
$$\{e_ie_i;(e_ie_j+e_je_i)\}.$$
This leads to an identification of the associated sheaf of symmetric algebras
$$\text{Sym}_{๐ช_X}(\text{Quad}_๐ฑ^{})๐ช_X[Y_1,Y_2,Y_3,Y_{12},Y_{13},Y_{23}],$$
where $`e_ie_iY_i`$ and $`e_ie_j+e_je_iY_{ij}`$, and therefore
$$\text{Quad}_V:=\text{Spec}\left(\text{Sym}_{๐ช_X}(\text{Quad}_๐ฑ^{})\right)\text{Spec}(๐ช_X[Y_1,Y_2,Y_3,Y_{12},Y_{13},Y_{23}])=๐ธ_X^6.$$
Consider the universal quadratic bundle $`(๐,๐ช)`$ where $`๐`$ is the pullback of $`V`$ by $`\text{Quad}_VX.`$ The universal quadratic form $`๐ช`$ is given by
$$(x_1,x_2,x_3)\mathrm{\Sigma }_iY_i.(x_i)^2+\mathrm{\Sigma }_{i<j}Y_{ij}.x_i.x_j$$
and moreover the global bilinear form on $`๐`$ given by
$$๐(๐ช):((x_1,x_2,x_3),(x_1^{},x_2^{},x_3^{}))\mathrm{\Sigma }_iY_i.x_i.x_i^{}+Y_{12}.x_2.x_1^{}+Y_{23}.x_3.x_2^{}+Y_{13}.x_1.x_3^{}$$
induces $`๐ช`$ (the bilinear form โassociated in the usual senseโ to $`๐ช`$, viz. $`b_๐ช`$ is not $`๐(๐ช)`$ but in fact its symmetrisation). Therefore, by (2d), Theorem 2.2, we get an isomorphism of vector bundles
$$\psi _{๐(๐ช)}:C_0(๐,๐ช=q_{๐(๐ช)})\mathrm{\Lambda }^{even}(๐)=:๐$$
which, according to (2a) and (2f) of the same Theorem, carries the ordered Poincarรฉ-Birkhoff-Witt basis
$$\{1;e_1^{}.e_2^{},e_2^{}.e_3^{},e_3^{}.e_1^{}\}$$
onto the corresponding ordered basis of the even exterior algebra (=even Clifford algebra of the zero quadratic form on $`๐`$) given by
$$\{w^{}=1^{}=1;e_1^{}e_2^{},e_2^{}e_3^{},e_3^{}e_1^{}\}.$$
The choices $`e_3^{}.e_1^{}`$ and $`e_3^{}e_1^{}`$ instead of the usual $`e_1^{}.e_3^{}`$ and $`e_1^{}e_3^{}`$ are deliberateโfor example, $`\psi _{๐(๐ช)}`$ would carry
$$\{1;e_1^{}.e_2^{},e_2^{}.e_3^{},e_1^{}.e_3^{}\}$$
onto
$$\{w^{}=1^{}=1;e_1^{}e_2^{},e_2^{}e_3^{},e_1^{}e_3^{}+Y_{13}.w^{}\}$$
which depends on $`Y_{13}.`$ Thus the even Clifford algebra bundle $`C_0(๐,๐ช=q_{๐(๐ช)})`$ induces via $`\psi _{๐(๐ช)}`$ a $`w^{}`$-unital algebra structure on the pullback bundle $`๐`$ of $`W:=\mathrm{\Lambda }^{even}(V)`$ (where $`w`$ corresponds to 1 in $`\mathrm{\Lambda }^{even}(V)`$). But by definition, this algebra structure corresponds precisely to an $`X`$-morphism
$$\theta :\text{Quad}_V\text{Id-}w\text{-Sp-Azu}_W.$$
The isomorphism $`\mathrm{\Theta }`$ is now given by the composition of the following $`X`$-morphisms (cf. Def.5.2, Part A, ):
$$\text{Quad}_V\times _X\text{L}_w\stackrel{\theta \times \text{ID}}{}\text{Id-}w\text{-Sp-Azu}_W\times _X\text{L}_w\stackrel{\text{SWAP}()}{}$$
$$\text{L}_w\times _X\text{Id-}w\text{-Sp-Azu}_W\stackrel{\text{ACTION}}{}\text{Id-}w\text{-Sp-Azu}_W.$$
The association of $`๐ช`$ with $`๐(๐ช)`$ also defines a splitting of the exact sequence $`(\mathrm{})`$ of page 5.1 above, so that more generally, given a valued point $`q(\text{Quad}_V)(T)`$, we may associate uniquely a valued point $`b(q)(\text{Bil}_V)(T)`$ which induces it. That this association is not $`\text{GL}_V`$-equivariant is reflected in the lack of equivariance of the isomorphism $`\mathrm{\Theta }`$ (Theorem 5.4).
###### Theorem 5.1
Let $`T`$ be an $`X`$-scheme. Let $`q`$ be a $`T`$-valued point of $`\text{Quad}_V๐ธ_X^6`$ which is identified uniquely with a 6-tuple
$$(\lambda _1,\lambda _2,\lambda _3,\lambda _{12},\lambda _{13},\lambda _{23})$$
corresponding to the quadratic form
$$(x_1,x_2,x_3)\mathrm{\Sigma }_i\lambda _ix_i^2+\mathrm{\Sigma }_{i<j}\lambda _{ij}x_ix_j.$$
Let $`\underset{ยฏ}{t}`$ be a $`T`$-valued point of $`\text{L}_w(๐ธ_X^3,+)`$ which is identified uniquely with a 3-tuple $`(t_1,t_2,t_3)`$ that corresponds to the $`T`$-valued point of $`\text{Stab}_w`$ given by the $`(4\times 4)`$-matrix
$$\left(\begin{array}{cccc}1& t_1& t_2& t_3\\ & & & \\ 0& & I_3& \end{array}\right)$$
where $`I_3`$ is the $`(3\times 3)`$-identity matrix. Then in terms of the global basis
$$\{w^{}=1^{}=1;ฯต_1^{}:=e_1^{}e_2^{},ฯต_2^{}:=e_2^{}e_3^{},ฯต_3^{}:=e_3^{}e_1^{}\}$$
induced from that of $`W=\mathrm{\Lambda }^{even}(V),`$ the multiplication table for the specialised algebra structure
$$\mathrm{\Theta }(q,\underset{ยฏ}{t})=\underset{ยฏ}{t}.\theta (q)$$
on the pullback bundle $`W_T`$ with unit $`w^{}=w_T`$ is given as follows:
1. $`ฯต_1^{}.ฯต_1^{}=(t_1\lambda _{12}\lambda _1\lambda _2t_1^2).w^{}+(\lambda _{12}2t_1).ฯต_1^{};`$
2. $`ฯต_2^{}.ฯต_2^{}=(t_2\lambda _{23}\lambda _2\lambda _3t_2^2).w^{}+(\lambda _{23}2t_2).ฯต_2^{};`$
3. $`ฯต_3^{}.ฯต_3^{}=(t_3\lambda _{13}\lambda _1\lambda _3t_3^2).w^{}+(\lambda _{13}2t_3).ฯต_3^{};`$
4. $`ฯต_1^{}.ฯต_2^{}=(\lambda _2\lambda _{13}\lambda _2t_3t_1t_2).w^{}t_2ฯต_1^{}t_1ฯต_2^{}\lambda _2ฯต_3^{};`$
5. $`ฯต_2^{}.ฯต_3^{}=(\lambda _3\lambda _{12}\lambda _3t_1t_2t_3).w^{}\lambda _3ฯต_1^{}t_3ฯต_2^{}t_2ฯต_3^{};`$
6. $`ฯต_3^{}.ฯต_1^{}=(\lambda _1\lambda _{23}\lambda _1t_2t_1t_3).w^{}t_3ฯต_1^{}\lambda _1ฯต_2^{}t_1ฯต_3^{};`$
7. $`ฯต_2^{}.ฯต_1^{}=(\lambda _2t_3(\lambda _{12}t_1)(\lambda _{23}t_2)).w^{}+(\lambda _{23}t_2)ฯต_1^{}+(\lambda _{12}t_1)ฯต_2^{}+\lambda _2ฯต_3^{};`$
8. $`ฯต_3^{}.ฯต_2^{}=(\lambda _3t_1(\lambda _{13}t_3)(\lambda _{23}t_2)).w^{}+\lambda _3ฯต_1^{}+(\lambda _{13}t_3)ฯต_2^{}+(\lambda _{23}t_2)ฯต_3^{};`$
9. $`ฯต_1^{}.ฯต_3^{}=(\lambda _1t_2(\lambda _{12}t_1)(\lambda _{13}t_3)).w^{}+(\lambda _{13}t_3)ฯต_1^{}+\lambda _1ฯต_2^{}+(\lambda _{12}t_1)ฯต_3^{}`$.
#### Proof of Theorem 5.1.
For clarity, let $`_q`$ denote the multiplication in the algebra $`C_0(V_T,q)`$, and for uniformity, let $`ฯต_0:=w.`$ Since $`q=q_{b(q)}`$, we have by (2d), Theorem 2.2, the isomorphism
$$\psi _{b(q)}:C_0(V_T,q)\mathrm{\Lambda }^{even}(V_T)=W_T.$$
Let $`_{b(q)}`$ denote the product in the algebra structure $`\theta (q)`$ thus induced on $`W_T.`$ Since the $`ฯต_i^{}`$ are a basis for $`W_T`$, it is enough to compute the products $`ฯต_i^{}_{b(q)}ฯต_j^{}`$ for $`1i,j3.`$ For example, consider the product $`ฯต_2^{}_{b(q)}ฯต_1^{}.`$ Using the properties of the multiplication in $`C(V_T,q)`$, and the properties of the isomorphism $`\psi _{b(q)}`$ from (2), Theorem 2.2, we get the following:
$$\begin{array}{cc}\hfill ฯต_2^{}_{b(q)}ฯต_1^{}& =\psi _{b(q)}\left(\{\psi _{b(q)}^1(e_2^{}e_3^{})\}_q\{\psi _{b(q)}^1(e_1^{}e_2^{})\}\right)\hfill \\ & =\psi _{b(q)}\left((e_2^{}_qe_3^{})_q(e_1^{}_qe_2^{})\right)\hfill \\ & =\psi _{b(q)}\left((\lambda _{23}(1^{})e_3^{}_qe_2^{})_q(\lambda _{12}(1^{})e_2^{}_qe_1^{})\right)\hfill \\ & =\psi _{b(q)}\left(\lambda _{23}\lambda _{12}(1^{})\lambda _{23}e_2^{}_qe_1^{}\lambda _{12}e_3^{}_qe_2^{}+(e_3^{}_qe_2^{})_q(e_2^{}_qe_1^{})\right)\hfill \\ & =\psi _{b(q)}(\lambda _{23}\lambda _{12}(1^{})\lambda _{23}(\lambda _{12}(1^{})e_1^{}_qe_2^{})\hfill \\ & \lambda _{12}(\lambda _{23}(1^{})e_2^{}_qe_3^{})+e_3^{}_q(e_2^{}_qe_2^{})_qe_1^{})\hfill \\ & =(\lambda _{12}\lambda _{23})w^{}+\lambda _{23}ฯต_1^{}+\lambda _{12}ฯต_2^{}+\lambda _2ฯต_3^{}.\hfill \end{array}$$
In a similar fashion, the other products may be computed; this amounts to computing $`\theta `$ on $`T`$-valued points. The following result is needed to compute $`\mathrm{\Theta }`$ from $`\theta .`$
###### Lemma 5.2
Let $`_{(b(q),\underset{ยฏ}{t})}`$ denote the multiplication in the algebra $`\mathrm{\Theta }(q,\underset{ยฏ}{t})=\underset{ยฏ}{t}.\theta (q)`$ and as before, $`_{b(q)}`$ denote the multiplication in $`\theta (q).`$ Then we have
$`\underset{ยฏ}{t}(ฯต_i^{})=t_iw^{}+ฯต_i^{}`$ for $`1i3;`$
$`(\underset{ยฏ}{t})^1(ฯต_i^{})=t_iw^{}+ฯต_i^{}`$ for $`1i3;`$
$`ฯต_i^{}_{(b(q),\underset{ยฏ}{t})}ฯต_j^{}=\underset{ยฏ}{t}(ฯต_i^{}_{b(q)}ฯต_j^{})t_jฯต_i^{}t_iฯต_j^{}t_it_jw^{}.`$
While the first two of the above formulae follow easily by direct computation, the third follows by using the first two alongwith the following:
$$ฯต_i^{}_{(b(q),\underset{ยฏ}{t})}ฯต_j^{}=\underset{ยฏ}{t}\left((\underset{ยฏ}{t}^1(ฯต_i^{}))_{b(q)}(\underset{ยฏ}{t}^1(ฯต_j^{}))\right).$$
We may now compute the multiplication in the algebra $`\mathrm{\Theta }(q,\underset{ยฏ}{t})=\underset{ยฏ}{t}.\theta (q)`$ by making use of the formulas listed in the above lemma and the expressions for the products of the form $`ฯต_i^{}_{b(q)}ฯต_j^{}`$ whose computation had already been illustrated before the lemma. Q. E. D., Theorem 5.1.
### 5.3 Computation of the Isomorphism arising from a Similarity
We continue with the notations introduced above. In the following we study the lack of equivariance of the isomorphism $`\mathrm{\Theta }`$ relative to $`\text{GL}_V`$ and show that it satisfies a curious โtwistedโ version of equivariance. Firstly we consider the morphism of $`X`$-groupschemes
$$\mathrm{\Lambda }^{even}:\text{GL}_V\text{Stab}_w\text{ (given on valued points by) }g\mathrm{\Lambda }^{even}(g).$$
Recall that
$$\mathrm{\Lambda }^{even}(V)=:A_0\text{Id-}w\text{-Sp-Azu}_W(X)$$
is the even graded part of the Clifford algebra of the zero quadratic form on $`V.`$ A simple computation reveals the following result.
###### Lemma 5.3
For each $`X`$-scheme $`T`$, define the map
$$\text{GL}(V_T)\text{Stab}((A_0)_T):g\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& B(g)\end{array}\right)\text{Stab}(w_T)$$
where $`B(g):=\text{det}(g)\left(E_{12}E_{23}(g^1)^tE_{23}E_{12}\right)`$ with
$$E_{12}=\left(\begin{array}{ccc}0& 1& 0\\ 1& 0& 0\\ 0& 0& 1\end{array}\right)\text{ and }E_{23}=\left(\begin{array}{ccc}1& 0& 0\\ 0& 0& 1\\ 0& 1& 0\end{array}\right).$$
Then the above maps define a morphism of $`X`$-groupschemes which in fact is none other than
$$\mathrm{\Lambda }^{even}:\text{GL}_V\text{Stab}_w;$$
in other words: $`B(g)=\mathrm{\Lambda }^2(g).`$
Recall from Lemma 5.1, Part A, , that $`\text{Stab}_w`$ is the semidirect product of $`\text{Stab}_{A_0}`$ and $`\text{L}_w`$, so that $`\text{Stab}_{A_0}`$ naturally acts on $`\text{L}_w`$ by โouter conjugationโ. Let $`\text{GL}_V`$ act on $`\text{L}_w`$ via the homomorphism $`\mathrm{\Lambda }^{even}`$ i.e., for $`g\text{GL}(V_T)`$ and $`\underset{ยฏ}{t}\text{L}_w(T)`$,
$$g.\underset{ยฏ}{t}:=\mathrm{\Lambda }^{even}(g).\underset{ยฏ}{t}:=\mathrm{\Lambda }^{even}(g)\underset{ยฏ}{t}\mathrm{\Lambda }^{even}(g^1).$$
Any element $`h\text{Stab}(w_T)`$ can be uniquely written as
$$h=h_sh_l=h_l^{}h_s\text{ where }h_s\text{Stab}((A_0)_T)\text{ and }h_l,h_l^{}\text{L}_w(T).$$
Then the relation between $`h_l`$ and $`h_l^{}`$ can be written as
$$h_l^{}=h_s.h_l\text{ or }h_l=h_s^1.h_l^{}$$
where โ.โ stands for the action of $`\text{Stab}_{A_0}`$ on $`\text{L}_w.`$ Thus one has a $`\text{GL}_V`$-action on $`\text{Quad}_V\times _X\text{L}_w`$ induced by the diagonal embedding
$$\text{GL}_V\stackrel{\mathrm{\Delta }}{}\text{GL}_V\times _X\text{GL}_V.$$
Since $`\text{Id-}w\text{-Sp-Azu}_W`$ comes with a canonical action of $`\text{Stab}_w`$ on it, we let $`\text{GL}_V`$ act on $`\text{Id-}w\text{-Sp-Azu}_W`$ via $`\mathrm{\Lambda }^{even}.`$ The following result describes the lack of $`\text{GL}_V`$-equivariance of the isomorphism $`\mathrm{\Theta }.`$
###### Theorem 5.4
Let $`T`$ be an $`X`$-scheme. For $`T`$-valued points $`g,q,\underset{ยฏ}{t}`$ respectively of $`\text{GL}_V`$, $`\text{Quad}_V`$, and $`\text{L}_w`$, there exists a unique $`T`$-valued point of $`\text{L}_w`$ given by an isomorphism $`h_l^{}(g,q)`$ of $`๐ช_T`$-algebra bundles
$$h_l^{}(g,q):g.\mathrm{\Theta }(q,\underset{ยฏ}{t})\stackrel{}{}\mathrm{\Theta }(g.q,g.\underset{ยฏ}{t}).$$
Further, $`h_l^{}(g,q)`$ satisfies the formula
$$h_l^{}(gg^{},q)=h_l^{}(g,g^{}.q)(g.h_l^{}(g^{},q)).$$
Therefore $`\mathrm{\Theta }`$ satisfies a โtwistedโ version of $`\text{GL}_V`$-equivariance. The next theorem, which was originally motivated by the proof of this โtwisted equivarianceโ, will be of central importance to us for the rest of this section.
###### Theorem 5.5
Given a similarity
$$g:(V_T,q)_l(V_T,q^{})$$
with multiplier $`l\mathrm{\Gamma }(T,๐ช_T^{})`$, let $`h(g,l,q,q^{})`$ be the automorphism of $`(W_T,w_T)`$ given by the composition of the following isomorphisms:
$$W_T\stackrel{(\psi _{b(q)})^1()}{}C_0(V_T,q)\stackrel{C_0(g,l)()}{}C_0(V_T,q^{})\stackrel{\psi _{b(q^{})}()}{}W_T$$
where the algebra bundle isomorphism $`C_0(g,l)`$ comes from (1), Prop.2.8 and the linear isomorphisms $`\psi _{b(q)}`$ and $`\psi _{b(q^{})}`$ come from (2d), Theorem 2.2. In terms of actions, this means that
$$h(g,l,q,q^{}).\theta (q)=\theta (q^{}).$$
Write $`h(g,l,q,q^{})\text{Stab}(w_T)`$ uniquely as a product
$$h(g,l,q,q^{})=h_s(g,l,q,q^{})h_l(g,l,q,q^{})$$
with the first factor in $`\text{Stab}_{A_0}(T)`$ and the second in $`\text{L}_w(T)`$ as explained earlier. Then $`h_s(g,l,q,q^{})`$ depends only on $`g`$ and $`l`$ and not on $`q`$ or $`q^{}.`$ In fact, one has
$$h_s(g,l,q_1,q_2)=h_s(g,l):=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l^1\mathrm{\Lambda }^2(g)\end{array}\right)q_1,q_2\text{Quad}(V_T).$$
#### Proof:
We directly compute the $`๐ช_T`$-linear automorphism $`h(g,l,q,q^{})`$ of $`W_T`$ as follows. Of course, this automorphism fixes $`w^{}=w_T.`$ So we need to only compute the images of the three remaining basis elements
$$ฯต_1^{}=e_1^{}e_2^{},ฯต_2^{}=e_2^{}e_3^{}\text{ and }ฯต_3^{}=e_3^{}e_1^{}$$
in terms of the basis elements $`w^{}`$ and $`ฯต_i^{}.`$ Let $`q`$ and $`l(g.q)=q^{}`$ respectively correspond to the 6-tuples
$$(\mu _1,\mu _2,\mu _3,\mu _{12},\mu _{13},\mu _{23})\text{ and }(\nu _1,\nu _2,\nu _3,\nu _{12},\nu _{13},\nu _{23})\mathrm{\Gamma }(T,๐ช_{T}^{}{}_{}{}^{6}).$$
(We caution the reader that $`l(g.q)(lg).q=l^2(g.q)`$!) Let
$$g\text{GL}(V_T)\text{GL}_3(\mathrm{\Gamma }(T,๐ช_T))$$
be given by the matrix $`(g_{ij}).`$ Observe that the $`\nu `$ are polynomials in the $`\mu `$ and $`g_{ij}.`$ In the following computation, for the sake of clarity, we denote the product in $`C(V_T,q)`$ by $`_q.`$ For example, we have
$$\begin{array}{cc}\hfill h(g,l,q,q^{})ฯต_1& =\psi _{b(q^{})}C_0(g,l)\left((\psi _{b(q)})^1(e_1^{}e_2^{})\right)\hfill \\ & =\psi _{b(q^{})}\left(C_0(g,l)\left(e_1^{}_qe_2^{}\right)\right)\text{ (by (2f), Theorem }\text{2.2}\text{)}\hfill \\ & =\psi _{b(q^{})}\left(l^1(g(e_1^{})_q^{}g(e_2^{}))\right)\text{ (by (1), Prop.}\text{2.8}\text{)}\hfill \\ & =l^1\psi _{b(q^{})}\left((g_{11}e_1^{}+g_{21}e_2^{}+g_{31}e_3^{})_q^{}(g_{12}e_1^{}+g_{22}e_2^{}+g_{32}e_3^{})\right)\hfill \\ & =l^1\psi _{b(q^{})}((g_{11}g_{12}\nu _1+g_{21}g_{22}\nu _2+g_{31}g_{32}\nu _3+\hfill \\ & +g_{21}g_{12}\nu _{12}+g_{31}g_{22}\nu _{23}+g_{11}g_{32}\nu _{13})w^{}+\hfill \\ & +(g_{11}g_{22}g_{21}g_{12})e_1^{}_q^{}e_2^{}+(g_{21}g_{32}g_{31}g_{22})e_2^{}_q^{}e_3^{}+\hfill \\ & +(g_{31}g_{12}g_{11}g_{32})e_3^{}_q^{}e_1^{})\hfill \\ & =l^1\left(P_1(g,l,q,q^{})w^{}+C_{33}(g)ฯต_1^{}+C_{13}(g)ฯต_2^{}+C_{23}(g)ฯต_3^{}\right)\hfill \\ & \text{ (by (2f), Theorem }\text{2.2}\text{)}\hfill \end{array}$$
where $`P_1(g,l,q,q^{})`$ is the polynomial in the $`\nu `$ and $`g_{ij}`$ (as computed in the previous step) and where $`C_{ij}(g)`$ represents the cofactor determinant of the element $`g_{ij}`$ of the matrix $`g=(g_{ij}).`$ Similarly one computes the values of $`h(g,l,q,q^{})ฯต_2`$ and $`h(g,l,q,q^{})ฯต_3.`$ Then the matrix of $`h(g,l,q,q^{})`$ is given by
$$h(g,l,q,q^{})=\left[\begin{array}{cccc}1& l^1P_1(g,l,q,q^{})& l^1P_2(g,l,q,q^{})& l^1P_3(g,l,q,q^{})\\ 0& l^1C_{33}(g)& l^1C_{31}(g)& l^1C_{32}(g)\\ 0& l^1C_{13}(g)& l^1C_{11}(g)& l^1C_{12}(g)\\ 0& l^1C_{23}(g)& l^1C_{21}(g)& l^1C_{22}(g)\end{array}\right]$$
which implies that
$$h_s(g,l,q,q^{})=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& l^1C_{33}(g)& l^1C_{31}(g)& l^1C_{32}(g)\\ 0& l^1C_{13}(g)& l^1C_{11}(g)& l^1C_{12}(g)\\ 0& l^1C_{23}(g)& l^1C_{21}(g)& l^1C_{22}(g)\end{array}\right]\text{ depends only on }g\text{ and }l.$$
Next define the matrix
$$\widehat{g}=\left[\begin{array}{ccc}g_{33}& g_{13}& g_{23}\\ g_{31}& g_{11}& g_{21}\\ g_{32}& g_{12}& g_{22}\end{array}\right]\text{ so that }h_s(g,l,q,q^{})=\left[\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l^1C(\widehat{g})^t\end{array}\right]$$
where $`C(\widehat{g})`$ is the cofactor matrix of $`\widehat{g}.`$ Now if $`E_{12}`$ and $`E_{23}`$ are the matrices defined in Lemma 5.3 above, premultiplying by $`E_{ij}`$ has the effect of interchanging the $`i`$th and $`j`$th rows, while postmultiplying has a similar effect on the columns. Thus we get
$$\widehat{g}=E_{12}E_{23}(g^t)E_{23}E_{12}$$
from which it follows that
$$C(\widehat{g})^t=\text{ Adjoint }(\widehat{g})=\text{ det }(\widehat{g}).(\widehat{g})^1=\text{ det }(g).\left(E_{12}E_{23}(g^1)^tE_{23}E_{12}\right),$$
showing that $`C(\widehat{g})^t=\mathrm{\Lambda }^2(g)`$ by Lemma 5.3. Q.E.D., Theorem 5.5.
#### Proof of Theorem 5.4.
Note that $`g`$ is an isometry from $`(V_T,q)`$ to $`(V_T,g.q)`$ and hence according to (1), Prop.2.8, induces the algebra isomorphism
$$C_0(g,l=1):C_0(V_T,q)C_0(V_T,g.q).$$
Let $`h(g,q):=h(g,1,q,g.q)`$ where $`h(g,l,q,q^{})`$ was defined in Theorem 5.5 above. As explained in page 5.3, there are two canonical decompositions of $`\text{Stab}(w_T)`$, which lead to unique (ordered) decompositions of $`h(g,q)`$ as
$$h_l^{}(g,q)h_s(g,q)\text{ and }h_s(g,q)h_l(g,q).$$
By Theorem 5.5 above,
$$h_s(g,q_1)=h_s(g,q_2)=h_s(g,1)=\mathrm{\Lambda }^{even}(g)=:h_s(g)q_1,q_2\text{Quad}(V_T)$$
and hence we get:
$$\begin{array}{cc}\hfill \mathrm{\Theta }(g.q,g.\underset{ยฏ}{t})& :=(g.\underset{ยฏ}{t}).\theta (g.q)\hfill \\ & =(\mathrm{\Lambda }^{even}(g)\underset{ยฏ}{t}\mathrm{\Lambda }^{even}(g^1)).(h(g,1,q,g.q).\theta (q))\hfill \\ & =(h_s(g)\underset{ยฏ}{t}h_s^1(g)).(h(g,q).\theta (q))\hfill \\ & =((h_s(g)\underset{ยฏ}{t}h_s^1(g))(h_s(g,q)h_l(g,q))).\theta (q)\hfill \\ & =(h_s(g)\underset{ยฏ}{t}h_l(g,q)).\theta (q)\hfill \\ & =(h_s(g)h_l(g,q)).(\underset{ยฏ}{t}.\theta (q))\hfill \\ & =(h_s(g,q)h_l(g,q)).(\mathrm{\Theta }(q,\underset{ยฏ}{t}))\hfill \\ & =(h_l^{}(g,q)h_s(g,q)).(\mathrm{\Theta }(q,\underset{ยฏ}{t}))\hfill \\ & =h_l^{}(g,q).(h_s(g,q).\mathrm{\Theta }(q,\underset{ยฏ}{t}))\hfill \\ & =h_l^{}(g,q).(g.\mathrm{\Theta }(q,\underset{ยฏ}{t})).\hfill \end{array}$$
Note that $`h_l(g,q)`$ was explicitly computed in the proof of Theorem 5.5 above to be
$$h_l(g,q)=\left[\begin{array}{cccc}1& P_1(g,1,q,g.q)& P_2(g,1,q,g.q)& P_3(g,1,q,g.q)\\ \mathrm{๐}& & I_3& \end{array}\right]\text{L}_w(T).$$
The formula for $`h_l^{}(g_1g_2,q)`$ stated in the theorem is gotten thus:
$$\begin{array}{cc}\hfill h_l^{}(g_1g_2,q)& =(h_l^{}(g_1g_2,q)h_s(g_1g_2))h_s^1(g_1g_2)\hfill \\ & =h(g_1g_2,q)h_s^1(g_1g_2).\hfill \end{array}$$
Now by (3) of Prop.2.8 it follows that
$$\begin{array}{cc}\hfill h_l^{}(g_1g_2,q)& =(h(g_1,g_2.q)h(g_2,q))h_s^1(g_1g_2)\hfill \\ & =h_l^{}(g_1,g_2.q)h_s(g_1,g_2.q)h_s(g_2,q)h_l(g_2,q)h_s^1(g_1g_2)\hfill \\ & =h_l^{}(g_1,g_2.q)h_s(g_1)h_s(g_2)h_l(g_2,q)h_s^1(g_1g_2)\hfill \\ & =h_l^{}(g_1,g_2.q)h_s(g_1g_2)h_l(g_2,q)h_s^1(g_1g_2)\hfill \\ & =h_l^{}(g_1,g_2.q)((g_1g_2).h_l(g_2,q)).\hfill \end{array}$$
On the other hand
$$g_2^1.h_l^{}(g_2,q)=h_s^1(g_2)(h_l^{}(g_2,q)h_s(g_2))=h_s^1(g_2)(h_s(g_2)h_l(g_2,q))=h_l(g_2,q)$$
and therefore
$$\begin{array}{cc}\hfill h_l^{}(g_1g_2,q)& =h_l^{}(g_1,g_2.q).((g_1g_2).(g_2^1.h_l^{}(g_2,q)))\hfill \\ & =h_l^{}(g_1,g_2.q).(g_1.h_l^{}(g_2,q)).\hfill \end{array}$$
Finally, one has to show the uniqueness of $`h_l^{}(g,q)\text{L}_w(T).`$ Suppose $`h_l\text{L}_w(T)`$ is also an algebra isomorphism
$$h_l:g.\mathrm{\Theta }(q,\underset{ยฏ}{t})\stackrel{}{}\mathrm{\Theta }(g.q,g.\underset{ยฏ}{t}),$$
which means
$$h_l.(g.\mathrm{\Theta }(q,\underset{ยฏ}{t}))=\mathrm{\Theta }(g.q,g.\underset{ยฏ}{t}).$$
Notice that while showing the โtwistedโ equivariance of $`\mathrm{\Theta }`$ above, we have also proved that
$$\mathrm{\Theta }(g.q,g.\underset{ยฏ}{t})=h(g,q)\mathrm{\Theta }(q,\underset{ยฏ}{t}).$$
Therefore we get
$$\begin{array}{cc}\hfill (h_lh_s(g)).(\underset{ยฏ}{t}.\theta (q))& =h(g,q).\mathrm{\Theta }(q,\underset{ยฏ}{t})\hfill \\ \hfill (h_lh_s(g)).(\underset{ยฏ}{t}.\theta (q))& =(h_s(g)h_l(g,q)).\mathrm{\Theta }(q,\underset{ยฏ}{t})\hfill \\ \hfill (h_s^1(g)h_lh_s(g)).(\underset{ยฏ}{t}.\theta (q))& =(h_l(g,q)\underset{ยฏ}{t}).\theta (q)\hfill \\ \hfill \mathrm{\Theta }(q,(g^1.h_l)\underset{ยฏ}{t})& =\mathrm{\Theta }(q,h_l(g,q)\underset{ยฏ}{t}).\hfill \end{array}$$
But since $`\mathrm{\Theta }`$ is an isomorphism by Theorem 3.10, this implies that
$$(g^1.h_l)\underset{ยฏ}{t}=h_l(g,q)\underset{ยฏ}{t}$$
which gives
$$h_l=h_s(g)h_l(g,q)h_s^1(g)=h_l^{}(g,q).$$
Q.E.D., Theorem 5.4.
### 5.4 Conclusion of Proof of Injectivity
We remind the reader that towards the end of ยง4, we had reduced the proof of the injectivity of Theorem 3.1 to that of Theorem 3.4, and had indicated in page 4 that it would be enough to prove the latter in the case when $`V`$ and $`I`$ are both freeโwhich has been the case in this section so far. Starting with an isomorphism of algebra bundles
$$\varphi :C_0(V,q)C_0(V,q^{})$$
we arrive at the element $`g_l^\varphi \text{GL}(V)`$ as defined in page 4; to briefly recall this, firstly $`g\text{GL}(V)`$ was defined by the following commuting diagram:
$$\begin{array}{ccccccc}(C_0(V,q))^{}& \underset{}{\overset{((\psi _{b(q)})^{})^1}{}}& (\mathrm{\Lambda }^{even}(V))^{}& \stackrel{\text{surjection}}{}& (\mathrm{\Lambda }^2(V))^{}& \stackrel{=}{}& \\ \varphi ^{}& & \left(\varphi _{\mathrm{\Lambda }^{ev}}\right)^{}& & \left(\varphi _{\mathrm{\Lambda }^2}\right)^{}& & \\ (C_0(V,q^{}))^{}& \underset{}{\overset{((\psi _{b(q^{})})^{})^1}{}}& (\mathrm{\Lambda }^{even}(V))^{}& \stackrel{\text{inclusion}}{}& (\mathrm{\Lambda }^2(V))^{}& \stackrel{=}{}& \end{array}$$
$$\begin{array}{ccccc}(\mathrm{\Lambda }^2(V))^{}& \underset{}{\overset{(\eta ^{})^1}{}}& V(\text{det}(V))^1& \underset{}{\overset{\text{det}(V)}{}}& V\\ \left(\varphi _{\mathrm{\Lambda }^2}\right)^{}& & \left(g^{}\right)^1& & g^1& & \\ (\mathrm{\Lambda }^2(V))^{}& \underset{}{\overset{(\eta ^{})^1}{}}& V(\text{det}(V))^1& \underset{}{\overset{\text{det}(V)}{}}& V\end{array}$$
Secondly, we had defined
$$g_l^\varphi :=(l^1\sqrt{\gamma (l)})g.$$
Our current special choices of bilinear forms $`b(q)`$ and $`b(q^{})`$ that induce $`q`$ and $`q^{}`$ respectively do not affect the generality, as was observed in the proof of Prop.3.2. We shall now show that $`g_l^\varphi `$ is a similitude from $`(V,q)`$ to $`(V,q^{})`$ with multiplier $`l`$ and that this similitude induces $`\varphi `$ i.e., with the notations of (1), Prop. 2.8, that $`C_0(g_l^\varphi ,l)=\varphi .`$ We proceed with the proof which will follow from several lemmas.
###### Lemma 5.6
Consider the element
$$h_sh_l=h_l^{}h_s=h:=\varphi _{\mathrm{\Lambda }^{ev}}(\text{Stab}_w)(X)$$
written uniquely as an ordered product in two different ways with $`h_l,h_l^{}(\text{L}_w)(X)`$ and $`h_s(\text{Stab}_{A_0})(X)`$ as explained in page 5.3; let $`B`$ be the matrix corresponding to $`\varphi _{\mathrm{\Lambda }^2}`$, and let the matrices $`E_{ij}`$ be as defined in Lemma 5.3. Then we have matrix representations:
$$h_s=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& B\end{array}\right)\text{ and }g_l^\varphi =\left(l^1\sqrt{\gamma (l)}\right)E_{23}E_{12}((B)^t)^1E_{12}E_{23}$$
The proof of the above lemma follows from the fact that the matrix of the canonical isomorphism
$$\eta :\mathrm{\Lambda }^2(V)V^{}\text{det}(V)$$
is given by $`E_{23}E_{12},`$ which can be verified by a simple computation.
###### Lemma 5.7
We have the formulae
$$h_s(\text{Identity},l^1,q^{},l^1q^{})=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l\times I_3\end{array}\right)$$
and
$$h_s(\text{Identity},l^1,q^{},l^1q^{}).\theta (q^{})=\theta (l^1q^{}).$$
The identity map on $`V`$ is obviously a similarity with multiplier $`l^1`$ from $`(V,q^{})`$ to $`(V,l^1q^{}).`$ Hence the above lemma follows by taking $`T=X`$, $`g=\text{Identity}`$, and the $`l^1`$ and the $`q^{}`$ at hand for the $`l`$ and the $`q`$ of Theorem 5.5 (caution: the $`q^{}`$ there would have to be replaced by $`l^1q^{}`$!). This can also be seen directly from the multiplication tables for
$$\theta (l^1q^{})=\mathrm{\Theta }(l^1q^{},I_4)\text{ and }\theta (q^{})=\mathrm{\Theta }(q^{},I_4)$$
written out in Theorem 5.1, where we must take $`T=X`$ and $`\underset{ยฏ}{t}=I_4`$ i.e., $`t_i=0i.`$ We observe from the multiplication table that each of the coefficients of $`ฯต_i`$ for $`1i3`$ is a single $`\lambda `$, whereas each coefficient of $`w=1=ฯต_0`$ is a product of two $`\lambda `$s, and this observation implies the lemma above.
As the reader might have noticed, there are two crucial facts about the identifications in this section; namely, firstly, for any $`X`$-scheme $`T`$, each of the maps $`\psi _{b(q)}`$ (for different $`q`$) identify $`(C_0(V_T,q),1)`$ with the same $`(W_T,w_T)`$ and secondly, relative to the bases chosen, all these identifying maps have trivial determinant. The latter is also true of the identification $`\eta `$, since it is given by the matrix $`E_{23}E_{12}`$ (as was noted after Lemma 5.6). It therefore follows that
$$\text{det}(\varphi )=\text{det}(\varphi _{\mathrm{\Lambda }^{ev}})=\text{det}(\varphi _{\mathrm{\Lambda }^2})=\text{det}(g^{})=\text{det}(g)=\text{det}(B^1).$$
But we had chosen $`l\mathrm{\Gamma }(X,๐ช_X^{})`$ such that
$$\gamma (l):=(l^3).\text{det}(\varphi _{\mathrm{\Lambda }^2})=l^3\text{det}(B).$$
Using these facts alongwith Lemma 5.6 above, a straightforward computation gives the following.
###### Lemma 5.8
We have the equality
$$\text{det}(g_l^\varphi )=\sqrt{\gamma (l)}$$
from which it follows that
$$B(g_l^\varphi )=l\times B$$
where $`B(g_l^\varphi )`$ and $`B`$ are as defined in Lemmas 5.3 and 5.6 respectively. In particular,
$$\text{det}^2(g_l^\varphi )=\text{det}(\varphi _{\mathrm{\Lambda }^2})$$
when $`\text{det}(\varphi _{\mathrm{\Lambda }^2})`$ is itself a square and for the cases 1 and 2 of page Case 1. where we had chosen $`l:=1.`$
###### Lemma 5.9
$`g_l^\varphi `$ is a similitude from $`(V,q)`$ to $`(V,q^{})`$ with multiplier $`l.`$
The hypothesis $`\varphi :C_0(V,q)C_0(V,q^{})`$ is an algebra isomorphism translates in terms of actions into $`h.\theta (q)=\theta (q^{})`$ where $`h=\varphi _{\mathrm{\Lambda }^{ev}}(\text{Stab}_w)(X).`$ Let
$$h(g_l^\varphi ,q):=h(g_l^\varphi ,1,q,g_l^\varphi .q)$$
where $`h(g,l,q,q^{})`$ was defined in Theorem 5.5 above. Then we have
$$\begin{array}{cc}\hfill \mathrm{\Theta }(g_l^\varphi .q,I_4)& :=\theta (g_l^\varphi .q)=h(g_l^\varphi ,q).\theta (q)=h(g_l^\varphi ,q).(h^1.\theta (q^{}))\hfill \\ & =\left(h_l^{}(g_l^\varphi ,q)h_s(g_l^\varphi ,q)h_{s}^{}{}_{}{}^{1}(h_l^{})^1\right).\theta (q^{})\hfill \\ & =\left(h_l^{}(g_l^\varphi ,q)\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& B(g_l^\varphi )\end{array}\right)\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& B^1\end{array}\right)(h_l^{})^1\right).\theta (q^{})\hfill \\ & \\ & \text{(by Theorem }\text{5.5}\text{; Lemmas }\text{5.3}\text{ \& }\text{5.6}\text{)}\hfill \\ & \\ & =\left(h_l^{}(g_l^\varphi ,q)\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l\times I_3\end{array}\right)(h_l^{})^1\right).\theta (q^{})\text{(by Lemma }\text{5.8}\text{)}\hfill \\ & =\left(h_l^{}(g_l^\varphi ,q)h_s(\text{Identity},l^1,q^{},l^1q^{})(h_l^{})^1\right).\theta (q^{})\text{(by Lemma }\text{5.7}\text{)}\hfill \\ & =(h_l^{}(g_l^\varphi ,q)h_l^{\prime \prime }).(h_s(\text{Identity},l^1,q^{},l^1q^{}).\theta (q^{}))\hfill \\ & \\ & \text{(since }\text{Stab}_w\text{ is a semidirect product)}\hfill \\ & \\ & =(h_l^{}(g_l^\varphi ,q)h_l^{\prime \prime }).\theta (l^1q^{}))\text{(by Lemma }\text{5.7}\text{)}\hfill \\ & =:\mathrm{\Theta }(l^1q^{},(h_l^{}(g_l^\varphi ,q)h_l^{\prime \prime })).\hfill \end{array}$$
But since $`\mathrm{\Theta }`$ is an isomorphism (Theorem 3.10), this implies the claim of the above lemma namely, that
$$g_l^\varphi .q=l^1q^{}\text{ and further that }h_l^{}(g_l^\varphi ,q)=(h_l^{\prime \prime })^1.$$
###### Lemma 5.10
The similarity
$$g_l^\varphi :(V,q)_l(V,q^{})$$
induces $`\varphi `$ i.e., with the notations of (1), Prop.2.8, $`C_0(g_l^\varphi ,l)=\varphi .`$
We have $`C_0(g_l^\varphi ,l)=\varphi `$ iff
$$\begin{array}{cc}\hfill h(g_l^\varphi ,l,q,q^{})& :=\psi _{b(q^{})}C_0(g_l^\varphi ,l)\psi _{b(q)}^1\hfill \\ & =\psi _{b(q^{})}\varphi \psi _{b(q)}^1=:\varphi _{\mathrm{\Lambda }^{ev}}=:h.\hfill \end{array}$$
Now using successively Theorem 5.5, Lemma 5.3, Lemma 5.8 and Lemma 5.6, we get the following sequence of equalities:
$$h_s(g_l^\varphi ,l,q,q^{})=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l^1\times \mathrm{\Lambda }^2(g_l^\varphi )\end{array}\right)=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l^1\times B(g_l^\varphi )\end{array}\right)=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& B\end{array}\right)=h_s.$$
Therefore the present hypotheses translated in terms of actions give
$$\begin{array}{cc}\hfill h(g_l^\varphi ,l,q,q^{}).\theta (q)=\theta (q^{})& =h.\theta (q)\hfill \\ \hfill h_s(g_l^\varphi ,l,q,q^{}).(h_l(g_l^\varphi ,l,q,q^{}).\theta (q))& =h_s.(h_l.\theta (q))\hfill \\ \hfill \mathrm{\Theta }(q,h_l(g_l^\varphi ,l,q,q^{}))& =\mathrm{\Theta }(q,h_l).\hfill \end{array}$$
But $`\mathrm{\Theta }`$ being an isomorphism (Theorem 3.10), the last equality implies that
$$h_l(g_l^\varphi ,l,q,q^{})=h_l\text{ which gives }h(g_l^\varphi ,l,q,q^{})=h.$$
###### Lemma 5.11
For a similarity $`g\text{Sim}[(V,q),(V,q^{})]`$ with multiplier $`l`$ and the induced isomorphism
$$C_0(g,l)\text{Iso}[C_0(V,q),C_0(V,q^{})]$$
given by (1), Prop.2.8, we have the equality
$$\text{det}((C_0(g,l))_{\mathrm{\Lambda }^2})=l^3\text{det}^2(g).$$
Therefore the map
$$\text{Sim}[(V,q),(V,q^{})]\text{Iso}[C_0(V,q),C_0(V,q^{})]:gC_0(g,l)$$
maps the subsets
$$\text{Iso}[(V,q),(V,q^{})]\text{ and }\text{S-Iso}[(V,q),(V,q^{})]$$
respectively into the subsets
$$\text{Iso}^{}[C_0(V,q),C_0(V,q^{})]\text{ and }\text{S-Iso}[C_0(V,q),C_0(V,q^{})].$$
In the case $`q^{}=q`$, if $`C_0(g,l)`$ is the identity on $`C_0(V,q)`$, then
$$g=l^1\text{det}(g)\times \text{Id}_V,$$
and further if $`g\text{O}(V,q)`$ then
$$g=\text{det}(g)\times \text{Id}_V\text{ with }\text{det}^2(g)=1.$$
By definition,
$$(C_0(g,l))_{\mathrm{\Lambda }^{ev}}=\psi _{b(q^{})}C_0(g,l)\psi _{b(q)}^1,$$
and the latter isomorphism is $`h(g,l,q,q^{})`$ from Theorem 5.5 which further gives a formula for $`h_s(g,l,q,q^{}).`$ Now using the facts that $`\psi _{b(q)}`$ and $`\psi _{b(q^{})}`$ have trivial determinant (as noted before Lemma 5.8) we get assertion (1):
$$\begin{array}{cc}\hfill \text{det}\left((C_0(g,l))_{\mathrm{\Lambda }^2}\right)& =\text{det}\left((C_0(g,l))_{\mathrm{\Lambda }^{ev}}\right)=\text{det}(h(g,l,q,q^{}))\hfill \\ & =\text{det}(h_s(g,l,q,q^{}))=l^3\text{det}^2(g).\hfill \end{array}$$
If $`q=q^{}`$ and $`C_0(g,l)`$ is the identity, then the same argument in fact shows that
$$l^1\mathrm{\Lambda }^2(g)=I_3$$
and by using the formula in Lemma 5.3 for $`B(g)=\mathrm{\Lambda }^2(g)`$, we get
$$g=l^1\text{det}(g)I_3;$$
taking determinants in the last equality gives $`\text{det}^2(g)=l^3`$, so that when $`g\text{O}(V,q)`$ i.e., $`l=1`$,
$$\text{det}^2(g)\mu _2(\mathrm{\Gamma }(X,๐ช_X))$$
and assertion (2) follows.
###### Lemma 5.12
The following map is a bijection:
$$\text{S-Iso}[(V,q),(V,q^{})]\text{S-Iso}[C_0(V,q),C_0(V,q^{})]:gC_0(g,1,q,q^{}).$$
Given
$$\varphi \text{S-Iso}[C_0(V,q),C_0(V,q^{})],$$
by definition 3.3 we have $`\text{det}(\varphi _{\mathrm{\Lambda }^2})=1`$, so by Lemma 5.8
$$\text{det}(g_l^\varphi )=\sqrt{\gamma (l)}:=1$$
for our choice under Case 1 on page Case 1.. Therefore the corresponding element
$$g_l^\varphi \text{S-Iso}[(V,q),(V,q^{})]$$
and is, according to Lemma 5.10, such that
$$C_0(g_l^\varphi ,l=1,q,q^{})=\varphi $$
which gives the surjectivity. As for the injectivity, if
$$g_1,g_2:(V,q)_1(V,q^{})$$
are isometries with determinant 1 such that
$$C_0(g_1,1,q,q^{})=C_0(g_2,1,q,q^{}),$$
then we have
$$h(g_1,1,q,q^{})=h(g_2,1,q,q^{})\text{ so that }h_s(g_1,1,q,q^{})=h_s(g_2,1,q,q^{})$$
whence by Theorem 5.5 and Lemma 5.3
$$B(g_1)=B(g_2)g_1=g_2.$$
Q.E.D., Theorem 3.4 and injectivity of Theorem 3.1.
#### Proof of Theorem 3.5.
Taking $`q^{}=q`$ in Theorem 3.4 gives the commutative diagram of groups and homomorphisms as asserted in the statement of the theorem. We continue with the notations above. For $`g\text{GO}(V,q,I)`$ with multiplier $`l`$, that the equality
$$\text{det}(C_0(g,l,I))=\text{det}\left((C_0(g,l,I))_{\mathrm{\Lambda }^2}\right)=l^3\text{det}^2(g)$$
holds (locally, hence globally) was shown in (1), Lemma 5.11. Assertion (2) of the same Lemma shows the following (locally, and hence globally): if $`C_0(g,l,I)`$ is the identity on $`C_0(V,q,I)`$, then
$$g=l^1\text{det}(g).\text{Id}_V,$$
and further if $`g\text{O}(V,q,I)`$ then
$$g=\text{det}(g).\text{Id}_V\text{ with }\text{det}^2(g)=1.$$
The map
$$\mathrm{\Gamma }(X,๐ช_X^{})\text{GO}(V,q,I)$$
is the natural one given by sending $`\lambda `$ to the similarity $`\lambda .\text{Id}_V`$ with multiplier $`\lambda ^2.`$ It follows from the formula in (1), Prop.2.8 that
$$C_0(\lambda .\text{Id}_V,\lambda ^2,I)=\text{Identity}.$$
This gives exactness at $`\text{GO}(V,q,I)`$ and at $`\text{O}(V,q,I).`$ We proceed to prove assertion (b). Let
$$\varphi \text{Aut}(C_0(V,q,I)),$$
and consider the self-similarity
$$g_l^\varphi =s_{2k+1}^+(\varphi )\text{ with multiplier }l=\text{det}(\varphi )^{2k+1}.$$
For the moment, assume that $`V`$ and $`I`$ are trivial over $`X.`$ Fix a global basis $`\{e_1,e_2,e_3\}`$ for $`V`$ and set $`e_i^{}=g_l^\varphi (e_i).`$ It follows from Kneserโs definition of the half-discriminant $`d_0`$โsee formula (3.1.4), Chap.IV, โthat
$$d_0(q,\{e_i\})=d_0(q,\{e_i^{}\})\text{det}^2(g_l^\varphi ).$$
Since we have
$$g_l^\varphi .q=l^1q,$$
a simple computation shows that
$$d_0(q,\{e_i^{}\})=l^3d_0(q,\{e_i\}).$$
The hypothesis that $`q\kappa (x)`$ is semiregular means that the image of the element
$$d_0(q,\{e_i\})\mathrm{\Gamma }(X,๐ช_X)$$
in $`\kappa (x)`$ is nonzero. Since $`X`$ is integral, we therefore deduce that
$$\text{det}^2(g_l^\varphi )=l^3.$$
On the other hand, we know that
$$\text{det}^2(g_l^\varphi )l^3=\text{det}(\varphi ).$$
It follows that
$$\text{det}^{12k+7}(\varphi )=1k,$$
which forces $`\text{det}(\varphi )=1.`$ In general, even if $`V`$ and $`I`$ are not necessarily trivial, since this equality holds over a covering of $`X`$ which trivialises both $`V`$ and $`I`$, it also holds over all of $`X.`$ Q.E.D., Theorem 3.5.
## 6 Surjectivity of Theorem 3.1: Bilinear Forms as Specialisations
In this section we reduce the proof of Theorem 3.8 to Theorem 3.9. We prove the latter and using it alongwith Theorem 5.1, deduce Theorem 3.11. The surjectivity of Theorem 3.1 is also established.
#### Reduction of Proof of Theorem 3.8 to the case $`I=๐ช_X.`$
We adopt the notations introduced just before Theorem 3.8. Let $`T`$ be an $`X`$-scheme. Given a bilinear form
$$b\text{Bil}_{(V,I)}(T)=\mathrm{\Gamma }(T,\text{Bil}_{(V_T,I_T)}),$$
consider the linear isomorphism
$$\psi _b:C_0(V_T,q_b,I_T)๐ช_T\mathrm{\Lambda }^2(V_T)(I_T)^1=W_T$$
of (2d), Theorem 2.2. Let $`A_b`$ denote the algebra bundle structure on $`W_T`$ with unit $`w_T=1`$ induced via $`\psi _b`$ from the even Clifford algebra $`C_0(V_T,q_b,I_T)`$. By definition, $`A_b\text{Assoc}_{W,w}(T)`$ and we get a map of $`T`$-valued points
$$\mathrm{{\rm Y}}^{}(T):\text{Bil}_{(V,I)}(T)\text{Assoc}_{W,w}(T):bA_b.$$
This is functorial in $`T`$ because of (3), Theorem 2.2, and hence defines an $`X`$-morphism
$$\mathrm{{\rm Y}}^{}:\text{Bil}_{(V,I)}\text{Assoc}_{W,w}.$$
The morphism $`\mathrm{{\rm Y}}^{}`$ is $`\text{GL}_V`$-equivariant due to (2j), Theorem 2.2. Notice that the schemes $`\text{Bil}_{(V,I)},\text{Bil}_{(V,I)}^{sr}`$ and $`\text{Assoc}_{W,w}`$ are well-behaved relative to $`X`$ with respect to base-change. In fact, so are $`\text{Azu}_{W,w}`$ and $`\text{SpAzu}_{W,w}`$, as may be recalled from Theorems 2.14 and 2.15 of page 2.14. In view of these observations, by taking a trivialisation for $`I`$ over $`X`$, we may reduce to the case when $`I`$ is trivial. We treat this case next.
#### Reduction of Proof of Theorem 3.8 for $`I=๐ช_X`$ to Theorem 3.9.
We first recall the following crucial fact (see (1), Prop.3.2.4, Chap.IV ): The even Clifford algebra of a semiregular quadratic form is an Azumaya algebra. Using this fact and the definition of $`\mathrm{{\rm Y}}^{}`$, we see that the morphism $`\mathrm{{\rm Y}}^{}`$ restricted to $`\text{Bil}_V^{sr}`$ factors through $`\text{Azu}_{W,w}`$ by a morphism $`\mathrm{{\rm Y}}^{sr}`$ such that the following diagram is commutative
$$\begin{array}{ccc}\text{Bil}_V& \stackrel{\mathrm{{\rm Y}}^{}}{}& \text{Assoc}_{W,w}\\ & & & & \\ \text{Bil}_V^{sr}& \underset{\mathrm{{\rm Y}}^{sr}}{}& \text{Azu}_{W,w}\end{array}$$
where the vertical arrows are the canonical open immersions. The above diagram base changes well in view of (2), Theorem 2.14, Prop.2.1 and (3), Theorem 2.2. Notice that since the structure morphism $`\text{Bil}_VX`$ is an affine morphism, and since the same is true of
$$\text{Assoc}_{W,w}X,$$
it is also true of $`\mathrm{{\rm Y}}^{}.`$ In particular, $`\mathrm{{\rm Y}}^{}`$ is quasi-compact and separated, and therefore has a schematic image by case (1) of Prop.2.12. The same is true of each of the two vertical arrows and of $`\mathrm{{\rm Y}}^{sr}`$ in view of Prop.2.1 and (1) of Theorem 2.14. Further, as noted in Prop.2.1,
$$\text{Bil}_V^{sr}\text{Bil}_V$$
is schematically dominant and therefore by (5), Prop.2.13, the limiting scheme of the former in the latter is the latter itself. So using the commutative diagram above, the transitivity of the schematic image (assertion (3), Prop.2.13), and the definition of $`\text{SpAzu}_{W,w}`$ (assertion (1), Theorem 2.15), we see that in order to prove (1), Theorem 3.8, it is enough to show that
$`\mathrm{{\rm Y}}^{sr}`$ is schematically dominant and surjective, and $`\mathrm{{\rm Y}}^{}`$ is a closed immersion.
We now claim that the above properties are equivalent to
$`\mathrm{{\rm Y}}^{sr}`$ is proper and $`\mathrm{{\rm Y}}^{}`$ is a closed immersion.
Suppose (\**) holds. To show (\*), we only need to show that $`\mathrm{{\rm Y}}^{sr}`$ is surjective and schematically dominant. From (\**) it follows that
$$\mathrm{{\rm Y}}_K^{sr}:=\mathrm{{\rm Y}}^{sr}_XK$$
is functorially injective and proper for each algebraically closed field $`K`$ with an $`X`$-morphism
$$\text{Spec}(K)X.$$
That both the $`K`$-schemes
$$\text{Bil}_V^{sr}_XK\text{ and }\text{Azu}_{W,w}_XK$$
are integral and smooth of the same dimension follows from the smoothness of relative dimension 9 and geometric irreducibility $`/X`$ of $`\text{Bil}_V^{sr}`$ (which is obvious), and of $`\text{Azu}_{W,w}`$ from (3), Theorem 2.14. Since $`\mathrm{{\rm Y}}_K^{sr}`$ is differentially injective at each closed point, it has to be a smooth morphism by Theorem 17.11.1 of EGA IV and thus has to be an open map.
But by (\**) it is also proper and hence a closed map. Thus $`\mathrm{{\rm Y}}_K^{sr}`$ is bijective etale, and hence an isomorphism. This also gives that $`\mathrm{{\rm Y}}^{sr}`$ is surjective. Now from Cor.11.3.11 of EGA IV and from the flatness of $`\text{Bil}_V^{sr}`$ over $`X`$, it follows that $`\mathrm{{\rm Y}}^{sr}`$ is itself flat, and hence schematically dominant since it is faithfully flat (being already surjective). Therefore (\**)$``$(\*).
####
The property of a morphism being proper is local on the target (see for e.g., (f), Cor.4.8, Chap.IV, ) and the same is true of the property of being a closed immersion. Therefore, in verifying (\**), we may assume that $`V`$ is free over $`X`$ so that
$$W=\mathrm{\Lambda }^{even}(V)$$
is also free over $`X`$ and and $`w`$ is part of a global basis. We are now in the situation of Theorem 3.9. Granting it, we see immediately from the multiplication table that (\**) holds. For the table shows that the composition of the following $`X`$-morphisms
$$\text{Bil}_V\stackrel{\mathrm{{\rm Y}}^{}}{}\text{Assoc}_{W,w}\stackrel{\text{CLOSED}}{}\text{Alg}_W$$
is a closed immersion, which implies that $`\mathrm{{\rm Y}}^{}`$ is also a closed immersion. Further, the multiplication table also shows that both $`\mathrm{{\rm Y}}^{}`$ and $`\mathrm{{\rm Y}}^{sr}`$ satisfy the valuative criterion for properness, and are therefore proper. Thus the conditions (\**) are verified. So we have reduced the proof of (1), Theorem 3.8 to Theorem 3.9.
####
As for statement (2) of Theorem 3.8, firstly, the involution $`\mathrm{\Sigma }`$ of $`\text{Assoc}_{W,w}`$ defines a unique involution (also denoted by $`\mathrm{\Sigma }`$) on the scheme of specialisations $`\text{SpAzu}_{W,w}`$ (leaving the open subscheme $`\text{Azu}_{W,w}`$ invariant) because of the defining property of the schematic image involved; for we may verify that an automorphism of a scheme $`T`$ which leaves an open subscheme $`U`$ stable will also leave stable the limiting scheme of $`U`$ in $`T`$ (of course we assume here that the canonical open immersion $`UT`$ is a quasi-compact open immersion, which ensures the existence of the limiting scheme). Secondly, a glance at the multiplication table of Theorem 3.9 keeping in view the definition of opposite algebra shows that indeed the induced $`\mathrm{\Sigma }\text{Aut}_X\left(\text{Bil}_V\right)`$ takes the $`T`$-valued point $`B=(b_{ij})`$ to $`\text{transpose}(B)=(b_{ji}).`$ Finally, assertion (3) of Theorem 3.8 is a consequence of (1) taking into account (3), Theorem 2.15.
#### Proof of Theorem 3.9.
Given $`B=(b_{ij})\text{Bil}_V(T)`$, by our definition above, $`(\mathrm{{\rm Y}}^{}(T))(B)=A_B`$ is the algebra structure induced from the linear isomorphism
$$\psi _B:C_0(V_T,q_B)\mathrm{\Lambda }^{even}(V_T)$$
of (2d), Theorem 2.2. The stated multiplication table for $`A=A_B`$ is a consequence of straightforward calculation, keeping in mind (2f), Theorem 2.2 and the standard properties of the multiplication in the even Clifford algebra $`C_0(V_T,q_B).`$ Q.E.D., Theorems 3.9 & 3.8.
#### Proof of Theorem 3.11.
The proof follows by comparing the multiplication table relative to $`\mathrm{\Theta }`$ as computed in Theorem 5.1 with the multiplication table relative to $`\mathrm{{\rm Y}}`$ of Theorem 3.9 computed above. Q.E.D., Theorem 3.11.
#### Proofs of assertions in (a), Theorem 3.6 and Surjectivity part of Theorem 3.1.
Let $`W`$ be the rank 4 vector bundle underlying the specialised algebra $`A`$ and $`w\mathrm{\Gamma }(X,W)`$ be the global section corresponding to $`1_A.`$ We choose an affine open covering $`\{U_i\}_i`$ of $`X`$ such that $`W|U_i`$ is trivial and $`w|U_i`$ is part of a global basis $`i.`$ Therefore on the one hand, for each $`i`$, we can find a linear isomorphism
$$\zeta _i:\mathrm{\Lambda }^{even}\left(๐ช_X^3|U_i\right)W|U_i$$
taking $`1_{\mathrm{\Lambda }^{even}}`$ onto $`w|U_i.`$ The $`(w|U_i)`$-unital algebra structure $`A|U_i`$ induces via $`\zeta _i`$ an algebra structure $`A_i`$ on $`\mathrm{\Lambda }^{even}\left(๐ช_X^3|U_i\right)`$ (so that $`\zeta _i`$ becomes an algebra isomorphism). Recall that $`A_i`$ is also a specialised algebra structure by (3), Theorem 2.15. Hence by Theorem 3.8 applied to $`X=U_i`$, $`V=๐ช_{U_i}^3`$ and $`I=๐ช_{U_i}`$, we can also find an $`๐ช_{U_i}`$-valued quadratic form $`q_i`$ on $`๐ช_X^3|U_i`$ induced from a bilinear form $`b_i`$ so that the algebra structure $`A_i`$ is precisely the one induced by the linear isomorphism
$$\psi _{b_i}:C_0\left(๐ช_X^3|U_i,q_i\right)\mathrm{\Lambda }^{even}\left(๐ช_X^3|U_i\right)$$
given by (2d) of Theorem 2.2. For each pair of indices $`(i,j)\times `$, let $`\zeta _{ij}`$ and $`\varphi _{ij}`$ be defined so that the following diagram commutes:
$$\begin{array}{ccccc}C_0(๐ช_X^3|U_{ij},q_i|U_{ij})& \underset{}{\overset{\psi _{b_i}|U_{ij}}{}}& \mathrm{\Lambda }^{ev}(๐ช_X^3|U_{ij})& \underset{}{\overset{\zeta _i|U_{ij}}{}}& A|U_{ij}\\ \varphi _{ij}& & \zeta _{ij}& & =& & \\ C_0(๐ช_X^3|U_{ij},q_j|U_{ij})& \underset{\psi _{b_j}|U_{ij}}{\overset{}{}}& \mathrm{\Lambda }^{ev}(๐ช_X^3|U_{ij})& \underset{\zeta _j|U_{ij}}{\overset{}{}}& A|U_{ij}\end{array}$$
The above diagram means that the algebras $`A_i`$ glue along $`U_{ij}:=U_iU_j`$ via $`\zeta _{ij}`$ to give (an algebra bundle isomorphic to) $`A`$, and in the same vein, the even Clifford algebras $`C_0(๐ช_X^3|U_i,q_i)`$ glue along the $`U_{ij}`$ via $`\varphi _{ij}`$ to give $`A`$ as well. Now consider the similarity
$$g_{l_{ij}}^{\varphi _{ij}}=s_1^+(\varphi _{ij}):(๐ช_X^3|U_{ij},q_i|U_{ij})_{l_{ij}}(๐ช_X^3|U_{ij},q_j|U_{ij})$$
with multiplier
$$l_{ij}:=\text{det}(\varphi _{ij})^1$$
given by (c), Theorem 3.4. Since $`s_1^+`$ is multiplicative, and since $`\varphi _{ij}`$ satisfy the cocycle condition, it follows that $`s_1^+(\varphi _{ij})`$ also satisfy the cocycle condition and therefore glue the $`๐ช_X^3|U_i`$ along the $`U_{ij}`$ to give a rank 3 vector bundle $`V`$ on $`X.`$ While the $`q_i`$ do not glue to give an $`๐ช_X`$-valued quadratic form on $`V`$, the facts that the multipliers $`\{l_{ij}\}`$ form a cocycle for
$$I:=\text{det}^1(A)$$
and that $`s_{(1)}^+`$ is a section together imply, taking into account the uniqueness in (1), Prop.2.8, that actually the $`q_i`$ glue to give an $`I`$-valued quadratic form $`q`$ on $`V`$ and that $`C_0(V,q,I)A.`$ We shall now revert to the notations of Section 5. By Theorem 5.5, we have
$$h_s(g_{l_{ij}}^{\varphi _{ij}},l_{ij},q_i|U_{ij},q_j|U_{ij})=\left(\begin{array}{cc}1& \mathrm{๐}\\ \mathrm{๐}& l_{ij}^1\mathrm{\Lambda }^2(g_{l_{ij}}^{\varphi _{ij}})\end{array}\right)$$
which means that
$$(\varphi _{ij})_{\mathrm{\Lambda }^2}=\text{det}(\varphi _{ij})\times \mathrm{\Lambda }^2(g_{l_{ij}}^{\varphi _{ij}}).$$
This immediately implies part (1) of assertion (a) of Theorem 3.6, from which parts (2)โ(4) can be deduced using the standard properties of the determinant and the perfect pairings between suitable exterior powers of a bundle.
#### Proofs of assertions in (b), Theorem 3.6.
We first prove (b1). Let $`A`$ be a given specialisation, and let $`AC_0(V,q,I)`$ as in part (a) of Theorem 3.6 with $`I=\text{det}^1(A).`$ By the injectivity part of Theorem 3.1, we have
$$C_0(V,q,I)AC_0(V^{},q^{},๐ช_X)$$
iff there exists a twisted discriminant bundle $`(L,h,J)`$ and an isomorphism
$$(V,q,I)(V^{},q^{},๐ช_X)(L,h,J).$$
The latter implies that $`IJL^2`$ and hence $`\text{det}(A)2.\text{Pic}(X).`$ On the other hand, if this last condition holds, we could take for $`L`$ a square root of $`J:=I^1`$, alongwith an isomorphism $`h:L^2J`$ and we would have by Prop.2.9 an algebra isomorphism
$$\gamma _{(L,h,J)}:C_0(VL,qh,๐ช_X)C_0\left((V,q,I)(L,h,J)\right)C_0(V,q,I)A.$$
For the proof of (b2), suppose that the line subbundle $`๐ช_X.1_AA`$ is a direct summand of $`A.`$ We may choose a splitting
$$A๐ช_X.1_A(A/๐ช_X.1_A).$$
Using assertion (1) of (a), Theorem 3.6, we see that there exists a rank 3 vector bundle $`V`$ on $`X`$ such that
$$\begin{array}{cc}\hfill A& ๐ช_X.1_A(A/๐ช_X.1_A)\hfill \\ & ๐ช_X.1_A(\mathrm{\Lambda }^2(V)I^1)๐ช_X.1\mathrm{\Lambda }^2(V)I^1=:W\hfill \end{array}$$
where $`I:=\text{det}^1(A)`$ and the last isomorphism is chosen so as to map $`๐ช_X.1_A`$ isomorphically onto $`๐ช_X.1.`$ Therefore if
$$(W,w):=(๐ช_X.1\mathrm{\Lambda }^2(V)I^1,1),$$
then by the above identification $`A`$ induces an element of $`\text{SpAzu}_{W,w}(X)`$, and since
$$\mathrm{{\rm Y}}:\text{Bil}_{(V,I)}\text{SpAzu}_{W,w}$$
is an $`X`$-isomorphism by (1), Theorem 3.8, it follows that there exists an $`I`$-valued global quadratic form $`q=q_b`$ induced from an $`I`$-valued global bilinear form $`b`$ on $`V`$ such that the algebra structure $`\mathrm{{\rm Y}}(b)A.`$ (We recall that $`\mathrm{{\rm Y}}(b)`$ is the algebra structure induced from the linear isomorphism
$$\psi _b:C_0(V,q=q_b,I)๐ช_X.1\mathrm{\Lambda }^2(V)I^1=W$$
of (2d), Theorem 2.2, which preserves 1 by (2a) of the same Theorem). The proof of (b3) follows from a combination of those of (b1) and (b2). Q.E.D., Theorem 3.6 and surjectivity of Theorem 3.1.
## 7 Specialised Algebras on Self-Dual Bundles
In this section, we investigate the specialised algebras when the underlying bundle is self-dual and prove Theorem 3.19. We first have the following general result, which yields part of the assertions in (a), Theorem 3.19 that concern only rank 4 vector bundles.
###### Proposition 7.1
Let $`X`$ be any scheme, $`n`$ an integer $`2`$ and $`W`$ a rank $`n^2`$ vector bundle over $`X`$ with nowhere-vanishing global section $`w.`$ Then $`X`$ is irreducible (resp. reduced) iff $`\text{Azu}_{W,w}`$ is irreducible (resp. reduced) iff $`\text{SpAzu}_{W,w}`$ is irreducible (resp. reduced).
#### Proof:
We first make certain observations when $`X`$ is any reduced scheme and $`W`$ is a rank $`n^2`$ vector bundle over $`X`$ with nowhere-vanishing global section $`w.`$ It is not hard to see that the structure morphism
$$\text{Azu}_{W,w}X$$
is in fact a morphism of finite presentation. Hence, in view of Prop.17.5.7, EGA IV , assertion (3) of Theorem 2.14 implies that $`\text{Azu}_{W,w}`$ is reduced. Since $`\text{SpAzu}_{W,w}`$ is the schematic image of $`\text{Azu}_{W,w}`$, it follows from (1), Prop.2.13 that $`\text{SpAzu}_{W,w}`$ is reduced as well. When $`X`$ is integral, we next verify that $`\text{Azu}_{W,w}`$ is also integral. Consider any affine open subscheme
$$U=\text{Spec}(R)X$$
such that $`W|U`$ is trivial and $`w|U`$ is part of a global basis. There are $`(w|U)`$-unital Azumaya algebra structures on $`W|U`$ which are isomorphic to the $`(n\times n)`$-matrix algebra over $`R.`$ Consider the orbit morphism
$$\text{Stab}_{w|U}\text{Azu}_{W|U,w|U}$$
corresponding to one such algebra structure. Assertions (2) and (3) of Theorem 2.14 show that the topological image of this morphism is dense. Further, $`\text{Stab}_{w|U}`$ is integral since it is $`๐ธ_U^{12}`$ and since $`U`$ is integral. Thus
$$\text{Azu}_{W|U,w|U}$$
is integral. Recall (e.g., Prop.2.1.6 & Cor.2.1.7, Chap.0.2, EGA I ) that a nonempty topological space, whose set of irreducible components is locally finite, is locally irreducible iff each irreducible component is open; further it is irreducible iff it is locally irreducible and connected. Now since $`\text{Azu}_{W,w}`$ can be covered by irreducible open subschemes which pairwise intersect (as $`X`$ is irreducible), it follows that $`\text{Azu}_{W,w}`$ is integral as well.
Putting the above facts together with (1), Prop.2.13 shows that $`\text{SpAzu}_{W,w}`$ is also integral if $`X`$ is integral. Using again the facts that $`\text{Azu}_{W,w}`$ behaves well under base change (Theorem 2.14) and that the schematic image of an irreducible scheme is irreducible (by (1), Prop.2.13), we may infer from the foregoing that $`X`$ is irreducible iff $`\text{Azu}_{W,w}`$ is irreducible iff $`\text{SpAzu}_{W,w}`$ is irreducible. Q.E.D., Prop.7.1.
#### Proof of assertions (a)โ(d) of Theorem 3.19.
The proofs follow essentially from the properties of
$$(\mathrm{})\text{SpAzu}_{W,w}X\text{ and }\text{Azu}_{W,w}X$$
as mentioned in Theorems 2.15 and 2.14. Part of the assertions in (a) are valid more generally and were proved in Prop.7.1. We indicate proofs for the not-so-obvious assertions, especially (c) and for the implication
$$X\text{ irreducible }D_X\text{ irreducible }$$
in statement (a). The converse implication would follow from the fact that $`D_XX`$ is topologically surjective. Since the structure morphisms $`(\mathrm{})`$ are smooth, they are faithfully flat (hence surjective). Therefore in view of (b), which is actually the result of good base-change properties of $`\text{SpAzu}_{W,w}`$ and $`\text{Azu}_{W,w}`$ relative to $`X`$, the irreducible components of $`D_X`$ and of $`\text{SpAzu}_{W,w}`$ are induced from those of $`X`$ by pullbackโprovided we check the particular case when $`X`$ is irreducible and reduced.
So let $`X`$ be integral and first assume that $`W`$ is trivial and $`w`$ is part of a global basis. Without loss of generality, we may take
$$(W,w)=(\mathrm{\Lambda }^{even}(V),1)$$
for $`V`$ a rank 3 trivial vector bundle over $`X.`$ After fixing a suitable basis for $`V`$, we may define the morphism $`\mathrm{\Theta }`$, which by Theorem 3.10 is an isomorphism that maps the closed subset
$$(\text{Quad}_V\times _X\text{L}_w)\backslash (\text{Quad}_V^{sr}\times _X\text{L}_w)$$
onto
$$D_0:=\text{SpAzu}_{W,w}\backslash \text{Azu}_{W,w}.$$
Therefore the irreducibility of $`D_0`$ is equivalent to that of the closed subset
$$\text{Quad}_V\backslash \text{Quad}_V^{sr}.$$
Recall from the discussion on semiregular forms (page 2.3, Section 2) that the open subscheme $`\text{Quad}_V^{sr}`$ corresponds to localisation by the polynomial $`P_3.`$ This polynomial is irreducible as an element of
$$R[\zeta _i,\zeta _{ij}]R[\text{Quad}_V]$$
when $`X=\text{Spec}(R)`$ is affine, though it is not clear if it is a prime element (unless we assume something more e.g., $`R`$ a UFD). The closed subset
$$\text{Quad}_V\backslash \text{Quad}_V^{sr}$$
may be given the canonical closed subscheme structure $`Z(P_3)`$ corresponding to the vanishing of $`P_3.`$ Let $`q^{(2)}`$ be the global quadratic form on $`V`$ given by
$$(x_1,x_2,x_3)x_1x_2.$$
It can be checked that $`q^{(2)}`$ is not semiregular, but that its restriction to the rank two (direct summand) vector subbundle generated by $`\{e_1,e_2\}`$ is regular. Therefore the $`X`$-valued point corresponding to $`q^{(2)}`$ lands topologically inside the closed subset underlying $`Z(P_3).`$ Consider the orbit morphism
$$O(q^{(2)}):\text{GL}_V\text{Quad}_V$$
corresponding to this $`X`$-valued point which also lands topologically inside $`Z(P_3).`$ It will follow from assertion (2), Theorem 3.20, that the topological image of $`O(q^{(2)})`$ is dense in $`Z(P_3).`$ On the other hand this topological image is irreducible, since $`\text{GL}_V๐ธ_X^9.`$ It follows that $`\text{Quad}_V\backslash \text{Quad}_V^{sr}`$, and hence $`D_0`$, is irreducible in the case when $`W`$ is trivial and $`w`$ is part of a global basis over $`X.`$ Since the reduced closed subscheme structure on $`Z(P_3)`$ is given by the radical of the ideal $`(P_3)`$, it follows that $`\text{rad}(P_3)`$ is the minimal prime divisor of $`(P_3)`$ and by Krullโs Hauptidealsatz this prime has height 1. Therefore, the codimension of $`D_0`$ is also 1 in the present case. Now consider the case of a general $`(W,w)`$, choose an affine open covering
$$\{U_i=\text{Spec}(R_i)\}_{iI}$$
of $`X`$ such that $`W_i:=W|U_i`$ is trivial and $`w_i:=w|U_i`$ is part of a global basis $`i.`$ The subset
$$D_i:=\text{SpAzu}_{W_i,w_i}\backslash \text{Azu}_{W_i,w_i}$$
is irreducible for each $`i`$ by the preceding discussion. On the other hand, by Theorems 2.14 and 2.15, the subsets $`D_i`$ form an open cover of
$$D:=\text{SpAzu}_{W,w}\backslash \text{Azu}_{W,w}.$$
Since $`X`$ is irreducible, we have $`D_iD_j\mathrm{}`$ when $`ij.`$ Thus $`D`$ is locally irreducible and connected, and hence irreducible (for e.g., by Cor.2.1.7, Chap.0, EGA I ). Since $`X`$ is integral and noetherian, assertion (2) of Theorem 2.15 implies that $`\text{SpAzu}_{W,w}`$ is integral and noetherian as well. Therefore the codimension of $`D`$ in $`\text{SpAzu}_{W,w}`$ is atleast 1. On the other hand (for e.g., by Prop.14.2.3, Chap.0, EGA IV, ) this codimension is bounded above by
$$1=\text{Codim}(D_i,\text{SpAzu}_{W_i,w_i})$$
for any $`i`$, since
$$\text{SpAzu}_{W_i,w_i}\text{SpAzu}_{W,w}$$
is an open subset whose intersection with $`D`$ is precisely $`D_i.`$
#### Proof of assertion (e) of Theorem 3.19.
Let $`X`$ be an integral separated quasi-compact scheme and let $`\pi :\text{SpAzu}_{W,w}X`$ be the structure morphism. We need the following result.
###### Lemma 7.2
The canonical map $`H^0(X,๐ช^{})H^0(\text{SpAzu}_{W,w},๐ช^{})`$ induced by $`\pi `$ is an isomorphism.
For the proof, we first consider the case when $`X=\text{Spec}(R)`$ is affine, $`W`$ is trivial and further $`w`$ is part of a global basis for $`W.`$ In this case, by Theorem 3.10, $`\text{SpAzu}_{W,w}`$ may be identified with
$$๐ธ_X^9=\text{Spec}(R[X_1,\mathrm{},X_9])$$
and the assertion of the Lemma is trivial since $`R[X_1,\mathrm{},X_9]^{}=R^{}`$, $`R`$ being an integral domain.
Since $`X`$ is quasi-compact, we can always find a finite affine open covering which trivializes $`W`$ and restricted to each member of which $`w`$ becomes part of a global basis. We now proceed by induction on the number $`n`$ of open sets in such an affine covering $`๐ฑ`$ of $`X.`$ The case $`n=1`$ was dealt with in the previous paragraph. Consider $`n=2`$ i.e., let $`๐ฑ=\{V_0,V_1\}`$ be a covering by two members. We have the following natural commutative diagram with exact rows and with the vertical arrows given by the natural maps induced by $`\pi `$, where $`๐ฐ:=\pi ^1๐ฑ`$ is the Zariski-open affine covering of $`\text{SpAzu}_{W,w}`$ induced from $`๐ฑ`$ via $`\pi .`$ We write $`H^0()`$ for $`H^0(,๐ช^{})`$ for typographical reasons.
$$\begin{array}{ccccccc}0& & H^0(\text{SpAzu}_{W,w})& & H^0(U_0)\times H^0(U_1)& & H^0(U_{01})\\ & & & & & & & & & & \\ 0& & H^0(X)& & H^0(V_0)\times H^0(V_1)& & H^0(V_{01})\end{array}$$
Here $`V_{01}:=V_0V_1`$ and $`U_{01}:=\pi ^1(V_{01}).`$ Note that both $`V_{01}`$ and $`U_{01}`$ are affine since $`X`$ is separated and since $`\pi `$ is an affine morphism. Also note that $`U_i`$ (resp. $`U_{ij}`$) may be identified with $`\text{SpAzu}_{W_i,w_i}`$ (resp. with $`\text{SpAzu}_{W_{ij},w_{ij}}`$) due to the good base-change property (3) of Theorem 2.15, where $`W_i:=W|V_i`$ etc. By the previous paragraph, the middle and right vertical arrows are isomorphisms, hence so is the first. Suppose therefore that Lemma 7.2 is true for any integral separated quasi-compact $`X`$ with an affine open covering of cardinality $`n`$ which trivializes $`W`$ and to each member of which $`w`$ restricts to part of a basis. Let
$$๐ฑ=\{V_i^{}|0in\}$$
be such a covering with $`(n+1)`$ members and let $`๐ฐ:=\pi ^1๐ฑ`$ be the covering of $`\text{SpAzu}_{W,w}`$ induced by $`\pi `$ from $`๐ฑ.`$ Let $`V_0`$ denote the union of the first $`n`$ open sets $`V_i^{}`$, $`U_0:=\pi ^1(V_0)`$, $`V_1:=V_n^{}`$ and $`U_1:=\pi ^1(V_1).`$ Then writing out a commutative diagram as done above for the case of a covering with 2 members, and using the induction hypotheses (for $`n`$ for $`V_0`$, for $`n=1`$ for $`V_1`$ and again for $`n`$ for $`V_{01}`$), we conclude the proof of Lemma 7.2.
We now proceed with the proof of assertion (e) of Theorem 3.19. So let $`X`$ be locally-factorial. We want to show that $`\text{Pic}(\pi )`$ is an isomorphism. Recall that we have canonical isomorphisms $`\text{Pic}(T)H^1(T,๐ช_T^{})`$ (Ex.4.5, Ch.III, ) and $`\text{Pic}(T)\text{CL}(T)`$ for $`T`$ locally-factorial (Cor.6.16, Ch.II, ). Here CL denotes the Weil divisor class group. We shall use these identifications in what follows without particular mention.
Consider first the case when $`X=\text{Spec}(R)`$ is affine, $`W`$ is trivial and further $`w`$ is part of a global basis for $`W.`$ In this case, by Theorem 3.10, $`\text{SpAzu}_{W,w}`$ may be identified with $`๐ธ_X^9`$ and so the assertion (e) is a consequence of Prop.6.6, Ch.II, .
Next, let $`๐ฑ=\{V_0,V_1\}`$ be a Zariski-open affine covering of $`X`$ which trivializes $`W`$ and makes $`w|V_i`$ part of a global basis for $`W|V_i`$ ($`i=0,1`$). We have the following commutative diagram with exact rows given by Mayer-Vietoris sequences (Ex.2.24, Ch.III, ) and with the vertical arrows given by the natural maps induced by $`\pi `$, where $`๐ฐ:=\pi ^1๐ฑ`$ is the Zariski-open affine covering of $`\text{SpAzu}_{W,w}`$ induced from $`๐ฑ`$ via $`\pi .`$ We again write $`H^i()`$ for $`H^i(,๐ช^{})`$ for typographical reasons.
$$\begin{array}{ccccccc}H^0(U_0)\times H^0(U_1)& & H^0(U_{01})& & H^1(\text{SpAzu}_{W,w})& \stackrel{(=)}{}& \\ & & & & & & & \\ H^0(V_0)\times H^0(V_1)& & H^0(V_{01})& & H^1(X)& \stackrel{(=)}{}& \end{array}$$
$$\begin{array}{ccccccc}& \stackrel{(=)}{}& H^1(\text{SpAzu}_{W,w})& & H^1(U_0)\times H^1(U_1)& & H^1(U_{01})\\ & & & & & & & & & \\ & \stackrel{(=)}{}& H^1(X)& & H^1(V_0)\times H^1(V_1)& & H^1(V_{01})\end{array}$$
Here $`V_{01}:=V_0V_1`$ and $`U_{01}:=\pi ^1(V_{01}).`$ By Lemma 7.2, the first and second vertical arrows are isomorhisms. By the previous paragraph, the fourth and fifth vertical arrows are also isomorphisms. Therefore the 5-Lemma implies that the central vertical arrow is an isomorphism too. We now conclude the proof of assertion (e) by induction on the number of members in an open covering of $`X`$ (just as in the proof of Lemma 7.2).
#### Proof of assertion (f) of Theorem 3.19.
Let us remind the reader that the pullbacks
$$W^{}:=W_X\text{SpAzu}_{W,w}\text{ and }W_X\text{Azu}_{W,w}$$
are naturally endowed with $`(w_X\text{SpAzu}_{W,w})`$-unital associative algebra structures $`A^{}`$ and $`A^{}|\text{Azu}_{W,w}`$ with which they become respectively the universal specialisation and universal Azumaya algebra relative to the pair $`(W,w)`$โfor details see the proof of Theorem 3.4, Part I, . By (c) and (d), the natural map
$$\text{Pic}(\text{SpAzu}_{W,w})\text{Pic}(\text{Azu}_{W,w})$$
is surjective and its kernel is generated by the image of $`.(D_X).`$ Given an isomorphism
$$\varphi :W^{}\stackrel{}{}W,$$
we consider its pullback $`\varphi ^{}`$ to $`\text{SpAzu}_{W,w}`$, and let
$$(X^{},A^{},\varphi ^{}):=(\text{SpAzu}_{W,w},W_X\text{SpAzu}_{W,w},\varphi _X\text{SpAzu}_{W,w}).$$
Note that $`D_X=X^{}\backslash U(X^{},A^{}).`$ Consider the composition of the following morphisms of vector bundles:
$$W^{}_X^{}W^{}A^{}_X^{}A_{}^{}{}_{}{}^{op}\stackrel{(ab)(xaxb)}{}\text{End}(A^{})W_{}^{}{}_{}{}^{}_X^{}W^{}\stackrel{\varphi ^{}\times Id}{}W^{}_X^{}W^{}.$$
The above composite gives an endomorphism of the vector bundle $`W^{}W^{}`$ which is an isomorphism precisely at the local rings of the points of $`U(X^{},A^{}).`$ Therefore, the induced element
$$s(\varphi ^{})\mathrm{\Gamma }(X^{},๐ช_X^{})\text{H}^0(X^{},\text{End}(\text{det}(W^{}W^{})))$$
goes into the maximal ideal of the local ring at each point of $`D(X^{},A^{})`$ and to a unit of the local ring at each point of $`U(X^{},A^{}).`$ It follows that $`n.(D_X)`$ is principal for some $`n1.`$ Now $`A^{}|\text{Azu}_{W,w}`$ is an Azumaya algebra bundle, and as seen in Theorem 2.10 (cf. also the discussion following Prop. 3.12) it is isomorphic to the even Clifford algebra of a canonically obtained rank 3 quadratic bundle on $`\text{Azu}_{W,w}`$ with values in the structure sheaf. Therefore it follows from (b1), Theorem 3.6 that
$$\text{det}(W_X\text{SpAzu}_{W,w})$$
maps to an element of $`2.\text{Pic}(\text{Azu}_{W,w}).`$ The last assertion in (f) is now again a consequence of (b1), Theorem 3.6.
#### Proof of assertion (g) of Theorem 3.19.
Suppose that $`s`$ is a section to $`\pi ^{}:\text{Azu}_{W,w}X`$ corresponding to an Azumaya algebra structure on $`W`$ with identity $`w.`$ By assertions (e) and (f) (notice that $`W`$ is self-dual!) of Theorem 3.19, the homomorphism
$$\text{Pic}(\pi ^{}):\text{Pic}(X)\text{Pic}(\text{Azu}_{W,w})$$
is surjective. But $`\pi ^{}s=\text{Identity}`$ implies that $`\text{Pic}(\pi ^{})`$ is injective as well. The rest of the assertions in (g) now follow from (f). Q.E.D., Theorem 3.19.
## 8 Stratification of the Variety of Specialisations
In this section, we prove Prop.3.13 and Theorem 3.20.
#### Proof of Prop.3.13.
Fix an $`S`$-basis $`\{e_1,e_2,e_3\}`$ for $`V`$, and with respect to this basis, let $`q^1`$ denote the quadratic form given by
$$(x_1e_1+x_2e_2+x_3e_3)x_1x_2+x_3^2.$$
It is easy to see that this quadratic form is semiregular. We show that any semiregular quadratic form $`q`$ can be moved to $`q^1`$ i.e., that $`g\text{GL}(V)\text{ such that }gq=q^1.`$ By Prop.3.17, Chap.IV, , there exists a basis $`\{e_1^{},e_2^{},e_3^{}\}`$ for $`V`$ such that $`q`$ restricted to the submodule generated by $`e_1^{}`$ and $`e_2^{}`$ is regular and further such that
$$q(e_3^{})S^{},b_q(e_1^{},e_2^{})=1\text{ and }b_q(e_1^{},e_3^{})=0=b_q(e_2^{},e_3^{}).$$
Let $`g^{}\text{GL}(V)`$ be the automorphism that maps $`e_i^{}`$ onto the $`e_i`$ for each $`i`$ and consider the quadratic form $`q^{}:=g^{}q.`$ Then by definition of the $`\text{GL}(V)`$-action on the set $`\text{Quad}(V)`$ of quadratic $`S`$-forms on $`V`$ we have
$$q^{}(e_3)S^{},b_q^{}(e_1,e_2)=1\text{ and }b_q^{}(e_1,e_3)=0=b_q^{}(e_2,e_3).$$
So if we assume that
$$q^{}(e_i)=\lambda _i(q(e_i^{})=\lambda _i),$$
then we would have
$$q^{}(x_1e_1+x_2e_2+x_3e_3)=\lambda _1x_1^2+\lambda _2x_2^2+\lambda _3x_3^2+x_1x_2x_iS.$$
Thus it is enough to show that $`q^{}`$ can be moved to $`q^1.`$ We look for an invertible matrix
$$g^{\prime \prime }=(u_{ij})\text{GL}(V)\text{ such that }g^{\prime \prime }q^{}=q^1.$$
Writing this condition equivalently as $`q^{}=(g^{\prime \prime })^1q^1`$ and comparing the polynomials in the $`x_i`$ gives 6 equations in terms of the $`u_{ij}`$ and the $`\lambda _i`$ which are to be satisfied. We choose the $`u_{ij}`$ as follows. First set
$$u_{11}=u_{22}=0\text{ and let }u_{12}S^{}$$
be a free parameter. Since every element of $`S`$ has square roots in $`S`$, it makes sense to choose
$$u_{31}=\pm \sqrt{\lambda _1}\text{ and }u_{32}=\pm \sqrt{\lambda _2}.$$
We let
$$\alpha =1+2u_{31}u_{32},\beta =12u_{31}u_{32}\text{ and }u_{21}=\beta /u_{12}.$$
Since $`q^{}=gq`$ and since $`q`$ is semiregular, $`q^{}`$ is also semiregular. Its half-discriminant relative to the present basis of $`V`$ is (remembering that $`\lambda _3S^{}`$)
$$d_q^{}(e_1,e_2,e_3)=\lambda _3.(4\lambda _1\lambda _21)S^{}.$$
This implies that
$$\alpha \beta =14\lambda _1\lambda _2S^{}\alpha ,\beta S^{}.$$
Therefore it makes sense to define
$$u_{33}=\pm \sqrt{(\beta \lambda _3)/\alpha },u_{13}=2u_{31}u_{33}u_{12}/\beta \text{ and }u_{23}=2u_{32}u_{33}/u_{12}.$$
Note that $`u_{33}S^{}.`$ A computation shows that the determinant of the matrix $`g^{\prime \prime }=(u_{ij})`$ defined above is $`u_{33}\alpha S^{}`$ and hence $`g^{\prime \prime }`$ is invertible. It is also easily checked that $`g^{\prime \prime }q^{}=q^1.`$ Q.E.D., Prop.3.13.
#### Proof of assertion (2) of Theorem 3.20.
(We shall not prove assertion (1) since it is well-known). Recall from the discussion on semiregular forms (page 2.3, Section 5) that the open subscheme $`\text{Quad}_V^{sr}`$ corresponds to localisation by the polynomial $`P_3`$ and that this polynomial is prime as an element of
$$k[\zeta _i,\zeta _{ij}]k[\text{Quad}_V].$$
Here a quadratic form $`q`$ corresponding to the point $`(\lambda _i,\lambda _{ij})๐ธ_k^6`$ is given by
$$(x_1,x_2,x_3)\mathrm{\Sigma }_i\lambda _ix_i^2+\mathrm{\Sigma }_{i<j}\lambda _{ij}x_ix_j.$$
For ease of readability (and typesetting !) let us denote the closure $`\overline{T}`$ of a subset $`T`$ (given the reduced closed subscheme structure) by $`T`$ in what follows. Since $`\text{Quad}_V^{(1)}`$ is the same as the variety underlying the open subscheme $`\text{Quad}_V^{sr}`$ of semiregular quadratic forms, that
$$\text{Quad}_V^{(1)}=\text{Quad}_V$$
follows from the fact that $`\text{Quad}_V`$ is irreducible. By assertion (1) of the present Theorem, $`\text{Quad}_V`$ is the disjoint union of the $`\text{Quad}_V^{(i)}`$; therefore
$$\text{Quad}_V^{(1)}\backslash \text{Quad}_V^{(1)}$$
is the disjoint union of
$$\{\text{Quad}_V^{(i)}|2i4\}$$
and also equals the closed subset $`Z(P_3)`$ defined by the vanishing of $`P_3.`$ An explicit computation shows that the dimension of the stabilizer of $`q^{(2)}`$ in $`\text{GL}(V)`$ is 4. Since $`\text{Quad}_V^{(2)}`$ is an open dense subvariety of
$$\text{Quad}_V^{(2)}V(P_3),$$
its closure is thus 5-dimensional. But since $`P_3`$ is an irreducible polynomial, $`Z(P_3)`$ is also an irreducible 5-dimensional subvariety. It follows that
$$\text{Quad}_V^{(2)}=\text{Quad}_V^{(1)}\backslash \text{Quad}_V^{(1)}.$$
Since $`\text{Quad}_V^{(2)}`$ is smooth in its closure ($`=Z(P_3)`$ as seen above), the singularities of its closure are contained in
$$\text{Quad}_V^{(3)}\text{Quad}_V^{(4)}$$
which consists of quadratic forms that are perfect squares i.e., squares of linear forms. These singularities may be identified with points
$$(\lambda _i,\lambda _{ij})๐ธ_k^6\text{Quad}_V$$
at which all the partial derivatives of $`P_3`$ vanish. A simple computation shows that this set is the symmetric determinantal variety given by the vanishing of the $`(2\times 2)`$-minors of the matrix of the symmetric bilinear form associated to the generic quadratic form given by
$$(x_1,x_2,x_3)\mathrm{\Sigma }_i\zeta _ix_i^2+\mathrm{\Sigma }_{i<j}\zeta _{ij}x_ix_j;$$
but it can also be shown that this set precisely corresponds to the perfect squares. Therefore
$$\text{ Sing }(\text{Quad}_V^{(2)})=\text{Quad}_V^{(2)}\backslash \text{Quad}_V^{(2)}.$$
That
$$\text{Quad}_V^{(i+1)}=\text{Quad}_V^{(i)}\backslash \text{Quad}_V^{(i)}\text{ for }i=2$$
follows from the above and the obvious fact that any quadratic form can be specialised to the zero quadratic form. The case $`i=3`$ is trivial. To see that
$$\text{Quad}_V^{(3)}$$
is smooth if char($`k`$)=2, we first note from the above and assertion (1) of the present Theorem that the closure of $`\text{Quad}_V^{(3)}`$ consists of perfect squares. Since char($`k`$)=2, under the identification
$$\text{Quad}_V๐ธ_k^6,$$
the perfect squares are seen to correspond to the copy of $`๐ธ_k^3`$ in $`๐ธ_k^6`$ given by the vanishing of the last three coordinates $`\lambda _{ij},1i<j3.`$ To see that the zero quadratic form is a singularity of
$$\text{Quad}_V^{(3)}$$
if char($`k`$) $`2`$, we note from the above that $`\text{Quad}_V^{(3)}`$ is defined by the same equations that define the singularities of $`Z(P_3)`$ and is a certain symmetric determinantal variety. It is a well-known factโan application of Standard Monomial Theory (for e.g., see or )โthat the ideal defining these equations is itself reduced, i.e., it is the ideal of the variety. Checking the Jacobian criterion now shows that the zero quadratic form is indeed a singular point.
#### Proof of assertion (3) of Theorem 3.20.
The proof will follow from a series of lemmas. Recall that on page 5.2 we had defined the $`X`$-morphism
$$\theta :\text{Quad}_V\text{Id-}w\text{-Sp-Azu}_W$$
which was used to define the isomorphism $`\mathrm{\Theta }.`$
###### Lemma 8.1
There are exactly 4 $`H`$-orbits in SpAzu.
By Theorem 3.10, any point of SpAzu is of the form $`\underset{ยฏ}{t}\theta (q).`$ Its $`H`$-orbit is $`H\theta (q).`$ There exists a unique $`i`$, $`(1i4)`$, and some $`g\text{GL}(V)`$ such that $`q=gq^{(i)}.`$ Consider the algebra isomorphism $`C_0(g,1)`$ of (1), Prop.2.8 and the induced isomorphism $`h(g,1,q^{(i)},q)`$ of Theorem 5.5. By definition,
$$\begin{array}{cc}\hfill \theta (q)& =\theta (gq^{(i)})=h(g,1,q^{(i)},q)\theta (q^{(i)})\hfill \\ & H\theta (q)=H\theta (q^{(i)})\hfill \end{array}$$
so that there are atmost 4 orbits. To see that there are atleast 4, we use the following result.
###### Lemma 8.2
For $`q\text{Quad}_V`$, we have
$$\text{L}_w\mathrm{\Theta }((GL(V)q)\times \{I_4\})=\mathrm{\Theta }((GL(V)q)\times \text{L}_w)=H\theta (q).$$
On the one hand, we have
$$\begin{array}{cc}\hfill \mathrm{\Theta }((GL(V)q)\times \text{L}_w)& =\left\{\underset{ยฏ}{t}\theta (gq)\right|\underset{ยฏ}{t}\text{L}_w\text{ and }gGL(V)\}\hfill \\ & =\left\{\underset{ยฏ}{t}(h(g,1,q,gq)\theta (q))\right|\underset{ยฏ}{t}\text{L}_w\text{ and }gGL(V)\}\hfill \\ & H\theta (q).\hfill \end{array}$$
Conversely take any $`h\theta (q)H\theta (q).`$ By Theorem 3.10, there exists a unique $`\underset{ยฏ}{t^{}}\text{L}_w`$ and a unique $`q^{}\text{Quad}_V`$ such that
$$h\theta (q)=\underset{ยฏ}{t^{}}\theta (q^{}).$$
Therefore
$$((\underset{ยฏ}{t^{}})^1.h)\theta (q)=\theta (q^{}).$$
Since $`k`$ is algebraically- and hence quadratically closed, by (b), Theorem 3.4, there exists $`gGL(V)`$ such that $`q^{}=gq`$ and
$$(\underset{ยฏ}{t^{}})^1.h=h(g,1,q,q^{}).$$
Hence
$$h\theta (q)=\underset{ยฏ}{t^{}}\theta (gq)=\mathrm{\Theta }(gq,\underset{ยฏ}{t^{}})H\theta (q)\mathrm{\Theta }((GL(V)q)\times \text{L}_w).$$
This settles Lemma 8.2. As for Lemma 8.1, if $`q,q^{}\text{Quad}_V`$ are such that their $`GL(V)`$-orbits are distinct, then because $`\mathrm{\Theta }`$ is an isomorphism, Lemma 8.2 shows that $`H\theta (q)`$ is distinct from $`H\theta (q^{}).`$
###### Lemma 8.3
For each $`i`$, with $`1i4`$,
$$\text{Quad}_V^{(i)}\times \text{L}_w=\text{Quad}_V^{(i)}\times \text{L}_w.$$
If $`f:XY`$ is a smooth morphism and $`UY`$ is an open subset, then
$$f^1(U)=f^1(U).$$
Since $`\text{L}_w\text{Spec}(k)`$ is smooth, so is the induced morphism
$$\text{Quad}_V^{(i)}\times \text{L}_w\text{Quad}_V^{(i)}.$$
Taking $`f`$ to be this morphism and $`U=\text{Quad}_V^{(i)}`$ gives Lemma 8.3.
###### Lemma 8.4
The $`GL(V)`$-stratification of $`\text{Quad}_V`$ induces a $`GL(V)`$-stratification of
$$\text{Quad}_V\times \text{L}_w$$
(the $`GL(V)`$-action on $`\text{L}_w`$ taken to be trivial) with strata given by
$$(\text{Quad}_V\times \text{L}_w)^{(i)}:=\text{Quad}_V^{(i)}\times \text{L}_w,(1i4).$$
To prove Lemma 8.4, the only thing that needs to be checked is that
$$(\text{Quad}_V\times \text{L}_w)^{(i+1)}=(\text{Quad}_V\times \text{L}_w)^{(i)}\backslash (\text{Quad}_V\times \text{L}_w)^{(i)}.$$
This follows by applying Lemma 8.3 twice:
$$\begin{array}{cc}\hfill (\text{Quad}_V\times \text{L}_w)^{(i+1)}& =\text{Quad}_V^{(i+1)}\times \text{L}_w\hfill \\ & =\text{Quad}_V^{(i+1)}\times \text{L}_w\hfill \\ & =(\text{Quad}_V^{(i)}\backslash \text{Quad}_V^{(i)})\times \text{L}_w\hfill \\ & =\text{Quad}_V^{(i)}\times \text{L}_w\backslash \text{Quad}_V^{(i)}\times \text{L}_w\hfill \\ & =(\text{Quad}_V\times \text{L}_w)^{(i)}\backslash (\text{Quad}_V\times \text{L}_w)^{(i)}.\hfill \end{array}$$
Now according to Lemma 8.2, we have
$$\text{SpAzu}^{(i)}=\mathrm{\Theta }(GL(V)q^{(i)}\times \text{L}_w)=\mathrm{\Theta }((\text{Quad}_V\times \text{L}_w)^{(i)}).$$
This combined with Lemma 8.4 and the fact that $`\mathrm{\Theta }`$ is an isomorphism (Theorem 3.10) completes the proof of assertion (3) of Theorem 3.20. Q.E.D., Theorem 3.20.
#### Acknowledgements
The author is grateful for the Postdoctoral Fellowship of the Graduiertenkolleg Gruppen und Geometrie under support from the Deutschen Forschungsgemeinschaft and the State of Niedersachsen at the Mathematisches Institut Gรถttingen where this paper was written.
It is a pleasure to thank the National Board for Higher Mathematics, Department of Atomic Energy, Government of India, for its Postdoctoral Fellowship (August 2002โAugust 2003) at the Chennai Mathematical Institute, Chennai, India, during the latter half of which certain special cases of the results of this work were obtained.
|
warning/0506/hep-ph0506170.html
|
ar5iv
|
text
|
# SU-4252-810 Toy model for two chiral nonets
## I Introduction
The last few years have seen a renewal of interest kyotoconf \- hss3 in the low energy scalar sector of QCD. Many physicists now believe in the existence of the light, broad $`I=J=0`$ resonance, sigma in the 500-600 MeV region as well as a light broad $`I,J=1/2,0`$ resonance, kappa in the 700-900 MeV region. Together with the well established $`f_0(980)`$ and $`a_0(980)`$ scalar resonances, these comprise a putative nonet of โelementary particlesโ. Furthermore, this nonet seems likely to have a quark structure like $`qq\overline{q}\overline{q}`$ rather than the conventional $`q\overline{q}`$ Jaffe . This of course raises the question of where are the conventional $`q\overline{q}`$ p-wave scalars expected in the quark model. Arguments have been given mixing that the experimental data are better fit when the two scalar nonets mix with each other and the resulting โlevel repulsionโ, pushes the conventional scalars to higher masses than otherwise expected.
In order to further explore the feature of mixing between $`q\overline{q}`$ type and $`qq\overline{q}\overline{q}`$ type states it seems interesting to consider a linear SU(3)$`\times `$ SU(3) sigma model which contains also the pseudoscalar nonet partners of these two scalar nonets. Parenthetically, we remark that while the non- linear sigma model GL ; CW and its extension to the chiral perturbation theory program CPT are often more efficient for systematic calculations, linear sigma models have a very long history of furnishing important insights into the nature of strong hadron dynamics. The SU(2) linear sigma model was first given in ref. GL . It was used as a basis for understanding the current algebra treatment of $`\pi \pi `$ scattering near threshold in ref. w . The SU(3) version was given in the first of ref. l . A detailed application to the low energy pseudoscalar mass spectrum was given SU1 before QCD in which, among other things, it was shown how a U(1)<sub>A</sub> violating term natural in the SU(3) model could solve the $`\eta ^{}`$ problem. Such a term was later discovered to arise from instanton effects tH1 . The connection was pointed out in ref. MS and emphasized by โt Hooft tH2 .
The model containing two different chiral nonets to be discussed here was proposed in section V of ref. BFMNS01 and an initial treatment, neglecting flavor symmetry breaking, was given. A discussion, taking the flavor symmetry breaking into account has very recently been presented in ref. nr . Actually, it turns out that the model is very complicated since many different terms can be included and various assumptions about the nature of the symmetry breaking can be made. In this paper we will set up the formalism for treating consequences of the model which hold (at tree level) just due to the symmetry structure of the model and will give a numerical treatment using what might be the simplest choice of symmetry breaking terms.
Section II begins with a review of the flavor transformation properties of the two chiral nonet fields, $`M`$ and $`M^{}`$ which are used in the model. Each contains nine pseudoscalar and nine scalar fields. Under chiral SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> transformations both fields transform in an identical manner. Thus a chiral Lagrangian which respects only this symmetry cannot directly distinguish between a โtwo quarkโ (i.e. $`q\overline{q}`$) or a โfour quarkโ scalar, for example. However, it is noted again that the U(1)<sub>A</sub> transformation actually counts the number of quarks in these mesons and provides a way to distinguish them. In order to make use of this, the Lagrangian should of course be set up appropriately. We implement this by requiring that the Lagrangian mock up the anomalous U(1)<sub>A</sub> equation of the underlying QCD and that the analogs of the quark mass terms also mock up the U(1)<sub>A</sub> transformation properties of the quark mass terms in the underlying theory. A reasonable initial thought on which terms to include in the Lagrangian is to restrict it to be renormalizable. It is noted, with details in Appendix A, that the renormalizable $`MM^{}`$ Lagrangian has however very many more terms than does the renormalizable single $`M`$ Lagrangian. An alternate way, which still satisfies generality, is to consider any number of terms, renormalizable or not, and just use the information which follows from the symmetry behavior of the Lagrangian.
In order to exploit this symmetry information we derive, in section III, vector type and axial vector type โgenerating equationsโ for the model. These can be differentiated with respect to the fields to yield many tree level Ward identities which are independent of the number of symmetric terms included in the Lagrangian. In addition to the analog of โtwo quarkโ condensates which occur in the single $`M`$ model, the present model also brings โfour quarkโ condensates into the picture.
In section IV, we derive predictions for the mass spectrum which follow from this symmetry approach. The characteristic feature is mixing betweenโtwo quarkโ and โfour quarkโ mesons with the same quantum numbers. Assuming isospin invariance, predictions are made for the $`\pi \pi ^{}`$ mixing sector, the $`KK^{}`$ mixing sector, the strange scalar $`\kappa \kappa ^{}`$ mixing sector and the sector involving mixing of the four isocalar pseudoscalars ($`\eta `$ type particles). It is shown how to formulate the first three of these mixing sectors in a parallel and economical way.
In section V, the mass spectrum relations are compared with experiment. First the three $`2\times 2`$ mixing sectors are treated. The inputs are taken to be the six masses of the well known and not so well known particles, the pion and kaon decay constants and a model parameter denoted $`x_\pi `$, which is the squared mass of the unmixed (or โbareโ) pion. These are enough to determine all the relevant parameters of these three systems. The pseudoscalar mixing is very sensitively dependent on $`x_\pi `$; as it increases from the experimental value, $`m_\pi ^2`$ the four quark components of the pion and the kaon increase. On the other hand, the scalar $`\kappa `$ has a large four quark component. This feature thus provides some support for a more exotic structure of the low lying scalars. Another interesting feature of the present model, discussed in this section, is that it permits one to estimate the strength of a four quark vacuum condensate. Finally, section V contains a brief summary, the connection with other results on the same model and directions for future work.
## II Symmetries and Lagrangian
First, let us briefly review BFMNS01 the fields of the model and their transformation properties. The schematic structure for the matrix $`M(x)`$ realizing a $`q\overline{q}`$ composite in terms of quark fields $`q_{aA}(x)`$ can be written
$$M_a^b=\left(q_{bA}\right)^{}\gamma _4\frac{1+\gamma _5}{2}q_{aA},$$
(1)
where $`a`$ and $`A`$ are respectively flavor and color indices. Our convention for matrix notation is $`M_a^bM_{ab}`$. Then $`M`$ transforms under chiral SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> as
$$MU_LMU_R^{},$$
(2)
where $`U_L`$ and $`U_R`$ are unitary, unimodular matrices associated with the transformations on the left handed ($`q_L=\frac{1}{2}\left(1+\gamma _5\right)q`$) and right handed ($`q_R=\frac{1}{2}\left(1\gamma _5\right)q`$) quark projections. For the discrete transformations charge congugation $`C`$ and parity $`P`$ one verifies
$$C:MM^T,P:M(๐ฑ)M^{}(๐ฑ).$$
(3)
The U(1)<sub>A</sub> transformation acts as $`q_{aL}e^{i\nu }q_{aL}`$, $`q_{aR}e^{i\nu }q_{aR}`$ and results in:
$$Me^{2i\nu }M.$$
(4)
Next, consider the $`qq\overline{q}\overline{q}`$ type fields. One interesting model Isgur postulates that the light scalars are โmoleculesโ made out of two pseudoscalar mesons. The chiral realization of this picture would result in the following schematic structure:
$$M_a^{(2)b}=ฯต_{acd}ฯต^{bef}\left(M^{}\right)_e^c\left(M^{}\right)_f^d.$$
(5)
One can verify that $`M^{(2)}`$ transforms exactly in the same way as $`M`$ under SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub>, $`C`$ and $`P`$. Under U(1)<sub>A</sub> it transforms as
$$M^{(2)}e^{4i\nu }M^{(2)},$$
(6)
which differs from Eq. (4). Another interesting approach Jaffe to explaining the light scalar mesons was formulated by Jaffe in the framework of the MIT bag model. It was observed that the spin-spin (hyperfine) piece of the one gluon exchange interaction between quarks gives an exceptionally strong binding to an s-wave $`qq\overline{q}\overline{q}`$ scalar state. The scalar states of this type may be formally written as bound states of a โdual quarkโ and โdual antiquarkโ. There are two possibilities if the dual antiquark is required to belong to a $`\overline{3}`$ representation of flavor SU(3). In the first case it belongs to a $`\overline{3}`$ of color and is a spin singlet. This has the schematic chiral realization,
$`L^{gE}=ฯต^{gab}ฯต^{EAB}q_{aA}^TC^1{\displaystyle \frac{1+\gamma _5}{2}}q_{bB},`$
$`R^{gE}=ฯต^{gab}ฯต^{EAB}q_{aA}^TC^1{\displaystyle \frac{1\gamma _5}{2}}q_{bB},`$ (7)
where $`C`$ is the charge conjugation matrix of the Dirac theory. A suitable form for the $`M`$ matrix is:
$$M_g^{(3)f}=\left(L^{gA}\right)^{}R^{fA}.$$
(8)
$`M^{(3)}`$ can be seen to transform in the same way as $`M^{(2)}`$ under SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub>, $`C`$, $`P`$ and U(1)<sub>A</sub>. In the second case the dual antiquark belongs to a $`6`$ representation of color and has spin 1. It has the corresponding schematic chiral realization:
$`L_{\mu \nu ,AB}^g=L_{\mu \nu ,BA}^g=ฯต^{gab}q_{aA}^TC^1\sigma _{\mu \nu }{\displaystyle \frac{1+\gamma _5}{2}}q_{bB},`$
$`R_{\mu \nu ,AB}^g=R_{\mu \nu ,BA}^g=ฯต^{gab}q_{aA}^TC^1\sigma _{\mu \nu }{\displaystyle \frac{1\gamma _5}{2}}q_{bB},`$ (9)
where $`\sigma _{\mu \nu }=\frac{1}{2i}[\gamma _\mu ,\gamma _\nu ]`$. This choice leads to an $`M`$ matrix
$$M_g^{(4)f}=\left(L_{\mu \nu ,AB}^g\right)^{}R_{\mu \nu ,AB}^f,$$
(10)
where the dagger operation includes a factor $`(1)^{\delta _{\mu 4}+\delta _{\nu 4}}`$. $`M^{(4)}`$ also transforms like $`M^{(2)}`$ and $`M^{(3)}`$ under all of SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub>, $`C`$, $`P`$ and U(1)<sub>A</sub>. The specific form favored by the MIT bag model calculation actually corresponds to a particular linear combination of $`M^{(3)}`$ and $`M^{(4)}`$. Furthermore one can verify that $`M^{(2)}`$ in Eq. (5) is related by a Fierz transformation to a linear combination of $`M^{(3)}`$ and $`M^{(4)}`$. Thus only two of $`M^{(2)}`$, $`M^{(3)}`$ and $`M^{(4)}`$ are linearly independent. In any event, at the present effective Lagrangian level, there are no quantum numbers to distinguish $`M^{(2)}`$, $`M^{(3)}`$, and $`M^{(4)}`$ from each other so we may as well just denote an arbitrary linear combination of them to be our $`qq\overline{q}\overline{q}`$ field, $`M^{}`$. Note that $`M`$ and $`M^{}`$ are distinguished from each other by their different U(1)<sub>A</sub> transformation properties. These fields may be decomposed into hermitian scalar (S) and pseudoscalar ($`\varphi `$) nonets as,
$`M`$ $`=`$ $`S+i\varphi ,`$
$`M^{}`$ $`=`$ $`S^{}+i\varphi ^{}.`$ (11)
We will be interested in the situation where non-zero vacuum values of the diagonal components of $`S`$ and $`S^{}`$ may exist. These will be denoted by,
$$S_a^b=\alpha _a\delta _a^b,S_a^b=\beta _a\delta _a^b.$$
(12)
In the iso-spin invariant limit, $`\alpha _1=\alpha _2`$ and $`\beta _1=\beta _2`$ while in the SU(3) invariant limit, $`\alpha _1=\alpha _2=\alpha _3`$ and $`\beta _1=\beta _2=\beta _3`$.
The Lagrangian density which defines our model is
$$=\frac{1}{2}\mathrm{Tr}\left(_\mu M_\mu M^{}\right)\frac{1}{2}\mathrm{Tr}\left(_\mu M^{}_\mu M^{}\right)V_0(M,M^{})V_{SB},$$
(13)
where $`V_0(M,M^{})`$ stands for a general function made from SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> (but not necessarily U(1)<sub>A</sub>) invariants formed out of $`M`$ and $`M^{}`$. Furthermore $`V_{SB}`$ is taken to be a flavor symmetry breaking term which should mock up the quark mass terms which perform this function in the fundamental QCD Lagrangian. Other physical particles (including glueballs) could be added for more realism, but Eq. (13) is already quite complicated.
To get an initial indication of what is happening in this kind of model the drastically simplified case where the quark mass effective term, $`V_{SB}`$ is absent and where $`V_0`$ is simply given by:
$$V_0=c_2\mathrm{Tr}\left(MM^{}\right)+c_4\mathrm{Tr}\left(MM^{}MM^{}\right)+d_2\mathrm{Tr}\left(M^{}M^{}\right)+e\mathrm{Tr}\left(MM^{}+M^{}M^{}\right),$$
(14)
was treated in sec.V of ref. BFMNS01 . Here $`c_2`$, $`c_4`$ and $`d_2`$ are positive real constants. The $`M`$ matrix field is chosen to have a wrong sign mass term so that there will be spontaneous breakdown of chiral symmetry. A pseudoscalar octet is thus massless. The mixing between the $`M`$ and $`M^{}`$ is controlled by the parameter $`e`$. The first feature found for this simplified model was that the analog, $`S_{}^{}{}_{a}{}^{a}`$ of the $`qq\overline{q}\overline{q}`$ condensate in QCD acquired a small non-zero value due to the mixing between $`S`$ and $`S^{}`$. The main question is the level ordering. Since the light pseudoscalars (e.g. $`\pi ^+=\varphi _1^2`$) are naturally identified, before mixing, with the $`q\overline{q}`$ field M, one wonders whether the two quark rather than the four quark scalars arenโt the lightest ones. It was found however that it is natural (but not unique) in the model to have the energy level pattern in ascending order- pseudoscalar Nambu-Goldstone boson with primarily $`q\overline{q}`$ structure, scalar with primarily $`qq\overline{q}\overline{q}`$ structure, pseudoscalar with primarily $`qq\overline{q}\overline{q}`$ structure and scalar with primarily $`q\overline{q}`$ structure. These refer to degenerate octets which are each mixtures of $`M`$ and $`M^{}`$ states. This seems to be similar to the expected experimental pattern and gives us some motivation to proceed further.
The next question is what terms to include in the Lagrangian Eq. (13). A natural first attempt would be to consider a renormalizable model in which $`V_0`$ contains all the SU(3)$`\times `$ SU(3) invariant terms up to four powers of the fields. These are listed in Appendix A. It is seen that there are 21 terms of this type. This is a rather large number and while not impossible to handle suggests trying another tack. We will just allow $`V_0`$ to contain all possible terms which are SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> symmetric and use the information provided by this symmetry. This is more general and also allows for non-renormalizable terms. The price to be paid is that we only get information which follows just from the symmetry structure. In an earlier treatment SU1 of the single chiral nonet case, it was found that the results obtained were essentially those which could be obtained from the โcurrent algebraโ approach. Furthermore, we will try to make use of the fact that $`M`$ and $`M^{}`$ have different U(1)<sub>A</sub> transformation properties. We thus demand that the Lagrangian without $`V_{SB}`$ mock up the anomalous U(1)<sub>A</sub> equation of QCD,
$$\delta =G,$$
(15)
where $`\delta `$ denotes the axial U(1) variation and $`G`$ is proportional to the product of the QCD field strength tensor and its dual. This can be achieved by making all of the terms in $`V_0`$, except for a limited number, U(1)<sub>A</sub> invariant. The special terms will be constructed to satisfy Eq. (15). An example of a term which is not U(1)<sub>A</sub> invariant is the mixing term used in the simplified model above: $`\mathrm{Tr}(M^{}M^{})+\mathrm{h}.\mathrm{c}.`$. However a mixing term of the type:
$$ฯต_{abc}ฯต^{def}M_d^aM_e^bM_f^c+\mathrm{h}.\mathrm{c}.$$
(16)
is U(1)<sub>A</sub> invariant and hence possibly the most important one.
An SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> invariant but not U(1)<sub>A</sub> invariant term which mocks up Eq. (15) can be seen U1A to be
$$_{anom}=\frac{iG}{12}\mathrm{ln}(\frac{\mathrm{det}M}{\mathrm{det}M^{}}).$$
(17)
Here, $`G`$ is being formally considered as an effective pseudoscalar glueball field in the effective Lagrangian. To get an $`\eta ^{}(958)`$ mass term in the effective lagrangian framework one can U1A include a wrong sign mass term for G: $`cG^2/2`$ in the Lagrangian which of course does not change the flavor symmetry structure. Then integrating out $`G`$ yields the effective $`\eta (960)`$ mass term:
$$_\eta =c_3[\mathrm{ln}(\frac{\mathrm{det}M}{\mathrm{det}M^{}})]^2,$$
(18)
where $`c_3=1/(288c)`$. The nature of this term becomes more apparent when one goes to the non-linear realization where $`M\alpha _1\mathrm{exp}(i\varphi /\alpha _1)`$. For the present paper we shall consider this to be the only SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> invariant but not U(1)<sub>A</sub> invariant term. However, it is not at all unique when we consider a model with two chiral nonets. For example one can also include something like the non- U(1)<sub>A</sub> invariant mixing term $`\mathrm{Tr}(M^{}M^{})+\mathrm{h}.\mathrm{c}.`$ by writing a candidate Lagrangian piece:
$$\frac{iG}{12}[\gamma _1\mathrm{ln}(\frac{\mathrm{det}M}{\mathrm{det}M^{}})+\gamma _2\mathrm{ln}(\frac{\mathrm{Tr}(MM^{})}{\mathrm{Tr}(M^{}M^{})}],$$
(19)
and proceeding as above. In order to properly mock up the anomaly in this case it is necessary hsuss that the real numbers $`\gamma _1`$ and $`\gamma _2`$ satisfy
$$\gamma _1+\gamma _2=1.$$
(20)
The generalization to more than two such terms is evident. It may be noted that the $`MM^{}`$ mixing term resulting from Eq. (19) mixes only the pseudoscalar fields and not the scalar ones.
Finally, let us consider the flavor symmetry breaking terms. To get more restrictions, we assume that such a term should mock up both the SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> and U(1)<sub>A</sub> transformation properties of the quark mass terms in the fundamental QCD Lagrangian. It is convenient to introduce a diagonal matrix,
$$A=\mathrm{diag}(A_1,A_2,A_3),$$
(21)
which is proportional to the diagonal matrix made from the three light quark masses, $`diag(m_u,m_d,m_s)`$ (See MS for further details). Then, from Eq. (1), we note an obvious choice for a flavor symmetry breaking term,
$$V_{SB}=\mathrm{Tr}[A(M+M^{})]=2\mathrm{T}\mathrm{r}(AS),$$
(22)
which transforms like $`(3,3^{})+(3^{},3)`$ under SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub>. Under the U(1)<sub>A</sub> transformation of Eq. (4), it goes to $`e^{2i\nu }\mathrm{Tr}(AM)+\mathrm{h}.\mathrm{c}.`$. Note that the similar simple possibility, $`2\mathrm{T}\mathrm{r}(AS^{})`$ does not correctly mock up the U(1)<sub>A</sub> transformation property of the QCD mass term. However Eq. (22) is not at all unique in correctly mocking up the quark mass term. An interesting term which does mock up the quark mass term also involves mixing and has the form,
$$ฯต_{abc}ฯต^{def}A_d^aM_e^bM_{}^{}{}_{f}{}^{c}+\mathrm{h}.\mathrm{c}.$$
(23)
This term mixes both scalars and pseudoscalars but with opposite signs.
For what follows, it is convenient to record the behaviors of the fields under infinitesimal transformations. Let us write the infinitesimal vector (L+R) and axial vector (L-R) transformations of $`\varphi `$ and $`S`$ as,
$`\delta _V\varphi =[E_V,\varphi ],\delta _A\varphi =i[E_A,S]_+,`$
$`\delta _VS=[E_V,S],\delta _AS=i[E_A,\varphi ]_+.`$ (24)
Here, unitarity demands that the infinitesimal matrices obey,
$$E_V^{}=E_V,E_A^{}=E_A.$$
(25)
If we demand that the transformations be unimodular, so that the U(1)<sub>A</sub> transformation is not included (the U(1)<sub>V</sub> transformation is trivial for mesons), we should also impose $`\mathrm{Tr}(E_A)=0`$. However we will not do this so the effects of U(1)<sub>A</sub> will also be included. The transformation properties of the $`qq\overline{q}\overline{q}`$ type fields are:
$`\delta _V\varphi ^{}=[E_V,\varphi ^{}],\delta _A\varphi ^{}=i[E_A,S^{}]_++2iS^{}\mathrm{Tr}(E_A),`$
$`\delta _VS^{}=[E_V,S^{}],\delta _AS^{}=i[E_A,\varphi ^{}]_+2i\varphi ^{}\mathrm{Tr}(E_A).`$ (26)
The extra terms for the axial transformations reflect the different U(1)<sub>A</sub> transformation properties of $`M`$ and $`M^{}`$.
## III Generating equations
We shall consider, in this paper, tree level predictions for the Lagrangian of Eq.(13) in which the only U(1)<sub>A</sub> violating term in $`V_0`$ is that of Eq.(18). The only term in $`V_{SB}`$ will be taken to be the simplest one given in Eq. (22). In this minimal picture, there is no symmetry breaking associated with the $`qq\overline{q}\overline{q}`$ fields in $`M^{}`$. The symmetry breaking in the physical states (which contain two quark as well as four quark components) is due to the mixing terms which, as we have already seen in Eq. (16), can be invariant under SU(3)$`\times `$ SU(3)$`\times `$ U(1)<sub>A</sub>.
The method of treatment, as used earlier SU1 to discuss the model containing only the field $`M`$, is based on two generating equations which reflect the invariance of $`V_0`$ under vector and axial vector transformations. Differentiating them once, relates two point vertices (masses) with one point vertices. Differentiating them twice relates three point vertices (trilinear couplings) with masses and so on. These are essentially tree level Ward identities.
Under the infinitesimal vector and axial vector transformations we have,
$`\delta _VV_0`$ $`=`$ $`\{\mathrm{Tr}({\displaystyle \frac{V_0}{\varphi }}\delta _V\varphi +{\displaystyle \frac{V_0}{S}}\delta _VS)+(\varphi ,S)(\varphi ^{},S^{})\}=0,`$
$`\delta _AV_0`$ $`=`$ $`\{\mathrm{Tr}({\displaystyle \frac{V_0}{\varphi }}\delta _A\varphi +{\displaystyle \frac{V_0}{S}}\delta _AS)+(\varphi ,S)(\varphi ^{},S^{})\}=_\eta ,`$ (27)
wherein the non-zero value of the axial variation equation reflects the presence in $`V_0`$ of the single $`U(1)_A`$ non-invariant term of Eq. (18). Using Eqs. (24) and (26) as well as the arbitrariness of the variations $`E_V`$ and $`E_A`$ yields the matrix generating equations,
$`\{[\varphi ,{\displaystyle \frac{V_0}{\varphi }}]+[S,{\displaystyle \frac{V_0}{S}}]+(\varphi ,S)(\varphi ^{},S^{})\}=0,`$
$`\{[\varphi ,{\displaystyle \frac{V_0}{S}}]_+[S,{\displaystyle \frac{V_0}{\varphi }}]_++(\varphi ,S)(\varphi ^{},S^{})\}=1[2\mathrm{T}\mathrm{r}(\varphi ^{}{\displaystyle \frac{V_0}{S^{}}}S^{}{\displaystyle \frac{V_0}{\varphi ^{}}})8c_3i\mathrm{ln}({\displaystyle \frac{\mathrm{det}M}{\mathrm{det}M^{}}})],`$ (28)
where, in addition, the form of Eq. (18) was used. To get constraints on the particle masses we will differentiate these equations once with respect to each of the four matrix fields: $`\varphi ,\varphi ^{},S,S^{}`$ and evaluate the equations in the ground state. Thus we also need the โminimumโ condition,
$$\frac{V_0}{S}+\frac{V_{SB}}{S}=0,\frac{V_0}{S^{}}+\frac{V_{SB}}{S^{}}=0.$$
(29)
Using our present choice of Eq. (22) as the only flavor symmetry breaker and Eq. (12), this becomes
$$\frac{V_0}{S_a^a}=2A_a,\frac{V_0}{S_{}^{}{}_{a}{}^{a}}=0.$$
(30)
Now let us differentiate successively the vector generating equation with respect to $`S_a^b`$ and to $`S_{}^{}{}_{a}{}^{b}`$. This gives with the help of Eq.(30), the following two relations:
$`(\alpha _a\alpha _b){\displaystyle \frac{^2V_0}{S_b^aS_a^b}}+(\beta _a\beta _b){\displaystyle \frac{^2V_0}{S_{}^{}{}_{b}{}^{a}S_a^b}}`$ $`=`$ $`2(A_aA_b),`$
$`(\alpha _a\alpha _b){\displaystyle \frac{^2V_0}{S_b^aS_{}^{}{}_{a}{}^{b}}}+(\beta _a\beta _b){\displaystyle \frac{^2V_0}{S_{}^{}{}_{b}{}^{a}S_{}^{}{}_{a}{}^{b}}}`$ $`=`$ $`0.`$ (31)
The first of these equations relates the mass mixing transition with the unprimed scalar squared masses while the second of these relates the mass mixing transition with the primed scalar squared masses. It may be seen that information is obtained only for particles with different upper and lower SU(3) tensor indices. In the isospin invariant limit (where $`\alpha _1=\alpha _2`$ etc.), information will be obtained only for the kappa type particles (e.g. $`\kappa ^+=S_1^3`$ when mixing is neglected). If isospin violation information is inserted, information may be obtained also about the isovector scalars like $`a_0^+(980)`$ (which is represented by $`S_1^2`$ when mixing is neglected). Next, let us differentiate successively the axial vector generating equation with respect to $`\varphi `$ and to $`\varphi ^{}`$. It is neater to write the results first for the case when fields with different upper and lower tensor indices are involved:
$`(\alpha _a+\alpha _b){\displaystyle \frac{^2V_0}{\varphi _b^a\varphi _a^b}}+(\beta _a+\beta _b){\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{b}{}^{a}\varphi _a^b}}`$ $`=`$ $`2(A_a+A_b),`$
$`(\alpha _a+\alpha _b){\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{b}{}^{a}\varphi _a^b}}+(\beta _a+\beta _b){\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{b}{}^{a}\varphi _{}^{}{}_{a}{}^{b}}}`$ $`=`$ $`0`$ (32)
Next, let us write the corresponding equations for the case when the upper and lower tensor indices on each field are the same.
$`\alpha _b{\displaystyle \frac{^2V_0}{\varphi _a^a\varphi _b^b}}+\beta _b{\displaystyle \frac{^2V_0}{\varphi _a^a\varphi _{}^{}{}_{b}{}^{b}}}`$ $`=`$ $`{\displaystyle \underset{g}{}}\beta _g{\displaystyle \frac{^2V_0}{\varphi _a^a\varphi _{}^{}{}_{g}{}^{g}}}{\displaystyle \frac{8c_3}{\alpha _a}},`$
$`\alpha _b{\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{a}{}^{a}\varphi _b^b}}+\beta _b{\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{a}{}^{a}\varphi _{}^{}{}_{b}{}^{b}}}`$ $`=`$ $`{\displaystyle \underset{g}{}}\beta _g{\displaystyle \frac{^2V_0}{\varphi _{}^{}{}_{a}{}^{a}\varphi _{}^{}{}_{g}{}^{g}}}.`$ (33)
Note that the axial generating equation provides information on the masses of all the pseudoscalars. Further differentiations will relate a large number of trilinear and quadrilinear coupling constants to the meson masses and to the quark mass coefficients, $`A_a`$.
To fully characterize the system we will also require some knowledge of the axial vector and vector currents SU1 obtained by Noetherโs method:
$`(J_\mu ^{axial})_a^b`$ $`=`$ $`(\alpha _a+\alpha _b)_\mu \varphi _a^b+(\beta _a+\beta _b)_\mu \varphi _{}^{}{}_{a}{}^{b}+\mathrm{},`$
$`(J_\mu ^{vector})_a^b`$ $`=`$ $`i(\alpha _a\alpha _b)_\mu S_a^b+i(\beta _a\beta _b)_\mu S_{}^{}{}_{a}{}^{b}+\mathrm{},`$ (34)
where the three dots stand for terms bilinear in the fields.
## IV Predictions for mass spectrum
Here we consider the predictions for the mass spectrum of the model with the Lagrangian given in Eq. (13), whose potential contains any SU(3)$`{}_{L}{}^{}\times `$ SU(3)$`{}_{R}{}^{}\times `$ U(1)<sub>A</sub> invariant terms whatsoever, amended with the SU(3)$`{}_{L}{}^{}\times `$ SU(3)<sub>R</sub> but not $`U(1)_A`$ invariant term of Eq. (18) as well as the term, Eq.(22) which transforms exactly like the QCD quark mass term. A characteristic feature is mixing between fields with the same quantum numbers. Specifically, there is information about mixing between $`\pi `$ and $`\pi ^{}`$, between $`K`$ and $`K^{}`$, between $`\kappa `$ and $`\kappa ^{}`$ and among among the four $`\eta `$ type (isosinglet) states. We will take these up in turn. Note that we will be working in the isotopic spin invariant limit ispinviolation .
### IV.1 The $`\pi \pi ^{}`$ system
For compactness let us denote,
$`x_\pi `$ $`=`$ $`{\displaystyle \frac{2A_1}{\alpha _1}},`$
$`y_\pi `$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _{}^{}{}_{2}{}^{1}\varphi _{}^{}{}_{1}{}^{2}}},`$
$`z_\pi `$ $`=`$ $`{\displaystyle \frac{\beta _1}{\alpha _1}}.`$ (35)
Here we have introduced the total potential $`V=V_0+V_{SB}`$. However, since the second derivatives of $`V_{SB}`$ vanish with our present choice of flavor symmetry breaker we may just use $`V_0`$. Substituting $`a=1,b=2`$ into both of Eqs. (32) enables us to write the (non-diagonal) matrix of squared $`\pi `$ and $`\pi ^{}`$ masses as:
$$(M_\pi ^2)=\left[\begin{array}{cc}x_\pi +z_\pi ^2y_\pi & z_\pi y_\pi \\ z_\pi y_\pi & y_\pi \end{array}\right].$$
It is clear that $`z_\pi `$ is a measure of the mixing between $`\pi `$ and $`\pi ^{}`$ since the matrix becomes diagonal in the limit when $`z_\pi `$ is set to zero. So we see that $`x_\pi `$ would be the squared pion mass in the single M model and $`y_\pi `$ represents the squared mass of the โbareโ $`\pi ^{}`$. Denoting the eigenvalues of this matrix by $`m_\pi ^2`$ and $`m_\pi ^{}^2`$, we read off the product and sum rules:
$`m_\pi ^2m_\pi ^{}^2=x_\pi y_\pi ,`$
$`m_\pi ^2+m_\pi ^{}^2=x_\pi +y_\pi (1+z_\pi ^2).`$ (36)
Assuming that the values of $`m_\pi `$ and $`m_\pi ^{}`$ are known, the first of these equations expresses $`y_\pi `$ in terms of $`x_\pi `$. Then the second of these equations also expresses $`z_\pi ^2`$ in terms of $`x_\pi `$. The value of $`x_\pi `$ is not known but its range is restricted to be,
$$m_\pi ^2x_\pi m_\pi ^{}^2.$$
(37)
This range may be derived by expressing $`z_\pi ^2`$ in terms of $`x_\pi `$ as mentioned and requiring $`z_\pi ^20`$.
The transformation between the diagonal fields (say $`\pi ^+`$ and $`\pi ^+`$) and the original pion fields is defined as:
$$\left[\begin{array}{c}\pi ^+\\ \pi ^+\end{array}\right]=\left[\begin{array}{cc}\mathrm{cos}\theta _\pi & \mathrm{sin}\theta _\pi \\ \mathrm{sin}\theta _\pi & \mathrm{cos}\theta _\pi \end{array}\right]\left[\begin{array}{c}\varphi _1^2\\ \varphi _{}^{}{}_{1}{}^{2}\end{array}\right].$$
The explicit diagonalization gives an expression for the mixing angle $`\theta _\pi `$:
$$\mathrm{tan}(2\theta _\pi )=\frac{2y_\pi z_\pi }{y_\pi (1z_\pi ^2)x_\pi },$$
(38)
which evidently is also known, up to a sign choice for $`z_\pi `$, once $`x_\pi `$ is specified.
The mixing angle, $`\theta _\pi `$ can also be connected to the experimentally known value of the pion decay constant (i.e. the amplitude for the $`\pi ^+`$ meson to decay to two leptons). Substituting the expressions from Eq. (IV.1) for $`\varphi _1^2`$ and $`\varphi _{}^{}{}_{1}{}^{2}`$ in terms of the physical fields $`\pi ^+`$ and $`\pi ^+`$ into Eq. (34) yields,
$`(J_\mu ^{axial})_1^2`$ $`=`$ $`F_\pi _\mu \pi ^++F_\pi ^{}_\mu \pi ^++\mathrm{},`$
$`F_\pi `$ $`=`$ $`(\alpha _1+\alpha _2)\mathrm{cos}\theta _\pi (\beta _1+\beta _2)\mathrm{sin}\theta _\pi ,`$
$`F_\pi ^{}`$ $`=`$ $`(\alpha _1+\alpha _2)\mathrm{sin}\theta _\pi +(\beta _1+\beta _2)\mathrm{cos}\theta _\pi .`$ (39)
We can then obtain $`\alpha _1`$ (in the isospin invariant limit) as,
$$\alpha _1=\frac{F_\pi }{2(\mathrm{cos}\theta _\pi z_\pi \mathrm{sin}\theta _\pi )}.$$
(40)
We then successively obtain $`A_1`$ from the definition of $`x_\pi `$, Eq. (35) and $`\beta _1`$ from the definition of $`z_\pi `$, Eq, (35). To sum up, specifying $`x_\pi `$ and the experimental quantities $`m_\pi ,m_\pi ^{}`$ and $`F_\pi `$ determines all the other parameters of the $`\pi \pi ^{}`$ system.
### IV.2 The $`KK^{}`$ system
The treatment of this system is almost exactly analogous to that of the $`\pi \pi ^{}`$ system above when one defines the analogous variables,
$`x_K`$ $`=`$ $`{\displaystyle \frac{2(A_3+A_1)}{\alpha _3+\alpha _1}},`$
$`y_K`$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _{}^{}{}_{3}{}^{1}\varphi _{}^{}{}_{1}{}^{3}}},`$
$`z_K`$ $`=`$ $`{\displaystyle \frac{\beta _3+\beta _1}{\alpha _3+\alpha _1}}.`$ (41)
Substituting $`a=1,b=3`$ into both of Eqs. (32) enables us to write the (non-diagonal) matrix of squared $`K`$ and $`K^{}`$ masses as:
$$(M_K^2)=\left[\begin{array}{cc}x_K+z_K^2y_K& z_Ky_K\\ z_Ky_K& y_K\end{array}\right].$$
This is observed to be identical to the expression for $`(M_\pi ^2)`$ in Eq. (IV.1) when one simply substitutes everywhere $`K`$ for $`\pi `$ and $`K^{}`$ for $`\pi ^{}`$. Similarly, the four equations (36), (37), (IV.1) and (38) continue to hold when one substitutes everywhere $`K`$ for $`\pi `$ and $`K^{}`$ for $`\pi ^{}`$. Similarly, the $`K^+`$ decay constant, $`F_K`$ is now defined from,
$`(J_\mu ^{axial})_1^3`$ $`=`$ $`F_K_\mu K^++F_K^{}_\mu K^++\mathrm{},`$
$`F_K`$ $`=`$ $`(\alpha _1+\alpha _3)\mathrm{cos}\theta _K(\beta _1+\beta _3)\mathrm{sin}\theta _K,`$
$`F_K^{}`$ $`=`$ $`(\alpha _1+\alpha _3)\mathrm{sin}\theta _K+(\beta _1+\beta _3)\mathrm{cos}\theta _K..`$ (42)
We can then obtain $`\alpha _3+\alpha _1`$ (in the isospin invariant limit) as,
$$\alpha _3+\alpha _1=\frac{F_K}{\mathrm{cos}\theta _Kz_K\mathrm{sin}\theta _K}.$$
(43)
We then successively obtain $`A_3+A_1`$ from the definition of $`x_K`$ and $`\beta _3+\beta _1`$ from the definition of $`z_K`$. To sum up, specifying $`x_K`$ and the experimental quantities $`m_K,m_K^{}`$ and $`F_K`$ determines all the other parameters of the $`KK^{}`$ system.
### IV.3 The $`\kappa \kappa ^{}`$ system
Again, we can treat this system in an exactly analogous way to the $`\pi \pi ^{}`$ and $`KK^{}`$ cases if we define the analogous quantities:
$`x_\kappa `$ $`=`$ $`{\displaystyle \frac{2(A_3A_1)}{\alpha _3\alpha _1}},`$
$`y_\kappa `$ $`=`$ $`{\displaystyle \frac{^2V}{S_{}^{}{}_{3}{}^{1}S_{}^{}{}_{1}{}^{3}}},`$
$`z_\kappa `$ $`=`$ $`{\displaystyle \frac{\beta _3\beta _1}{\alpha _3\alpha _1}}.`$ (44)
In this case, however, the vector generating equations in Eqs. (31) with the choices $`a=1`$ and $`b=3`$ are used. The transformation between the diagonal and original strange scalar fields is given by,
$$\left[\begin{array}{c}\kappa ^+\\ \kappa ^+\end{array}\right]=\left[\begin{array}{cc}\mathrm{cos}\theta _\kappa & \mathrm{sin}\theta _\kappa \\ \mathrm{sin}\theta _\kappa & \mathrm{cos}\theta _\kappa \end{array}\right]\left[\begin{array}{c}S_1^3\\ S_{}^{}{}_{1}{}^{3}\end{array}\right],$$
where the mixing angle is determined by the diagonalization:
$$tan(2\theta _\kappa )=\frac{2y_\kappa z_\kappa }{y_\kappa (1z_\kappa ^2)x_\kappa }.$$
(45)
We may define $`\kappa `$ โdecay constantsโ as,
$`F_\kappa `$ $`=`$ $`(\alpha _3\alpha _1)\mathrm{cos}\theta _\kappa (\beta _3\beta _1)\mathrm{sin}\theta _\kappa ,`$
$`F_\kappa ^{}`$ $`=`$ $`(\alpha _3\alpha _1)\mathrm{sin}\theta _\kappa +(\beta _3\beta _1)\mathrm{cos}\theta _\kappa ,`$ (46)
although there is no direct experimental information available about them.
Now let us consider the $`\pi \pi ^{}`$, $`KK^{}`$ and $`\kappa \kappa ^{}`$ systems together. Using the first two we can get all of $`A_1,A_3,\alpha _1,\alpha _3,\beta _1,\beta _3`$ from the experimental masses of $`\pi ,\pi ^{},K,K^{}`$, the experimental decay constants $`F_\pi ,F_K`$ and the assumed values of $`x_\pi `$ and $`x_K`$, as seen above. This means that $`x_\kappa `$ and $`z_\kappa `$ may be read off directly from Eqs. (44) while $`y_\kappa `$ can be found from the product rule $`m_\kappa ^2m_\kappa ^{}^2=x_\kappa y_\kappa `$ if $`m_\kappa `$ and $`m_\kappa ^{}`$ are furnished. Thus all the parameters of the $`\kappa \kappa ^{}`$ system are known, given the input masses and the values of $`x_\pi `$ and $`x_K`$. However we have not yet made use of the sum rule analogous to the second of Eqs. (36). This provides another way to calculate $`z_\kappa `$ so we get the consistency condition:
$$(\frac{\beta _3\beta _1}{\alpha _3\alpha _1})^2=\frac{x_\kappa (m_\kappa ^2+m_\kappa ^{}^2x_\kappa )}{m_\kappa ^2m_\kappa ^{}^2}1$$
(47)
Since the quantities in this equation depend on both $`x_\pi `$ and $`x_K`$, the solution can determine the value of $`x_K`$ for each choice of $`x_\pi `$. In other words, if $`x_\pi `$ is specified, the parameters of the $`\pi \pi ^{}`$, the $`KK^{}`$ and the $`\kappa \kappa ^{}`$ systems are all determined in the present model.
### IV.4 The $`\eta `$ system
This system is more complicated because, even in the isotopic spin invariant limit, there are four different $`I=0`$ pseudoscalars which can mix with each other. These may be put together as a column vector according to,
$$\mathrm{\Phi }_0=\left[\begin{array}{c}\frac{\varphi _1^1+\varphi _2^2}{\sqrt{2}}\\ \varphi _3^3\\ \frac{\varphi _{}^{}{}_{1}{}^{1}+\varphi _{}^{}{}_{2}{}^{2}}{\sqrt{2}}\\ \varphi _{}^{}{}_{3}{}^{3}\end{array}\right].$$
(48)
The part of the Lagrangian describing the masses of the $`I=0`$ pseudoscalars is then: $`=(1/2)\mathrm{\Phi }_0^T(M_\eta ^2)\mathrm{\Phi }_0`$, where $`(M_\eta ^2)`$ is a symmetric $`4\times 4`$ matrix. Relations among the matrix elements follow by using both of Eqs. (33). These connect the transition masses both to the โbareโ unprimed particle masses and to the โbareโ primed particle masses. The use of isospin invariance relations like the ones given in Appendix B may also be useful. Eventually, the matrix elements of $`(M_\eta ^2)`$ depend on four new quantities in addition to the ones appearing in the above three subsystems. The resulting matrix elements are listed below:
$`\left(M_\eta ^2\right)_{11}`$ $`=`$ $`{\displaystyle \frac{2A_1}{\alpha _1}}{\displaystyle \frac{16c_3}{\alpha _1^2}}{\displaystyle \frac{\beta _1^2m_\pi ^2m_\pi ^{}^2}{2A_1\alpha _1}}+2\left({\displaystyle \frac{\beta _1}{\alpha _1}}\right)^2{\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}+4\left({\displaystyle \frac{\beta _1\beta _3}{\alpha _1^2}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`+2\left({\displaystyle \frac{\beta _3}{\alpha _1}}\right)^2{\displaystyle \frac{^2V}{\varphi _{}^{}{}_{3}{}^{3}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{12}`$ $`=`$ $`{\displaystyle \frac{8\sqrt{2}c_3}{\alpha _1\alpha _3}}{\displaystyle \frac{\beta _1^2m_\pi ^2m_\pi ^{}^2}{\sqrt{2}A_1\alpha _3}}+\left({\displaystyle \frac{2\sqrt{2}\beta _1^2}{\alpha _1\alpha _3}}\right){\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}+\left({\displaystyle \frac{2\sqrt{2}\beta _1\beta _3}{\alpha _1\alpha _3}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{13}`$ $`=`$ $`{\displaystyle \frac{\beta _1m_\pi ^2m_\pi ^{}^2}{2A_1}}+2\left({\displaystyle \frac{\beta _1}{\alpha _1}}\right){\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}+2\left({\displaystyle \frac{\beta _3}{\alpha _1}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{14}`$ $`=`$ $`\left({\displaystyle \frac{\sqrt{2}\beta _1}{\alpha _1}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}+\left({\displaystyle \frac{\sqrt{2}\beta _3}{\alpha _1}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{3}{}^{3}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{22}`$ $`=`$ $`{\displaystyle \frac{2A_3}{\alpha _3}}{\displaystyle \frac{8c_3}{\alpha _3^2}}{\displaystyle \frac{\alpha _1\beta _1^2m_\pi ^2m_\pi ^{}^2}{A_1\alpha _3^2}}+4\left({\displaystyle \frac{\beta _1}{\alpha _3}}\right)^2{\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}`$
$`\left(M_\eta ^2\right)_{23}`$ $`=`$ $`{\displaystyle \frac{\alpha _1\beta _1m_\pi ^2m_\pi ^{}^2}{\sqrt{2}A_1\alpha _3}}+\left({\displaystyle \frac{2\sqrt{2}\beta _1}{\alpha _3}}\right){\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}`$
$`\left(M_\eta ^2\right)_{24}`$ $`=`$ $`\left({\displaystyle \frac{2\beta _1}{\alpha _3}}\right){\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{33}`$ $`=`$ $`{\displaystyle \frac{\alpha _1m_\pi ^2m_\pi ^{}^2}{2A_1}}+2{\displaystyle \frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2}}`$
$`\left(M_\eta ^2\right)_{34}`$ $`=`$ $`\sqrt{2}{\displaystyle \frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}}}`$
$`\left(M_\eta ^2\right)_{44}`$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _{}^{}{}_{3}{}^{3}\varphi _{}^{}{}_{3}{}^{3}}}`$ (49)
The four new quantities are $`c_3`$, discussed earlier, and the โbareโ primed squared masses:
$$\frac{^2V}{(\varphi _{}^{}{}_{1}{}^{1})^2},\frac{^2V}{\varphi _{}^{}{}_{1}{}^{1}\varphi _{}^{}{}_{3}{}^{3}},\frac{^2V}{\varphi _{}^{}{}_{3}{}^{3}\varphi _{}^{}{}_{3}{}^{3}}.$$
These four quantities may be found by inputing the masses of four isosinglet pseudoscalars. The net result is that all four systems discussed will be completely described if all the experimental masses and the decay constants, $`F_\pi ,F_K`$ are specified together with an assumed value for $`x_\pi `$.
## V Comparison with experiment and discussion
In the preceding section we gave the tree level formulas resulting from the $`MM^{}`$ model with any SU(3)$`{}_{L}{}^{}\times `$ SU(3)$`{}_{R}{}^{}\times `$ U(1)<sub>A</sub> invariant terms together with a single โinstantonโ type term which mocks up the U(1)<sub>A</sub> anomaly and the simplest structure which mocks up the quark mass terms. Isotopic spin invariance was also assumed. Information is provided for only the pseudoscalar nonets and the strange scalar particles. Information about the scalar isotriplets can be obtained by including isospin violation effects while information about the scalar isosinglets requires either assuming some specific form for the invariant interaction terms or computing other physical quantities. These will be discussed elsewhere. Now we will input the experimental masses to try to learn what the model has to say about the quark structure of the various mesons being described. In particular we are interested in the mixing angles like $`\theta _\pi `$, governing admixtures of $`q\overline{q}`$ and $`qq\overline{q}\overline{q}`$ in the physical states and the four quark โcondensateโ strengths, $`\beta _a`$ which are associated with this mixing in the present model.
The well known lowest pseudoscalar nonet masses and decay constants will be taken, for definiteness (considering the ambiguity as to which member of a non trivial isospin multiplet to choose), to be:
$`m_\pi =0.137\mathrm{GeV}`$ , $`m_K=0.496\mathrm{GeV},`$
$`m_\eta =0.548\mathrm{GeV}`$ , $`m_\eta ^{}=0.958\mathrm{GeV},`$
$`F_\pi =0.131\mathrm{GeV}`$ , $`F_K=0.160\mathrm{GeV}.`$ (50)
Next, let us consider what are the suitable experimental inputs for the masses of the excited mesons, $`\pi ^{},K^{}.\kappa ^{}`$ and for the $`\kappa `$ meson itself. In the latest Review of particle properties rpp there are two dotted (i.e. considered established) candidates for excited pions below 2 GeV: the $`\pi (1300)`$ and the $`\pi (1800)`$. These particles could have four quark components and/or radially excited two quark components. In fact, judging from an investigation of excited baryons baryons , it is likely that both types are present. Clearly, however, for our present investigation it seems reasonable to assume that the four quark component is the dominant one and to choose the lower mass object as the more suitable one. Similarly there are two undotted (non established) excited kaon candidates: the K(1460) and the K(1830). We will again choose the lower value. As candidates for an excited strange scalar there is a dotted $`K_0^{}(1430)`$ and an undotted $`K_0^{}(1950)`$ and we again choose the lower value. In the case of the low mass strange scalar there is an undotted $`K_0^{}(800)`$ candidate, which we will interpret, with the help of BFSS1 , to be closer to 900 MeV. We summarize these choices:
$`m_\pi ^{}=1.30\mathrm{GeV}`$ , $`m_K^{}=\mathrm{\hspace{0.17em}1.46}\mathrm{GeV},`$
$`m_\kappa =0.90\mathrm{GeV}`$ , $`m_\kappa ^{}=1.42\mathrm{GeV}.`$ (51)
For the excited $`\eta `$ type pseudoscalar particles the Review of particle properties lists, below 2 GeV, the possible masses (all in GeV):
$$1.294,1.410,1.476,1.760.$$
(52)
The first three of these are dotted but the fourth is undotted. Here it seems more difficult to a priori choose which are most relevant so we shall study all possible pairings in a systematic way.
First let us discuss the $`\pi \pi ^{}`$, $`KK^{}`$ and $`\kappa \kappa ^{}`$ systems. After using the inputs of Eq. (50), all features of these systems in our model will, as already discussed, be determined by specifying $`x_\pi `$. Table 1 shows the predicted physical parameters for three values of $`x_\pi `$. For orientation we note that in the chiral model with a single field, M one has
$`\alpha _1F_\pi /2=0.0655\mathrm{GeV}`$ , $`\alpha _3F_K\alpha _1=0.0945\mathrm{GeV},`$
$`A_1{\displaystyle \frac{\alpha _1}{2}}m_\pi ^2=6.15\times 10^4\mathrm{GeV}^3`$ , $`A_3{\displaystyle \frac{F_K}{2}}m_K^2=0.01866\mathrm{GeV}^3,`$
$`\beta _10`$ , $`\beta _20.`$ (53)
The single M model corresponds to the choice $`x_\pi =m_\pi ^2`$. Increasing $`x_\pi `$ has the effect of increasing the admixture of the โfour quarkโ field component in the physical pion. The โquark mass ratioโ, $`A_3/A_1`$ = 30.3 in the single M model is not very different from the value of 31.2 obtained using the values in the $`x_\pi =0.019\mathrm{Gev}^2`$ column. The $`\overline{q}q`$ meson condensates $`\alpha _1`$ and $`\alpha _3`$ are also very similar. Of course the โfour quarkโ meson condensates $`\beta _1`$ amd $`\beta _2`$ are zero without $`M^{}`$. Despite the similarities, the $`6.4^o`$ mixing angle already corresponds to about an 11 percent โfour quarkโ admixture in the physical pion wave function. Considering that the accuracy of current algebra predictions for low energy pion physics is roughly ten percent, it seems that this choice of $`x_\pi `$ is the most plausible one. One sees from the second and third columns that relatively small increases in $`x_\pi `$ lead to large increases in four quark admixture for the pion and the kaon. Interestingly, the behavior of the four quark admixture in the strange scalar meson $`\kappa `$ is quite different. When the pseudoscalars are closer to pure โtwo quarkโ states in the model the scalar has a large four quark admixture ($`34.1^o`$, with the choice of $`x_\pi `$ in the first column). Thus the result is consistent with having a fairly large four quark component in the light scalars.
The analogs of the two quark condensates $`\alpha _1=\alpha _2`$ and $`\alpha _3`$ are approximately equal, in agreement with the usual assumption that the vacuum is approximately SU(3) symmetric. The analogs of the four quark condensates in this model are roughly an order of magnitude smaller than the similarly normalized two quark condensates. They are furthermore seen to deviate appreciably from SU(3) symmetry. It should be noted, as discussed in ref, BFSS2 for example, that the tensor indices for the primed mesons really correspond to โdual quarkโ or diquark indices in accordance with,
$$Q_aฯต_{abc}\overline{q}^b\overline{q}^c.$$
(54)
Thus in terms of the usual quarks,
$$\beta _1\overline{d}d\overline{s}s,\beta _2\overline{u}u\overline{s}s,\beta _1\overline{d}d\overline{u}u.$$
(55)
Now consider the mixing of the four $`\eta `$ type fields in the model. The basis is given in Eq.(48) while the elements of the $`4\times 4`$ mass squared matrix are given in Eq.(49). The orthogonal transformation matrix, $`K`$ which relates the mass eigenstate fields, $`\mathrm{\Phi }`$ to the original ones is defined by
$$\mathrm{\Phi }_0=K\mathrm{\Phi }.$$
(56)
As discussed in the previous section, there are, after using the symmetry information, four new unknown parameters characterizing the $`\eta `$ system. Thus taking the four mass eigenvalues from experiment could in principle determine, together with results from the $`\pi \pi ^{}`$, $`K=K^{}`$ and $`\kappa \kappa ^{}`$ systems, everything about the $`\eta `$ system for a given value of $`x_\pi `$. However there is no guarantee that there will be an exact solution for all choices of experimental parameters. This is the case, in fact, so we will search numerically for a choice of โtheoreticalโ masses which will best fit the experimental inputs. The criterion for goodness of fit will be taken to be the smallness of the quantity:
$$\chi \underset{i}{}|m_i^{\mathrm{exp}.}m_i^{\mathrm{theo}.}|/m_i^{\mathrm{exp}.}.$$
(57)
As shown in Eq. (52), there are three established candidates and one not yet established candidate below 2 GeV for the two excited $`\eta `$ states. This yields six possible scenarios for choosing them. The quantity $`\chi `$ for each choice is shown in Table 2 for three values of the parameter $`x_\pi `$. It may be observed that the fits typically get worse with increasing $`x_\pi `$, so it is reasonable to consider the choice 0.019 GeV<sup>2</sup> for this quantity as we did previously. The smallest values of $`\chi `$ are found for scenarios 5 and 6. However these both involve the $`\eta (1760)`$ state which is the one not yet established. The smallest value of $`\chi `$ using only established states is scenario 2. This case corresponds to an exact fit with eta type masses in GeV (experimental values in parentheses for comparison):
$`0.533(0.548),0.963(0.958),`$
$`1.327(1.294),1.716(1.476).`$ (58)
The detailed content of all the $`\eta `$ mass eigenstates can be read off from the matrix $`K^1`$. For scenario 2 we have,
$$K^1=\left[\begin{array}{cccc}0.570& 0.750& 0.023& 0.333\\ 0.329& 0.573& 0.142& 0.737\\ 0.704& 0.267& 0.309& 0.581\\ 0.267& 0.192& 0.940& 0.088\end{array}\right].$$
Thus, in the present model there is an 89 percent probability ($`(K^1)_{11}^2+(K^1)_{12}^2`$) that the $`\eta (548)`$ is a quark-antiquark state and an eleven percent probability that it is a four quark state. As expected, the $`\eta (548)`$ is most likely to be in an $`\overline{s}s`$ state. In the case of the $`\eta (958)`$, there is a 44 percent probability for it to be in a quark antiquark state. There is a 54 percent probability for it to be in the four quark state $`\varphi _{}^{}{}_{3}{}^{3}`$. This situation has some plausibility since in terms of ordinary quarks, the latter state has the content $`\overline{u}u\overline{d}d`$ and it should be most energetically favorable to bind a four quark state made without strange quarks.
The other scenarios which donโt employ the unconfirmed $`\eta (1760)`$ state (numbers 1 and 4) have contents very similar to the one in Eq. (V). On the other hand the three scenarios employing the $`\eta (1760)`$ have a rather different content, which seems unusual: scenarios 3, 5 and 6 make the $`\eta (958)`$ almost completely $`\varphi _{}^{}{}_{3}{}^{3}`$.
In scenario 2, which seems the most reasonable choice, we notice that the $`\eta (1295)`$ has a 43 percent probability of being in a four quark state while the $`\eta (1475)`$ has an 89 percent probability of being in a four quark state. To sum up, the value $`x_\pi `$ =0.019 GeV<sup>2</sup> leads to fairly small four quark content in the light pseudoscalars- $`\pi ,K,\eta `$ at the same time that the light scalar $`\kappa `$ has an appreciable four quark component. The โexcitedโ $`\eta `$โs are predominantly four quark states. The $`\eta (960)`$ is mainly two quark in content but has a non trivial four quark piece.
The results obtained here provide supporting evidence for the feature, illustrated in the first treatment of this model BFMNS01 , that the lightest scalars, unlike the lightest pseudoscalars, have appreciable four quark components. That model neglected quark masses and used the simplified choice of terms shown in Eq. (14). The more recent treatment of ref. nr , includes two additional invariant terms beyond those in Eq. (14) (although not all the renormalizable terms shown in Appendix A) as well as four types of quark mass splitting terms. Our results for the present treatment, where quark masses are included and which holds for any possible SU(3)$`{}_{L}{}^{}\times `$ SU(3)$`{}_{R}{}^{}\times `$ U(1)<sub>A</sub> conserving terms, are also in qualitative agreement for the $`\pi `$-type, K-type, $`\eta `$-type and $`\kappa `$-type states with that treatment. Roughly, this may be expected since the present approach includes any choice of invariant terms. However, we only used here the single quark mass splitting term of Eq. (22). Thus the results seem qualitatively robust with respect to the treatment of the mass splittings.
An interesting feature of our model is the presence of โfour quarkโ condensates as signaled by the non-zero values of the $`\beta _a`$. To make a rough estimate of what this corresponds to in quark language we proceed as follows. In ref. MS it was pointed out that the mass formulas of the single M linear sigma model could be transformed to the โcurrent algebraโ ones ca by the replacements:
$$A_a=m_a\mathrm{\Lambda }^2,\alpha _a=\frac{\overline{q}_aq_a}{2\mathrm{\Lambda }^2},$$
(59)
where the $`m_a`$ are the (โcurrentโ type) quark masses and $`\mathrm{\Lambda }`$ is the QCD scale factor. Taking $`A_1`$ = $`6.19\times 10^4\mathrm{GeV}^2`$ from the left column of Table 1 and $`m_1`$ 5 MeV we get $`\mathrm{\Lambda }0.35`$ GeV (and $`\overline{q}_aq_a0.016\mathrm{GeV}^3`$). In the case of the four quark condensate, as one sees from the discussion in the Introduction, there are several ways to couple the four quarks together to make scalars. We are assuming that one such way has been selected. For that case, it is reasonable to expect, on dimensional grounds, that
$$|\overline{d}d\overline{s}s|\mathrm{\Lambda }^5\beta _14\times 10^5\mathrm{GeV}^6.$$
(60)
In comparing the scalar masses with experiment there are expected to be, as discussed in the first four sections of ref. BFMNS01 , non-negligible corrections due to the use of unitary models for the pseudoscalar- pseudoscalar scattering based on this Lagrangian. We plan to report on this elsewhere. This should also enable us to study the isosinglet scalar masses. For both isosinglet scalars and pseudoscalars, the inclusion of possible glueball states is another interesting topic we plan to pursue. The additional symmetry breaking terms like those in Eqs. (19) and (23) seem also to be worth investigating.
## Acknowledgments
We are happy to thank A. Abdel-Rehim, D. Black, M. Harada, S. Moussa, S. Nasri and F. Sannino for many helpful related discussions. The work of A.H.F. has been supported by the 2004 Crouse Grant from the School of Arts and Sciences, SUNY Institute of Technolgy. The work of R.J. and J.S. is supported in part by the U. S. DOE under Contract no. DE-FG-02-85ER 40231.
## Appendix A Renormalizable model
The twenty one SU(3)$`\times `$ SU(3) invariant renormalizable terms in $`V_0`$ of Eq. (13) which can be made out of $`M`$ and $`M^{}`$ are:
$`V_0=`$ $``$ $`c_2\mathrm{Tr}(MM^{})+\stackrel{~}{c}_3(\mathrm{det}M+\mathrm{h}.\mathrm{c}.)+c_4^a\mathrm{Tr}(MM^{}MM^{})+c_4^b\left(\mathrm{Tr}(MM^{})\right)^2`$ (61)
$`+`$ $`d_2\mathrm{Tr}(M^{}M^{})\mathrm{Tr}(MM^{})+d_3(\mathrm{det}M^{}+\mathrm{h}.\mathrm{c}.)+d_4^a\mathrm{Tr}(M^{}M^{}M^{}M^{})+d_4^b\left(\mathrm{Tr}(M^{}M^{})\right)^2`$
$`+`$ $`e_2(\mathrm{Tr}(MM^{})+\mathrm{h}.\mathrm{c}.)`$
$`+`$ $`e_3^a(ฯต_{abc}ฯต^{def}M_d^aM_e^bM_f^c+\mathrm{h}.\mathrm{c}.)+e_3^b(ฯต_{abc}ฯต^{def}M_d^aM_e^bM_f^c+\mathrm{h}.\mathrm{c}.)`$
$`+`$ $`e_4^a\mathrm{Tr}(MM^{}M^{}M^{})+e_4^b\mathrm{Tr}(MM^{}M^{}M^{})`$
$`+`$ $`e_4^c[\mathrm{Tr}(MM^{}MM^{})+\mathrm{h}.\mathrm{c}.]+e_4^d[\mathrm{Tr}(MM^{}MM^{})+\mathrm{h}.\mathrm{c}.]+e_4^e[\mathrm{Tr}(M^{}M^{}M^{}M^{})+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`e_4^f\mathrm{Tr}(MM^{})\mathrm{Tr}(M^{}M^{})+e_4^g\mathrm{Tr}(MM^{})\mathrm{Tr}(M^{}M^{})+e_4^h[(\mathrm{Tr}(M^{}M^{}))^2+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`e_4^i[\mathrm{Tr}(MM^{})\mathrm{Tr}(MM^{})+\mathrm{h}.\mathrm{c}.]+e_4^j[\mathrm{Tr}(M^{}M^{})\mathrm{Tr}(M^{}M^{})+h.c.].`$
Notice that among these terms, those with the coefficients $`c_2,c_4^a,c_4^b,d_4^a,d_4^b,e_3^a,e_4^a,e_4^b,e_4^f,e_4^g`$ and $`e_4^h`$ are $`U(1)_A`$ invariant. It also may be of some interest to write down the twenty one renormalizable terms, linear in the matrix A, which transform like the QCD quark mass terms under SU(3)$`\times `$ SU(3). Again, for this listing, the U(1)<sub>A</sub> transformation property of the mass terms in the fundamental QCD Lagrangian is respected only for the terms with the coefficients $`k_1,k_3,k_4,k_9,k_{11},k_{12},k_{17},k_{21}`$.
$`V_{SB}=`$ $`+`$ $`k_1[\mathrm{Tr}(AM)+\mathrm{h}.\mathrm{c}.]+k_2[\mathrm{Tr}(AM^{})+\mathrm{h}.\mathrm{c}.]`$ (62)
$`+`$ $`k_3[\mathrm{Tr}(AMM^{}M)+\mathrm{h}.\mathrm{c}.]+k_4[\mathrm{Tr}(AMM^{}M^{})+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_5[\mathrm{Tr}(AMM^{}M^{})+\mathrm{h}.\mathrm{c}.]+k_6[\mathrm{Tr}(AMM^{}M)+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_7[\mathrm{Tr}(AM^{}M^{}M^{})+\mathrm{h}.\mathrm{c}.]+k_8[\mathrm{Tr}(AM^{}M^{}M)+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_9[\mathrm{Tr}(AM^{}M^{}M)+\mathrm{h}.\mathrm{c}.]+k_{10}[\mathrm{Tr}(AM^{}M^{}M^{})+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_{11}[\mathrm{Tr}(AM)+\mathrm{h}.\mathrm{c}.]\mathrm{Tr}(MM^{})`$
$`+`$ $`k_{12}[\mathrm{Tr}(AM)+\mathrm{h}.\mathrm{c}.]\mathrm{Tr}(M^{}M^{})`$
$`+`$ $`k_{13}[\mathrm{Tr}(AM)\mathrm{Tr}(MM^{})+\mathrm{h}.\mathrm{c}.]+k_{14}[\mathrm{Tr}(AM)\mathrm{Tr}(M^{}M^{})+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_{15}[\mathrm{Tr}(AM^{})+\mathrm{h}.\mathrm{c}.]\mathrm{Tr}(MM^{})`$
$`+`$ $`k_{16}[\mathrm{Tr}(AM^{})+\mathrm{h}.\mathrm{c}.]\mathrm{Tr}(M^{}M^{})`$
$`+`$ $`k_{17}[\mathrm{Tr}(AM^{})\mathrm{Tr}(MM^{})+\mathrm{h}.\mathrm{c}.]+k_{18}[\mathrm{Tr}(AM^{})\mathrm{Tr}(M^{}M^{})+\mathrm{h}.\mathrm{c}.]`$
$`+`$ $`k_{19}A_a^bฯต_{bcd}ฯต^{aef}M_e^cM_f^d+\mathrm{h}.\mathrm{c}.`$
$`+`$ $`k_{20}A_a^bฯต_{bcd}ฯต^{aef}M_{}^{}{}_{e}{}^{c}M_{}^{}{}_{f}{}^{d}+\mathrm{h}.\mathrm{c}.`$
$`+`$ $`k_{21}A_a^bฯต_{bcd}ฯต^{aef}M_e^cM_{}^{}{}_{f}{}^{d}+\mathrm{h}.\mathrm{c}.`$
## Appendix B Some isospin relations
We give examples of relations which follow from isotopic spin invariance:
$`{\displaystyle \frac{^2V}{\varphi _2^2\varphi _2^2}}`$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _1^1\varphi _1^1}},`$
$`{\displaystyle \frac{^2V}{\varphi _2^2\varphi _3^3}}`$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _1^1\varphi _3^3}},`$
$`{\displaystyle \frac{^2V}{\varphi _1^2\varphi _2^1}}`$ $`=`$ $`{\displaystyle \frac{^2V}{\varphi _1^1\varphi _1^1}}{\displaystyle \frac{^2V}{\varphi _1^1\varphi _2^2}}.`$ (63)
Similar relations hold when V is differentiated with respect to two primed fields and with respect to one primed and one unprimed field.
|
warning/0506/quant-ph0506196.html
|
ar5iv
|
text
|
# Multiplicativity of completely bounded p-norms implies a new additivity result
## 1 Introduction
Quantum channels are represented by completely positive, trace preserving (CPT) maps on $`M_d`$, the space of $`d\times d`$ matrices. Results and conjectures about additivity and superadditivity of various types of capacity play an important role in quantum information theory.
In this paper, we present a new additivity result which can be stated in terms of a type of minimal conditional entropy defined as
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })=\underset{\psi ๐^d๐^d}{inf}(S[(\mathrm{\Phi })(|\psi \psi |)]S\left[(\mathrm{Tr}_2(|\psi \psi |)]\right)`$ (1.1)
where $`S(Q)=\mathrm{Tr}Q\mathrm{log}Q`$ is the von Neumann entropy. The shorthand CB stands for โcompletely boundedโ which will be explained later. We will show that this CB minimal conditional entropy is additive, i.e.,
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A\mathrm{\Phi }_B)=S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A)+S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_B).`$ (1.2)
The expression (1.1) for $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ should be compared to those for two important types of capacity. To facilitate this, it is useful to let $`\gamma _{12}=(\mathrm{\Phi })\left(|\psi \psi |\right)`$, and observe that its reduced density matrices are $`\gamma _1=\mathrm{Tr}_2(\mathrm{\Phi })\left(|\psi \psi |\right)`$, and $`\gamma _2=\mathrm{Tr}_1(\mathrm{\Phi })\left(|\psi \psi |\right)`$. We can now rewrite (1.1) as
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ $`=\underset{\psi }{sup}\left[S(\gamma _1)S(\gamma _{12})\right].`$ (1.3)
The capacity of a quantum channel for transmission of classical information when assisted by unlimited entanglement (as in, e.g., dense coding) is given by
$`C_{EA}(\mathrm{\Phi })`$ $`=\underset{\psi }{sup}\left[S(\gamma _1)+S(\gamma _2)S(\gamma _{12})\right].`$ (1.4)
The capacity for transmission of quantum information without additional resources is the coherent information,
$`C_Q(\mathrm{\Phi })`$ $`=\underset{\psi }{sup}\left[S(\gamma _2)S(\gamma _{12})\right]`$ (1.5)
In these expressions, the supremum is taken over all normalized vectors $`\psi `$ in $`๐^d๐^d`$ and $`\gamma _{12}`$ depends on both $`\psi `$ and $`\mathrm{\Phi }`$. It has been established that $`C_{EA}(\mathrm{\Phi })`$ is additive , but $`C_Q(\mathrm{\Phi })`$ is not additive in general . To understand the difference between $`C_Q(\mathrm{\Phi })`$ and $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$, use the trace-preserving property of $`\mathrm{\Phi }`$ to rewrite $`\gamma _1=\mathrm{Tr}_2\left(|\psi \psi |\right)`$ and $`\gamma _2=\mathrm{\Phi }\left[\mathrm{Tr}_1\left(|\psi \psi |\right)\right]`$. The additive quantity (1.3) contains $`\gamma _1`$ which is independent of $`\mathrm{\Phi }`$, while the non-additive quantity (1.5) contains $`\gamma _2`$ which depends upon $`\mathrm{\Phi }`$.
We do not have a completely satisfactory physical interpretation of the the CB entropy, although an operational meaning can be found. It appears to provide a measure of how well a channel preserves entanglement. In particular, if $`\mathrm{\Phi }`$ is entanglement breaking, $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })>0`$ (although the converse does not hold). Recently, Horodecki, Oppenheim and Winter gave an elegant interpretation of quantum conditional information which we discuss in the context of our results in Section 5.
The additivity (1.2) will follow from the multiplicativity (2.5) of the quantity
$`\omega _p(\mathrm{\Phi })\underset{\psi ๐^d๐^d}{sup}{\displaystyle \frac{(\mathrm{\Phi })\left(|\psi \psi |\right)_p}{\mathrm{Tr}_2\left(|\psi \psi |\right)|_p}}=\underset{\psi ๐^d๐^d}{sup}{\displaystyle \frac{\gamma _{12}_p}{\gamma _1_p}}.`$ (1.6)
We will see that this is a type of CB norm. Recall that one of several equivalent criteria for a map $`\mathrm{\Phi }`$ to be completely positive is that for all integers $`d`$, the map $`_d\mathrm{\Phi }`$ takes positive semi-definite matrices to positive semi-definite matrices. (We use $``$ to denote the identity map $`(\rho )=\rho `$ to avoid confusion with the identity matrix $`\mathrm{I}`$.) One can similarly define other concepts, such as completely isometric, in terms of the maps $`_d\mathrm{\Phi }`$. The completely bounded (CB) norm is thus
$`\mathrm{\Phi }_{\mathrm{CB}}=\underset{d}{sup}_d\mathrm{\Phi }.`$ (1.7)
However, this depends on the precise definition of the norm on the right side of (1.7) or, equivalently, on the norms used to regard $`\mathrm{\Phi }`$ and $`_d\mathrm{\Phi }`$ as maps between Banach spaces. The appropriate definitions for the situations considered here are described in Sections 3.1 and 3.3.
In the process of deriving our results, we obtain a number of related results of independent interest. For example, we show that when $`\mathrm{\Phi }`$ is a CP map, both $`\mathrm{\Phi }_{qp}`$ and the corresponding CB norm are attained for a positive semi-definite matrix, extending a result in . The strong subadditivity (SSA) inequality for quantum entropy
$`S(Q_{123})+S(Q_3)S(Q_{23})+S(Q_{13})`$ (1.8)
is the basis for Holevoโs proof of additivity of $`C_{EA}(\mathrm{\Phi })`$ and the proof of (1.2) given in Section 2.3. In Section 6 we use operator space methods to obtain a new proof of SSA.
This paper is organized as follows. Section 2 is concerned with our main result, (1.2). After some background, we present two different proofs. In Section 3, which is divided into six subsections, we introduce notation and summarize results about CB norms and operator spaces used in the paper. Only the basic notation in Section 3.1 and the Minkowski inequalities in Section 3.4 are needed for the main result, Theorem 11. A subtle distinction between the norms used to define $`\mathrm{\Phi }_{\mathrm{CB}}`$ and $`_d\mathrm{\Phi }_{qp}`$ often used in quantum information (e.g., ) is described in the penultimate paragraph of Section 3.2.
In Section 4, we prove multiplicativity of the CB norm for maps $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$. When $`qp`$, we also show that the CB norm equals $`\mathrm{\Phi }_{qp}`$, yielding a proof of multiplicativity for the latter. In Section 5, we explicitly give $`\mathrm{\Phi }_{\mathrm{CB}}`$ and $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ for simple examples, including the depolarizing channel; prove that $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })>0`$ for EBT maps; and discuss physical interpretations. In Section 6, we use the Minkowski inequalities for the CB norms to obtain a new proof of SSA. We also show that the minimizer implicit in $`X_{12}_{(1,p)}`$ converges to $`X_1`$.
## 2 Additivity of CB entropy
### 2.1 Multiplicativity questions in quantum information theory
We are interested in CB norms when $`\mathrm{\Phi }`$ is a map $`L_q(M_d)L_p(M_d)`$ where $`L_p(M_d)`$ denotes the Banach space of $`d\times d`$ matrices with the Schatten norm $`A_p=\left(\mathrm{Tr}|A|^p\right)^{1/p}`$. One then defines the norm
$`\mathrm{\Phi }_{qp}\underset{A}{sup}{\displaystyle \frac{\mathrm{\Phi }(A)_p}{A_q}}`$ (2.1)
Watrous and Audenaert independently showed that this norm is unchanged if the supremum in (2.1) is restricted to positive semi-definite matrices, resolving a question raised in . Thus,
$`\mathrm{\Phi }_{qp}=\underset{A>0}{sup}{\displaystyle \frac{\mathrm{\Phi }(A)_p}{A_q}}`$ (2.2)
In quantum information theory, the norm $`\nu _p(\mathrm{\Phi })=\mathrm{\Phi }_{1p}`$ plays an important role. It has been conjectured (see also ) that
$`\nu _p(\mathrm{\Phi }_A\mathrm{\Phi }_B)=\nu _p(\mathrm{\Phi }_A)\nu _p(\mathrm{\Phi }_B)`$ (2.3)
in the range $`1p2`$. Proof of this conjecture would imply additivity of minimal entropy which has been shown to be equivalent to several other important and long-standing conjectures . We note here only that $`S_{\mathrm{min}}(\mathrm{\Phi })=\underset{\rho ๐}{inf}S[\mathrm{\Phi }(\rho )]`$ where $`๐=\{\rho :\rho >0,\mathrm{Tr}\rho =1\}`$ denotes the set of density matrices. Note that $`\nu _p(\mathrm{\Phi })=\underset{\rho ๐}{sup}\mathrm{\Phi }(\rho )_p`$. Amosov, Holevo and Werner showed that the additivity of minimal entropy
$`S_{\mathrm{min}}(\mathrm{\Phi }_A\mathrm{\Phi }_B)=S_{\mathrm{min}}(\mathrm{\Phi }_A)+S_{\mathrm{min}}(\mathrm{\Phi }_B)`$ (2.4)
would follow if (2.3) can be proved.
In this paper, we consider instead $`\mathrm{\Phi }_{CB,1p}`$ for which the expression in (1.7) reduces to $`\omega _p(\mathrm{\Phi })`$, and show that it is multiplicative, i.e., that
$`\omega _p(\mathrm{\Phi }_A\mathrm{\Phi }_B)=\omega _p(\mathrm{\Phi }_A)\omega _p(\mathrm{\Phi }_B).`$ (2.5)
We first show that (2.5) implies our new additivity result, providing a motivation for the technical material needed to prove (2.5). We subsequently found another proof which does not use CB norms; this is presented in Section 2.3. However, the CB proof given next provides an indication of the potential of this machinery for quantum information.
### 2.2 Proof of additivity from CB multiplicativity
We define a function of a self adjoint matrix with spectral decomposition $`A=_k\lambda _k|\varphi _k\varphi _k|`$ as $`f(A)=_kf(\lambda _k)|\varphi _k\varphi _k|`$. We will need functions of the form $`f(t)=t^p\mathrm{log}t`$ defined on $`[0,\mathrm{})`$ so that $`f(0)=0`$ for $`p>0`$ and $`Q^p\mathrm{log}Q`$ is $`0`$ on $`\mathrm{ker}(Q)`$. For any $`Q>0`$ we define the entropy as $`S(Q)=\mathrm{Tr}Q\mathrm{log}Q`$ and note that $`S\left(\frac{Q}{\mathrm{Tr}Q}\right)=\frac{1}{\mathrm{Tr}Q}S(Q)+\mathrm{log}\mathrm{Tr}Q`$.
We will often use the notation $`\gamma _{12}`$ for density matrices in the tensor product $`M_dM_nM_{dn}`$ and $`\gamma _1=\mathrm{Tr}_2\gamma _{12}`$, for the corresponding reduced density matrix in $`M_d`$. (The partial trace $`\mathrm{Tr}_2`$ denotes the trace on $`M_n`$. One can similarly define $`\gamma _2=\mathrm{Tr}_1\gamma _{12}`$ The density matrix $`\gamma _{12}`$ can be regarded as a probability distribution on $`๐_d๐_n`$ in which case $`\gamma _1`$ and $`\gamma _2`$ are the non-commutative analogues of its marginals.) We first prove a technical result.
###### Lemma 1
The function $`u(p,\gamma _{12})\frac{1}{p1}\left(1\frac{\mathrm{Tr}_{12}\gamma _{12}^p}{\mathrm{Tr}_1\gamma _1^p}\right)`$ is well-defined for $`p>1`$ and $`\gamma _{12}`$ a density matrix. It can be extended by continuity to $`p1`$ and this extension satisfies
$`u(1,\gamma _{12})={\displaystyle \frac{d}{dp}}{\displaystyle \frac{\mathrm{Tr}_{12}\gamma _{12}^p}{\mathrm{Tr}_1\gamma _1^p}}|_{p=1}=S(\gamma _{12})S(\gamma _1).`$ (2.6)
Moreover, $`u(p,\gamma _{12})`$ is uniformly bounded in $`\gamma _{12}`$ for $`p[1,2]`$ and the continuity at $`p=1`$ is uniform in $`\gamma _{12}`$.
Proof: It is well-known and straightforward to verify that, for any density matrix $`\rho `$ in $`M_m`$, $`lim_{p1}\frac{1}{p1}\left(1\mathrm{Tr}\rho ^p\right)=S(\rho )`$ and that $`0S(\rho )\mathrm{log}m`$. It then follows that (2.6) holds; the convergence is uniform in $`\gamma _{12}`$ because the set of density matrices is compact. By the mean value theorem, for any fixed $`p,\gamma _{12}`$ one can find $`\stackrel{~}{p}`$ with $`1\stackrel{~}{p}p`$ such that $`u(p,\gamma _{12})=\frac{d}{dp}\frac{\mathrm{Tr}_{12}\gamma _{12}^{\stackrel{~}{p}}}{\mathrm{Tr}_1\gamma _1^{\stackrel{~}{p}}}|_{p=\stackrel{~}{p}}`$. Combining this with the fact that $`\gamma \gamma ^{\stackrel{~}{p}}\gamma ^2`$ for any density matrix and $`\stackrel{~}{p}(1,2]`$ gives the following bound
$`|u(p,\gamma _{12})|`$ $`=`$ $`\left|{\displaystyle \frac{\mathrm{Tr}_{12}\gamma _{12}^{\stackrel{~}{p}}\mathrm{log}\gamma _{12}\mathrm{Tr}_1\gamma _1^{\stackrel{~}{p}}\mathrm{Tr}_1\gamma _1^{\stackrel{~}{p}}\mathrm{log}\gamma _1\mathrm{Tr}_{12}\gamma _{12}^{\stackrel{~}{p}}}{\mathrm{Tr}_1\gamma _1^{\stackrel{~}{p}}}}\right|`$ (2.7)
$``$ $`{\displaystyle \frac{S(\gamma _{12})+S(\gamma _1)}{\mathrm{Tr}_1\gamma _1^2}}\mathrm{๐๐๐}`$ (2.8)
which is uniform in $`p`$ for $`p(1,2]`$.
The quantity $`S_{\mathrm{cond}}(\gamma _{12})S(\gamma _{12})S(\gamma _1)`$ is called the conditional entropy. Motivated by (2.6), we define the C.B. minimal entropy as
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })=\underset{\psi ๐^d๐^d}{inf}S_{\mathrm{cond}}\left[(\mathrm{\Phi })\left(|\psi \psi |\right)\right]`$ (2.9)
and observe that it satisfies the following.
###### Theorem 2
For any CPT map $`\mathrm{\Phi }`$,
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ $`=`$ $`\underset{p1+}{lim}{\displaystyle \frac{1\left[\omega _p(\mathrm{\Phi })\right]^p}{p1}}`$ (2.10)
where $`\omega _p(\mathrm{\Phi })`$ is given by (1.6).
Proof: With $`\gamma _{12}=(\mathrm{\Phi })\left(|\psi \psi |\right)`$, one finds
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ $`=`$ $`\underset{|\psi ๐^d๐^d}{inf}u(1,\gamma _{12})=\underset{\psi }{inf}\underset{p1+}{lim}u(p,\gamma _{12})`$
$`=`$ $`\underset{\psi }{inf}\underset{p1+}{lim}{\displaystyle \frac{1}{p1}}\left(1{\displaystyle \frac{\mathrm{Tr}_{12}\gamma _{12}^p}{\mathrm{Tr}_1\gamma _1^p}}\right)`$
$`=`$ $`\underset{p1+}{lim}\underset{\psi }{inf}{\displaystyle \frac{1}{p1}}\left(1{\displaystyle \frac{\mathrm{Tr}_{12}\gamma _{12}^p}{\mathrm{Tr}_1\gamma _1^p}}\right)`$
$`=`$ $`\underset{p1+}{lim}{\displaystyle \frac{1}{p1}}\left(1\underset{\psi }{sup}{\displaystyle \frac{\mathrm{Tr}_{12}\gamma _{12}^p}{\mathrm{Tr}_1\gamma _1^p}}\right)`$
where the interchange of $`lim_{p1+}`$ and $`inf_\psi `$ is permitted by the uniformity in $`\gamma _{12}`$ of the continuity of $`u(p,\gamma _{12})`$ at $`p=1`$.
###### Theorem 3
For all pairs of CPT maps $`\mathrm{\Phi }_A,\mathrm{\Phi }_B`$,
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A\mathrm{\Phi }_B)=S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A)+S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_B)`$
Proof: The result follows easily from the observations above and (2.5).
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A\mathrm{\Phi }_B)`$ $`=`$ $`\underset{p1+}{lim}{\displaystyle \frac{1\left[\omega _p(\mathrm{\Phi }_A\mathrm{\Phi }_B)\right]^p}{p1}}`$
$`=`$ $`\underset{p1+}{lim}{\displaystyle \frac{1\left[\omega _p(\mathrm{\Phi }_A)\right]^p\left[\omega (\mathrm{\Phi }_B)\right]^p}{p1}}`$
$`=`$ $`\underset{p1+}{lim}{\displaystyle \frac{1\left[\omega _p(\mathrm{\Phi }_A)\right]^p}{p1}}+\left(\underset{p1+}{lim}\left[\omega _p(\mathrm{\Phi }_A)\right]^p\right)\underset{p1+}{lim}{\displaystyle \frac{1\left[\omega _p(\mathrm{\Phi }_B)\right]^p}{p1}}`$
$`=`$ $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_A)+S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_B)`$
where we used $`lim_{p1+}\left[\omega _p(\mathrm{\Phi }_A)\right]^p=1`$. QED
This result relies on (2.5) which is a special case of Theorem 11 with $`q=1`$. Recently, Jencova found a simple direct proof of (2.5).
### 2.3 Proof of CB additivity from SSA
Recall that any CPT map $`\mathrm{\Phi }`$ can be represented in the form
$`\mathrm{\Phi }(\rho )=\mathrm{Tr}_EU_{AE}\rho \tau _EU_{AE}^{}`$ (2.13)
with $`U_{AE}`$ unitary and $`\tau _E`$ a pure reference state on the environment. The following key result follows from standard purification arguments (which are summarized in Appendix A).
###### Lemma 4
Let the CPT map $`\mathrm{\Phi }`$ have a representation as in (2.13). One can find a reference system $`R`$ and a pure state $`|\psi _{RA}\psi _{RA}|`$ such that $`\mathrm{Tr}_R|\psi _{RA}\psi _{RA}|=\rho `$. Define $`\gamma _{REA}=(I_RU_{AE})\left(|\psi _{RA}\psi _{RA}|\tau _E\right)(I_RU_{AE})^{}`$. Then $`\gamma _{REA}`$ is also pure and
$`S(\gamma _{EA})S(\gamma _E)=S(\gamma _R)S(\gamma _{RA})`$ (2.14)
where the reduced density matrices are defined via partial traces.
It follows from (1.8) that the conditional entropy is subadditive, i.e., for any state $`\gamma _{E_1E_2A_1A_2}`$,
$`S(\gamma _{E_1E_2A_1A_2})S(\gamma _{E_1E_2})`$ $``$ $`S(\gamma _{E_1A_1})S(\gamma _{E_1})+S(\gamma _{E_2A_2})S(\gamma _{E_2})`$ (2.15)
This was proved by Nielsen and appears as Theorem 11.16 in . It follows easily from the observation that (2.15) is the sum of the following pair of inequalities, which are special cases of SSA
$`S(\gamma _{E_1E_2A_1A_2})+S(\gamma _{E_1})`$ $``$ $`S(\gamma _{E_1A_1})+S(\gamma _{E_1E_2A_2})`$
$`S(\gamma _{E_1E_2A_2})+S(\gamma _{E_2})`$ $``$ $`S(\gamma _{E_1E_2})+S(\gamma _{E_2A_2}).`$
Now define
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })=\underset{\psi }{inf}\left(S\left[(\mathrm{\Phi })\left(|\psi \psi |\right)\right]S\left[\mathrm{Tr}_A|\psi \psi |\right]\right),`$ (2.16)
Let $`\mathrm{\Psi }_{RA_1A_2}`$ denote the minimizer for $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ and
$`\gamma _{R_1R_2A_1A_2E_1E_2}=(I_RU_{A_1E_1A_2E_2})\left(|\psi _{RA_1A_2}\psi _{RA_1A_2}|\tau _{E_1E_2}\right)(I_RU_{A_1E_1A_2E_2})^{}.`$ (2.17)
Then
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_1\mathrm{\Phi }_2)`$ $`=S(\gamma _{R_1R_2A_1A_2})S(\gamma _{R_1R_2})`$
$`=S(\gamma _{E_1E_2})S(\gamma _{E_1E_2A_1A_2})`$ (2.18)
$`S(\gamma _{E_1})S(\gamma _{E_1A_1})+S(\gamma _{E_2})S(\gamma _{E_2A_2}).`$
Next, use the lemma to find purifications $`\psi _{RA}^{}`$ and $`\psi _{RA}^{\prime \prime }`$ so that the last line above
$`=S(\gamma _{R_1A_1}^{})S(\gamma _{R_1}^{})+S(\gamma _{R_2A_2}^{\prime \prime })S(\gamma _{R_2}^{\prime \prime })`$ (2.19)
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_1)+S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_2).`$
The reverse inequality can be obtained using product $`\mathrm{\Psi }`$.
## 3 Completely bounded norms
### 3.1 Definitions
For the applications in this paper, we can define the completely bounded (CB) norm of a map $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ as
$`\mathrm{\Phi }_{\mathrm{CB},qp}\underset{d}{sup}_d\mathrm{\Phi }_{(\mathrm{},q)(\mathrm{},p)}=\underset{d}{sup}\left(\underset{Y}{sup}{\displaystyle \frac{(_d\mathrm{\Phi })(Y)_{(\mathrm{},p)}}{Y_{(\mathrm{},q)}}}\right).`$ (3.1)
with
$`Y_{(\mathrm{},p)}Y_{L_{\mathrm{}}(M_d;L_p(M_n))}=\underset{A,BM_d}{sup}{\displaystyle \frac{(A\mathrm{I}_n)Y(B\mathrm{I}_n)_p}{A_{2p}B_{2p}}}.`$ (3.2)
Effros and Ruan introduced the norm $`Y_{(1,p)}`$. Pisier subsequently used complex interpolation between them to define a norm $`Y_{(t,p)}`$ for any $`1<t<\mathrm{}`$. He showed (Theorem 1.5 in ) that the norm obtained by this procedure satisfies
$`Y_{(t,p)}Y_{L_t(M_d;L_p(M_n))}=\underset{\begin{array}{c}Y=(A\mathrm{I}_n)Z(B\mathrm{I}_n)\\ A,BM_d\end{array}}{inf}A_{2t}B_{2t}Z_{(\mathrm{},p)},`$ (3.3)
which we can regard as its definition. The vector space $`M_dM_n`$ equipped with the norm (3.3) is a Banach space which we denote by $`L_t(M_d;L_p(M_n))`$. Given an operator $`\mathrm{\Omega }:L_t(M_d;L_q(M_m))L_s(M_d^{};L_p(M_n))`$, the usual norm for linear maps from one Banach space to another becomes
$`\mathrm{\Omega }\mathrm{\Omega }_{(t,q)(s,p)}=\underset{QM_dM_m}{sup}{\displaystyle \frac{\mathrm{\Omega }(Q)_{(s,p)}}{Q_{(t,q)}}}.`$ (3.4)
Theorem 1.5 and Lemma 1.7 in Pisier show that one can use this norm to obtain another expression for the CB norm
$`\mathrm{\Phi }_{\mathrm{CB},qp}\underset{d}{sup}_d\mathrm{\Phi }_{(t,q)(t,p)}=\underset{d}{sup}\left(\underset{Y}{sup}{\displaystyle \frac{(_d\mathrm{\Phi })(Y)_{(t,p)}}{Y_{(t,q)}}}\right)`$ (3.5)
valid for all $`t1`$. In effect, we can replace $`\mathrm{}`$ in (3.1) by any $`t1`$. In working with the CB norm, we will find it convenient to choose $`t=q`$ when $`qp`$ and $`t=p`$ when $`qp`$. Thus our working definition of the CB norm is (3.5) with $`t=\mathrm{min}\{q,p\}`$. For the applications considered in Sections 2 and 5, this becomes $`t=q=1`$.
Remark: When $`X>0`$, Hรถlderโs inequality implies
$`AXB^{}_p\sqrt{AXA^{}_pBXB^{}_p}\mathrm{max}\{AXA^{}_p,BXB^{}_p\}`$
and the unitary invariance of the norm implies that $`AXA^{}_p=|A|X|A|_p`$. Therefore, when $`X0`$, we can replace any expression of the form $`sup_{A,B}AXB^{}_p`$ by $`sup_{A>0}AXA_p`$ irrespective of what other restrictions may be placed upon $`A,B`$. We will show that for CP maps, the CB norm is unchanged if the supremum is taken over $`Y>0`$. (See Section 3.2, and Theorem 12 and Corollary 14 in Section 4.) Thus, when working with CP maps, one can generally assume that $`A=B>0`$ in expressions for $`Y_{(q,p)}`$.
When $`Y>0`$ combining (3.2) and (3.3) gives the identity,
$`Y_{(p,p)}=\underset{\begin{array}{c}B>0\\ \mathrm{Tr}B=1\end{array}}{inf}\underset{\begin{array}{c}A>0\\ \mathrm{Tr}A=1\end{array}}{sup}(A\mathrm{I}_n)^{\frac{1}{2p}}(B\mathrm{I}_n)^{\frac{1}{2p}}Y(B\mathrm{I}_n)^{\frac{1}{2p}}(A\mathrm{I}_n)^{\frac{1}{2p}}_p`$ (3.6)
for all $`p1`$. Since Theorem 7 implies that $`Y_{(p,p)}=Y_p`$, this gives a variational expression for the usual $`p`$-norm on $`M_{dn}M_dM_n`$. The choice $`n=1`$ yields a max-min principle for the $`p`$-norm on $`M_d`$.
The Banach space $`L_t(M_d;L_p(M_n))`$ is a special case of a more general Banach space $`L_t(M_d;E)`$ for which a norm is defined on $`d\times d`$ matrices with entries in an operator space $`E`$ as described in Section 3.3. Because we use here only operators $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ rather than the general situation of operators $`\mathrm{\Omega }:EF`$ between Banach spaces $`E,F`$, we give explicit expressions only for norms on $`L_t(M_d;L_p(M_n))`$. On a few occasions we need to consider spaces $`L_t(M_d;E)`$ with $`E=L_q(M_m;L_p(M_n))`$; we denote the norm on these space by $`Y_{(t,q,p)}`$. In general we will only encounter triples with two distinct indices and will not need additional expressions for these norms. Such cases as $`Y_{(q,q,p)}`$ reduce to $`L_q(M_{dm};L_p(M_n))`$ via the isomorphism between $`M_{dm}M_nM_dM_mM_n`$; most situations require only comparisons via Minkowski type inequalities given in Section 3.4. In section 3.5 we show that $`Y_{(1,p,1)}=Y_{(1,p)}`$; this is needed only for the application in Section 6.
### 3.2 An important lemma
We illustrate the use of (3.3) by proving the following lemma, which is a special case of a more general result in . It plays a key role in the multiplicativity results of Section 4.2 for $`qp`$. Although not needed for our main result, it also has important implications when $`qp`$. We first define $`\mathrm{\Phi }_{qp}^+=\underset{Q>0}{sup}{\displaystyle \frac{\mathrm{\Phi }(Q)_p}{Q_q}}`$.
###### Lemma 5
Let $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ be a CP map. Then for every $`r1`$ the map $`\mathrm{\Phi }_d:L_q(M_m;L_r(M_d))L_p(M_n;L_r(M_d))`$ satisfies
$`\mathrm{\Phi }_d_{(q,r)(p,r)}\mathrm{\Phi }_{qp}^+`$ (3.7)
Proof of Lemma: For any $`Q`$ (3.3) implies that one can find $`A,Y`$ such that $`Q=(A\mathrm{I})Y(B\mathrm{I})`$ and $`Q_{(q,r)}=A_{2q}B_{2q}Y_{(\mathrm{},r)}`$. Since $`\mathrm{\Phi }`$ is completely positive, one can find $`K_j`$ satisfying (3.29). Let $`V_A`$ denote the block row vector with elements $`(K_1A\mathrm{I}_d,K_2A\mathrm{I}_d,\mathrm{},K_mA\mathrm{I}_d)`$, and similarly for $`B`$. Then
$`(\mathrm{\Phi }_d)(Q)=V_A(\mathrm{I}_\nu Y)V_B^{}={\displaystyle \underset{j}{}}(K_jA\mathrm{I}_d)Y(BK_j^{}\mathrm{I}_d).`$ (3.8)
(Note that $`\mathrm{I}_\nu Y`$ denotes a block diagonal matrix with $`Y`$ along the diagonal with $`\mathrm{I}_\nu `$ the identity in an additional reference space used to implement the representation (3.29). $`Y`$ itself is in the tensor product space $`M_mM_d`$ on which $`\mathrm{\Phi }_d`$ acts; $`K_j`$ and $`A`$ are in $`M_m`$. We can extend $`V`$ to an element of $`M_\nu M_mM_d`$ by adding rows of zero blocks; i.e., to $`_{i,j=1}^\nu \delta _{i1}|ij|K_jA\mathrm{I}_d`$.) Therefore, applying (3.3) on this extended space gives
$`(\mathrm{\Phi }_d)(Q)_{(p,r)}`$ $``$ $`V_A^{}V_A_p^{1/2}V_B^{}V_B_p^{1/2}\mathrm{I}_\nu Y_{(\mathrm{},\mathrm{},r)}`$
$`=`$ $`{\displaystyle \underset{j}{}}K_j^{}A^{}AK_j_p{\displaystyle \underset{j}{}}K_j^{}B^{}BK_j_p^{1/2}Y_{(\mathrm{},r)}`$
$`=`$ $`\mathrm{\Phi }(A^{}A)_p^{1/2}\mathrm{\Phi }(A^{}A)_p^{1/2}Y_{(\mathrm{},r)}`$
$``$ $`\mathrm{\Phi }_{qp}^+A^{}A_q^{1/2}B^{}B_q^{1/2}Y_{(\mathrm{},r)}`$
$`=`$ $`\mathrm{\Phi }_{qp}^+Q_{(q,r)}`$
where we used $`A^{}A_q=\left(A_{2q}\right)^2`$. QED
The following corollary implies that for any $`p,q`$, the norm $`\mathrm{\Phi }_{qp}`$ is achieved on a positive semi-definite matrix $`Q>0`$. This was proved earlier by Watrous , resolving a question raised in . In Section 4, we will see that a similar result holds for CB norms of CP maps. This is stated as Theorem 12 for $`qp`$ and Corollary 14 for $`qp`$.
###### Corollary 6
Let $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ be a CP map. Then for all $`q,p1`$, the norm $`\mathrm{\Phi }_{qp}=\mathrm{\Phi }_{qp}^+`$
Proof: The choice $`d=1`$ in Lemma 5 gives $`\mathrm{\Phi }_{qp}\mathrm{\Phi }_{qp}^+`$. Since the reverse inequality always holds, the result follows.
Note that one can similarly conclude that $`sup_d\mathrm{\Phi }_d_{(q,t)(p,t)}=\mathrm{\Phi }_{qp}^+`$ so that nothing would be gained by defining an alternative to the CB norm in this way. In Section 5 we show that the depolarizing channel gives an explicit example of a map with $`\mathrm{\Phi }_{\mathrm{CB},1p}>\mathrm{\Phi }_{1p}`$. It is worth commenting on the difference between this result and the proof by Amosov, Holevo and Werner that $`\mathrm{\Phi }_{(1,1)(p,p)}=\mathrm{\Phi }_{1p}`$. In the latter, the identity is viewed as an isometry from one Banach space $`L_q(M_d)`$ to another, $`L_p(M_d)`$. In the case of the CB norm, the identity is viewed as a map from the Banach space $`L_t(M_d)`$ onto itself. Thus, we consider $`_d\mathrm{\Phi }`$ with $`_d:L_t(M_d)L_t(M_d)`$ and $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$, for which we need to consider what norm should be used on the domain $`M_dM_m`$ if $`tq`$ or on the range $`M_dM_n`$ if $`tp`$? When $`qp`$ this question is unavoidable. One needs a norm which acts like $`L_t`$ on $`M_d`$ and $`L_p`$ on $`M_n`$, and (3.2) provides such a norm. Some of the motivation for the definitions used here is sketched in the next section.
For a discussion of the stability properties of $`\mathrm{\Phi }_{(q,q)(p,p)}`$ see Kitaev and Watrous . Note that in the case $`q=p`$, the two types of norms for the extension $`\mathrm{\Phi }_d`$ coincide and our results imply that for CP maps $`\mathrm{\Phi }_{\mathrm{CB},\mathrm{p}\mathrm{p}}=\mathrm{\Phi }_d_{(p,p)(p,p)}=\mathrm{\Phi }_{pp}^+`$. However, for measuring the difference between channels , one is primarily interested in maps of the form $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ which are not CP.
### 3.3 Operator spaces
The Banach space $`E=L_p(M_n)`$ together with the sequence of norms on the spaces $`L_{\mathrm{}}(M_d;L_p(M_n))`$ with $`d=1,2,\mathrm{}`$ form what is known as an operator space. More generally, an operator space is a Banach space $`E`$ and a sequence of norms defined on the spaces $`M_d(E)`$, whose elements are $`d\times d`$ matrices with elements in $`E`$, with certain properties that guarantee that $`E`$ can be embedded in $`()`$, the bounded operators on some Hilbert space $``$. Alternatively, one can begin with a subspace $`E()`$; then the norm in $`M_d(E)`$ is given by the inclusion $`M_d(E)M_d\left(()\right)(^d)`$ consistent with interpreting an element of $`M_d(E)`$ as a block matrix. (Usually such a situation is considered a concrete operator space in contrast to an abstract operator space given by matrix norms satisfying Ruanโs axioms .) The only operator spaces we use in this paper are those with $`E=L_p(M_n)`$ and, occasionally, $`E=L_t(M_d;L_p(M_n))`$. Although a concrete representation for even these spaces is not known, the explicit expressions for the norms given in Sections 3.1 and 3.5 suffice for many purposes. (The reader who wishes to explore the literature should be aware that most of it is written in terms of $`L_t(M_d;E)`$ rather than $`L_t(M_d;L_p(M_n))`$ and that the notation $`S_t(M_d;E)`$ (for Schatten norm) is more common than $`L_t`$.)
For maps from $`()`$ to $`(๐ฆ)`$ complete boundedness is just uniform boundedness for the sequence of norms of $`_d\mathrm{\Phi }`$ . This notion is built in a manner analogous to the familiar notion of complete positivity. In a similar way, one can define other โcompleteโ notions, such as complete isometry based on the behavior of $`_d\mathrm{\Phi }`$.
The particular type of operator space considered here is called a โvector-valued $`L_p`$ spaceโ. We have already remarked on the need to define a norm on $`L_t(M_d;L_p(M_n))`$ to give a non-commutative generalization of the classical Banach space $`\mathrm{}_t(\mathrm{}_p)`$. Unfortunately, such naive generalizations as $`\left(_{jk}Y_{jk}_p^t\right)^{1/t}`$ or $`\left(\mathrm{Tr}_1\left(\mathrm{Tr}_2|Y|^p\right)^{t/p}\right)^{1/t}`$ do not even define norms. The norms described in Section 3.1, although difficult to work with, yield an elegant structure with the following properties.
1. for the subalgebra of diagonal matrices the norm on $`L_t(M_d;L_p(M_n))`$ reduces to that on $`\mathrm{}_t(\mathrm{}_q)`$.
2. When $`Y=AB`$ is a tensor product, $`Y_{t,p}=A_tB_p=\left(\mathrm{Tr}|A|^t\right)^{1/t}\left(\mathrm{Tr}|B|^p\right)^{1/p}`$.
3. The Banach space duality between $`L_p`$ and $`L_p^{}`$ with $`\frac{1}{p}+\frac{1}{p^{}}=1`$ generalizes to
$`L_q(M_d;L_p(M_n))^{}=L_q^{}(M_d;L_p^{}(M_n)).`$ (3.11)
4. The collection of norms on $`\{L_t(M_d;L_p(M_n))\}`$ can be obtained from some (abstract) embedding of $`L_p(M_d)`$ into $`()`$ providing the operator space structure of $`L_p(M_d)`$.
5. The structure of $`L_t(M_d;L_p(M_n))`$ can be used to develop a theory of vector-valued non-commutative integration which generalizes the theory of non-commutative integration developed by Segal and Nelson .
Although not used explicitly, properties (c) and (e) play an important role in our results. Consequences of (e) described in Section 3.4 play a key role in the proofs in Section 4 and Section 6. Theorem 10, which gives the simple expression (1.6) for the CB norm in the case $`1p`$, is an immediate consequence of a fundamental duality theorem.
For general information on operator spaces, see Paulsen , Effros and Ruan or Pisier . The theory of non-commutative vector valued $`L_p`$ spaces was developed by Pisier in two monographs and . Additional developments can be found in and .
### 3.4 Fubini and Minkowski generalizations
Because vector valued $`L_p`$-spaces permit the development of a consistent theory of vector-valued non-commutative integration, one would expect generalizations of fundamental integration theorems. This is indeed the case, and analogues of both Fubiniโs theorem and Minkowskiโs inequality play an important role in the results that follow.
First, Theorem 1.9 in gives a non-commutative version of Fubiniโs theorem.
###### Theorem 7
For any $`1p\mathrm{}`$, the isomorphisms $`L_p(M_d;L_p(M_n))L_p(M_dM_n)L_p(M_{dn})`$ hold in the sense of complete isometry, which implies that for all $`WM_dM_n`$,
$`W_{L_p(M_d;L_p(M_n))}=W_{L_p(M_n;L_p(M_d))}=W_p=\left(\mathrm{Tr}W^p\right)^{1/p}.`$ (3.12)
The next result, which is Theorem 1.10 in , will lead to non-commutative versions of Minkowskiโs inequality and deals with the flip map $`F`$ which takes $`ABBA`$ and is then extended by linearity to arbitrary elements of a tensor product space so that $`W_{12}W_{21}`$.
###### Theorem 8
For $`qp`$, the flip map $`F:L_q(M_d;L_p(M_n))L_p(M_n;L_q(M_d))`$ is a complete contraction.
The fact that $`F`$ is a contraction yields an analogue of Minkowskiโs inequality for matrices.
$`W_{21}_{(p,q)}=F(W_{12})_{(p,q)}W_{12}_{(q,p)}\text{for}qp.`$ (3.13)
The fact that $`F`$ is a complete contraction means that $`F`$ is also a contraction which yields a triple Minkowski inequality
$`W_{132}_{(q,p,q)}W_{123}_{(q,q,p)}`$ (3.14)
when $`qp`$.
Remark: To see why we regard (3.13) as a non-commutative version of Minkowskiโs inequality, recall the usual $`\mathrm{}_p(\mathrm{}_q)`$ version. For $`t1`$, $`\left[_j\left(_k|a_{jk}|\right)^t\right]^{1/t}_k\left(_j|a_{jk}|^t\right)^{1/t}`$, and Carlen and Lieb extended this to positive semi-definite matrices
$`\left[\mathrm{Tr}_1\left(\mathrm{Tr}_2Q_{12}\right)^t\right]^{1/t}\mathrm{Tr}_2\left(\mathrm{Tr}_1Q_{12}^t\right)^{1/t}`$ (3.15)
As in the case of the classical inequalities, (3.15) holds for $`t1`$ and the reverse inequality holds for $`t1`$. Moreover, it follows that for $`R0`$
$`\left[\mathrm{Tr}_1\left(\mathrm{Tr}_2R_{12}^q\right)^{p/q}\right]^{1/p}\left[\mathrm{Tr}_2\left(\mathrm{Tr}_1R_{12}^p\right)^{q/p}\right]^{1/q}\text{for}qp.`$ (3.16)
To see that (3.16) and (3.15) are equivalent, let $`t=p/q`$, and $`Q_{12}=R_{12}^p`$. Then raising both sides of (3.16) to the $`q`$-th power yields (3.15).
In general, the quantity $`\left[\mathrm{Tr}_1\left(\mathrm{Tr}_2R^p\right)^{q/p}\right]^{1/q}`$ does not define a norm. Carlen and Lieb conjectured that $`\mathrm{Tr}_1\left(\mathrm{Tr}_2R^p\right)^{1/p}`$ does define a norm for $`1p2`$, but proved it only in the case $`p=2`$. (For $`p>2`$ it can be shown not to be a norm.) Their conjecture is that
$`\mathrm{Tr}_3\left[\mathrm{Tr}_2\left(\mathrm{Tr}_1Q_{123}\right)^t\right]^{1/t}\mathrm{Tr}_{1,3}\left(\mathrm{Tr}_2Q_{123}^t\right)^{1/t}`$ (3.17)
which is very similar in form to (3.14) with $`q=1,p=t`$.
### 3.5 More facts about $`L_q(M_d;L_p(M_n))`$ norms
We now state two additional formulas for norms on $`L_q(M_d;L_p(M_n))`$. Although not needed for the main result, some consequences are needed for Theorem 12 and in Section 6. For detailed proofs see .
We state both under the assumption $`1qp\mathrm{}`$ and $`\frac{1}{q}=\frac{1}{p}+\frac{1}{r}`$. Then
$`Y_{(p,q)}Y_{L_p(M_d;L_q(M_n))}`$ $`=\underset{A,BM_d}{sup}{\displaystyle \frac{(A\mathrm{I}_n)Y(B\mathrm{I}_n)_q}{A_{2r}B_{2r}}}`$ (3.18)
and
$`Y_{(q,p)}Y_{L_q(M_d;L_p(M_n))}`$ $`=\underset{\begin{array}{c}Y=(A\mathrm{I}_n)Z(B\mathrm{I}_n)\\ A,BM_d\end{array}}{inf}A_{2r}B_{2r}Z_p`$ (3.19)
Moreover, when $`Y>0`$ is positive semi-definite, one can restrict both optimizations to $`A=B>0`$. In the case $`X>0`$, $`q=1`$, (3.18) becomes
$`X_{12}_{(p,1)}`$ $`=`$ $`\underset{A>0}{sup}{\displaystyle \frac{(A\mathrm{I}_n)X_{12}(A\mathrm{I}_n)_1}{A_{2p^{}}^2}}`$ (3.20)
$`=`$ $`\underset{A>0}{sup}{\displaystyle \frac{\mathrm{Tr}A^2X_1}{A^2_p^{}}}=X_1_p`$
and (3.19) can be rewritten as
$`X_{(1,p)}`$ $`=\underset{\begin{array}{c}A>0\\ X=(A\mathrm{I}_n)Z(A\mathrm{I}_n)\end{array}}{inf}A_{2p^{}}^2Z_p`$ (3.21)
$`=\underset{B>0,B_1=1}{inf}(B^{1/2p^{}}\mathrm{I}_n)X(B^{1/2p^{}}\mathrm{I}_n)_p`$
$`=\underset{B>0,B_1=1}{inf}(B^{\frac{1}{2}(1\frac{1}{p})}\mathrm{I}_n)X(B^{\frac{1}{2}(1\frac{1}{p})}\mathrm{I}_n)_p`$ (3.22)
In Section 6, we will also need
$`W_{132}_{(1,p,1)}`$ $`=`$ $`W_{132}_{L_1(M_d;L_p(M_n;L_1(M_m)))}`$
$`=`$ $`\underset{\begin{array}{c}AM_d,A>0\\ W_{132}=(A\mathrm{I}_{32})Z_{132}(A\mathrm{I}_{32})\end{array}}{inf}A_{2p^{}}^2Z_{132}_{(p,p,1)}`$
$`=`$ $`\underset{B_1>0,B_1_1}{inf}(B_1^{1/2p^{}}\mathrm{I}_3\mathrm{I}_2)W_{132}(B_1^{1/2p^{}}\mathrm{I}_3\mathrm{I}_2)_{(p,p,1)}`$
$`=`$ $`\underset{B_1>0,B_1_1}{inf}(B_1^{1/2p^{}}\mathrm{I}_3)W_{13}(B_1^{1/2p^{}}\mathrm{I}_3)_{(p,p)}`$
$`=`$ $`W_{13}_{(1,p)}`$
where (3.5) is proved in and the reductions which follow used (3.20) and (3.18).
###### Lemma 9
When $`1qp\mathrm{}`$ and $`X`$ is a contraction, then
$`C^{}XD_{(q,p)}\left(C^{}C_{(q,p)}D^{}D_{(q,p)}\right)^{1/2}`$ (3.25)
Proof: It follows from (3.19) that one can find $`A,BM_d`$ and $`Y,ZM_{dn}`$ such that $`A,B>0`$, $`A_{2r}=B_{2r}=1`$, $`Y,Z>0`$ and
$`C^{}C`$ $`=(A\mathrm{I}_n)Y(A\mathrm{I}_n)`$ $`C^{}C_{(q,p)}`$ $`=(A\mathrm{I}_n)Y(A\mathrm{I}_n)_p`$
$`D^{}D`$ $`=(B\mathrm{I}_n)Z(B\mathrm{I}_n)`$ $`D^{}D_{(q,p)}`$ $`=(B\mathrm{I}_n)Z(B\mathrm{I}_n)_p.`$
Moreover, there are partial isometries, $`V,W`$ such that $`C=VY^{1/2}(A\mathrm{I}_n)`$ and $`D=WZ^{1/2}(B\mathrm{I}_n)`$. Then
$`C^{}XD=(A\mathrm{I}_n)Y^{1/2}V^{}XWZ^{1/2}(B\mathrm{I}_n)`$ (3.26)
and it follows from (3.19) and Hรถlderโs inequality that
$`C^{}XD_{(q,p)}`$ $``$ $`(A\mathrm{I}_n)Y^{1/2}V^{}XWZ^{1/2}(B\mathrm{I}_n)_p`$
$``$ $`(A\mathrm{I}_n)YA\mathrm{I}_n)_p^{1/2}V^{}XW_{\mathrm{}}(B\mathrm{I}_n)ZB\mathrm{I}_n)_p^{1/2}`$
$`=`$ $`C^{}C_{(q,p)}^{1/2}D^{}D_{(q,p)}^{1/2}\mathrm{๐๐๐}`$
### 3.6 State representative of a map
A linear map $`\mathrm{\Phi }:M_dM_d`$ can be associated with a block matrix in which the $`j,k`$ block is the matrix $`\mathrm{\Phi }\left(|e_je_k|\right)`$ in the standard basis. This is often called the โChoi-Jamiolkowski matrixโ or โstate representativeโ in quantum information theory and will be denoted $`X_\mathrm{\Phi }`$. Thus,
$`X_\mathrm{\Phi }`$ $`=`$ $`{\displaystyle \underset{jk}{}}|e_je_k|\mathrm{\Phi }\left(|e_je_k|\right)`$ (3.28)
Choi showed that the map $`\mathrm{\Phi }`$ is CP if and only if $`X_\mathrm{\Phi }`$ is positive semi-definite. Conversely given a (positive semi-definite) $`d^2\times d^2`$ matrix $`X`$, one can use (3.28) to define a CP map $`\mathrm{\Phi }`$. In addition, Choi showed that the eigenvectors of $`X_\mathrm{\Phi }`$ can be rearranged to yield operators, $`K_j`$ such that
$`\mathrm{\Phi }(Q)={\displaystyle \underset{j}{}}K_jQK_j^{}.`$ (3.29)
This result representation was obtained independently by Kraus and can be recovered from that of Stinespring .
For every CP map $`\mathrm{\Phi }`$ with Choi matrix $`X_\mathrm{\Phi }`$, it follows from (3.28) that
$`(A\mathrm{I})X_\mathrm{\Phi }(A\mathrm{I})_p`$ $`=`$ $`{\displaystyle \underset{jk}{}}A|e_je_k|A\mathrm{\Phi }\left(|e_je_k|\right)_p`$
$`=`$ $`(\mathrm{\Phi })\left(|\psi _A\psi _A|\right)_p`$
where the last equality follows if we choose $`|\psi _A=_jA|e_j|e_j`$
###### Theorem 10
For any CP map $`\mathrm{\Phi }`$,
$`\mathrm{\Phi }_{\mathrm{CB},1p}=X_\mathrm{\Phi }_{(\mathrm{},p)}=\underset{\psi =1}{sup}{\displaystyle \frac{(\mathrm{\Phi })\left(|\psi \psi |\right)_p}{\mathrm{Tr}_2(|\psi \psi |)_p}}\omega _p(\mathrm{\Phi })`$ (3.31)
Proof: This result requires a fundamental duality result proved by Blecher and Paulsen and by Effros and Ruan and described in Section 2.3 of . It states that
$`\mathrm{\Phi }_{\mathrm{CB},1p}=\mathrm{\Phi }^{}_{\mathrm{CB},p^{}\mathrm{}}=X_\mathrm{\Phi }_{(\mathrm{},p)}`$ (3.32)
Using (3.2) gives
$`\mathrm{\Phi }_{\mathrm{CB},1p}`$ $`=`$ $`\underset{A>0}{sup}{\displaystyle \frac{(A\mathrm{I})X_\mathrm{\Phi }(A\mathrm{I})_p}{A^2_p}}`$
$`=`$ $`\underset{\psi }{sup}{\displaystyle \frac{(\mathrm{\Phi })\left(|\psi \psi |\right)_p}{\mathrm{Tr}_2(|\psi \psi |)_p}}.`$
Since the ratio is unchanged if $`|\psi `$ is multiplied by a constant, one can restrict the supremum above to $`\psi =1`$. QED
## 4 Multiplicativity for CB norms
### 4.1 $`1qp`$
We now prove multiplicativity of the CB norm for maps $`\mathrm{\Phi }:L_q(M_m)L_p(M_m)`$ with $`qp`$.
###### Theorem 11
Let $`qp`$ and $`\mathrm{\Phi }_A:L_q(M_{m_A})L_p(M_{n_A})`$ and $`\mathrm{\Phi }_B:L_q(M_{m_B})L_p(M_{n_B})`$ be CP and CB. Then
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}=\mathrm{\Phi }_A_{\mathrm{CB},qp}\mathrm{\Phi }_B_{\mathrm{CB},qp}.`$ (4.1)
Proof: Let $`Q_{CAB}`$ be in $`M_dM_{m_A}M_{m_B}`$ and $`R_{CAB}=(_d_{m_A}\mathrm{\Phi }_B)(Q_{CAB})`$. Then using (3.14), one finds
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}=\underset{d}{sup}\underset{Q_{CAB}}{sup}{\displaystyle \frac{(_d\mathrm{\Phi }_A\mathrm{\Phi }_B)Q_{CAB}_{(q,p,p)}}{Q_{CAB}_{(q,q,q)}}}`$
$`=`$ $`\underset{Q_{CAB}}{sup}{\displaystyle \frac{(_d\mathrm{\Phi }_A_{n_B})R_{CAB}_{(q,p,p)}}{R_{CAB}_{(q,q,p)}}}{\displaystyle \frac{(_d_{m_A}\mathrm{\Phi }_B)(Q_{CAB})_{(q,q,p)}}{Q_{CAB}_{(q,q,q)}}}`$
$``$ $`\underset{R_{CBA}}{sup}{\displaystyle \frac{(_d_{n_B}\mathrm{\Phi }_A)R_{CBA}_{(q,p,p)}}{R_{CBA}_{(q,p,q)}}}{\displaystyle \frac{R_{CBA}_{(q,p,q)}}{R_{CAB}_{(q,q,p)}}}`$
$`\times \underset{Q_{CAB}}{sup}{\displaystyle \frac{(_d_{m_A}\mathrm{\Phi }_B)(Q_{CAB})_{(q,q,p)}}{Q_{CAB}_{(q,q,q)}}}`$
$``$ $`_{n_B}\mathrm{\Phi }_A_{\mathrm{CB},(p,q)(p,p)}\mathrm{\Phi }_B_{\mathrm{CB},qp}`$
$`=`$ $`\mathrm{\Phi }_A_{\mathrm{CB},qp}\mathrm{\Phi }_B_{\mathrm{CB},qp}.`$
For the last two lines, we used $`_n\mathrm{\Phi }_A_{\mathrm{CB},(p,q)(p,p)}`$ to denote the CB norm of $`_n\mathrm{\Phi }_A:L_p(M_n;L_q(M_m))L_p(M_n;L_p(M_m))`$ and then applied Corollary 1.2 in , which states that this is the same as the CB norm of $`\mathrm{\Phi }:L_q(M_m)L_p(M_m)`$.
To prove the reverse direction, we need a slight modification of the standard strategy of showing that the bound can be achieved with a tensor product. It can happen that the CB norm itself is not attained for any finite $`I_d\mathrm{\Phi }`$ norm. Therefore, we first show that any finite product can be achieved, and then use the fact that the CB norm can be approximated arbitrarily closely by such a product.
Thus, we begin with the observation that for any $`d`$ and $`X,Y`$ in the unit balls for $`L_q(M_dM_m)`$ and $`L_q(M_dM_n)`$, there exist $`Q,R>0`$ in the unit ball of $`L_{2q}(M_d)`$ such that
$`(Q1_m)[_d\mathrm{\Phi }_A(X)](Q\mathrm{I}_m)_q=(_d\mathrm{\Phi }_A(X))_{L_q(M_d;L_p(M_m))}`$ (4.5)
and
$`(R\mathrm{I}_n)(_d\mathrm{\Phi }_B(Y))(R\mathrm{I}_n)_q=[\mathrm{\Phi }_B(Y)]_{L_q(M_d;L_p(M_n))}.`$ (4.6)
Then, using Theorem 7, one finds
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}[_{M_{d^2}}(\mathrm{\Phi }_A\mathrm{\Phi }_B)](XY)_{L_q(M_{d^2};L_p(M_{mn}))}`$ (4.7)
$``$ $`(QR\mathrm{I}_{mn})[_{d^2}(\mathrm{\Phi }_A\mathrm{\Phi }_B)](XY)(QR\mathrm{I}_{mn})_q`$
$`=`$ $`(Q\mathrm{I})[_{M_d}\mathrm{\Phi }_A(X)](Q\mathrm{I})_q(R\mathrm{I})[_{M_d}\mathrm{\Phi }_B(Y)](R\mathrm{I})_q`$
$`=`$ $`(_{M_d}\mathrm{\Phi }_A(X))_{L_q(M_d;L_p(M_m))}(_{M_d}\mathrm{\Phi }_B(X))_{L_q(M_d;L_p(M_n))}`$
Given $`ฯต>0`$, one can find $`d,X,Y`$ such that $`\mathrm{\Phi }_A_{\mathrm{CB},qp}<ฯต+(_{M_d}\mathrm{\Phi }_A(X))_{L_q(M_d;L_p(M_m))}`$ and $`\mathrm{\Phi }_B_{\mathrm{CB},qp}<ฯต+(_{M_d}\mathrm{\Phi }_B(X))_{L_q(M_d;L_p(M_n))}`$. Inserting this in (4.7) above gives
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}\mathrm{\Phi }_A_{\mathrm{CB},qp}\mathrm{\Phi }_B_{\mathrm{CB},qp}ฯต\left(\mathrm{\Phi }_A_{\mathrm{CB},qp}+\mathrm{\Phi }_B_{\mathrm{CB},qp}\right)+O(ฯต^2)`$
Since $`ฯต>0`$ is arbitrary, we can conclude that
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}\mathrm{\Phi }_A_{\mathrm{CB},qp}\mathrm{\Phi }_B_{\mathrm{CB},qp}.\mathrm{๐๐๐}`$
The next result implies that for CP maps, it suffices to restrict the supremum in the CB norm to positive semi-definite matrices.
###### Theorem 12
When $`qp`$ and $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ is CP, $`_d\mathrm{\Phi }_{(q,q)(q,p)}`$ is achieved with a positive semi-definite matrix, i.e., $`_d\mathrm{\Phi }_{(q,q)(q,p)}=_d\mathrm{\Phi }_{(q,q)(q,p)}^+`$.
Proof: First use the polar decomposition of $`QM_{dm}`$ to write $`Q=Q_1^{}Q_2`$ with $`Q_1=|Q|^{1/2}U,Q_2=|Q|^{1/2}`$ where $`U`$ is a partial isometry and $`|Q|=(Q^{}Q)^{1/2}`$. The matrix
$`\left(\begin{array}{c}Q_1^{}\\ Q_2^{}\end{array}\right)\left(\begin{array}{cc}Q_1& Q_2\end{array}\right)=\left(\begin{array}{cc}Q_1^{}Q_1& Q_1^{}Q_2\\ Q_2^{}Q_1& Q_2^{}Q_2\end{array}\right)=\left(\begin{array}{cc}U^{}|Q|U& Q\\ |Q|& Q^{}\end{array}\right)>0`$ (4.8)
is positive semi-definite. Since $`\mathrm{\Phi }`$ is CP, so is $`\mathrm{\Phi }`$ which implies that
$`\left(\begin{array}{cc}(\mathrm{\Phi })(U^{}|Q|U)& (\mathrm{\Phi })(Q)\\ (\mathrm{\Phi })(Q^{})& (\mathrm{\Phi })(|Q|)\end{array}\right)>0`$ (4.9)
is positive semi-definite. We now use the fact that a $`2\times 2`$ block matrix $`\left(\begin{array}{cc}A& C\\ C^{}& B\end{array}\right)`$ with $`A,B>0`$ is positive semi-definite if and only if $`C=A^{1/2}XB^{1/2}`$ with $`X`$ a contraction. Applying this to (4.9) gives
$`(\mathrm{\Phi })(Q)=[(\mathrm{\Phi })(U^{}|Q|U)]^{1/2}X[(\mathrm{\Phi })(|Q|)]^{1/2}`$ (4.10)
with $`X`$ a contraction. Therefore, it follows from (3.25) that
$`(\mathrm{\Phi })(Q)_{(q,p)}`$ $`=`$ $`[(\mathrm{\Phi })(U^{}|Q|U)]^{1/2}X[(\mathrm{\Phi })(|Q|)]^{1/2}_{(q,p)}`$
$``$ $`\left((\mathrm{\Phi })(U^{}|Q|U)_{(q,p)}(\mathrm{\Phi })(|Q|)_{(q,p)}\right)^{1/2}`$
$``$ $`\mathrm{\Phi }_{(q,q)(q,p)}^+\left(|Q|_qU^{}|Q|U_q\right)^{1/2}`$
$`=`$ $`\mathrm{\Phi }_{(q,q)(q,p)}^+|Q|_q\mathrm{๐๐๐}`$
### 4.2 $`qp`$
###### Theorem 13
Let $`qp`$ and $`\mathrm{\Phi }_A:L_q(M_{m_A})L_p(M_{n_A})`$, $`\mathrm{\Phi }_B:L_q(M_{m_B})L_p(M_{n_B})`$ be maps which are both CP. Then
a) $`\mathrm{\Phi }_{\mathrm{CB},qp}=\mathrm{\Phi }_{qp}=\mathrm{\Phi }_{qp}^+`$ (4.12)
b) $`\mathrm{\Phi }_A\mathrm{\Phi }_B_{qp}=\mathrm{\Phi }_A_{qp}\mathrm{\Phi }_B_{qp}`$ (4.13)
c) $`\mathrm{\Phi }_A\mathrm{\Phi }_B_{\mathrm{CB},qp}=\mathrm{\Phi }_A_{\mathrm{CB},qp}\mathrm{\Phi }_B_{\mathrm{CB},qp}`$ $`.`$ (4.14)
Combining part (a) with Corollary 6 implies that it suffices to restrict the supremum in the CB norm to positive semi-definite matrices.
###### Corollary 14
When $`qp`$ and $`\mathrm{\Phi }:L_q(M_m)L_p(M_n)`$ is CP, $`_d\mathrm{\Phi }_{\mathrm{CB},qp}`$ is achieved with a positive semi-definite matrix.
Proof of Theorem 13: To prove part (a), observe that
$`\mathrm{\Phi }_{\mathrm{CB},qp}`$ $`=`$ $`\underset{d}{sup}\left(\underset{W_{AB}M_dM_m}{sup}{\displaystyle \frac{(_d\mathrm{\Phi })(W_{AB})_p}{W_{AB}_{(p,q)}}}\right)`$
$`=`$ $`\underset{d}{sup}\left(\underset{W_{AB}}{sup}{\displaystyle \frac{(_d\mathrm{\Phi })(W_{AB})_p}{W_{BA}_{(q,p)}}}{\displaystyle \frac{W_{BA}_{(q,p)}}{W_{AB}_{(p,q)}}}\right)`$
$``$ $`\underset{d}{sup}\underset{W_{BA}M_mM_d}{sup}{\displaystyle \frac{(\mathrm{\Phi }_d)(W_{BA})_p}{W_{BA}_{(q,p)}}}`$
$``$ $`\mathrm{\Phi }_{qp}^+`$ (4.16)
The first inequality follows from the fact that the second ratio in (4.2) is $`1`$ by (3.13) and the last inequality then follows from (3.7). When $`d=1`$, the supremum over $`W`$ of the ratio in (4.2) is precisely $`\mathrm{\Phi }_{qp}`$ which implies $`\mathrm{\Phi }_{\mathrm{CB},qp}\mathrm{\Phi }_{qp}`$. This proves part (a).
To prove part (b), write $`\mathrm{\Phi }_A\mathrm{\Phi }_B=(\mathrm{\Phi }_A)(\mathrm{\Phi }_B)`$ and for any $`Q_{AB}M_{m_A}M_{m_B}`$, let $`R_{AB}=(\mathrm{\Phi }_B)(Q)`$. Then
$`\mathrm{\Phi }_A\mathrm{\Phi }_B_{qp}=\underset{Q}{sup}{\displaystyle \frac{(\mathrm{\Phi }_A\mathrm{\Phi }_B)(Q)_p}{Q_q}}`$
$``$ $`\underset{Q}{sup}{\displaystyle \frac{(\mathrm{\Phi }_A)(R_{AB})_p}{R_{AB}_{(q,p)}}}{\displaystyle \frac{R_{AB}_{(q,p)}}{R_{BA}_{(p,q)}}}{\displaystyle \frac{(\mathrm{\Phi }_B)(Q_{BA})_{(p,q)}}{Q_q}}`$
$``$ $`\underset{R}{sup}{\displaystyle \frac{(\mathrm{\Phi }_A)(R)_{(p,p)}}{R_{(q,p)}}}\underset{Q_{BA}}{sup}{\displaystyle \frac{(\mathrm{\Phi }_B)(Q_{BA})_{(p,q)}}{Q_{BA}_{q,q}}}`$
$``$ $`\mathrm{\Phi }_A_{qp}\mathrm{\Phi }_B_{qp}`$
where we used (3.13), Fubini, and $`R_{BA}=(\mathrm{\Phi }_B)(Q_{BA})`$. This proves (b).
Part (c) then follows immediately from (a) and (b). QED
## 5 Applications of CB entropy
### 5.1 Examples and bounds
It is well-known that conditional information can be negative as well as positive. Therefore, it is not surprising that (1.1) can also be either positive or negative, depending on the channel $`\mathrm{\Phi }`$. As in Section 1, we adopt the convention that $`\gamma _{12}=(\mathrm{\Phi })\left(|\psi \psi |\right)`$. One has the general bounds
$`S(\gamma _1)S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })S(\gamma _1)`$ (5.1)
which imply
$`\mathrm{log}d`$ $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })\mathrm{log}d.`$ (5.2)
The lower bound in (5.2) follows from the definition (1.3) and the positivity of the entropy $`S(\gamma _{12})>0`$; the upper bound follows from subadditivity $`S(\gamma _{12})S(\gamma _1)+S(\gamma _2)`$. The upper bound is attained if and only if the output $`(\mathrm{\Phi })\left(|\psi \psi |\right)`$ is always a product. The lower bound in (5.2) is attained for the identity channel, and the upper bound for the completely noisy channel $`\mathrm{\Phi }(\rho )=(\mathrm{Tr}\rho )\frac{1}{d}I`$.
Next, consider the depolarizing channel $`\mathrm{\Omega }_\mu (\rho )=\mu \rho +(1\mu )(\mathrm{Tr}\rho )\frac{1}{d}I`$. This channel satisfies the covariance condition $`U\mathrm{\Omega }(\rho )U^{}=\mathrm{\Omega }(U\rho U^{})`$ for all unitary $`U`$. Lemma 2 in the appendix of can therefore be used to show that the minimal CB entropy is achieved when $`\gamma _1=\mathrm{Tr}_2(\mathrm{\Omega })(|\psi \psi |)`$ is the maximally mixed state $`\frac{1}{d}\mathrm{I}`$ so that $`|\psi `$ is maximally entangled and
$`\mathrm{\Omega }_{\mathrm{CB},1p}=S(X_\mathrm{\Omega })\mathrm{log}d`$ (5.3)
Moreover, the state $`(\mathrm{\Omega }_\mu )(|\psi \psi |)`$ has one non-degenerate eigenvalue $`\frac{1+(d^21)\mu }{d^2}`$ and the eigenvalue $`\frac{1\mu }{d^2}`$ with multiplicity $`d^21`$. From this one finds
$`\omega _p(\mathrm{\Omega }_\mu )`$ $`=d^{\frac{p+1}{p}}\left[(1\mu +d^2\mu )^p+(d^21)(1\mu )^p\right]^{1/p}`$ (5.4)
and
$`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Omega }_\mu )`$ $`=\frac{1\mu }{d^2}\mathrm{log}\frac{1\mu }{d^2}(d^21)\frac{1\mu }{d^2}\mathrm{log}\frac{1\mu }{d^2}\mathrm{log}d`$ (5.5)
$`=\mathrm{log}d\frac{1}{d^2}[(1\mu +d^2\mu )\mathrm{log}(1\mu +d^2\mu )+(d^21)(1\mu )\mathrm{log}(1\mu )`$
In the case of qubits, $`d=2`$ and (5.4) becomes
$`\mathrm{\Omega }_\mu _{\mathrm{CB},1p}=\omega _p(\mathrm{\Omega }_\mu )`$ $`=2^{(p+1)/p}\left[(1+3\mu )^p+3(1\mu )^p\right]^{1/p}`$ (5.6)
which can be compared to
$`\mathrm{\Omega }_\mu _{1p}=\nu _p(\mathrm{\Omega }_\mu )`$ $`=2^1\left[(1+\mu )^p+(1\mu )^p\right]^{1/p}.`$ (5.7)
The strict convexity of $`f(x)=x^p`$ implies that for $`\mu >0`$,
$`(1+\mu )^p=\left(\frac{(1+3\mu )+(1\mu )}{2}\right)^p<{\displaystyle \frac{1}{2}}\left[(1+3\mu )^p+3(1\mu )^p\right]`$
from which it follows that $`\mathrm{\Omega }_\mu _{\mathrm{CB},1p}>\mathrm{\Omega }_\mu _{1p}`$. This confirms that, in general, the CB norm $`\mathrm{\Phi }_{\mathrm{CB},1p}`$ of a map $`\mathrm{\Phi }`$ is strictly greater than $`\mathrm{\Phi }_{1p}`$. (This can be seen directly for the identity map $``$ which corresponds to $`\mu =1`$.) For qubits, one can verify explicitly that $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ is achieved with a maximally entangled state and that it decreases monotonically with $`\mu `$. Numerical work shows that $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ changes from positive to negative at $`\mu =0.74592`$, which is also the cut-off for $`C_Q(\mathrm{\Phi })=0`$.
The Werner-Holevo channel is $`\mathrm{\Phi }_{\mathrm{WH}}(\rho )=\frac{1}{d1}\left[(\mathrm{Tr}\rho )\mathrm{I}\rho ^T\right]`$. One finds that $`\gamma _{12}`$ has exactly $`\left(\genfrac{}{}{0pt}{}{d}{2}\right)`$ non-zero eigenvalues $`\frac{1}{d1}(a_j^2+a_k^2)`$ with $`j<k`$ and $`a_j^2`$ the eigenvalues of $`\gamma _1`$. One can then use the concavity of $`x\mathrm{log}x`$ to show that $`S(\gamma _{12})S(\gamma _1)+\mathrm{log}\frac{d1}{2}`$, which implies that $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_{\mathrm{WH}})=\mathrm{log}\frac{d1}{2}`$ is achieved with a maximally entangled input. Moreover, $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_{\mathrm{WH}})=1`$ for $`d=2`$, and $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi }_{\mathrm{WH}})=0`$ for $`d=3`$. One can also use the covariance property $`\mathrm{\Phi }_{\mathrm{WH}}(U\rho U^{})=\overline{U}\mathrm{\Phi }_{\mathrm{WH}}(\rho )U^T`$ and Lemma 2 of to see that $`\omega _p(\mathrm{\Phi }_{\mathrm{WH}})`$ is achieved with a maximally entangled state, and verify that
$`\omega _p(\mathrm{\Phi }_{\mathrm{WH}})=\left(\frac{2}{d1}\right)^{1\frac{1}{p}}>\left(\frac{1}{d1}\right)^{1\frac{1}{p}}=\nu _p(\mathrm{\Phi }_{\mathrm{WH}}).`$ (5.8)
This gives another example for which the CB norm is strictly greater than $`\mathrm{\Phi }_{1p}`$.
However, the CB norm is not always attained on a maximally entangled state. Consider for example the non-unital qubit map $`\mathrm{\Phi }(\rho )=\lambda \rho +\left(\frac{(1\lambda )}{2}I+\frac{t}{2}\sigma _3\right)\mathrm{Tr}\rho `$, and the one-parameter family of pure bipartite states $`|\psi _a=\sqrt{a}|00+\sqrt{1a}|11`$ where $`0a1`$. In this case
$`\gamma _{12}`$ $`=`$ $`(I\mathrm{\Phi })(|\psi _a\psi |)`$
$`=`$ $`{\displaystyle \frac{1}{2}}\left(\begin{array}{cccc}a(1+t+\lambda )& 0& 0& 2\lambda \sqrt{a(1a)}\\ 0& (1a)(1+t\lambda )& 0& 0\\ 0& 0& a(1t\lambda )& 0\\ 2\lambda \sqrt{a(1a)}& 0& 0& (1a)(1t+\lambda )\end{array}\right)`$
Numerical computations show that for $`p>1`$, $`\frac{\gamma _{12}_p}{\gamma _1_p}`$ is maximized at values $`a>1/2`$ when $`t>0`$, and values $`a<1/2`$ when $`t<0`$. Since the state $`|\psi _a`$ is maximally entangled only when $`a=1/2`$, this demonstrates that the CB norm $`\omega _p(\mathrm{\Phi })`$ is achieved at a non-maximally entangled state for this family of maps.
### 5.2 Entanglement breaking and preservation
The class of channels for which $`(\mathrm{\Phi })(\rho )`$ is separable for any input is called entanglement breaking (EB). Those which are also trace preserving are denoted EBT. These maps were introduced in by Holevo who wrote them in the form $`\mathrm{\Phi }(\rho )=_kR_k\mathrm{Tr}\rho E_k`$ with each $`R_k`$ a density matrix and $`\{E_k\}`$ a POVM, i.e., $`E_k0`$ and $`_kE_k=I`$. They were studied in where several equivalent conditions were proved. The next result shows that EBT channels always have positive minimal CB entropy. Therefore, a channel for which $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ is negative always preserves some entanglement.
###### Lemma 15
If $`\mathrm{\Phi }:M_mM_n`$ is an EBT map, then for all $`p1`$ and positive semi-definite $`QM_nM_m`$,
$`(_n\mathrm{\Phi })(Q)_p\mathrm{Tr}_2Q_p=Q_1_p`$ (5.9)
###### Theorem 16
If $`\mathrm{\Phi }`$ is an EBT map, then $`\omega _p(\mathrm{\Phi })1`$ and $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ is positive.
Theorem 16 follows immediately from Lemma 15 and Theorem 2 of Section 2.2. The converse does not holds, i.e., $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })0`$ does not imply that $`\mathrm{\Phi }`$ is EBT. For the depolarizing channel, it is known that $`\mathrm{\Omega }_\alpha `$ is EBT if and only if $`|\alpha |\frac{1}{3}`$; however, as reported above, $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Omega }_\alpha )>0`$ for $`0<\alpha <0.74592`$. For $`d>3`$, the WH channel also has positive CB entropy, although it can not break all entanglement because it is known that $`\nu _p(\mathrm{\Phi }_{\mathrm{WH}})`$ is not multiplicative for sufficiently large $`p`$.
The proof of Lemma 15 is similar to Kingโs argument for showing multiplicativity of the maximal $`p`$-norm for EBT maps, and is based on the following inequality due to Lieb and Thirring
$`\mathrm{Tr}(C^{}DC)\mathrm{Tr}(CC^{})^pD^p`$ (5.10)
for $`p1`$ and $`D>0`$ positive semi-definite.<sup>1</sup><sup>1</sup>1The proof in the Appendix of is based on the concavity of $`A\mathrm{Tr}(BA^{1/m}B)^m`$ for $`m1`$ and $`A,B0`$. This was first proved by Epstein ; it is also a special case of Lemma 1.14 in , which is proved using complex interpolation in the operator space framework. Araki gave another proof of (5.10), and a simple proof based on Hรถlderโs inequality was given by Simon in Theorem I.4.9 of .
Proof of Lemma 15: By assumption, we can write $`\mathrm{\Phi }(\rho )=_kR_k\mathrm{Tr}\rho E_k`$ with each $`R_k`$ a density matrix and $`\{E_k\}`$ a POVM. Then
$`(_n\mathrm{\Phi })(Q)`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{\kappa }{}}}[\mathrm{Tr}_2(\mathrm{I}X_k)Q]R_k`$ (5.11)
$`=`$ $`{\displaystyle \underset{k=1}{\overset{\kappa }{}}}G_kR_k`$
where $`G_k=_k[\mathrm{Tr}_2(\mathrm{I}X_k)Q]`$. Note that
$`\mathrm{Tr}_2Q={\displaystyle \underset{k=1}{\overset{\kappa }{}}}[\mathrm{Tr}_2(\mathrm{I}X_k)Q]={\displaystyle \underset{k=1}{\overset{\kappa }{}}}G_k`$ (5.12)
With $`|e_k`$ the canonical basis in $`๐_\kappa `$ we define the following matrices in $`M_\kappa M_nM_n`$.
$`R={\displaystyle \underset{k}{}}|e_ke_k|\mathrm{I}_nR_k=\left(\begin{array}{cccc}\mathrm{I}_nR_1& 0& \mathrm{}& 0\\ 0& \mathrm{I}_nR_2& \mathrm{}& 0\\ \mathrm{}& & \mathrm{}& \mathrm{}\\ 0& \mathrm{}& 0& \mathrm{I}_nR_\kappa \end{array}\right)`$ (5.13)
and
$`V=\stackrel{~}{V}\mathrm{I}_n={\displaystyle \underset{k}{}}|e_ke_1|G_k^{1/2}\mathrm{I}_n=\left(\begin{array}{cccc}\sqrt{G_1}& 0& \mathrm{}& 0\\ \sqrt{G_2}& 0& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& & \mathrm{}\\ \sqrt{G_\kappa }& 0& \mathrm{}& 0\end{array}\right)\mathrm{I}_n`$ (5.14)
where we adopt the convention of using the subscripts $`3,1,2`$ for $`M_\kappa ,M_n,M_n`$ respectively so that the partial traces $`\mathrm{Tr}_1`$ and $`\mathrm{Tr}_2`$ retain their original meaning. It follows that
$`|e_1e_1|(_n\mathrm{\Phi })(Q)=V^{}RV.`$ (5.15)
Applying (5.10) one finds
$`(_n\mathrm{\Phi })(Q)_p^p`$ $`=`$ $`\mathrm{Tr}(V^{}RV)^p=\mathrm{Tr}_{312}(V^{}RV)^p`$
$``$ $`\mathrm{Tr}_{312}(VV^{})^pR^p`$
$`=`$ $`{\displaystyle \underset{k}{}}\mathrm{Tr}_{12}[(VV^{})^p]_{kk}(\mathrm{I}_nR_k)^p`$
$`=`$ $`{\displaystyle \underset{k}{}}\mathrm{Tr}_1[(\stackrel{~}{V}\stackrel{~}{V}^{})^p]_{kk}\mathrm{Tr}_2(R_k)^p`$ (5.17)
where $`[(\stackrel{~}{V}\stackrel{~}{V}^{})^p]_{kk}=\mathrm{Tr}_3(\stackrel{~}{V}\stackrel{~}{V}^{})^p(|e_ke_k|\mathrm{I}_n)`$ is the $`k`$-th block on the diagonal of $`(\stackrel{~}{V}\stackrel{~}{V}^{})^p`$ and $`[(VV^{})^p]_{kk}=[(\stackrel{~}{V}\stackrel{~}{V}^{})^p]_{kk}\mathrm{I}_n`$. Since $`R_k`$ is a density matrix, $`\mathrm{Tr}_2(R_k)^p1`$. (In fact, we could assume wlog that $`R_k=|\theta _k\theta _k|`$ so that $`R_k^p=R_k`$ and $`\mathrm{Tr}_2(R_k)^p=1`$.) Therefore,
$`(_n\mathrm{\Phi })(Q)_p^p`$ $``$ $`{\displaystyle \underset{k}{}}\mathrm{Tr}_1[(\stackrel{~}{V}\stackrel{~}{V}^{})^p]_{kk}`$
$`=`$ $`\mathrm{Tr}_{31}(\stackrel{~}{V}\stackrel{~}{V}^{})^p=\mathrm{Tr}_{31}(\stackrel{~}{V}^{}\stackrel{~}{V})^p`$
$`=`$ $`{\displaystyle \underset{k}{}}\mathrm{Tr}_1G_k=\mathrm{Tr}_2Q\mathrm{๐๐๐}`$
### 5.3 Operational interpretation
Recently Horodecki, Oppenheim and Winter (HOW) obtained results which give an important operational meaning to quantum conditional information, consistent with both positive or negative values. Applying their results to the expression $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })=S(\gamma _{AB})S(\gamma _A)`$ with $`\gamma _{AB}=(\mathrm{\Phi })\left(|\psi \psi |\right)`$ where $`|\psi `$ is the minimizer in (1.1) gives the following interpretation:
* A channel for which $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })>0`$ always breaks enough entanglement so that some EPR pairs must be added to enable Alice to transfer her information to Bob.
* A channel for which $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })<0`$ leaves enough entanglement in the optimal state so that some EPR pairs remain after Alice has transferred her information to Bob.
For example, as discussed in Section 5.1 the depolarizing channel is entanglement breaking for $`\mu [\frac{1}{3},\frac{1}{3}]`$; for $`\mu (\frac{1}{3},0.74592)`$ it always breaks enough entanglement to require input of EPR pairs to transfer Bobโs corrupted state back to Alice; and for $`\mu >0.74592`$ maximally entangled states retain enough entanglement to allow the distillation of EPR pairs after Bobโs corrupted information is transferred to Alice.
Note, however, that the HOW interpretation is an asymptotic result in the sense that it is are based on the assumption of the availability of the tensor product state $`\gamma _{AB}^n`$ with $`n`$ arbitrarily large, and is related to the โentanglement of assistanceโ which is known not to be additive. One would also like to have an interpretation of the additivity of $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ so that the โone-shotโ formula $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$ represents the capacity of an asymptotic process which is not enhanced by entangled inputs. Thus far, the only scenarios for which we have found this to be true seem extremely contrived and artificial.
## 6 Entropy Inequalities
In this section, we show that operator space methods can be used to give a new proof of SSA (1.8). Although the strategy is straightforward, it requires some rather lengthy and tedious bounds on derivatives and norms. Our purpose is not to give another proof of SSA, but to demonstrate the fundamental role of Minkowski-type inequalities and provide some information on the behavior of the $`_{(1,p)}`$ near $`p=1`$.
Differentiation of inequalities of the type found in Section 3.4 often yields entropy inequalities. The procedure is as follows. Consider an inequality of the form $`g_L(p)g_R(p)`$ valid for $`p1`$ which becomes an equality at $`p=1`$. Then the function $`g(p)=g_R(p)g_L(p)0`$ for $`p1`$ and $`g(1)=0`$. This implies that the right derivative $`g^{}(1+)0`$ or, equivalently, that $`g_L^{}(1+)g_R^{}(1+)`$.
Applying this to (3.16) yields
$`S(Q_1)S(Q_{12})+S(Q_2)`$ (6.1)
which is the well-known subadditivity inequality $`S(Q_{12})S(Q_1)+S(Q_2)`$. Applying the same principle to conjecture (3.17) yields
$`S(Q_{23})+S(Q_3)S(Q_{123})+S(Q_{13})`$ (6.2)
which is equivalent to strong subadditivity (1.8). (Carlen and Lieb observed that the reverse of (3.17) holds when $`t1`$ and used the corresponding left derivative inequality $`g_L^{}(1)g_R^{}(1)`$ to obtain another proof of SSA.)
These entropy inequalities can also be obtained by differentiating the corresponding CB Minkowski inequalities (3.13) and (3.14). We will need the following.
###### Theorem 17
For any $`X=X_{12}`$ in $`M_mM_n`$, with $`X0`$ and $`\mathrm{Tr}X=1`$.
$`\frac{d}{dp}X_{12}_{(1,p)}^p|_{p=1}`$ $`=S(X_{12})+S(X_1).`$ (6.3)
Before proving this result, observe that (3.20) implies $`W_{12}_{(1,p)}=W_2_p`$ and (3.5) implies $`W_{132}_{(1,p,1)}=W_{13}_{(1,p)}`$. Then, when $`q=1`$, the inequalities (3.13) and (3.14) imply
$`W_2_p^p`$ $``$ $`W_{12}_{(1,p)}^p`$ (6.4)
$`W_{13}_{(1,p)}^p`$ $``$ $`W_{123}_{(1,1,p)}^p.`$ (6.5)
Now, under the assumption that $`W_{123}>0`$ and $`\mathrm{Tr}W_{123}=1`$, Theorem 17 implies
$`\frac{d}{dp}W_{21}_{(1,p)}^p|_{p=1}`$ $`=`$ $`S(W_2)`$
$`\frac{d}{dp}W_{12}_{(1,p)}^p|_{p=1}`$ $`=`$ $`S(W_{12})+S(W_1)`$
$`\frac{d}{dp}W_{123}_{(1,1,p)}^p|_{p=1}`$ $`=`$ $`S(W_{123})+S(W_{12})`$
$`\frac{d}{dp}W_{13}_{(1,p)}^p|_{p=1}`$ $`=`$ $`S(W_{13})+S(W_1)`$
Then usual subadditivity and SSA inequalities, (6.1) and (6.2) then follow from the principle, $`g_L^{}(1+)g_R^{}(1+)`$, above and (6.4) and (6.5) respectively.
Proof of Theorem 17: The basic strategy is similar to that in Section 2.2, but requires some additional details. Let $`X_1=\mathrm{Tr}_2X`$ and let $`Q`$ denote the orthogonal projection onto $`\mathrm{ker}(X_1)`$. Since $`QX_1Q=0`$, it follows that $`\mathrm{Tr}(Q\mathrm{I}_n)X(Q\mathrm{I}_n)=0`$. Since $`X`$ is positive semi-definite this implies that $`X=((\mathrm{I}_mQ)\mathrm{I}_n)X((\mathrm{I}_mQ)\mathrm{I}_n)`$. For fixed $`X`$ the functions
$`v(p,B)`$ $`=`$ $`X^{1/2}(B^{\frac{1}{p}1}\mathrm{I}_n)X^{1/2},\text{and}`$ (6.6)
$`w(p,B)`$ $`=`$ $`X_1^{\frac{1}{2}}B^{\frac{1}{p}1}X_1^{\frac{1}{2}}.`$ (6.7)
are well-defined for $`p>1`$, and $`B\beta (X_1)`$ where $`\beta (X_1)=\{B๐:\mathrm{ker}(B)\mathrm{ker}(X_1)\}`$. Since $`(B^{\frac{1}{2}(1\frac{1}{p})}\mathrm{I}_n)X(B^{\frac{1}{2}(1\frac{1}{p})}\mathrm{I}_n)_p=X^{\frac{1}{2}}(B^{\frac{1}{2}}B^{\frac{1}{p}}B^{\frac{1}{2}}\mathrm{I}_n)X^{\frac{1}{2}}_p`$, it follows from (3.22) and the remarks above that
$`X_{12}_{(1,p)}=\underset{B๐}{inf}v(p,B)_p=\underset{B\beta (X_1)}{inf}v(p,B)_p.`$ (6.8)
The set of density matrices $`๐`$ is compact, and $`v(p,B)_p`$ is bounded below and continuous, hence for each $`p>1`$ there is a (i.e., at least one) density matrix $`B(p)`$ which minimizes $`v(p,B)_p`$, so that
$`X_{12}_{(1,p)}=v(p,B(p))_p`$ (6.9)
Since $`p>1`$ and $`B(p)`$ is a density matrix, $`B(p)^{1+\frac{1}{p}}>\mathrm{I}_m`$ which implies
$`v(p,B)X\text{and}w(p,B)X_1.`$ (6.10)
Furthermore,
$`1=\mathrm{Tr}X\mathrm{Tr}v(p,B(p))(mn)^{\frac{1}{p}1}v(p,B(p))_p`$ (6.11)
(where the last inequality uses $`A_1d^{1\frac{1}{p}}A_p`$ for any positive semi-definite $`d\times d`$ matrix $`A`$ and any $`p1`$). Replacing $`B(p)`$ by another density matrix cannot decrease $`v(p,B(p))_p`$, hence
$`v(p,B(p))_pv(p,{\displaystyle \frac{1}{m}}\mathrm{I}_m)_p=m^{1\frac{1}{p}}X_pm^{1\frac{1}{p}}`$ (6.12)
Combining (6.11) and (6.12) shows that
$`\underset{p1+}{lim}\mathrm{Tr}(v(p,B(p))X)=0,`$ (6.13)
and, together with (6.10) implies that $`v(p,B(p))X`$. Also, for any $`B\beta (X_1)`$,
$`\mathrm{Tr}v(p,B)`$ $`=`$ $`\mathrm{Tr}_{12}X_{12}[B]^{\frac{1}{p}1}\mathrm{I}_n)`$ (6.14)
$`=`$ $`\mathrm{Tr}X_1[B]^{\frac{1}{p}1}=\mathrm{Tr}w(p,B)`$
so that $`lim_{p1+}\mathrm{Tr}(w(p,B(p))X_1)=0`$ and $`w(p,B(p))X_1`$.
Writing out the derivative on the left side of (6.3), we see that we need to show that
$`\underset{p1+}{lim}{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}v(p,B(p))^p1\right)=S(X)+S(X_1)`$ (6.15)
First note that for $`p>1`$.
$`{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}v(p,B(p))^p1\right){\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}v(p,X_1)^p1\right),`$ (6.16)
and a direct calculation shows that the right side of (6.16) converges to $`S(X)+S(X_1)`$ as $`p1+`$. Hence to prove (6.15) it is sufficient to show that
$`\underset{p1+}{lim\; inf}{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}v(p,B(p))^p1\right)S(X)+S(X_1)`$ (6.17)
Hรถlderโs inequality implies
$`1=X_1_1`$ $`=`$ $`B(p)^{\frac{1}{2}\frac{1}{2p}}\left(B(p)^{\frac{1}{2p}\frac{1}{2}}X_1B(p)^{\frac{1}{2p}\frac{1}{2}}\right)B(p)^{\frac{1}{2}\frac{1}{2p}}_1`$ (6.18)
$``$ $`B(p)^{\frac{1}{2}\frac{1}{2p}}_{2p/p1}^2B(p)^{\frac{1}{2p}\frac{1}{2}}X_1B(p)^{\frac{1}{2p}\frac{1}{2}}_p`$
$`=`$ $`w(p,B(p))_p`$
Combining this with (6.14) gives a bound on the numerator on the left in (6.15)
$`\mathrm{Tr}v(p,B(p))^p1`$ $``$ $`\mathrm{Tr}v(p,B(p))^p\mathrm{Tr}v(p,B(p))`$ (6.19)
$`\left[\mathrm{Tr}w(p,B(p))^p\mathrm{Tr}w(p,B(p))\right]`$
The mean value theorem for the function $`g(p)=x^p`$ implies that for some $`p_1,p_2[1,p]`$
$`{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}v(p,B(p))^p\mathrm{Tr}v(p,B(p))\right)`$ $`=`$ $`\mathrm{Tr}v(p,B(p))^{p_1}\mathrm{log}v(p,B(p))`$ (6.20a)
$`{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}w(p,B(p))^p\mathrm{Tr}w(p,B(p))\right)`$ $`=`$ $`\mathrm{Tr}w(p,B(p))^{p_2}\mathrm{log}w(p,B(p))`$ (6.20b)
The convergence in (6.13) and following (6.14) imply
$`\underset{p1}{lim}\mathrm{Tr}v(p,B(p))^{p_1}\mathrm{log}v(p,B(p))`$ $`=`$ $`S(X)`$ (6.21a)
$`\underset{p1}{lim}\mathrm{Tr}w(p,B(p))^{p_2}\mathrm{log}w(p,B(p))`$ $`=`$ $`S(X_1).`$ (6.21b)
Combining (6.20), (6.21)and (6.19) gives (6.17). QED
Remark: The proof above relies on the convergence of $`lim_{p1+}X^{\frac{1}{2}}(B^{\frac{1}{p}1}\mathrm{I}_n)X^{\frac{1}{2}}=X`$ and $`lim_{p1+}X^{\frac{1}{2}}X_1^{\frac{1}{2}}B^{\frac{1}{p}1}X_1^{\frac{1}{2}}=X_1`$, but tells us nothing at all about the behavior of $`B(p)`$ as $`p1+`$. By making a few changes at the end of this proof and exploiting Kleinโs inequality, we can also show that $`lim_{p1+}B(p)=X_1`$.
Kleinโs inequality states that
$`\mathrm{Tr}A\mathrm{log}A\mathrm{Tr}A\mathrm{log}B\mathrm{Tr}(AB)`$ (6.22)
with equality in the case $`\mathrm{Tr}A=\mathrm{Tr}B`$ if and only if $`A=B`$.
Now, replace (6.19) by
$`\mathrm{Tr}v(p,B(p))^p1`$ $`=`$ $`\mathrm{Tr}v(p,B(p))^p\mathrm{Tr}v(p,B(p))+(\mathrm{Tr}w(p,B(p))1).`$ (6.23)
Then use the mean value theorem for the function $`g_2(p)=y^{\frac{1}{p}}`$ to replace (6.20b) by
$`{\displaystyle \frac{1}{p1}}\left(\mathrm{Tr}w(p,B(p))1\right)`$ $`=`$ $`{\displaystyle \frac{1}{\stackrel{~}{p}^2}}\mathrm{Tr}X_1^{\frac{1}{2}}B(p)^{\frac{1}{\stackrel{~}{p}}1)}\mathrm{log}B(p)X_1^{\frac{1}{2}}.`$ (6.24)
We could use (6.22) with $`A=B(p)^{\frac{1}{2}(1\frac{1}{\stackrel{~}{p}})}X_1B(p)^{\frac{1}{2}(1\frac{1}{\stackrel{~}{p}})}`$ together with the fact that $`A`$ and $`w(\stackrel{~}{p},B(p))`$ have the same non-zero eigenvalues to bound the right side of (6.24) below by $`\frac{1}{\stackrel{~}{p}^2}S[w(\stackrel{~}{p},B(p)]+\mathrm{Tr}w(\stackrel{~}{p},B(p))1`$. However, because $`1<\stackrel{~}{p}<p`$ implies $`B^{1/\stackrel{~}{p}}>B^{1/p}`$, we cannot extend (6.12) and (6.14) to conclude that this converges to $`S(X_1)`$.
Instead, we first observe that the compactness of the set of density matrices $`๐`$ implies that we can find a sequence $`p_k1+`$ such that $`X_{12}_{(1,p)}=v(p_k,B(p_k))_{p_k}`$ and $`B_kB^{}๐`$. If $`B^{}`$ is not in $`\beta (X_1)`$, then the right side of the first line of (6.24) $`+\mathrm{}`$ giving a contradiction with (6.16). Hence $`B^{}\beta (X_1)`$. Therefore, (6.24) and (6.22) imply
$`\underset{p_k\mathrm{}}{lim}{\displaystyle \frac{1}{p_k1}}\left(\mathrm{Tr}w(p_k,B(p_k))1\right)`$ $`=`$ $`\mathrm{Tr}X_1\mathrm{log}B^{}S(X_1).`$ (6.25)
Inserting this in (6.23) yields
$`\underset{p_k\mathrm{}}{lim}{\displaystyle \frac{1}{p_k1}}\left(\mathrm{Tr}v(p_k,B(p_k))^p1\right)`$ $`=`$ $`S(X_{12})\mathrm{Tr}X_1\mathrm{log}B^{}`$ (6.26)
$``$ $`S(X_{12})+S(X_1).`$
Combining these results with (6.16), we conclude that equality holds in (6.26) and that
$`\mathrm{Tr}X_1\mathrm{log}B^{}=S(X_1)=\mathrm{Tr}X_1\mathrm{log}X_1.`$ (6.27)
We can now use the condition for equality in (6.22) to conclude that $`B^{}=X_1`$. Since this is true for the limit of any convergent sequence of minimizers $`B(p_k)`$ with $`p_k1`$, we have also proved the following which is of independent interest.
###### Corollary 18
For $`XM_mM_n`$ with $`X0`$ and $`\mathrm{Tr}X=1`$ and $`p(1,2]`$, let $`B(p)๐`$ minimize $`X_{(1,p)}`$, i.e., $`X^{\frac{1}{2}}(B^{\frac{1}{p}1}\mathrm{I}_n)X^{\frac{1}{2}}_p=X_{(1,p)}`$. Then $`\underset{p1+}{lim}B(p)=X_1\mathrm{Tr}_2X`$.
Acknowledgments: The work of M.J. was supported in part by National Science Foundation grant DMS-0301116. The work of C.K. was supported in part by the National Science Foundation under grant DMS-0400426. The work of M.B.R. was supported in part by the National Security Agency (NSA) and Advanced Research and Development Activity (ARDA) under Army Research Office (ARO) contract number DAAD19-02-1-0065; and by the National Science Foundation under Grant DMS-0314228.
This work had its genesis in a workshop in 2002 at the Pacific Institute for the Mathematical Sciences at which M.J. and M.B.R. participated. Part of this work was done while I.D. and M.B.R. were visiting the Isaac Newton Institute. The authors are grateful to these institutions for their hospitality and support. Finally, we thank Professor Andreas Winter for discussions about possible interpretations of $`S_{\mathrm{CB},\mathrm{min}}(\mathrm{\Phi })`$.
## Appendix A Purification
To make this paper self-contained and accessible to people in fields other than quantum information we summarize the results needed to prove Lemma 4.
Any density matrix in $`๐_d`$ can be written in terms of its spectral decomposition (restricted to $`[\mathrm{ker}(\gamma )]^{}`$) as $`\gamma =_{k=1}^m\lambda _k|\varphi _k\varphi _k|`$ where each eigenvalue $`\lambda _k>0`$ and counted in terms of its multiplicity so that the eigenvectors $`\{|\varphi _k\}`$ are orthonormal. If we then let $`\{|\chi _k\}`$ be any orthonormal basis of $`๐^m`$ and define $`|\mathrm{\Psi }๐^d๐^m`$ as
$`|\mathrm{\Psi }={\displaystyle \underset{k=1}{\overset{m}{}}}\sqrt{\lambda _k}|\varphi _k|\chi _k.`$ (A.1)
then $`\gamma =\mathrm{Tr}_2|\mathrm{\Psi }\mathrm{\Psi }|`$ and (A.1) is called a purification of $`\gamma `$.
Conversely, given a normalized vector $`|\mathrm{\Psi }๐^n๐^m`$, it is a straightforward consequence of the singular value decomposition that $`|\mathrm{\Psi }`$ can be written in the form
$`|\mathrm{\Psi }={\displaystyle \underset{k}{}}\mu _k|\varphi _k|\chi _k`$ (A.2)
with $`\{|\varphi _k\}`$ and $`\{|\chi _k\}`$ orthonormal sets in $`๐^n`$ and $`๐^m`$ respectively. (This is often called the โSchmidt decompositionโ in quantum information theory. For details and some history see Appendix A of .) It follows from (A.2) that the reduced density matrices $`\gamma _1=\mathrm{Tr}_2|\mathrm{\Psi }\mathrm{\Psi }|`$ and $`\gamma _2=\mathrm{Tr}_1|\mathrm{\Psi }\mathrm{\Psi }|`$ have the same non-zero eigenvalues. Although our interest here is for $`=๐^m`$, these results extend to infinite dimensions and yield the following
###### Corollary 19
When $`|\mathrm{\Psi }_{AB}`$ is a bipartite pure state in $`_A_B`$, then its reduced density matrices $`\gamma _A=\mathrm{Tr}_B|\mathrm{\Psi }\mathrm{\Psi }|`$ and $`\gamma _B=\mathrm{Tr}_A|\mathrm{\Psi }\mathrm{\Psi }|`$ have the same entropy, i.e., $`S(\gamma _A)=S(\gamma _B)`$.
|
warning/0506/math0506141.html
|
ar5iv
|
text
|
# Non-persistently recurrent points, qc-surgery and instability of rational maps with totally disconnected Julia sets
## Introduction and Statements
Fatouโs problem of the density of hyperbolic maps in the space of rational maps is one of the principal problems in the field of holomorphic dynamics. Due to Manรฉ, Sad and Sullivan \[MSS\] we can reformulate this problem in the following way:
If the Julia set $`J(R)`$ contains a critical point, then the rational map $`R`$ is a structurally unstable map.
For convenience we give the definition of the structural stability of a rational map. For other basic notations and definitions refer to the book of Milnor \[M\].
###### Definition 1
Let $`Rat_d`$ be the space of all rational maps of degree $`d`$ with the topology of coefficient convergence. A map $`RRat_d`$ is called structurally stable if there exists a neighborhood $`URat_d`$ of $`R`$ such that:
For any map $`R_1U`$ there exists a quasiconformal map $`f:\overline{}\overline{}`$ conjugating $`R`$ to $`R_1`$.
We give a condition โAssumption Gโ (see below) on the rational map with totally disconnected Julia set and with a critical point on $`J(R)`$ to be unstable. In a pioneer paper \[BH\], Branner and Hubbard prove that the Lebesgue measure of the Julia set is zero if there exists only one critical point on $`J(R).`$ Our result (see theorem A below) restricted on the Branner-Hubbard case is weaker, but it can be applied for maps with two or more critical points on $`J(R).`$ Let $`R`$ be a rational map with totally disconnected Julia set. Let us normalize $`R`$ so that the point $`z=\mathrm{}`$ becomes the attractive fixed point. Let $`Pc(R)`$ be postcritical set of the map $`R`$ and $`P(R)=Pc(R)J(R)`$ be postcritical set on the Julia set. Let $`S=\backslash \overline{_nR^n(Pc(R))},`$ then $`R:SS`$ is an unbranched autocovering.
###### Definition 2
We call a closed simple geodesic $`\gamma S`$ linked with $`P(R)`$ if the interior $`I(\gamma )`$ of $`\gamma `$ intersects $`P(R).`$
###### Assumptions "G"
Let $`R`$ be a rational map with totally disconnected Julia set. Assume there exists a simple closed geodesic $`\gamma S`$ such that:
The aim of this paper is to prove the following theorem.
###### Theorem A
Let $`R`$ be a rational map with totally disconnected Julia set satisfying Assumption โGโ. Then the map $`R`$ is not a structurally stable map (that is to say, is an unstable map.)
Apriori it is not clear when the Assumption G holds. We give a series of sufficient conditions on $`R`$ that imply Assumption โGโ. The next proposition translates Assumption โGโ into the language of โnon-persistently recurrent pointsโ on P(R).
###### Definition 3
A point $`xP(R)`$ is called persistently recurrent if any backward orbit $`U_0,U_1,\mathrm{}`$ of any neighborhood $`U_0`$ of $`x`$ along $`P(R)`$ hits a critical point infinitely many times.
###### Lemma (Sufficient condition)
Let $`R`$ be a rational map with totally disconnected Julia set and assume there exists a non-persistently recurrent point $`xP(R).`$ Then $`R`$ satisfies Assumption โGโ.
###### Demonstration Prof
Follows immediately from Definition 3.
Another sufficient condition is connected with the conical points of $`P(R).`$
###### Definition 4
let $`R`$ be a rational map, and denote by $`U(x_0,R^k,\delta )`$ the component of $`R^k(๐ป(R^k(x_0),\delta ))`$ that contains $`x_0,`$ where $`๐ป(R^k(x_0),\delta )`$ is the disk centered in $`R^k(x_0)`$ with radius $`\delta .`$ A point $`x_0`$ is conical if and only if there is a constant $`\delta >0,d,`$ and a sequence $`k_j\mathrm{}`$ such that
$$R^{k_j}:U(x_0,R^k,\delta )๐ป(R^k(x_0),\delta )$$
has degree no more than $`d.`$
Several other notions of conical point appear in the literature. One can see that the definition of conical point is somehow in the spirit of the notion of conical set of Lyubich and Minsky \[LM\]. Definition 4 above appears in \[P\], where Przytycki compares different notions of conical points. McMullen \[MM\] and independently Urbanski \[DMNU\] call a point conical if the mappings in Definition 4 can be chosen to be conformal.
###### Theorem 1
Let $`R`$ be a rational map with totally disconnected Julia set and assume there exists a conical point $`xP(R).`$ Then $`R`$ is an unstable map.
The following two results are immediate corollaries of the theorem 1.
###### Corollary 1
Let $`R`$ be a rational map with totally disconnected Julia set and assume that the postcritical set $`P(R)`$ contains a periodic point $`x.`$ Then $`R`$ is unstable map.
###### Demonstration Proof
By assumption, the periodic point $`xJ(R).`$ Hence $`x`$ is either parabolic or repelling. Now assume that $`R`$ is a structurally stable map, then $`x`$ should be repelling and hence conical. Applying Theorem 1 we are done.
###### Corollary 2
Let $`R`$ be a rational map with totally disconnected Julia set. Assume $`J(R)=P(R),`$ then R is an unstable map.
###### Demonstration Proof
In this case $`P(R)`$ contains all repelling periodic points and by Corollary 1 we are done.
## Proof of Theorem A
To prove Theorem A, we use a kind of quasiconformal surgery in the spirit of Shishikura \[Sh\].
Let $`\mathrm{\Delta }(r)`$ be a disk of the radius $`r`$ centered at $`z=0`$ and $`\mathrm{\Delta }=\mathrm{\Delta }(1).`$ Let $`A(p,q)=\{z:p<|z|<q`$ be a ring. Let $`A(p)=A(p,1),p<1`$ and $`a=\frac{1+3p}{4}A(p,\frac{1+p}{2})A(p)`$ be a point. Now we define a quasiconformal homeomorphism $`f_p:\mathrm{\Delta }\mathrm{\Delta }`$ as follows:
###### Remark 1
Note that the $`L_{\mathrm{}}`$ norm of $`\mu `$ depends only on the modulus of the ring $`A(p),`$ or in other words, if $`p_i\mathrm{\Delta }`$ converges to $`p_0\mathrm{\Delta },`$ then the $`L_{\mathrm{}}`$ norm of the dilatations $`\mu _i`$ are uniformly bounded away from $`1.`$
According to results of D. Sullivan (\[S\]) and C. McMullen, D. Sullivan (\[MS\]) the space of full orbits of the points on $`S`$ forms a Riemann surface $`S(R)`$ which is (conformally) the torus with a finite number of punctures: the punctures correspond to the full orbits of the critical points belonging to $`F(R).`$ Hence there exists a fundamental domain $`FF(R)`$ for the action $`R`$ on $`F(R).`$ We can choose the fundamental domain as follows:
Let $`O(F)`$ be the full orbit of the fundamental domain $`F.`$ Let $`\alpha S`$ be any geodesic, then $`\alpha `$ intersects a finite number, say $`n(\alpha ),`$ of elements of $`O(F),`$ say $`F_1(\alpha ),\mathrm{},F_{n(\alpha )}(\alpha ).`$
###### Remark 2
By the properties of the fundamental domain we can always assume that there exists $`i_0`$ so that the forward orbit $`O_+(F_{i_0}(\alpha ))=_{i1}R^i(F_{i_0}(\alpha ))`$ never intersects the interior of the geodesic $`\alpha .`$ For convenience we redefine $`F_1(\alpha )=F_{i_0}(\alpha ).`$
Let $`B(\alpha )S\{_{i=1}^{n(\alpha )}F_i(\alpha )`$ be an annulus containing $`\alpha `$ as a non-trivial curve with modulus $`m(\alpha )`$ of $`B(\alpha )`$ as large as possible. Note that $`B(\alpha )`$ is not unique. Now let $`\beta S`$ be an iterated preimage of $`\alpha `$ (that is there exists an integer $`k`$ such that $`R^k(\beta )=\alpha `$). If $`d(\beta )`$ is the degree of the covering $`R^k:\beta \alpha ,`$ then the hyperbolic length $`l(\beta )=d(\beta )l(\alpha ),`$ and $`m(\beta )\frac{m(\alpha )}{d(\beta )},`$ as well as $`n(\beta )d(\beta )n(\alpha ).`$
Let us start with any closed simple geodesic $`\gamma S`$ linked with $`P(R).`$ Now we associate a qc-homeomorphism $`f(\gamma ,p):\overline{}\overline{}`$ as follows:
Let $`I(\gamma )`$ be the interior of $`\gamma `$ and the point $`b`$ be the first hit in $`I(\gamma )`$ of the forward orbit of a critical point $`cI(\gamma .`$ Let $`h:I(\gamma )\mathrm{\Delta }`$ be the Riemann map with $`h(b)=0,h^{}(b)=1.`$ Now let $`p>0`$ be an number so that $`A(p)h(B(\gamma )).`$ Adjusting $`h`$ by a rotation we can construct a conformal map $`\varphi (\gamma ):I(\gamma )\mathrm{\Delta }`$ so that the point $`a=\frac{1+3p}{4}\varphi (F_1(\gamma )).`$ Then we set
$$f(\gamma ,p)=\{\begin{array}{cc}\varphi (\gamma )^1f_p\varphi (\gamma )\text{ on }I(\gamma ),\hfill & \\ id\text{ off of }I(\gamma )\hfill & \end{array}$$
Hence for any simple closed geodesic $`\gamma S`$ and a suitable number $`0<p<1`$ we can define a quasi-regular map $`P(\gamma ,p)=f(\gamma ,p)R:\overline{}\overline{}.`$
###### Lemma 1
Let $`\gamma S`$ be closed simple geodesic linked with $`P(R)`$ and $`1>p>0`$ be a suitable number. Then
###### Demonstration Proof
Follows immediately from the definition of $`f(\gamma ,p).`$
By the Riemann Mapping Theorem there exists a quasi-conformal homeomorphism $`f_\sigma `$ fixing the points $`0,1`$ and $`\mathrm{}`$ so that $`R(\gamma ,p)=f_\sigma Rf_\sigma ^1`$ is a rational map.
Let $`s(R)`$ be the number of critical points whose forward orbits converge to $`\mathrm{}.`$
###### Corollary 1
Let $`\gamma `$ and $`p`$ be as in Lemma 1 above. Assume that $`\gamma `$ has a sufficiently small spherical diameter. Then $`s(R(\gamma ,p)s(R)+1.`$
###### Demonstration Proof
Let the spherical diameter of $`\gamma `$ be so small that the interior $`I(\gamma )`$ does not contain any critical point of the Fatou set $`F(R).`$ Hence if the critical point $`cF(R),`$ then $`P^n(\gamma ,p)(c)=R^n(c)\mathrm{}.`$ Now let $`cJ(R)`$ be the critical point coming from the definition of $`f(\gamma ,p).`$ Then again by the construction, $`P^n(\gamma ,p)(c)\mathrm{}.`$
Now we are ready to prove Theorem A. Let $`\gamma _i_nR^n(\gamma )`$ be coming geodesics from the assumption. Let us redefine:
Now our aim is to show that there exists a subsequence $`\{\nu _{i_j}\}`$ with norms uniformly bounded away from 1. By Remark 1 and Lemma 1 it is enough to show that we can choose a subsequence $`p_{i_j}`$ uniformly bounded away from 1.
Let $`k_i`$ be integers so that $`R^{k_i}(\gamma _i)=\gamma .`$ Let $`B_i`$ be a component of $`R^{k_i}(B(\gamma ))`$ containing the geodesic $`\gamma _i.`$ Then by our assumptions there exists a constant $`C`$ so that the moduli $`m(B_i)=\frac{m(B(\gamma )}{d(\gamma _i)}C>0`$ are uniformly bounded. Let $`A_i=\varphi _i(B_i).`$
###### Lemma 2
There exists a subsequence $`\{i_j\}`$ and a number $`p<1`$ so that $`A(p)A_{i_j}`$ for any $`j.`$
###### Demonstration Proof
The argument is simple. Let $`A_{i_0}`$ be any annulus of minimal modulus. Then there exist conformal injections $`h_i:A_{i_0}A_i`$ such that $`h_i(\mathrm{\Delta })=\mathrm{\Delta }`$ and $`h_i(1)=1.`$ The family $`\{h_i\}`$ is normal so let $`\{h_{i_j}\}`$ be a convergent subsequence. Then the limit map $`h_{\mathrm{}}const.`$ Now let $`q<1`$ be so that $`A(q)A_{i_0},`$ then by the reflection principle $`\{h_{i_j}\}`$ converges to $`h_{\mathrm{}}`$ uniformly on $`A(\frac{q+1}{2}).`$ Hence $`h_{\mathrm{}}(A(\frac{q+1}{2}))h_{i_j}(A_{i_0})`$ for all large enough $`j.`$ Let $`p<1`$ be an integer so that $`A(p)h_{\mathrm{}}(A(\frac{q+1}{2}));`$ then by the discussion above
$$A(p)h_{i_j}(A_{i_0})A_{i_j}$$
for all large $`j.`$ The lemma is thus proved.
By Remark 1, Lemma 1 and Lemma 2 we have that the family of quasiconformal homeomorphisms $`\{f_{i_j}\}`$ is normal, and after passing to a subsequence we can assume that $`\{f_{i_j}\}`$ converges to a quasi-conformal homeomorphism $`f_{\mathrm{}}.`$ The Julia set $`J(R)`$ is a Cantor set hence the spherical diameter $`diam(\gamma _i)0.`$ Then the homeomorphisms $`f(\gamma _{i_j},p)`$ converge to the identity uniformly on $`\overline{},`$ and hence $`P_{i_j}R.`$
Again after passing to a subsequence we can assume that $`lim_j\mathrm{}R_{i_j}=R_{\mathrm{}},`$ where $`R_{\mathrm{}}`$ is a rational map of degree smaller or equal to the degree of $`R.`$ Then we can pass to the limit in the following equality:
$$f_{i_j}P_{i_j}f_{i_j}^1=R_{i_j}f_{\mathrm{}}Rf_{\mathrm{}}^1=R_{\mathrm{}}.$$
Now to obtain a contradiction assume that $`R`$ is a structurally stable map. Then $`R_{\mathrm{}}`$ is structurally stable ( a being a qc-deformation of $`R`$) and $`s(R)=s(R_{\mathrm{}}).`$ By construction $`R_{\mathrm{}}=lim_j\mathrm{}R_{i_j}`$ and by Corollary 1 $`s(R_{i_j})s(R)+1=s(R_{\mathrm{}})+1`$ which contradicts the structural stability of $`R_{\mathrm{}}.`$
## Proof of Theorem 1
Here we show that the existence of a conical point $`xP(R)`$ implies Assumption โGโ .
###### Lemma 3
Assume $`R`$ satisfies the assumptions of the Theorem 1. Then $`R`$ satisfies Assumption โGโ.
###### Demonstration Proof
Let $`x_0P(R)`$ be a conical point. Let integer $`d`$ and a sequence $`\{k_j\}`$ be as in the definition of the conical point. Let $`U(x_0,R^{k_j},\delta )`$ be the component of $`R^k(๐ป(R^{k_j}(x_0),\delta ))`$ that contains $`x_0,`$ where $`๐ป(R^k(x_0),\delta )`$ is the disk centered in $`R^k(x_0)`$ with the radius $`\delta .`$ After passing to a subsequence we can assume that the sequence $`R^{k_j}(x_0)`$ converges to a point $`yP(R).`$ The disks $`๐ป(R^{k_j}(x_0),\delta )`$ converges as well as to the disk $`๐ป(y,\delta ).`$ Now let $`\gamma S๐ป(y,\frac{\delta }{2})`$ with $`yI(\gamma ).`$ Then for large $`j`$ we have:
The lemma is proved.
An application of Theorem A completes the proof of Theorem 1.
R E F E R E N C E S
|
warning/0506/quant-ph0506058.html
|
ar5iv
|
text
|
# Algebraic invariants of five qubits
## I Introduction
Quantifying entanglement in multipartite systems is a fundamental issue in Quantum Information Theory. However, for systems with more than two parts, very little is know in this respect. A few useful entanglement measures for pure states of $`3`$ or $`4`$ qubits have been investigated Bre ; MW ; Ema , but one is still far from a complete understanding. Furthermore, for system of up to $`4`$ qubits, a complete classification of entanglement patterns and of corresponding invariants under local filtering operations (also known as SLOCC, Stochastic Local Operations assisted by Classical Communication) is know Ver ; LT3 . Klyachko Kly ; KS proposed to associate entanglement (of pure states) in a $`k`$-partite system (or perhaps , one should say โpure $`k`$-partiteโ entanglement) with the mathematical notion of semi-stability, borrowed from geometric invariant theory, which means that at least one SLOCC invariant is non zero. For such states, the absolute values of these invariants provide some kind of entanglement measure. However, even for system of $`k`$ qubits, the complexity of these invariants grows very rapidly with the number of parts. For $`k=2`$, they are given by simple linear algebra Per ; HHH . The case $`k=3`$ is already nontrivial but appears in the physics literature in DVC and boils down to a mathematical result which was known by 1880 LeP . The case $`k=4`$ is quite recent LT3 , and to the best of our knowledge, nothing was known for $`5`$-qubit systems<sup>1</sup><sup>1</sup>1Just after we posted the first version of this Note, A. Osterloh and J. Siewert informed us of their independent work Ost1 on the five qubits problem (see Conclusion for a short discussion). .
Our main result is a closed expression of the Hilbert series of the algebra of SLOCC invariants of pure $`5`$-qubit states. This result, which determines the number of linearly independent homogeneous invariants in any degree, was obtained through intensive symbolic computations relying on a very recent algorithm for multivariate residue calculations. We point out a few properties which can be read off from the series, and determine the simplest invariants, which are of degree $`4`$ and $`6`$ in the component of the states.
## II Hilbert series
Denote by $`V=^2`$ the local Hilbert space of a two state particle. The state space of a five particule system is $`=V^5`$, which will be regarded as the natural representation of the group of invertible local filtering operations, also known as reversible stochastic local quantum operations assisted by classical communication
$$G=G_{\mathrm{SLOCC}}=\mathrm{SL}(2,)^{\times 5},$$
that is, the group of $`5`$-tuples of complex unimodular $`2\times 2`$ matrices. We will denote by
$$|\mathrm{\Psi }=\underset{i_1,i_2,i_3,i_4,i_5=0}{\overset{1}{}}A_{i_1i_2i_3i_4i_5}|i_1|i_2|i_3|i_4|i_5$$
a state of the system. An element $`๐ =({}_{}{}^{k}g_{i}^{j})`$ of $`G`$ maps $`|\mathrm{\Psi }`$ to the state
$$|\mathrm{\Psi }^{}=๐ |\mathrm{\Psi }$$
whose components are given by
$$A_{i_1i_2i_3i_4i_5}^{}=\underset{๐ฃ}{}{}_{}{}^{1}g_{i_1}^{j_1}{}_{}{}^{2}g_{i_2}^{j_2}{}_{}{}^{3}g_{i_3}^{j_3}{}_{}{}^{4}g_{i_4}^{j_4}{}_{}{}^{5}g_{i_5}^{j_5}A_{j_1j_2j_3j_4j_5}$$
(1)
We are interested in the dimension of the space $`_d`$ of all $`G`$-invariant homogeneous polynomials of degree $`d=2m`$ ($`_d=0`$ for odd $`d`$) in the 32 variables $`A_{i_1i_2i_3i_4i_5}`$.
It is known that it is equal to the multiplicity of the trivial character of the symmetric group $`๐_{2m}`$ in the fifth power of its irreducible character labeled by the partition $`[m,m]`$
$$dim_d=\chi ^{2m}|(\chi ^{mm})^5.$$
(2)
The generating function of these numbers
$$h(t)=\underset{d0}{}dim_dt^d$$
(3)
is called the Hilbert series of the algebra $`=_d_d`$. Standard manipulations with symmetric functions allow to express it as a multidimensional residue:
$$h(t)=\frac{du_1}{2\pi \mathrm{i}u_1}\mathrm{}\frac{du_5}{2\pi \mathrm{i}u_5}\frac{A(๐ฎ)}{B(๐ฎ;t)}$$
(4)
where the contours are small circles around the origin,
$$A(๐ฎ)=\underset{i=1}{\overset{5}{}}\left(1+1/u_i^2\right)$$
(5)
and
$$B(๐ฎ;t)=\underset{a_i=\pm 1}{}(1tu_1^{a_1}u_2^{a_2}u_3^{a_3}u_4^{a_4}u_5^{a_5})$$
(6)
Such multidimensional residues are notoriously difficult to evaluate. After trying various approaches, we eventually succeded by means of a recent algorithm due to Guoce Xin Xin , in a Maple implementation. The result can be cast in the form
$$h(t)=\frac{P(t)}{Q(t)}$$
(7)
where $`P(t)`$ is an even polynomial of degree $`104`$ with non negative integer coefficients $`a_n`$
$$P(t)=\underset{k=0}{\overset{52}{}}a_{2k}t^{2k}$$
given in table 1,
and
$$Q(t)=(1t^4)^5(1t^6)(1t^8)^5(1t^{10})(1t^{12})^5.$$
On this expression, it is clear that a complete description of the algebra of $`G`$-invariant polynomials by generators and relations is out of reach of any computer system.
Nevertheless, inspection of the Hilbert series suggests the following kind of structure for this algebra. We know, since $`dimdimG=2^53\times 5=17`$, that there must exist a set of 17 algebraically independent invariants. The denominator of the series, which is precisely a product of 17 factors, makes it plausible that these invariants can be choosen as five polynomials of degree 4 (to be denoted by $`D_xD_y,D_z,D_t,D_u`$), one polynomial of degree $`6`$ ($`F`$), five polynomials of degree $`8`$ ($`H_1,H_2\mathrm{},H_5`$), one polynomial of degree $`10`$ ($`J`$) and five polynomials of degree $`12`$ ($`L_1,\mathrm{},L_5)`$. These 17 polynomials are called the primary invariants.
The numerator should then describe the secondary invariants, that is, a set of 3014400 homogeneous polynomials ($`1`$ of degree $`0`$ , 16 of degree $`8`$, 9 of degree $`10`$, $`82`$ of degree $`12`$ etc) such that any invariant polynomial can be uniquely expressed as a linear combination of secondary invariants, the coefficients being themselves polynomials in the primary invariants.
This picture, which is the simplest kind of description to be expected, is far too complex for physical applications. The best that can be done is to use the Hilbert series as a guide for finding explicitly a small set of reasonably simple invariants, in particular, the primary invariants of lowest degrees. We have computed the first primary invariants, those of degree 4 and 6, using methods from Classical Invariant Theory (see below).
## III The simplest invariants
### III.1 Transvectants and Cayleyโs Omega process
In order to apply the formalism of Classical Invariant Theory, a state $`|\mathrm{\Psi }`$ will be interpreted as a quintilinear form on $`^2`$ (called the ground form)
$$f:=\underset{i_1,i_2,i_3,i_4i_5=0}{\overset{1}{}}A_{i_1i_2i_3i_4i_5}x_{i_1}y_{i_2}z_{i_3}t_{i_4}u_{i_5}$$
A covariant of $`f`$ is a $`G`$-invariant polynomial in the coefficients $`A_{i_1i_2i_3i_4i_5}`$ and the variables $`x_i,y_i,z_i,t_i`$ and $`u_i`$. A complete set of covariants can be in principle computed from the ground form by means of the so-called Omega process (see Pea ; Olv for notations). Cayleyโs Omega process consists in applying iteratively differential operators called transvections and defined by
$$\begin{array}{ccc}\hfill (P,Q)^{ฯต_1\mathrm{}ฯต_5}& =& \mathrm{tr}\mathrm{\Omega }_x^{ฯต_1}\mathrm{}\mathrm{\Omega }_u^{ฯต_5}P(x^{},\mathrm{},u^{})Q(x^{\prime \prime },\mathrm{},u^{\prime \prime })\hfill \end{array}$$
where
$$\mathrm{\Omega }_x=det\left|\begin{array}{cc}\frac{}{x_{}^{}{}_{1}{}^{}}\hfill & \frac{}{x_{}^{}{}_{2}{}^{}}\hfill \\ \frac{}{x_{}^{\prime \prime }{}_{1}{}^{}}\hfill & \frac{}{x_{}^{\prime \prime }{}_{2}{}^{}}\hfill \end{array}\right|$$
and $`\mathrm{tr}:x^{},x^{\prime \prime }x`$.
### III.2 Degree 4
Regarding $`x`$ as a parameter, write $`f`$ as a quadrilinear binary form in the variables $`y_i,z_i,t_i`$ and $`u_i`$
$$f=A_{i_1i_2i_3i_4}^xy_{i_1}z_{i_2}t_{i_3}u_{i_4}$$
It is known that such a quadrilinear form admits an invariant of degree $`2`$ (called Cayleyโs hyperdeterminant Cay ; LT1 ; LT2 ) which is a quadratic binary form $`b_x`$ in the variables $`x=(x_1,x_2)`$. Hence, taking the discriminant of $`b_x`$ one obtains an invariant $`D_x`$ of degree $`4`$. We repeat this operation for the other binary variables and we obtain four other invariants $`D_y`$, $`D_z`$, $`D_t`$ and $`D_u`$. Evaluating the appropriate Jacobians with a computer algebra system gives the algebraic independance of the five invariants.
### III.3 Degree 6
We obtain the primary invariant of degree $`6`$ by a succession of transvections. First, we compute a triquadratic covariant of degree $`2`$
$$B_{22020}=(f,f)^{00101}.$$
This covariant allows to construct a cubico-quadrilinear covariant of degree $`3`$
$$C_{31111}=(B_{22020},f)^{01010}$$
which gives a triquadratic polynomial of degree $`4`$
$$D_{22200}=(C_{31111},f)^{10011}.$$
Hence, one obtains a quintilinear covariant of degree $`5`$
$$E_{11111}=(D_{22200},f)^{11100}.$$
Finally, we find the invariant of degree $`6`$
$$F=(E_{11111},f)^{11111}.$$
By computing the Jacobian, one finds that $`F`$ is algebraically independent of $`D_x,\mathrm{},D_u`$.
## IV Conclusion
From the Hilbert series, it appears that the algebra of polynomial invariants of a five qubit system has a very high complexity. Furthermore, as is already the case with smaller systems Ver ; LT2 ; LT3 , the knowledge of the invariants is not sufficient to classify entanglement patterns. In the case of four qubits or three qutrits, this classification can be achived due to hidden symmetries which have their roots in very subtle aspects of the theory of semi-simple Lie algebras (Vinbergโs theory VP ). However, such symmetries are absent in the case of 5 qubits. Then, the only known general approach for classifying orbits (entanglement patterns) requires the computation of the algebra of covariants, which is already almost intractable in the case of four qubits. It has 170 generators, which have been found LT3 , but the description of their algebraic relations (syzygies) is definitely out of reach. However, a closer look at the $`4`$-qubit system, reveals that the classification of Verstraete et al Ver ; Ver2 . can be reproduced by means of only a small set of covariants.
We hope that our results will allow the identification and the calculation of such a small set of invariants and covariants, sufficient to separate the physically relevant entanglement patterns, which are probably not so numerous. To illustrate this principle, let us consider a result of Osterloh and Siewert Ost1 . Having introduced a notion of filter which can be used to separate SLOCC orbits in the same way as covariants, these authors show that the four states
$`|\mathrm{\Phi }_1={\displaystyle \frac{1}{\sqrt{2}}}\left(|11111+|00000\right)`$
$`|\mathrm{\Phi }_2={\displaystyle \frac{1}{2}}\left(|11111+|11100+|00010+|00001\right)`$
$`|\mathrm{\Phi }_3={\displaystyle \frac{1}{\sqrt{6}}}\left(\sqrt{2}\right|11111+|11000+|00100+|00010`$
$`+|00001)`$
$`|\mathrm{\Phi }_4={\displaystyle \frac{1}{2\sqrt{2}}}\left(\sqrt{3}\right|11111+|10000+|01000+|00100`$
$`+|00010+|00001)`$
are in different orbits. As can be seen on Table 2, the orbits of these states are also distinguished by our covariants.
Finally, the investigation of entanglement measures requires an understanding of invariants under local unitary transformations (LUT) Gra . In a forthcoming paper, we will explain how to obtain LUT-invariants from SLOCC-covariants.
|
warning/0506/cond-mat0506670.html
|
ar5iv
|
text
|
# Complete spin polarization of electrons in semiconductor layers and quantum dots
## Abstract
We demonstrate that non-equilibrium electrons in thin nonmagnetic semiconductor layers or quantum dots can be fully spin polarized by means of simultaneous electrical spin injection and extraction. The complete spin polarization is achieved if the thin layers or quantum dots are placed between two ferromagnetic metal contacts with moderate spin injection coefficients and antiparallel magnetizations. The sign of the spin polarization is determined by the direction of the current. Aplications of this effect in spintronics and quantum information processing are discussed.
Spintronics is a new field of condensed matter physics based on manipulation of electron spins is solids Zut ; Aw . Injection of spin-polarized electrons into nonmagnetic semiconductors (NS) is of particular interest because of the relatively large spin-coherence lifetime, $`\tau _s`$, in NS Zut ; Aw and promising applications for ultrafast, low-power spin devices Zut ; Aw ; Datta ; Hot ; OBO and for spin-based quantum information processing (QIP) Aw ; QIP . Both the characteristics of the spintronic devices and fidelities of various spin-based QIP schemes dramatically improve when spin polarization $`P_n`$ of electrons in NS is high, i.e. when $`P_n100\%`$. Thus the main challenge is to achieve high values of $`P_n100\%`$ in NS both at room and low temperatures. A simple intuition, supported by all previous theoretical works Aron ; Mark ; Rash ; Flat ; BO ; OB , suggests that $`P_n`$ under electrical spin injection cannot exceed either the polarization of a spin source or the spin injection coefficient of a ferromagnet-semiconductor contact. Since conventional ferromagnetic metals (FMs) rootinely used in semiconductor technology have moderate spin polarizations $`P_n`$ 40%, the main strategy to improve spin injection has been identified as development of magnetic semiconductors and half-metals with high bulk spin polarizations Zut ; Aw . Nonetheless, the greatest value of electron spin polarization in NS to date, $`P_n`$ 32%, has been observed in spin injection experiments with conventional FM (Fe) contacts Jonk .
One of the obstacles for the spin injection from FM into NS is a high and wide potential Schottky barrier that normally forms in NS near FM-NS interfaces sze . The spin injection from FM into NS corresponds to a reverse current in the Schottky junction, which is usually negligible. Therefore, a thin heavily doped $`n^+`$-S layer between FM and NS has been used to increase the tunneling transparency of the barrier sze and the spin injection current OB ; BO ; Jonk . Since the spin injection is a tunneling of spin polarized electrons from FM into NS, which is a symmetric process, the spin selection should also occur in forward-biased FM-S junctions when electrons are emitted from NS into FM BO . In this case the spins are extracted from NS into FM and the tunneling FM-$`n^+`$-S contact works as a spin filter.
In this letter we consider a donor doped semiconductor ($`n`$-NS) layer placed between two spin selective contacts, for example, tunneling FM-$`n^+`$-NS junctions, Fig.1. We show that the electron spin polarization in this $`n`$-NS layer can achieve 100% even when the spin injection coefficients of the contacts, $`\gamma _L`$ and $`\gamma _R`$, are quite moderate. The only requirement is that $`\gamma _{L(R)}`$ should weakly depend on the current, $`J`$. Indeed, the currents of up-electrons with spin $`\sigma =`$ through the contacts can be written as
$$J_{}(0)=J(1+\gamma _L)/2,\text{ }J_{}(w)=J(1+\gamma _R)/2,$$
(1)
where $`J_{}(0)`$ and $`J_{}(w)`$ are the currents in the $`n`$-NS layer at the boundaries with the right and left contacts $`x=0`$ and $`x=w`$ (Fig.1). The continuity equation reads Aron ; Flat ; OB ; BO
$$dJ_{}/dx=q\delta n_{}/\tau _s,$$
(2)
where $`\delta n_{}=n_{}n_S/2`$, $`n_S`$ is the total electron density and $`n_{}`$ is the density of up-electrons. Obviously $`\delta n_{}(x)const`$ in the $`n`$-NS layer when its thickness $`wL_s`$, where $`L_s`$ is the spin diffusion length. Integrating Eq. (2) over $`x`$ from $`0`$ to $`w`$, we obtain $`J_{}(w)J_{}(0)=qw\delta n_{}/\tau _s`$, and using (1) and $`\delta n_{}(x)=\delta n_{}(x)`$ we find the spin polarization in thin $`n`$-NS layer:
$$P_n(x)=(\delta n_{}\delta n_{})/n_S(\gamma _R\gamma _L)J/J_w\overline{P}_n,$$
(3)
where $`J_w=qn_Sw/\tau _s`$. One can see from (3) that: $`P_n=0`$ when $`\gamma _R=\gamma _L`$, since in this case the currents of up- and down electrons through the right and left contacts are the same: $`J_{}(0)=J_{}(w)`$. On the contrary when $`\gamma _R\gamma _L`$ the value $`\left|P_n\right|1`$ when $`JJ_t`$ $`=J_w/(\gamma _R\gamma _L)`$. For example, if $`\gamma _L<0`$ and $`\gamma _R>0`$, i.e. magnetizations M<sub>1</sub> and M<sub>2</sub> in FM have opposite directions, Fig.1(a), $`P_n=(\left|\gamma _R\right|+\left|\gamma _L\right|)J/J_w`$, i.e. $`P_n1`$ as $`JJ_t=J_w/(\left|\gamma _R\right|+\left|\gamma _L\right|)`$. Thus, due to the difference of the currents of spin polarized electrons through the contacts, the spin density $`P_n`$ in the $`n`$-NS layer increases with the current and can reach $`100\%`$ even at small spin injection coefficients. One can see from (3) that the inversion of the current results in the opposite sign of $`P_n`$. These findings are valid for both degenerate and nondegenerate semiconductors.
In nondegenerate NS the diffusion constants and mobilities of up- and down-electrons are the same, $`D_{}=D_{}=D`$ and $`\mu _{}=`$ $`\mu _{}=`$ $`\mu `$, and the distribution of $`\delta n_{}(x)`$ for any $`w`$ reads Flat ; OB ; BO :
$$\delta n_{}(x)=(n_S/2)(c_1e^{x/L_1}+c_2e^{(wx)/L_2}),$$
(4)
where $`L_{1,2}=L_s\left(\sqrt{1+(J/2J_S)^2}J/2J_S\right)`$, $`L_s=\sqrt{D\tau _s}`$, $`J_S=qn_SL_s/\tau _s`$, and $`q`$ is the elementary charge. Using the boundary conditions (1) we obtain:
$`P_n(x)={\displaystyle \frac{J}{J_S}}[{\displaystyle \frac{L_s(\gamma _R\gamma _Le^{w/L_2})}{L_1\left(e^{w/L_2}e^{w/L_1}\right)}}e^{x/L_1}+`$
$`{\displaystyle \frac{L_s(\gamma _R\gamma _Le^{w/L_1})}{L_2\left(e^{w/L_2}e^{w/L_1}\right)}}e^{x/L_2}]`$ (5)
It follows from (5) that $`P_n(0)`$ or $`P_n(w)`$ reaches 1 at $`J=J_t(\gamma _L,\gamma _R,w)`$. The spatial dependence of $`P_n(x)`$ is very weak for $`wL_s`$ and $`P_n(x)\overline{P}_n`$ as in Eq. (3).
One can see from (3) and (5) that $`\gamma _R`$ and $`\gamma _L`$ determine a particular value of the threshold current, $`J_t`$, but it does not alter the main result: $`\left|P_n\right|1`$ as $`JJ_t`$. The only requirement is a relatively weak dependence of $`\gamma _R`$ and $`\gamma _L`$ on $`J`$. This condition can be fulfilled, for example, when thin, heavily-doped $`n^+`$-S layers are formed between $`n`$-NS and FM regions, Fig. 1. The parameters of $`n^+`$-S layers have to satisfy certain conditions OSP . In particular, the electron gas must be degenerate in a certain part of the $`n^+`$-S layer and the transition between the $`n^+`$-S and $`n`$-NS layers must be step-like. This situation is realized when the $`n^+`$-S layers have energy bandgaps narrower than that of the $`n`$-S region, Fig. 1(b), or when an additional $`\delta `$-doped acceptor layer is formed between the $`n^+`$-S and $`n`$-NS regions. Due to a high density of electrons in the $`n^+`$-S layer, $`\gamma `$ of such FM-$`n^+`$-S contacts weakly depends on $`J`$ up to the currents significantly exceeding $`J_t`$ OSP . Weak dependence of $`\gamma `$ on $`J`$ is also realized in FM-NS junctions with highly degenerate accumulation layers formed in NS near the FM-NS interface. This situation has been studied extensively in Refs. Aron ; Mark ; Rash ; Flat and has been realized expeimentally in Fe-InAs junctions Ohno .
We note that $`\gamma `$ Remk can strongly depend on $`J`$ BO ; OB in FM-NS junctions when the $`n`$-NS region and thin ($`\delta `$doped) $`n^+`$-layer are both nondegenerate semiconductors. In these FM-NS-FM junctions $`P_n`$ attains the value $`P_n=(\gamma _R\gamma _L)/(1\gamma _R\gamma _L)`$ for $`J>J_S`$ and $`wL_{1,2}`$ OBO . One can see that $`P_n=0.6`$ at $`\gamma _R=\gamma _L=0.33`$, i.e. even in this case $`P_n`$ can exceed the spin injection coefficients of the FM-S junctions.
In degenerate semiconductors the diffusion constants of up- and down-electrons are different and depend on electron densities. As a result, $`\delta n_{}(x)`$ is described by a diffusion-drift equation with a bi-spin diffusion constant $`D(P_n)`$ which goes to zero when $`\left|P_n\right|1`$ OSP . Therefore, in degenerate NS, the spatial variation of $`P_n(x)`$ is sharper and its current dependence is stronger than those in nondegenerate NS \[cf. yellow and red curves in Fig. 1(c)\]. For example, at $`w=0.1L_s`$ and currents $`J=J_t`$ the polarization changes in the ranges $`1>P_n(x)>0.984`$ or $`1>P_n(x)>0.954`$ for nondegenerate NS and $`1>P_n(x)>0.91`$ or $`1>P_n(x)>0.85`$ for degenerate NS when $`\gamma _R=\gamma _L=0.3`$ or $`\gamma _R=\gamma _L=0.1`$, respectively.
At high currents when $`J>J_t`$ \[see e.g. (3) and 5\] the value $`\left|P_n\right|=2\left|\delta n_{}\right|/n_S=\left|2n_{}n_S\right|/n_S`$ is higher than 1, i.e. spin density either near $`x=0`$ or near $`x=w`$ exceeds $`n_S`$. This means that the used condition of neutrality $`n_{}+`$ $`n_{}=n_S`$, i.e. $`\delta n_{}(x)=\delta n_{}(x)`$, is violated and a negative space charge accumulates near one of the boundaries. This charge will increase the voltage drop $`V_S`$ across $`n`$-NS region. This conclusion follows from numerical analysis of our system of equations which includes: Eq. (2), $`J=J_{}(x)+J_{}(x)=\mathrm{const}`$, $`J_\sigma =q\mu n_\sigma E+qD_\sigma dn_\sigma /dx`$, and the Poissonโs equation, $`\epsilon \epsilon _0dE/dx=q(n_sn_{}`$ $`n_{})`$.
The FM-$`n^+`$-$`n`$-$`n^+`$-FM heterostructures under consideration are supersensitive spin valves. Indeed, as we noticed above \[see (3) and (5)\], $`P_n0`$ and $`\delta n_{}(x)0`$ when $`w<L_s`$ and $`\gamma _R=\gamma _L`$, i.e. when the magnetizations M<sub>1</sub> and M<sub>2</sub> in the FM layers have the same direction. In this case, $`V_S`$ $`=V_{ohm}=Jw/q\mu n_S`$ at any current. When $`\gamma _R\gamma _L`$ the space charge arises in the $`n`$-S region at $`J>J_t`$ and $`V_{NS}`$ exceeds $`V_{ohm}`$ significantly. Thus, inversion of direction of M<sub>1</sub>or M<sub>2</sub> and also precession of electron spin in the $`n`$-S region have to change the voltage $`V_S`$ to a much greater extent than in the structures considered in Refs. OBO . The use of the FM-$`n^+`$-$`n`$-$`n^+`$-FM heterostructures in different spin-based devices Zut ; Aw ; Datta ; Hot ; OBO should result in dramatic improvement of their characteristics due to the high spin polarization of electrons in the $`n`$-NS layer.
A more complex FM-$`n^+`$-$`n`$-$`n^+`$-FM heterostructure is shown in Fig. 2(a). High spin polarization of electrons arises in a thin $`n`$-S layer 1 grown on highly resistive widebandgap semiconductor layer 3 when the current flows between ferromagnetic regions FM<sub>1</sub> and FM<sub>2</sub>. Such a heterostructure can be utilized modernize various spin-based nanodevices proposed in Refs. OBO such as ultraspeed magnetic sensors, transistors, square law detectors, and frequency multipliers. It can also be used for 100% spin polarization of electrons in quantum dots 5 localized in the layer 3 grown on a semiconductor substrate 4 having the same bandgap as the layer 1. This effect can be realized at very low temperatures even when the spin polarization inside the $`n`$-S layer 1 is less then 100%, e.g. $`P_n`$60%-80%. Indeed, let us consider quantum dots (QD) (regions (5) in Fig. 2(a) with band offset $`\mathrm{\Delta }E_c^/=E_{c5}E_{c3}<0`$) that are placed inside of a two-dimensional potential barrier (semiconductor layer (3) with band offset $`\mathrm{\Delta }E_c=E_{c3}E_{c1}>0`$). As we noted above, the spin polarization $`P_n`$ and density of up-electrons, $`n_{}`$, increase with current $`J`$ between FM<sub>1</sub>and FM<sub>2</sub> contacts when $`\gamma _L<0`$ and $`\gamma _R>0`$. This means that the quasi-Fermi level for the up-electrons $`F_{}`$ increases and $`F_{}`$ for down-electrons decreases with $`J`$. At a certain current $`F_{}`$ can exceed $`\mathrm{\Delta }E_c`$ and only the up-electrons will populate the layer 3 and will be captured by QDs. Thus, at very low temperatures the electron polarization in the QDs should be extremely close to 1, $`1P_n\mathrm{exp}[(E_{c3}F_{})/T]1`$. The spin polarization of electrons in QDs can be realized after their recombination with photogenerated holes or holes injected from p-region, shown in Fig.2a. This effect can be used for efficient polarization of nuclear spins in QDs. The sign of the spin polarization of electrons in the $`n`$-NS layer (1) and in QDs can be reversed by simple inversion of the current direction between FM<sub>1</sub>and FM<sub>2</sub> contacts, Fig. 2(a). Thus, the considered effects have a potential for changing the development strategy for various spin-based devices and QIP schemes.
This work is supported by NASA ITSR program (V. O and V. S.) and NSF (A. P.).
|
warning/0506/hep-ex0506014.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The concept of magnetic monopoles (MMs) goes back to the origin of magnetism. At the beginning of the 19th century there were discussions concerning the magnetic content of matter and the possible existence of isolated magnetic charges. In 1931 Dirac introduced the MM in order to explain the quantization of the electric charge . He established the relation between the elementary electric charge $`e`$ and a basic magnetic charge $`g`$: $`eg=n\mathrm{}c/2=ng_D`$, where $`n`$ is an integer, $`n=1,2,..`$; $`g_D=\mathrm{}c/2e=68.5e`$ is the unit Dirac charge. The existence of magnetic charges and of magnetic currents would symmetrize in form Maxwellโs equations, but the symmetry would not be perfect since $`eg`$ (but the couplings could be energy dependent and could merge in a common value at high energies) . There was no prediction for the MM mass; a rough estimate, obtained assuming that the classical monopole radius is equal to the classical electron radius yields $`m_M\frac{g^2m_e}{e^2}n4700m_en2.4GeV/c^2`$. From 1931 searches for โclassical Dirac monopolesโ were carried out at every new accelerator using simple setups, and recently also large collider detectors <sup>3-7</sup>.
Electric charge is naturally quantized in Grand Unified Theories (GUT) of the basic interactions; they imply the existence of GUT monopoles with calculable properties. The MMs appear in the Early Universe at the phase transition corresponding to the breaking of the unified group into subgroups, one of which is U(1) . The MM mass is related to the mass of the X, Y carriers of the unified interaction, $`m_Mm_X/G`$, where G is the dimensionless unified coupling constant at energies E $`m_X`$. If $`m_X10^{14}10^{15}`$ GeV and $`G0.025`$, $`m_M>10^{16}10^{17}`$ GeV. This is an enormous mass: MMs cannot be produced at any manโmade accelerator, existing or conceivable. They may have been produced only in the first instants of our Universe.
Larger MM masses are expected if gravity is brought into the unification picture, and in some SuperSymmetric models.
Multiply charged Intermediate Mass Monopoles (IMMs) may have been produced in later phase transitions in the Early Universe, when a semisimple gauge group yields a U(1) group . IMMs with m<sub>M</sub> $`10^7รท10^{13}`$ GeV may be accelerated to relativistic velocities in one galactic magnetic field domain. Very energetic IMMs could yield the highest energy cosmic rays .
The lowest mass MM is stable, since magnetic charge is conserved like electric charge. Thus the poles produced in the Early Universe should still exist as cosmic relics; their kinetic energy was affected by the Universe expansion and by travel through galactic and intergalactic magnetic fields.
GUT poles are best searched for underground in the penetrating cosmic radiation (CR). IMMs may be searched for at high altitude laboratories.
In this lecture we review the experimental situation on MM searches and briefly discuss the searches for nuclearites and Q-balls .
## 2 Properties of magnetic monopoles
The main properties of MMs are obtained from the Dirac relation.
\- If $`n`$ =1 and the basic electric charge is that of the electron, then the basic magnetic charge is $`g_D=\mathrm{}c/2e=137e/2`$. The magnetic charge is larger if $`n>1`$ and if the basic electric charge is $`e/3`$.
\- In analogy with the fine structure constant, $`\alpha =e^2/\mathrm{}c1/137`$, the dimensionless magnetic coupling constant is $`\alpha _g=g_D^2/\mathrm{}c34.25`$; since it is $`>1`$ perturbative calculations cannot be used.
\- Energy W acquired in a magnetic field B$`W=ng_DB\mathrm{}=n20.5`$ keV/G cm. In a coherent galacticโlength ($`\mathrm{}1`$ kpc, $`B3\mu `$G), the energy gained by a MM with $`g=g_D`$ is $`W1.8\times 10^{11}`$ GeV. Classical poles and IMMs in the CR may be accelerated to relativistic velocities. GUT poles should have low velocities, $`10^4<\beta <10^1`$.
\- MMs may be trapped in ferromagnetic materials by an image force, which could reach values of $`10`$ eV/ร
.
\- Electrically charged monopoles (dyons) may arise as quantumโmechanical excitations or as Mโp, M-nucleus composites.
\- The interaction of a MM magnetic charge with a nuclear magnetic dipole could lead to the formation of a Mโnucleus bound system. A monopoleโproton bound state may be produced via radiative capture. Monopoleโnucleus bound states may exist for nuclei with large gyromagnetic ratios.
\- Energy losses of fast poles. A fast MM with magnetic charge $`g_D`$ and velocity $`v=\beta c`$ behaves like an electric charge $`(ze)_{eq}=g_D\beta `$, Fig. 1.
\- Energy losses of slow poles ($`10^4<\beta <10^2`$) may be due to ionization or excitation of atoms and molecules of the medium (โelectronicโ energy loss) or to recoiling atoms or nuclei (โatomicโ or โnuclearโ energy loss). Electronic energy loss predominates for $`\beta >10^3`$.
\- Energy losses at very low velocities. MMs with $`v<10^4c`$ may lose energy in elastic collisions with atoms or with nuclei. The energy is released to the medium in the form of elastic vibrations and/or infraโred radiation .
Fig. 1 shows the energy loss in liquid hydrogen of a $`g=g_D`$ MM vs $`\beta `$ .
\- Energy loss of MMs in celestial bodies. For $`\beta `$ $`<10^4`$ the dE/dx in the Earth is due to poleโatom elastic scattering, eddy currents, and nuclear stopping power. MMs may be stopped by celestial bodies if they have:
Moon: $`\beta 5\times 10^5`$, Earth: $`\beta 10^4`$, Sun: $`\beta 10^3.`$
## 3 Monopole detectors
Monopole detectors are based on MM properties given by Diracโs relation.
\- Superconducting induction devices are sensitive to MMs of any velocity . A moving MM induces in a ring an electromotive force and a current change ($`\mathrm{\Delta }i`$). For a coil with N turns and inductance L, $`\mathrm{\Delta }i=4\pi Nng_D/L=2\mathrm{\Delta }i_o`$, where $`\mathrm{\Delta }i_o`$ is the current change corresponding to a change of one unit of the flux quantum of superconductivity. This method of detection is based only on the longโrange electromagnetic interaction between the magnetic charge and the macroscopic quantum state of a superconducting ring.
\- Scintillation counters for MMs have a threshold $`\beta 10^4`$, above which the light signal is larger than that of a minimum ionizing particle .
\- Gaseous detectors of various types have been used. MACRO used a gas mixture of 73% helium and 27% nโpentane . This allows exploitation of the Drell and Penning effects : a MM leaves a helium atom in a metastable state (He\*) with an excitation energy of $`20`$ eV. The ionization potential of nโpentane is $``$ 10 eV; the excited energy of the He\* is converted into ionization of the nโpentane molecule (Penning effect).
\- Nuclear track detectors (NTDs). The formation of an etchable track in a NTD is related to the Restricted Energy Loss (REL), the fraction of the energy loss localized in a cylindrical region of 10 nm diameter around the particle trajectory. It was shown that both the electronic and the nuclear energy losses are effective in producing etchable tracks in the CR39 NTD which has a threshold at $`z/\beta 5`$ ; it is the most sensitive NTD and it allows to search for MMs with $`g=g_D`$ for $`\beta `$ around $`10^4`$ and $`>10^3`$, the whole $`\beta `$-range of $`4\times 10^5<\beta <1`$ for MMs with $`g2g_D`$ . The Lexan and Makrofol polycarbonates are sensitive for $`z/\beta 50`$ .
## 4 โClassical Dirac monopolesโ
\- Accelerator searches. If MMs are produced at highโenergy accelerators, they would be relativistic and would ionize heavily. Examples of direct searches are the experiments performed with scintillators or NTDs. Experiments at the Fermilab $`\overline{p}p`$ collider established cross section limits of $`2\times 10^{34}`$ cm<sup>2</sup> for MMs with $`m_M<850`$ GeV . Searches at $`e^+e^{}`$ colliders excluded masses up to 45 GeV and later in the 45-102 GeV range ($`\sigma <5\times 10^{37}`$ cm<sup>2</sup>). Recently few high energy general purpose detectors used some subdetectors to search for Dirac MMs .
Fig. 2 summarizes the cross section limits vs MM mass obtained by direct and indirect experiments (solid lines and dashed lines) at the Fermilab $`\overline{p}p`$ collider, $`e^+e^{}`$ colliders, the ISR $`pp`$ collider . Most searches are sensitive to poles with magnetic charges $`g=ng_D/q`$ with $`0.5<n<5`$.
Examples of indirect searches are those performed at the CERN SPS and at Fermilab: the protons interacted in ferromagnetic targets, later the targets were placed in front of a superconducting solenoid with a field $`B>100`$ kG, large enough to extract and accelerate the MMs, to be detected in scintillators and in NTD sheets . An indirect experiment performed at the $`\overline{p}p`$ Tevatron collider, assumed that produced MMs could stop, be trapped and bound in the matter surrounding a collision region . Small Be and Al samples were passed through the 10 cm diameter bore of two superconducting coils, and the induced charge measured by SQUIDs. Limits m$`{}_{M}{}^{}>285`$ GeV were published for $`g=g_D`$ poles. It is difficult to establish the validity of the hypotheses made to interpret these results.
\- Multiโ$`\gamma `$ events. Five peculiar photon showers found in emulsion plates exposed to highโaltitude CRs, are characterized by an energetic narrow cone of tens of photons, without any incident charged particle . The total energy of the photons is $`10^{11}`$ GeV. The small radial spread of photons suggested a c.m. $`\gamma =(1\beta ^2)^{1/2}>10^3`$. The energies of the photons are too small to have $`\pi ^o`$ decays as their source. One possible explanation: a highโenergy $`\gamma `$โray, with energy $`>10^{12}`$ eV, produced a poleโantipole pair, which suffered bremsstrahlung and annihilation producing the final multiโ$`\gamma `$ events. Searches for multi-$`\gamma `$ events were performed in $`pp`$ collisions at the ISR at $`\sqrt{s}=53`$ GeV, at the $`\overline{p}p`$ 1.8 TeV collider and in $`e^+e^{}`$ collisions at LEP (Fig. 2). The D0 experiment searched for $`\gamma `$ pairs with high transverse energies; virtual pointlike MMs may rescatter pairs of nearly real photons into the final state via a box monopole diagram; they set a 95% CL limit of 870 GeV . At LEP the L3 coll. searched for $`Z\gamma \gamma \gamma `$ events; no deviation from QED predictions was observed, setting a 95% CL limit of 510 GeV . Many authors studied the effects from virtual monopole loops . The authors of Ref. criticized the underlying theory and believe that no significant limit can be obtained from present experiments.
\- Searches in bulk matter. Classical MMs could be produced by CRs and could stop at the Earth surface, where they may be trapped in ferromagnetic materials. Bulk matter searches used hundreds of kg of material, including meteorites, schists, ferromanganese nodules, iron ore and others. A superconducting coil through which the material was passed, yielded a monopole/nucleon ratio in the samples $`<1.2\times 10^{29}`$ at 90% CL .
Ruzicka and Zrelov summarized all searches for classical poles performed before 1980 . A more recent bibliography is given in Ref. . Possible effects arising from low mass MMs have been reported in Ref. .
## 5 GUT monopoles
As already stated, GUT theories of the electroweak and strong interations predict the existence of superheavy MMs produced in the Early Universe (EU) when the GUT gauge group breaks into separate groups, one of which is U(1). Assuming that the GUT group is SU(5) (which is excluded by proton decay experiments) one should have the following transitions:
$$\begin{array}{ccccc}& 10^{15}GeV& & 10^2GeV& \\ SU\left(5\right)& & SU\left(3\right)_C\times \left[SU\left(2\right)_L\times U\left(1\right)_Y\right]& & SU\left(3\right)_C\times U\left(1\right)_{EM}\\ & 10^{35}s& & 10^9s& \end{array}$$
(1)
MMs would be generated as topological point defects in the GUT phase transition, about one pole for each causal domain. In the standard cosmology this leads to too many poles (the monopole problem). Inflation would defer the GUT phase transition after large supercooling; in its simplest version the number of generated MMs would be very small. However the flux depends critically on several parameters, like the pole mass, the reheating temperature, etc. If the reheating temperature is large enough one would have MMs produced in high energy collisions, like $`e^+e^{}M\overline{M}`$.
Fig. 3 shows the structure of a GUT MM: a very small core, an electroweak region, a confinement region, a fermionโantifermion condensate (which may contain 4โfermion baryonโnumberโviolating terms); for $`r3`$ fm it behaves as a point particle generating a field $`B=g/r^2`$ .
A flux of cosmic GUT MMs may reach the Earth with a velocity spectrum in the range $`4\times 10^5<\beta <0.1`$, with possible peaks corresponding to the escape velocities from the Earth, the Sun and the Galaxy. Searches for such MMs in the CR performed with superconducting induction devices yielded a combined 90% CL limit of $`2\times 10^{14}`$cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup>, independent of $`\beta `$ . Direct searches were performed above ground and underground <sup>4,25-27</sup>. MACRO performed a search with different types of detectors (liquid scintillators, limited streamer tubes and NTDs) with an acceptance of $``$ 10,000 m<sup>2</sup>sr for an isotropic flux. No MM was detected; the 90% CL flux limits, shown in Fig. 4 vs $`\beta `$ for $`g=g_D`$, are at the level of $`1.4\times 10^{16}`$ cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup> for $`\beta >4\times 10^5`$ . The figure shows also the limits from the Ohya , Baksan, Baikal, and AMANDA experiments .
The interaction of the GUT monopole core with a nucleon can lead to a reaction in which the nucleon decays (monopole catalysis of nucleon decay), f. e. $`M+pM+e^++\pi ^0`$. The cross section for this process is very small, of the order of magnitude of the core size; but the catalysis process could proceed via the Rubakov-Callan mechanism with a $`\sigma `$ of the order of the strong interaction cross section . MACRO performed a dedicated search for nucleon decays induced by the passage of a GUT pole in the streamer tube system. The flux limits obtained, $`38\times 10^{16}`$ cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup>, depend on the MM velocity and on the catalysis cross section . Previous limits were at levels $`10^{15}`$ cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup> , except the Baikal limit which is $`6\times 10^{17}`$ cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup> for $`\beta 10^5`$ .
Indirect GUT MM searches used ancient mica, which has a high threshold. It is assumed that a pole passing through the Earth captures an Al nucleus and drags it through subterranean mica causing a trail of lattice defects, which survive as long as the mica is not reheated. Only small sheets were analyzed ($`13.5`$ and $`18`$ cm<sup>2</sup>), but should have been recording tracks for $`4รท9\times 10^8`$ years. The flux limits are $`10^{17}\text{cm}^2\text{s}^1`$sr<sup>-1</sup> for $`10^4<\beta <10^3`$ . There are reasons why these indirect experiments might not be sensitive: if MMs have a positive electric charge or protons attached, then Coulomb repulsion could prevent capture of heavy nuclei.
## 6 Cosmological and astrophysical bounds
Rough upper limits for a GUT monopole flux in the CR were obtained on the basis of cosmological and astrophysical considerations.
\- Limit from the mass density of the universe: it is obtained requiring that the present MM mass density be smaller than the critical density $`\rho _c`$ of the universe. For $`m_M10^{17}`$ GeV one has the limit: $`F=\frac{n_Mc}{4\pi }\beta <3\times 10^{12}h_0^2\beta (\text{cm}^2\text{s}^1\text{sr}^1)`$. It is valid for poles uniformely distributed in the universe. If poles are clustered in galaxies the limit is larger .
\- Limit from the galactic magnetic field (Parker limit). The $`3\mu `$G magnetic field in our Galaxy is probably due to the nonโuniform rotation of the Galaxy, which generates a field with a timeโscale of the order of the rotation period of the Galaxy $`(\tau 10^8`$ yr). An upper bound for the MM flux is obtained by requiring that the kinetic energy gained per unit time by MMs be less than the magnetic energy generated by the dynamo effect: $`F<10^{15}\text{cm}^2\text{s}^1`$ sr<sup>-1</sup> ; taking into account the almost chaotic nature of the field, with domains of $`\mathrm{}1`$ kpc, the limit becomes mass dependent . An extended โParker boundโ, obtained by considering the survival of an early seed field , yields $`F1.2\times 10^{16}(m_M/10^{17}GeV)\text{cm}^2\text{s}^1\text{sr}^1`$.
\- Limit from the intergalactic (IG) magnetic field. If $`B_{IG}3\times 10^8G`$ with a regeneration time $`\tau _{IG}10^9y`$, a more stringent bound is obtained; the limit is less reliable because the IG field is less known.
\- Limits from peculiar A4 stars and from pulsars may be stringent, but the assumptions made are not clear (see the pulsar PSR 1937+214) .
## 7 Intermediate mass magnetic monopoles
IMMs may appear as topological point defects at a later time in the Early Universe; f.e. the SO(10) GUT group would not yield directly a U(1) group
$$\begin{array}{ccccc}& 10^{15}GeV& & 10^9GeV& \\ SO\left(10\right)& & SU\left(4\right)\times SU\left(2\right)\times SU\left(2\right)& & SU\left(3\right)\times SU\left(2\right)\times U\left(1\right)\\ & 10^{35}s& & 10^{23}s& \end{array}$$
(2)
This would lead to MMs with masses of $`10^{10}`$ GeV; they would survive inflation, be stable, โdoubly chargedโ ($`g=2g_D`$) and do not catalyze nucleon decay . The structure of an IMM would be similar to that of a GUT MM, but the core would be larger (since R $``$ 1/$`m_M`$) and the outer cloud would not contain 4โfermion baryonโnumberโviolating terms.
Relativistic IMMs, $`10^7<m_M<10^{13}`$ GeV, could be present in the cosmic radiation, could be accelerated to large $`\gamma `$ in one coherent domain of the galactic field. Thus one would have to look for $`\beta 0.1`$ MMs.
Detectors at the Earth surface could detect MMs coming from above if they have $`m_M>10^510^6`$ GeV ; lower mass MMs may be searched for with detectors located at high mountain altitudes, balloons and satellites.
Few experimental results are available. Fig. 5 shows the situation on the flux upper limits for IMMs . The Cherenkov neutrino telescopes under ice and underwater are sensitive to fast ($`\gamma >>1`$) MMs coming from above.
The SLIM experiment, which searches for IMMs with NTDs at the Chacaltaya high altitude lab (5290 m a.s.l.) , is sensitive to $`g=2g_D`$ MMs in the whole range $`4\times 10^5<\beta <1`$.
## 8 Nuclearites and Q-balls
Strange Quark Matter (SQM) should consist of aggregates of u, d and s quarks in almost equal proportions; the number of s quarks should be lower than the number of u or d quarks and the SQM should have a positive integer charge. The overall neutrality of SMQ is ensured by an electron cloud which surrounds it, forming a sort of atom (see Fig. 6). SQM should have a constant density $`\rho _N=M_N/V_N3.5\times 10^{14}`$ g cm<sup>-3</sup>, larger than that of atomic nuclei, and it should be stable for all baryon numbers in the range between ordinary heavy nuclei and neutron stars (A $`10^{57}`$). Lumps of SQM with baryon number $`A<10^610^7`$ are usually called โstrangeletsโ; the word โnucleariteโ was introduced to indicate large lumps of SQM which could be present in the CR . SQM lumps could have been produced shortly after the Big Bang and may have survived as remnants; they could also appear in violent astrophysical processes, such as in neutron star collisions. SQM could contribute to the cold dark matter. The main energy loss mechanism for low velocity nuclearites is elastic or quasi-elastic collisions with the ambient atoms. The energy loss is large; therefore nuclearites should be easily detected in scintillators and CR39 NTDs . Nuclearites should have typical galatic velocities, $`\beta 10^3`$, and for masses larger than 0.1 g could traverse the earth. Most nuclearite searches were obtained as byproducts of CR MM searches; the flux limits are similar to those for MMs.
The most relevant direct flux limits for nuclearites come from three large area experiments: the first two use CR39 NTDs; one experiment was performed at mountain altitude (Mt. Norikura at 2770 m a.s.l.) , the 2nd at the depth of $`10^4`$ g cm<sup>-2</sup> in the Ohya mine ; the third experiment, MACRO, at an average depth of 3700 hg cm<sup>-2</sup>, used liquid scintillators besides NTDs . A 4th experiment (SLIM) is deployed at high altitudes. Indirect searches with old mica samples could yield the lowest limits, but they are affected by several uncertainties. Some exotic cosmic ray events were interpreted as due to incident nuclearites, f. e. the โCentauroโ events and the anomalous massive particles, but the interpretation is not unique . Supermassive nuclearites (M $``$ 1 ton) passing through Earth could induce epilinear earthquakes . Fig. 7 shows a compilation of limits for a flux of downgoing nuclearites compared with the dark matter (DM) limit, assuming a velocity at ground level $`\beta =10^3`$, corresponding to nuclearites of galactic or extragalactic origin. The MACRO limit is extended above the DM bound to show the transition to an isotropic flux for $`M_n>0.1`$ g ($`10^{23}`$ GeV). Some possible positive indications are discussed in Ref. .
Q-balls should be aggregates of squarks $`\stackrel{~}{q}`$, sleptons $`\stackrel{~}{l}`$ and Higgs fields . The scalar condensate inside a Q-ball core has a global baryon number Q (and may be also a lepton number). Protons, neutrons and may be electrons could be absorbed in the condensate. There could exist neutral and charged Q-balls. Supersymmetric Electrically Neutral Solitons (SENS) are generally massive and may catalyse proton decay. SENS may obtain a positive electric charge absorbing a proton in their interactions with matter yielding SECS (Supersymmetric Electrically Charged Solitons), which have a core electric charge, have generally lower masses and the Coulomb barrier could prevent the capture of nuclei. SECS have only integer charges because they are color singlets. A SENS which enters the earth atmosphere could absorb a nitrogen nucleus which would give it the positive charge of +7 (SECS with $`z=7`$). Other nuclear absorptions are prevented by Coulomb repulsion. If the Q-ball can absorb electrons at the same rate as protons, the positive charge of the absorbed nucleus may be neutralized by the charge of absorbed electrons. If, instead, the absorption of electrons is slow or impossible, the Q-ball carries a positive electric charge after the capture of the first nucleus in the atmosphere. Q-balls may be cold DM candidates. SECS with $`\beta 10^3`$ and $`M_Q<10^{13}`$ GeV could reach an underground detector from above, SENS also from below. SENS may be detected by their continuons emission of charged pions (energy loss $``$ 100 GeV g<sup>-1</sup>cm<sup>2</sup>), SECS may be detected by scintillators, NTDs and ionization detectors.
Note that we did not consider here the possibility of strongly interacting, colored, MMs, nuclearites and Q-balls.
## 9 Conclusions. Outlook
Direct and indirect accelerator searches for classical Dirac MMs placed limits at the level $`m_M>850`$ GeV with cross section upper values as shown in Fig. 2. Future improvements may come from experiments at the LHC .
Many searches were performed for GUT poles in the penetrating cosmic radiation. The 90% CL flux limits are at $`1.4\times 10^{16}`$ cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup> for $`\beta 4\times 10^5`$. It may be difficult to do much better since one would require refined detectors of considerably larger areas.
Present limits on Intermediate Mass Monopoles with high $`\beta `$ are relatively poor. Experiments at high altitudes and at neutrino telescopes should improve the situation. In particular stringent limits may be obtained by large neutrino telescopes for IMMs with $`\beta >0.5`$ coming from above.
As a byproduct of GUT MM searches some experiments obtained stringent limits on nuclearites and on Q-balls. Future experiments at neutrino telescopes and at high altitudes should perform searches for nuclearites and Q-balls of smaller masses.
## 10 Acknowledgements
We acknowledge the cooperation of many colleagues, in particular S. Cecchini, M. Cozzi, M. Giorgini, G. Mandrioli, S. Manzoor, V. Popa, M. Spurio, and others. We thank ms. Giulia Grandi for typing the manuscript.
|
warning/0506/astro-ph0506111.html
|
ar5iv
|
text
|
# The excitation of water in the S140 photon dominated region.
## 1 Introduction
The study of water is of great interest (e.g., Ceccarelli et al. 1996). Not only because $`\mathrm{H}_2\mathrm{O}`$ possesses a very large number of strong far-infrared (FIR) transitions, but also because it plays a crucial role in the chemistry of molecular clouds (Bergin et al. 1998) and can act as an important coolant in dense molecular clouds (Neufeld et al. 1995), in shocks (Draine et al. 1983; Neufeld $`\&`$ Melnick 1987) and circumstellar outflows (Chen $`\&`$ Neufeld 1995). Because of its high abundance and easy excitation in warm interstellar and circumstellar environments, water is a powerful tool to probe astrophysical conditions in a broad variety of sources, from protostars to molecular clouds.
Nonetheless, the large attenuation by terrestrial $`\mathrm{H}_2\mathrm{O}`$ molecules has made ground-based observations nearly impossible. Therefore, observations from space-born facilities are needed (e.g., Neufeld et al. 1999; Truong-Bach et al. 1999; Wright et al. 2000).
The launch of the Infrared Space Observatory (ISO) has made it possible to look for emission and absorption lines in warm ($``$ 100 K) molecular gas near young stellar objects (van Dishoeck et al. 1999), while the Submillimeter Wave Astronomy Satellite (SWAS) has probed $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ and $`{}_{}{}^{}\mathrm{O}_{2}^{}{}_{}{}^{}`$ residing in cold molecular clouds (Melnick et al. 2000), as has Odin (Hjalmarson et al. 2003). The future launch of Herschel, with on board the Heterodyne Instrument for the Far Infrared (HIFI), will provide even better information on the physical and chemical conditions in molecular clouds by means of observing many water lines with higher angular resolution and sensitivity.
In this paper we discuss the excitation of water in PDRs. As a testcase we take S140 because it has been probed extensively in the past with SWAS and will be examined in the future with HIFI. The results presented here are obtained with the use of a newly developed escape probability method described in Sect. 2 combined with the inhomogeneous numerical code of Spaans (1996), expanded in Spaans $`\&`$ van Dishoeck (1997). Section 3 describes briefly S140. Section 4 is used to describe the results from the 3D- (in)homogeneous models. We summarize and discuss the paper in Sect. 5.
## 2 The escape probability method
We calculate the transfer of line radiation of ortho- and para-$`\mathrm{H}_2\mathrm{O}`$ in a (in)homogeneous 3-dimensional spherical cloud by use of an escape probability approximation. Let $`n_i`$$`(x,y,z)`$ $`\mathrm{cm}^3`$ be the population density of the $`i^{\mathrm{th}}`$ level in point $`(x,y,z)`$. The equations of statistical equilibrium can then be written as:
$`n_i(x,y,z){\displaystyle \underset{ji}{\overset{l}{}}}R_{ij}(x,y,z)`$ $`=`$ $`{\displaystyle \underset{ji}{\overset{l}{}}}n_j(x,y,z)R_{ji}(x,y,z),`$ (1)
where $`l`$ is the total number of levels included. $`R_{ij}`$$`(x,y,z)`$ is expressible in terms of the Einstein $`A_{ij}`$ and $`B_{ij}`$ coefficients, and the collisional excitation ($`i`$ $`<`$ $`j`$) and de-excitation rates ($`i`$ $`>`$ $`j`$) $`C_{ij}`$:
$`R_{ij}(x,y,z)=\{\begin{array}{cc}A_{ij}+B_{ij}<J_{ij}>+C_{ij}(x,y,z),\hfill & \text{(i }>\text{ j)}\hfill \\ B_{ij}<J_{ij}>+C_{ij}(x,y,z),\hfill & \text{(i }<\text{ j)}\hfill \end{array}`$ (2)
$`<`$$`J_{ij}(x,y,z)`$$`>`$ is the mean integrated radiation field at frequency $`\nu _{ij}`$ corresponding to the transition from level $`i`$ to $`j`$ at position $`(x,y,z)`$ in the cloud, given by:
$`<J_{ij}(x,y,z)>`$ $`=`$ $`[1\beta _{ij}(x,y,z)]S_{ij}(x,y,z)`$ (3)
$`+\beta _{ij}(x,y,z)B_{ij}(\nu _{ij},T_{\mathrm{BB}}),`$
where $`\beta _{ij}`$(x,y,z) is the probability for a photon to escape and $`S_{ij}`$(x,y,z) is the source function, to be addressed later. The background radiation $`B_{ij}`$($`\nu _{ij}`$,$`T_{\mathrm{BB}}`$) consists of two terms: the 2.7 K microwave background and the infrared emission of dust at a temperature $`T_\mathrm{d}`$ and with an emission optical depth $`\tau _\mathrm{d}(\nu _{ij})`$:
$`B_{ij}(\nu _{ij},T_{\mathrm{BB}})`$ $`=`$ $`B(\nu _{ij},T=2.7\text{K})`$ (4)
$`+\tau _\mathrm{d}(\nu _{ij})B(\nu _{ij},T_\mathrm{d}),`$
where $`\tau _\mathrm{d}(\nu _{ij})`$ = $`\tau _{100\mu \mathrm{m}}`$(100 $`\mu `$m/$`\lambda `$). We adopt a value for $`\tau _{100\mu \mathrm{m}}`$ of 0.001 (Hollenbach et al. 1991). If all the photons escape (e.g., $`\beta _{ij}`$ = 1), then $`<`$$`J_{ij}(x,y,z)`$$`>`$ is $`B_{ij}`$($`\nu _{ij}`$,$`T_{\mathrm{BB}}`$); if none escape (e.g., $`\beta _{ij}`$ = 0) $`<`$$`J_{ij}(x,y,z)`$$`>`$ is the local source function $`S_{ij}`$$`(x,y,z)`$.
The source function reads as
$`S_{ij}(x,y,z)={\displaystyle \frac{2h\nu _{ij}^3}{c^2}}\left[{\displaystyle \frac{n_j(x,y,z)g_i}{n_i(x,y,z)g_j}}1\right]^1,`$ (5)
where $`g_i`$ and $`g_j`$ are the statistical weights of levels $`i`$ and $`j`$, respectively.
Since the set of $`l`$ statistical equilibrium equations is not independent, one equation has to be replaced by the conservation equation
$`n_x={\displaystyle \underset{j=0}{\overset{l}{}}}n_j,`$ (6)
where $`n_x`$ is the number density of species $`x`$ in all levels.
In a sphere, the probability that a photon emitted in the transition from level $`i`$ to level $`j`$ at position $`(x,y,z)`$ along a direction $`k`$ escapes the cloud is given by
$`\beta _{ij}(x,y,z,k)={\displaystyle \frac{1\mathrm{exp}(\tau _{ij}(x,y,z,k))}{\tau _{ij}(x,y,z,k)}},`$ (7)
with $`\tau _{ij}`$ the optical depth in the line. Note that the probability for a photon to escape in a point, p$`(x,y,z)`$, equals the sum of the escape probabilities over all directions, e.g., $`\beta _{ij}`$$`(x,y,z)`$ = $`\mathrm{\Sigma }_k`$ $`\beta _{ij}`$($`x`$,$`y`$,$`z`$,$`k`$). The number of directions is arbitrary, but a 6-ray approximation is implemented for the 3D models. Note that for a photon emitted in a plane-parallel medium, the probability to escape is given by the expression
$`\beta _{ij}(x,y,z,k)={\displaystyle \frac{1\mathrm{exp}(3\tau _{ij}(x,y,z,k))}{3\tau _{ij}(x,y,z,k)}}`$ (8)
The optical depth averaged over the line, in direction $`k`$ over a distance s = $`s_2`$($`x_2`$,$`y_2`$,$`z_2`$) - $`s_1`$($`x_1`$,$`y_1`$,$`z_1`$) is given by
$`\tau _{ij}(x,y,z,k)={\displaystyle \frac{A_{ij}c^3}{8\pi \nu _{ij}^3}}{\displaystyle \underset{s=s_1}{\overset{s=s_2}{}}}{\displaystyle \frac{n_i}{\mathrm{\Delta }\mathrm{v}_\mathrm{d}}}\left[{\displaystyle \frac{n_jg_i}{n_ig_j}}1\right]๐s,`$ (9)
with $`\mathrm{\Delta }`$$`\mathrm{v}_\mathrm{d}`$ = $`(\mathrm{v}_{\mathrm{th}}^2+\mathrm{v}_{\mathrm{turb}}^2)^{1/2}`$ the velocity dispersion. Turbulence in PDRs can be combined with the thermal speed of the gas into a Gaussian distribution. The resultant Doppler profile is
$`\phi (x)={\displaystyle \frac{e^{x^2}}{\pi ^{1/2}}},`$ (10)
where x = ($`\nu `$ \- $`\nu _{ij}`$)/$`\mathrm{\Delta }\nu _\mathrm{d}`$. The Doppler frequency width $`\mathrm{\Delta }\nu _\mathrm{d}`$ is given by $`\mathrm{\Delta }\nu _\mathrm{d}`$ = $`\mathrm{\Delta }\mathrm{v}_\mathrm{d}`$$`\frac{\nu _{ij}}{\mathrm{c}}`$. The turbulent dispersion, $`\mathrm{v}_{\mathrm{turb}}`$, is larger than the thermal velocity of the gas, $`\mathrm{v}_{\mathrm{th}}`$, and dominates the line broadening in PDRs.
Because interpretation of observed water lines requires knowledge of collisional excitation rates responsible for line formation, it is necessary to work with the most accurate collisional rate data available. The collisional rates are taken from Green et al. (1993) for collisions involving up to 8 levels for inelastic collisions between $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ and $`{}_{}{}^{}\mathrm{He}_{}^{}`$. To account for the different reduced mass of the collisional partner (e.g., $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$), the rate coefficients are scaled by a factor of 1.348. Rate coefficients of para- and ortho $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$O owing to collisions with ortho- and para-$`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ at kinetic temperatures from 20 to 140 K are taken from Phillips et al. (1996). In addition, rate coefficients for $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ in collisions with p-$`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ and o-$`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ for the low temperature regime (5-20 K) are taken from Grosjean et al. (2003) and Dubernet $`\&`$ Grosjean (2002), respectively. Clumps and their edges dominate the total water emissivity a few magnitudes of extinction into the PDR. The ambient electron abundance in these regions does not exceed 3 $`\times `$ $`10^7`$$`10^6`$ in our models. Using the $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$O-electron collision rates presented by Faure et al. (2004) we find that the contribution by electrons to the total collisional excitation rate is less than 1 $`\%`$ and is therefore ignored in our calculations. Note that when the electron abundance exceeds $`10^5`$, this contribution is equal to the collisional rates of $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$O with para-$`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$, and consequently has to be taken into account.
In PDRs, the ortho-to-para ratios of $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ have typical values in the range 1.5โ2.2 (Hasegawa et al. 1987; Ramsay et al. 1993; Chrysostomou et al. 1993; Hora $`\&`$ Latter 1996; Shupe et al. 1998), and are not constant because of gas temperature variations within the PDR. Therefore we adopt the expression for the ortho-to-para ratio (OPR), in thermal equilibrium, defined by
$`\mathrm{OPR}={\displaystyle \frac{(2I_\mathrm{o}+1)(2J+1)\mathrm{exp}\left(\frac{E_\mathrm{o}(J,K_\mathrm{a},K_\mathrm{c})}{kT}\right)}{(2I_\mathrm{p}+1)(2J+1)\mathrm{exp}\left(\frac{E_\mathrm{p}(J,K_\mathrm{a},K_\mathrm{c})}{kT}\right)}},`$ (11)
where $`I_\mathrm{o}`$ and $`I_\mathrm{p}`$ are the total nuclear spin, according to the hydrogen nuclear spins being parallel ($`I_\mathrm{o}`$ = 1, $``$$``$) or anti-parallel ($`I_\mathrm{p}`$ = 0, $``$$``$). The sum in the numerator (denominator) extends over all ortho (para) levels $`(J,K_\mathrm{a},K_\mathrm{c})`$ (Mumma et al. 1987). A similar expression holds for the ortho- and para-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ OPR, for which the statistical equilibrium (OPR = 3) value is attained at temperatures exceeding $``$ 60 K, but will differ from the statistical equilibrium value for lower temperatures.
This escape probability method is the first step towards a fully 3D-Monte Carlo radiative transfer code that calculates line intensities, as well as line profiles in different astrophysical environments.
## 3 The molecular cloud S140
The molecular cloud lies at a distance of $``$ 900 pc (Crampton $`\&`$ Fisher 1974) and is illuminated from the south-western side by the BOV star HD 211880. The star is located about $``$ 7 (1.85 pc) from the edge of the cloud. The cloud extends over more than 30 (8 pc) and contains a dense core in which star formation is occuring.
The molecular cloud consists of two parts: (i) the extended molecular cloud (EMC), and (ii) the interface region near the south-west edge of the cloud (PDR). Here focus is on the interface region, as this region is a far more interesting object to probe the behaviour of water with its high $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ abundance relative to the extended molecular cloud.
In the past, models were neither able to reproduce the observations of extended emission of $`[`$C i$`]`$ (Keene et al. 1985) and $`[`$C ii$`]`$ (Stutzki et al. 1988) deep into the cloud, nor the intense $`{}_{}{}^{13}\mathrm{CO}_{}^{}`$ 6-5 emission in some regions (Graf et al. 1990). This information, together with the $`{}_{}{}^{}\mathrm{CO}_{}^{}`$ maps by Falgarone $`\&`$ Perault (1988), led to the suggestion that interstellar clouds have a very inhomogeneous structure, so that the ultraviolet radiation can penetrate much deeper into the cloud.
Most previous models of S140 (e.g., Ashby et al. 2000) have assumed a spherically symmetric cloud with temperature and density power laws. Even though they can reproduce well the observed value of the intensity of the 557 GHz $`1_{10}`$ $``$ $`1_{01}`$ ground-state transition of ortho-$`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$$`{}_{}{}^{16}\mathrm{O}_{}^{}`$, being acquainted with the inhomogeneous structure of molecular clouds from observations and in order to explain and predict more accurately various line transitions, we make use of density, temperature and abundance (relative to $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$) distributions as calculated with the inhomogeneous code of Spaans (1996). This code is described first in Spaans (1996) and modified in Spaans $`\&`$ van Dishoeck (1997). This is the first time an attempt is made to model the water emission in S140 in an inhomogeneous way.
The Monte Carlo code calculates the chemical structure and thermal balance in PDRs, as well as the distribution of the water abundance and molecular hydrogen density simultaneously. A chemical network including 291 reactions between 51 species consisting of the elements hydrogen, carbon ($`{}_{}{}^{12}\mathrm{C}_{}^{}`$), oxygen, iron, magnesium, and PAHs (with charges -1, 0, +1) is used. The gas abundances for S140, are taken to be 2.0 $`\times `$ $`10^4`$, 5.0 $`\times `$ $`10^4`$, 2.5 $`\times `$ $`10^7`$ and 1.3 $`\times `$ $`10^6`$ for C, O, Fe and Mg, respectively. For carbon this implies that about 50$`\%`$ of the solar abundance is present in the gas phase, which is comparable with the values found in diffuse clouds (Cardelli et al. 1993). Mechanisms for converting the ultraviolet flux into gas heating are (1) photoelectric emission from small dust grains and large molecules like Polycyclic Aromatic Hydrocarbons (PAHs) (Bakes $`\&`$ Tielens 1994; Spaans et al. 1994), (2) neutral carbon ionization, (3) $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ photodissociation, (4) FUV pumping followed by collisional de-excitation of vibrationally excited $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ in the two level approximation, and (5) $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ formation from FUV-produced hydrogen atoms on grain surfaces. In addition, (6) cosmic-ray heating may also contribute deep into the cloud, and (7) O i may heat the gas through the collisional de-excitation of the $`{}_{}{}^{3}\mathrm{P}_{1}^{}`$ fine structure level excited by (dust) continuum radiation.
For the cooling, the infrared fine-structure lines from atoms $`{}_{}{}^{}\mathrm{C}_{}^{}`$, $`{}_{}{}^{}\mathrm{C}_{}^{+}`$, $`{}_{}{}^{}\mathrm{O}_{}^{}`$, $`{}_{}{}^{}\mathrm{Si}_{}^{}`$, and $`{}_{}{}^{}\mathrm{Fe}_{}^{}`$, if present in the chemical network, have been taken into account. The lowest 20 rotational levels of $`{}_{}{}^{}\mathrm{CO}_{}^{}`$ are calculated in statistical equilibrium and their contribution to the cooling is included. $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ rotational cooling is considered as well as cooling due to water through a cooling function (Neufeld $`\&`$ Melnick 1987). The contribution of water to the total cooling rate amounts to 10โ20$`\%`$ around the clump density of $`10^5`$ $`\mathrm{cm}^3`$ and for temperatures of 40โ60 K. The CO molecule dominates the cooling for these conditions.
The computed temperature, density and water abundance distribution is then used as input for the water line transfer code described previously in this paper. This approach leads to a more accurate treatment of the complex chemical and thermal structure of the PDR, than when a temperature and $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ abundance are simply imposed.
## 4 Models
This section consists of two parts. The first part describes the results from the homogeneous model, the second part describes the inhomogeneous models and the results thereof.
### 4.1 Homogeneous 3D model
The incident radiation field, taken to be $`I_{\mathrm{UV}}`$ = 140 with respect to the Draine (1978) field, is consistent with the enhancement at the edge of the cloud near the BOV star HD 211880. An isotropic component, $`I_{\mathrm{UV}}`$ = 1, of the average interstellar radiation field has also been taken into account. For the homogeneous PDR model we adopt a generic mean total hydrogen density of $`n_\mathrm{H}`$ = $`n`$($`{}_{}{}^{}\mathrm{H}_{}^{}`$) + 2$`n`$($`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$) = 2 $`\times `$ $`10^4`$ $`\mathrm{cm}^3`$ over a region 0.5 pc in extent, or $`A_\mathrm{V}`$ $``$ 20 mag, in agreement within a factor $``$ 1.5โ2 with observations (Hayashi et al. 1985; Zhou et al. 1994). For areas where $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ is present, hydrogen is totally molecular (the main part of the UV radiation is already absorbed), and therefore $`n_\mathrm{H}`$ = 2n($`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$). The grid size is taken to be 81 $`\times `$ 81 $`\times `$ 81, corresponding to a resolution of 0.006 pc.
The temperature distribution for the gas and dust is shown in Fig. 1. The gas (dust) temperature is $``$ 150 K (40 K) at the western edge of the cloud, and drops to about 30 K (30 K) deeper into the cloud. The dust temperature stays far below the gas temperature at the west side of the cloud, but becomes slightly larger or equal to the gas temperature farther into the cloud. It is worth noting that dust plays an important role in the excitation of water. It absorbs the interstellar FUV and visible radiation field and emits the absorbed energy in the far infrared, offering a source of excitation for water. Computed water abundances vary from $``$ $`10^{10}`$ to $``$ $`10^6`$, relative to the hydrogen density $`n_\mathrm{H}`$, with an average abundance of 1.4 $`\times `$ $`10^7`$. At the west side of the cloud, where the star is located, the abundance is lowered due to photodissociation of water by the FUV radiation field:
$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ \+ $`\mathrm{h}\nu `$ $``$ $`{}_{}{}^{}\mathrm{H}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{OH}_{}^{}`$,
$`{}_{}{}^{}\mathrm{OH}_{}^{}`$ \+ $`\mathrm{h}\nu `$ $``$ $`{}_{}{}^{}\mathrm{O}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{H}_{}^{}`$
Deeper into the cloud, the radiation field is attenuated by dust, resulting in a higher abundance.
At low temperatures $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ is formed mainly through the dissociative recombination of the $`{}_{}{}^{}\mathrm{H}_3\mathrm{O}_{}^{+}`$ ion:
$`{}_{}{}^{}\mathrm{H}_3\mathrm{O}_{}^{+}`$ \+ $`{}_{}{}^{}\mathrm{e}_{}^{}`$ $``$ $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{H}_{}^{}`$
A value of 0.3 is chosen for the branching ratio leading to $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ in the dissociative recombination of $`{}_{}{}^{}\mathrm{H}_3\mathrm{O}_{}^{+}`$ with a rate coefficient of 3.3 $`\times `$ $`10^7`$ $`(T/300)^{0.3}`$ $`\mathrm{cm}^3\mathrm{s}^1`$. The branching ratio is consistent with the results of Vejby-Christensen et al. (1997) but higher than the value of 0.05 suggested by Williams et al. (1996). At temperatures above $``$ 300 K, the formation of water is dominated by the neutral-neutral endoergic reaction sequence:
$`{}_{}{}^{}\mathrm{O}_{}^{}`$\+ $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ $``$ $`{}_{}{}^{}\mathrm{OH}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{H}_{}^{}`$,
$`{}_{}{}^{}\mathrm{OH}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{H}_{2}^{}{}_{}{}^{}`$ $``$ $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ \+ $`{}_{}{}^{}\mathrm{H}_{}^{}`$
(Elitzur $`\&`$ de Jong 1978; Elitzur $`\&`$ Watson 1978). The activation energy barriers are overcome by the thermal energy of the warm gas.
The emergent intensity integrated along a line of sight is calculated in the following way:
$`I=(4\pi )^1{\displaystyle _o^z}\mathrm{\Lambda }(z^{})๐z^{},`$ (12)
where
$`\mathrm{\Lambda }(\nu _{ij})=n_iA_{ij}h\nu _{ij}\beta (\tau _{ij}),`$ (13)
with $`n_i`$ the population density in the $`i^{\mathrm{th}}`$ level, $`A_{ij}`$ the Einstein A coefficient, h$`\nu _{ij}`$ the energy difference between the levels $`i`$ and $`j`$, $`\tau _{ij}`$ the optical depth averaged over the line, and $`\beta (\tau _{ij})`$ the escape probability at the optical depth $`\tau _{ij}`$ of the line.
Average intensities, peak intensities, as well as intensities averaged over the HIFI FWHM beam size at the transition frequency of the particular line for a beam centered on the peak emission are listed in Table 1. Figure 2 shows maps of the predicted distribution of the intensities for various transitions for ortho-, and para-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ as we would see them on the sky, convolved with a beam size of 0.3<sup>โฒโฒ</sup> ($``$ the angular extent that corresponds to one gridpoint).
To interpret the results one has to take into account the distribution of the gas and dust temperature ($`T_\mathrm{g}`$ and $`T_\mathrm{d}`$, respectively), the abundance of water in the cloud, the density of the medium, as well as optical depth effects. Since the overall mean density of hydrogen is 2 $`\times `$ $`10^4`$ $`\mathrm{cm}^3`$, far below the critical density for all the transitions (of the order of $`10^8`$ $`\mathrm{cm}^3`$ or higher), the lines are subthermally excited. This means that collisional de-excitation is less important than spontaneous emission, which results in a distribution of the level populations deviating from LTE.
The emission will originate from a depth where $`\tau _{ij}`$ $``$ 1. The place where $`\tau _{ij}`$ $``$ 1 varies for every transition. Optically thick lines, i.e., $`\tau _{ij}`$ $``$ 1, can be used to probe the physical conditions in the outer ($`\tau _{ij}`$ $`<`$ 1) layers of the cloud. Lines which are optically thin can be used to probe the physical conditions throughout the cloud. A peak in intensity for a certain transition will occur at places in the cloud where the temperature is high enough to excite the line and will depend on the column of $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ along the line of sight. Therefore at the edges of the cloud, where the column is rather low, the intensity drops.
The o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`1_{10}`$ $``$ $`1_{01}`$ transition peaks near the east edge of the cloud where the temperature of the gas drops to $``$ 60 K, the perfect regime to excite the line. Transitions with a higher energy difference, e.g., o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`3_{12}`$ $``$ $`2_{21}`$ , have their peak shifted towards the edge of the cloud where the temperature is higher. Note that HIFI, with its high angular resolution will be able to see this shift. Depending on the frequency of the transition, the beam will be $``$ 5 times smaller than the PDR.
The o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`2_{12}`$ $``$ $`1_{01}`$ transition is the strongest we found, with an average intensity of $``$ 5 $`\times `$ $`10^7\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1\mathrm{sr}^1`$.
### 4.2 Inhomogeneous 3D model with the EMC
The inhomogeneous nature of molecular clouds is well established from observations of extended $`[`$C i$`]`$ and $`[`$C ii$`]`$ (e.g., Keene et al. 1985; Plume et al. 1994) and has been modeled by various groups (e.g., Meixner $`\&`$ Tielens 1993; Spaans 1996; Stoerzer et al. 1996). It is therefore natural to explore the excitation of $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ in models incorporating high density clumps embedded in a low density interclump medium.
In the case of an inhomogeneous distribution, the radiative transfer is solved in three dimensions on a 81 $`\times `$ 81 $`\times `$ 81 grid. In this section we not only cover the excitation of water in the PDR but also in the EMC, in order to fully interpret the results. We define the molecular cloud core as that part of the EMC that encloses the PDR and lies in one SWAS beam. What is left outside the molecular cloud core is denoted as the EMC.
The clumpy models have three free parameters: (i) the volume filling factor $`F`$, which gives the fraction of the total volume that is occupied by the clumps, (ii) a clump size $`l_\mathrm{c}`$, which fixes the extinction through an individual clump, and (iii) the clump-interclump density ratio $`R_{\mathrm{ic}}`$. For the PDR a fixed incident radiation field has been taken $`I_{\mathrm{UV}}`$ = 140 with respect to the Draine (1978) field, as well as a mean total hydrogen density $`<`$$`n_\mathrm{H}`$$`>`$ = 2 $`\times `$ $`10^4`$ $`\mathrm{cm}^3`$, as in the case of the homogeneous PDR. The mean total hydrogen density $`<`$$`n_\mathrm{H}`$$`>`$ obeys $`<`$$`n_\mathrm{H}`$$`>`$ = $`F`$$`n_\mathrm{c}`$ \+ (1 - $`F`$)$`n_{\mathrm{ic}}`$. Here $`n_\mathrm{c}`$ and $`n_{\mathrm{ic}}`$ indicate the clump and interclump density, respectively. The enhancement factor for the EMC decreases to $`I_{\mathrm{UV}}`$ $``$ 30โ50 because of geometric dilution. The mean total hydrogen density $`<`$$`n_\mathrm{H}`$$`>`$ is taken to be $`10^3`$ $`\mathrm{cm}^3`$ over a region 10 pc in extent. The average hydrogen density of the EMC is not well constrained by observations, but the adopted value approximates those derived from the total $`{}_{}{}^{13}\mathrm{CO}_{}^{}`$ column density divided by the size of the cloud (Blair et al. 1978; Plume et al. 1994).
Table 2 gives the parameters used for the different models. Three models have been run for the PDR and one for the EMC. The PDR clump densities range from 3$`\times `$$`10^4`$ $`\mathrm{cm}^3`$ to 2$`\times `$$`10^5`$ $`\mathrm{cm}^3`$, whereas interclump densities vary between 3$`\times `$$`10^3`$ $`\mathrm{cm}^3`$ and 5$`\times `$$`10^3`$ $`\mathrm{cm}^3`$. Different volume filling factors were incorporated, ranging from 2.5 $`\%`$ to 20 $`\%`$. In the PDR models, the clump size has been taken constant, $`l_\mathrm{c}`$ = 0.03 pc. The clump size in the EMC is taken to be 20 times larger than the clump size in the PDR, equal in proportion to the ratio in scale size. The velocity dispersion is taken to be 1.2 $`\mathrm{km}\mathrm{s}^1`$ in the PDR (Zhou et al. 1994), and 0.7 $`\mathrm{km}\mathrm{s}^1`$ in the EMC, assuming a weaker turbulent velocity component in comparison with the $`\mathrm{v}_{\mathrm{turb}}`$ of the PDR. The larger PDR velocity width simply reflects our assumption that the embedded sources inject some additional turbulent motions (IRS1โ3, Blair et al. 1978; Beichman et al. 1979; Evans et al. 1989). This seems reasonable given that star-forming molecular cores tend to have larger (non-thermal) velocity widths than quiescent cores.
The clumps are distributed randomly in the interclump medium. The clump and interclump gas join smoothly with a density gradient. Due to the clumpiness the radiative transfer will differ from that of the homogeneous model. The UV radiation penetrates deeper into a clumpy region than into a cloud in which the gas is homogeneously distributed, leading to a lower water abundance because of enhanced photodissociation (Spaans $`\&`$ van Dishoeck 2001). The average water abundance in comparison with the homogeneous model drops by a factor $``$ 10. However, the resulting intensities depend sensitively on the optical depth, density and temperature structure along the line of sight. The inhomogeneity also affects the temperature structure, resulting in enhanced temperatures deeper into the interclump medium of the cloud, and lower clump temperatures. The result is an increase in the volume averaged gas temperature ($``$ 78 K for PDR I), compared with that of the homogeneous PDR model ($``$ 52 K). However, the mass-weighted average gas temperatures in the inhomogeneous models are 88 K, 70 K and 46 K for PDR I, PDR II and PDR III, respectively, and 52 K for the homogeneous model. This means that most of the mass of the cloud is located in cold clumps.
Intensities are computed in the same way as in the homogeneous model using Eq. (12) and (13). We calculate the total intensity along each line of sight. Average and peak intensities for various transitions for the PDR are shown in Table 3, and for the molecular cloud core, i.e., that part of the EMC lying at the southwest of the cloud which encloses the PDR, in Table 4. Fig. 3 and 4 show maps of the predicted distribution of the intensities for various transitions, and Fig. 5 shows the change in intensity along a W-E cut for two ortho-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ transitions. Note that the vast majority of the EMC is not contained in the SWAS beam. Snell et al. (2000) mapped a larger region of S140 with SWAS, including several positions on the molecular cloud. They find no significant 557 GHz water emission residing in these areas (e.g., outside the molecular cloud core), in agreement with our results. The contribution of the EMC (core) to the total signal is negligible in comparison with the PDR. The minor contribution of the EMC (core) is due to the low gas density, which results in little excitation of the $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ molecule. Freeze-out of water is not required to explain the SWAS non-detections away from the PDR. However, the derived physical conditions indicate that freeze-out of water does occur. From Bergin et al. (1995) we find a depletion factor of about 2โ3 (see Section 5).
The intensity for the o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`1_{10}`$ $``$ $`1_{01}`$ line in the homogeneous PDR model is a factor of a few lower than in PDR I and PDR II, and equals that of PDR III. The strongest transition we find is the o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`2_{12}`$ $``$ $`1_{01}`$ in all models. This transition has an upper state energy of $``$ 80 K above ground level. The gas temperatures in the models are sufficient to excite this transition. All other transitions are a few orders of magnitude weaker than the two lowest transitions. To excite these transitions much higher temperatures are needed. For para-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$, the lowest transition $`1_{11}`$ $``$ $`0_{00}`$ is found to be the strongest in all the models.
From the average intensities, antenna temperatures are derived for the 557 GHz transition as seen in Table 5, corrected for beam dilution in the case of the PDR with a beam dilution factor of 2. Note that the correct beam dilution factor in case of the PDR is uncertain. Since the antenna temperatures scale inversely in proportion to the beam dilution factor the results presented in Table 5 will differ when another beam dilution factor is incorporated.
The total antenna temperature is the weighted sum of the PDR and EMC contributions. PDR I and PDR II in Table 3 yield, when combined with Table 4, a value consistent with the observation of SWAS, i.e., $`T_\mathrm{A}^{}`$ $``$ 0.25 K, (Ashby et al. 2000; Snell et al. 2000) for the lowest lying transition of ortho-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$. At face value PDR I is in closest agreement with the observations.
Maps for the various transitions (line, not line + continuum) of the PDR I model are plotted in Figs. 3 and 4. It is seen that the o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`1_{10}`$ $``$ $`1_{01}`$ ground state transition peaks near the eastern edge of the cloud. The bulk of the emission resides in the individual clumps. These clumps are low-temperature-water reservoirs. At the west side of the cloud the water abundance drops due to dissociation, still, for higher transitions, we see that the majority of the emission is shifted towards the western edge of the cloud. The $`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ emission arises mainly from the warm, $``$ 60โ110 K, clump edges, where the gas temperature is higher than in the interiors of the clumps. Conversely, the clump edge water abundance is a factor of $``$ 10 lower compared to the insides of the clumps, but still a factor of $``$ 10 higher than in the interclump medium. The 557 GHz line traces mostly the total column of water, while it is evident from Table 3 that the higher excitation lines differ from one model to the other more strongly.
Again, HIFI will be able to confirm the predicted shift towards the western edge of the higher water transitions because of its good angular resolution. Also, HIFI can constrain the inhomogeneity of the medium through water line observations.
## 5 Summary and Discussion
In this paper we have examined the influence of inhomogeneity on the excitation of water in molecular clouds. Inhomogeneity causes photons to penetrate much deeper into the cloud leading to enhanced photodissociation of water. This gives rise to a lower water abundance. However, calculated intensities of PDR I and PDR II are a factor a few higher for the case of a homogeneous distribution. PDR III is in closest agreement with the intensities of the homogeneous model. This is not surprising since the parameters of PDR III resemble the homogeneous model most. This leads to the conclusion that the major part of the water line intensity emanates from the warm high density clump edges.
The total antenna temperature, $`T_\mathrm{A}^{}`$, is reproduced for the o-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$ $`1_{10}`$ $``$ $`1_{01}`$ line, observed by SWAS. However, the incorporated beam dilution factor is uncertain. The total signal is dominated by the signal of the PDR.
Water line emission is found also outside of the molecular cloud core, but is a few orders of magnitude weaker than in the PDR, in agreement with observations (Snell et al., 2000). In this paper, freeze-out of water onto dust grains has not been taken into account. Bergin et al. (1995) probed the evolution of molecular clouds, including depletion of atoms and molecules onto grain surfaces in the temperature and density regime we are interested in. They find that 50โ60$`\%`$ of the water vapour can be removed from the gas phase, when gas temperatures are 20โ25 K, dust temperatures 10 K and $`<`$$`n_\mathrm{H}`$$`>`$ = $`10^3`$ $`\mathrm{cm}^3`$, as in the case of the EMC. Consequently, depletion can result in an even lower intensity. Depletion of water molecules in the PDR will occur only inside the shielded clumps (not the irradiated clump edges). We find for $`n_\mathrm{H}`$ $``$ $`10^5`$ $`\mathrm{cm}^3`$, $`T_\mathrm{g}`$ = 25 K and $`T_\mathrm{d}`$ = 15 K that $``$ 90$`\%`$ of the water vapour is frozen out (Bergin et al. 1995).
Predictions have been made for the intensities and integrated antenna temperatures of various transitions that HIFI will probe in the future. It is found for S140 that the $`2_{12}`$ $``$ $`1_{01}`$ and $`1_{11}`$ $``$ $`0_{00}`$ transitions are the strongest for ortho- and para-$`{}_{}{}^{}\mathrm{H}_2\mathrm{O}_{}^{}`$, respectively, in all the models.
###### Acknowledgements.
We would like to thank the anonymous referee for constructive remarks on this work. We would like to thank Matt Ashby for helpful discussions.
|
warning/0506/quant-ph0506091.html
|
ar5iv
|
text
|
# Quantum damped oscillator II: Batemanโs Hamiltonian vs. 2D Parabolic Potential Barrier.
## 1 Introduction
In the previous paper we have investigated a quantization of a 1D damped harmonic oscillator defined by the following equation of motion
$$\ddot{x}+2\gamma \dot{x}+\kappa x=\mathrm{\hspace{0.33em}0},$$
(1.1)
where $`\gamma >0`$ denotes the damping constant. To quantize this system we follow an old observation of Bateman and double the number of degrees of freedom, that is together with (1.1) we consider
$$\ddot{y}2\gamma \dot{y}+\kappa y=\mathrm{\hspace{0.33em}0},$$
(1.2)
i.e. an amplified oscillator. The detailed historical review of the Bateman idea may be found in . For more recent papers see e.g. and . The enlarged system is a Hamiltonian one and it is governed by the following classical Bateman Hamiltonian:
$$H(x,y,p_x,p_y)=p_xp_y\gamma (xp_xyp_y)+\omega ^2xy,$$
(1.3)
where $`\omega =\sqrt{\kappa \gamma ^2}.`$<sup>1</sup><sup>1</sup>1Throughout the paper we shall consider the underdamped case, i.e. $`\kappa >\gamma ^2`$. Now, performing a linear canonical transformation $`(x,y,p_x,p_y)(x_1,x_2,p_1,p_2)`$:
$`x_1`$ $`=`$ $`{\displaystyle \frac{p_y}{\sqrt{\omega }}},p_1=\sqrt{\omega }y`$ (1.4)
$`x_2`$ $`=`$ $`\sqrt{\omega }x,p_2={\displaystyle \frac{p_x}{\sqrt{\omega }}},`$ (1.5)
and applying a standard symmetric Weyl ordering one obtains the following quantum Hamiltonian
$$\widehat{H}=\omega \widehat{๐ฉ}\widehat{๐ฑ}\gamma \widehat{๐ฉ}\widehat{๐ฑ},$$
(1.6)
where $`\widehat{๐ฑ}=(\widehat{x}_1,\widehat{x}_2)`$, $`\widehat{๐ฉ}=(\widehat{p}_1,\widehat{p}_2)`$ and we define two natural operations:
$$\widehat{๐ฉ}\widehat{๐ฑ}=\widehat{p}_1\widehat{x}_2\widehat{p}_2\widehat{x}_1,\widehat{๐ฉ}\widehat{๐ฑ}=\widehat{๐ฑ}\widehat{๐ฉ}=\frac{1}{2}\underset{k=1}{\overset{2}{}}(\widehat{x}_k\widehat{p}_k+\widehat{p}_k\widehat{x}_k).$$
Note, that $`[\widehat{๐ฉ}\widehat{๐ฑ},\widehat{๐ฉ}\widehat{๐ฑ}]=0`$. This operator was carefully analyzed in . In particular it was shown that the family of complex eigenvalues
$$\widehat{H}|๐ฃ_{nl}^\pm =E_{nl}^\pm |๐ฃ_{nl}^\pm ,$$
(1.7)
with
$$E_{nl}^\pm =\mathrm{}\omega l\pm i\mathrm{}\gamma (|l|+2n+1),$$
(1.8)
found already by Feshbach and Tikochinsky , corresponds to the poles of the resolvent operator $`\widehat{\mathrm{R}}(\widehat{H},z)=(\widehat{H}z)^1`$. Therefore, the corresponding generalized eigenvectors $`|๐ฃ_{nl}^\pm `$ may be interpreted as resonant states of the Bateman system. It shows that dissipation of energy is directly related to the presence of resonances.
In the present paper we continue to study this system but in a different representation. Let us observe that performing the linear canonical transformation $`(๐ฑ,๐ฉ)(๐ฎ,๐ฏ)`$:
$$๐ฑ=\frac{\gamma ๐ฎ๐ฏ}{\sqrt{2\gamma }},๐ฉ=\frac{\gamma ๐ฎ+๐ฏ}{\sqrt{2\gamma }},$$
(1.9)
one obtains for the Hamiltonian
$$\widehat{H}=\omega \widehat{๐ฏ}\widehat{๐ฎ}+\widehat{H}_{\mathrm{iho}},$$
(1.10)
where
$$\widehat{H}_{\mathrm{iho}}=\frac{1}{2}(\widehat{๐ฏ}^2\gamma ^2\widehat{๐ฎ}^2),$$
(1.11)
represents a Hamiltonian of a 2D isotropic inverted harmonic oscillator (iho) or, equivalently, a 2D potential barrier $`\gamma ^2\widehat{๐ฎ}^2`$. Now, $`\omega \widehat{๐ฏ}\widehat{๐ฎ}`$ generates an SO(2) rotation on $`(u_1,u_2)`$โplane. Therefore, in the rotating frame the problem is described by the following Schrรถdinger equation
$$i\mathrm{}\dot{\psi }_{\mathrm{rf}}=\widehat{H}_{\mathrm{iho}}\psi _{\mathrm{rf}},$$
(1.12)
where the rotating frame wave function $`\psi _{\mathrm{rf}}=\mathrm{exp}(i\omega \widehat{๐ฏ}\widehat{๐ฎ}t/\mathrm{})\psi `$.
A 1D inverted (or reversed) oscillator was studied by several authors in various contexts . Recently, this system was studied in the context of dissipation in quantum mechanics and a detailed analysis of its resonant states was performed in . The present paper is mostly devoted to analysis of a 2D iho. We find its energy eigenvectors and show that they are singular when one continues energy into complex plane. The complex poles correspond to resonant states of the 2D potential barrier .
Finally, we analyze the Bateman system in the hyperbolic representation used recently in by Blasone and Jizba. It turns out that this representation in not appropriate to describe resonant states and hence the family of generalized complex eigenvalues found in is not directly related to the spectral properties of the Bateman Hamiltonian. We stress that it does not prove that these representation are physically inequivalent. Clearly they are. Different representation lead to different mathematical realization which is connected with different functional spaces and different boundary conditions. These may lead to different analytical properties and hence some representation may display resonant states while others not.
From the mathematical point of view the natural language to analyze the spectral properties of Batemanโs system is the so called rigged Hilbert space approach to quantum mechanics . We show (cf. Section 4) that there are two dense subspaces $`\mathrm{\Phi }_\pm L^2(_๐ฎ^2)`$ such that restriction of the unitary group $`\widehat{U}(t)=e^{i\widehat{H}t/\mathrm{}}`$ to $`\mathrm{\Phi }_\pm `$ does no longer define a group but gives rise to two semigroups: $`\widehat{U}_{}(t)=\widehat{U}(t)|_\mathrm{\Phi }_{}`$ defined for $`t0`$ and $`\widehat{U}_+(t)=\widehat{U}(t)|_{\mathrm{\Phi }_+}`$ defined for $`t0`$. It means that the quantum damped oscillator corresponds to the following Gelโfand triplets:
$$\mathrm{\Phi }_\pm L^2(_๐ฎ^2)\mathrm{\Phi }_\pm ^{},$$
(1.13)
and hence it serves as a simple example of Arno Bohm theory of resonances .
## 2 2D inverted oscillator and complex eigenvalues
### 2.1 2D harmonic oscillator
Let us briefly recall the spectral properties of the 2D harmonic oscillator (see e.g. ):
$$\widehat{H}_{\mathrm{ho}}=\frac{\mathrm{}^2}{2}\mathrm{}_2+\frac{\mathrm{\Omega }^2}{2}\rho ^2,$$
(2.1)
where the 2D Laplacian reads
$$\mathrm{}_2=\frac{^2}{\rho ^2}+\frac{1}{\rho }\frac{}{\rho }+\frac{1}{\rho ^2}\frac{^2}{\phi ^2},$$
(2.2)
and $`(\rho ,\phi )`$ are standard polar coordinates on $`(u_1,u_2)`$โplane. The corresponding eigenvalue problem
$$\widehat{H}_{\mathrm{ho}}\psi _{nl}^{\mathrm{ho}}=\epsilon _{nl}^{\mathrm{ho}}\psi _{nl}^{\mathrm{ho}},$$
(2.3)
is solved by
$$\psi _{nl}^{\mathrm{ho}}(\rho ,\phi )=R_{nl}(\rho )\mathrm{\Phi }_l(\phi ),$$
(2.4)
where
$$\mathrm{\Phi }_l(\phi )=\frac{e^{il\phi }}{\sqrt{2\pi }},l=0,\pm 1,\pm 2,\mathrm{},$$
(2.5)
and the radial functions
$$R_{nl}(\rho )=C_{nl}(\sqrt{\mathrm{\Omega }/\mathrm{}}\rho )^{|l|}\mathrm{exp}(\mathrm{\Omega }\rho ^2/2\mathrm{}){}_{1}{}^{}F_{1}^{}(n,|l|+1,\mathrm{\Omega }\rho ^2/\mathrm{}),$$
(2.6)
where the normalization constant reads as follows
$$C_{nl}=\frac{\sqrt{2\mathrm{\Omega }/\mathrm{}}}{|l|!}\sqrt{\frac{(n+|l|)!}{n!}},n=0,1,2,\mathrm{}.$$
(2.7)
Finally, the corresponding eigenvalues $`\epsilon _{nl}^{\mathrm{ho}}`$ are given by the following formula
$$\epsilon _{nl}^{\mathrm{ho}}=\mathrm{}\mathrm{\Omega }(|l|+2n+1).$$
(2.8)
Note, that using well known relation between confluent hypergeometric function $`{}_{1}{}^{}F_{1}^{}`$ and generalized Laguerre polynomials
$$L_n^\mu (z)=\frac{\mathrm{\Gamma }(n+\mu +1)}{\mathrm{\Gamma }(n+1)\mathrm{\Gamma }(\mu +1)}_1F_1(n,\mu +1,z),$$
(2.9)
one may rewrite $`R_{nl}`$ alternatively as follows
$$R_{nl}(\rho )=\sqrt{2\mathrm{\Omega }/\mathrm{}}\sqrt{\frac{n!}{(n+|l|)!}}(\sqrt{\mathrm{\Omega }/\mathrm{}}\rho )^{|l|}\mathrm{exp}(\mathrm{\Omega }\rho ^2/2\mathrm{})L_n^{|l|}(\mathrm{\Omega }\rho ^2/\mathrm{}).$$
(2.10)
It is evident that the family $`\psi _{ln}^{\mathrm{ho}}`$ is orthonormal
$$\psi _{nl}^{\mathrm{ho}}|\psi _{n^{}l^{}}^{\mathrm{ho}}=\delta _{nn^{}}\delta _{ll^{}},$$
(2.11)
and complete
$$\underset{n=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\overline{\psi _{nl}^{\mathrm{ho}}(\rho ,\phi )}\psi _{nl}^{\mathrm{ho}}(\rho ^{},\phi ^{})=\frac{1}{\rho }\delta (\rho \rho ^{})\delta (\phi \phi ^{}),$$
(2.12)
where $`|`$ denotes the standard scalar product in the Hilbert space
$$=L^2(_+,\rho d\rho )L^2([0,2\pi ),d\phi ).$$
(2.13)
### 2.2 Scaling and complex eigenvalues
Let us note that $`\widehat{H}_{\mathrm{iho}}`$ defined in (1.11) corresponds to the Hamiltonian of the harmonic oscillator with purely imaginary frequency $`\mathrm{\Omega }=\pm i\gamma `$. The connection with a harmonic oscillator may be established by the following scaling operator
$$\widehat{V}_\lambda :=\mathrm{exp}\left(\frac{\lambda }{\mathrm{}}\widehat{๐ฏ}\widehat{๐ฎ}\right),$$
(2.14)
with $`\lambda `$. Using commutation relation $`[\widehat{u}_k,\widehat{v}_l]=i\mathrm{}\delta _{kl}`$, this operator may be rewritten as follows
$$\widehat{V}_\lambda =e^{i\lambda }\mathrm{exp}\left(i\lambda \rho \frac{}{\rho }\right),$$
(2.15)
and therefore it defines a complex dilation, i.e. the action of $`\widehat{V}_\lambda `$ on a function $`\psi =\psi (\rho ,\phi )`$ is given by
$$\widehat{V}_\lambda \psi (\rho ,\phi )=e^{i\lambda }\psi (e^{i\lambda }\rho ,\phi ).$$
(2.16)
In particular one easily finds:
$$\widehat{V}_\lambda \widehat{H}_{\mathrm{iho}}\widehat{V}_\lambda ^1=\frac{1}{2}e^{2i\lambda }\left(\mathrm{}^2\mathrm{}_2e^{4i\lambda }\gamma ^2\rho ^2\right).$$
(2.17)
Therefore, for $`e^{4i\lambda }=1`$, i.e. $`\lambda =\pm \pi /4`$, one has
$$\widehat{V}_{\pm \pi /4}\widehat{H}_{\mathrm{iho}}\widehat{V}_{\pm \pi /4}^1=\pm i\left(\frac{\mathrm{}^2}{2}\mathrm{}_2+\frac{\gamma ^2}{2}\rho ^2\right).$$
(2.18)
Now, let us introduce
$$๐ฒ_{nl}^\pm =\widehat{V}_{\pi /4}\psi _{nl}^{\mathrm{ho}},$$
(2.19)
that is
$$๐ฒ_{nl}^\pm (\rho ,\phi )=\sqrt{\pm i}\psi _{nl}^{\mathrm{ho}}(\sqrt{\pm i}\rho ,\phi ).$$
(2.20)
It is evident that
$$\widehat{H}_{\mathrm{iho}}๐ฒ_{nl}^\pm =\epsilon _{nl}^\pm ๐ฒ_{nl}^\pm ,$$
(2.21)
where
$$\epsilon _{nl}^\pm =\pm i\epsilon _{nl}^{\mathrm{ho}}=\pm i\mathrm{}\gamma (|l|+2n+1).$$
(2.22)
We stress that $`\widehat{V}_\lambda `$ is not unitary (for $`\lambda `$) and hence in general $`\widehat{V}_\lambda \psi `$ does not belong to $``$ even for $`\psi `$. In particular the generalized eigenvectors $`๐ฒ_{nl}^\pm `$ do not belong to $``$ (the radial part $`R_{nl}(\sqrt{\pm i}\rho )`$ is not an element from $`L^2(_+,\rho d\rho )`$).
###### Proposition 1
Two families of generalized eigenvectors $`๐ฒ_{nl}^\pm `$ satisfy the following properties:
1. they are bi-orthonormal
$$_0^{2\pi }_0^{\mathrm{}}\overline{๐ฒ_{nl}^\pm (\rho ,\phi )}๐ฒ_{n^{}l^{}}^{}(\rho ,\phi )\rho ๐\rho ๐\phi =\delta _{nn^{}}\delta _{ll^{}},$$
(2.23)
2. they are bi-complete
$$\underset{n=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\overline{๐ฒ_{nl}^\pm (\rho ,\phi )}๐ฒ_{nl}^{}(\rho ^{},\phi ^{})=\frac{1}{\rho }\delta (\rho \rho ^{})\delta (\phi \phi ^{}).$$
(2.24)
The proof follows immediately from orthonormality and completness of oscillator eigenfunctions $`\psi _{nl}^{\mathrm{ho}}`$.
## 3 Spectral properties of the Bateman Hamiltonian
Now, we solve the corresponding spectral problem for the Bateman Hamiltonian (1.10). Note that $`\widehat{H}`$ is bounded neither from below nor from above and hence its spectrum $`\sigma (\widehat{H})=(\mathrm{},\mathrm{})`$. The corresponding generalized eigenvectors satisfy
$$\widehat{H}\psi _{\epsilon ,l}=E_{\epsilon ,l}\psi _{\epsilon ,l},$$
(3.1)
where $`l`$ and $`\epsilon `$. Assuming the following factorized form of $`\psi _{\epsilon ,l}`$
$$\psi _{\epsilon ,l}(\rho ,\phi )=R_{\epsilon ,l}(\rho )\mathrm{\Phi }_l(\phi ),$$
(3.2)
one has
$$E_{\epsilon ,l}=\omega \mathrm{}l+\epsilon ,$$
(3.3)
with
$$\widehat{H}_{\mathrm{iho}}R_{\epsilon ,l}=\epsilon R_{\epsilon ,l}.$$
(3.4)
The above equation rewritten in terms of $`(\rho ,\phi )`$-variables takes the following form
$$\left(\frac{^2}{\rho ^2}+\frac{1}{\rho }\frac{}{\rho }\frac{|l|^2}{\rho ^2}+\frac{\gamma ^2}{\mathrm{}^2}\rho ^2+\frac{2\epsilon }{\mathrm{}^2}\right)R_{\epsilon ,l}=0,$$
(3.5)
and its solution reads as follows
$$R_{\epsilon ,l}(\rho )=N_{\epsilon ,l}(\sqrt{i\gamma /\mathrm{}}\rho )^{|l|}\mathrm{exp}(i\gamma \rho ^2/2\mathrm{}){}_{1}{}^{}F_{1}^{}(a,|l|+1,i\gamma \rho ^2/\mathrm{}),$$
(3.6)
with
$$a=\frac{1}{2}\left(|l|+1\frac{\epsilon }{i\gamma \mathrm{}}\right).$$
(3.7)
The normalization factor $`N_{\epsilon ,l}`$ is chosen such that
$$_0^{\mathrm{}}\overline{R_{\epsilon ,l}(\rho )}R_{\epsilon ^{},l}(\rho )\rho ๐\rho =\delta (\epsilon \epsilon ^{}).$$
(3.8)
It turns out (see Appendix A) that
$$N_{\epsilon ,l}=\sqrt{\frac{\gamma }{\pi |l|!}}(i)^a\mathrm{\Gamma }(a),$$
(3.9)
with $`a`$ defined in (3.7).
###### Proposition 2
The family of generalized eigenvectors $`\psi _{\epsilon ,l}`$ satisfy the following properties:
1. orthonormality
$$_0^{2\pi }_0^{\mathrm{}}\overline{\psi _{\epsilon ,l}(\rho ,\phi )}\psi _{\epsilon ^{},l^{}}(\rho ,\phi )\rho ๐\rho ๐\phi =\delta (\epsilon \epsilon ^{})\delta _{ll^{}},$$
(3.10)
2. completeness
$$\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}๐\epsilon \overline{\psi _{\epsilon ,l}(\rho ,\phi )}\psi _{\epsilon ,l}(\rho ^{},\phi ^{})=\frac{1}{\rho }\delta (\rho \rho ^{})\delta (\phi \phi ^{}).$$
(3.11)
Let us define another family of generalized energy eigenvectors
$$\chi _{\epsilon ,l}=๐ฏ\psi _{\epsilon ,l},$$
(3.12)
where the anti-unitary operator $`๐ฏ`$ is defined as follows
$$๐ฏ\psi _{\epsilon ,l}(\rho ,\phi )=\overline{R_{\epsilon ,l}(\rho )}\mathrm{\Phi }_l(\phi ).$$
(3.13)
It easy to show that Batemanโs Hamiltonian $`\widehat{H}`$ is $`๐ฏ`$โinvariant
$$๐ฏ\widehat{H}๐ฏ^{}=\widehat{H}.$$
(3.14)
Moreover, if $`\psi (t)=\widehat{U}(t)\psi _0`$, then $`๐ฏ\psi (t)=\widehat{U}(t)(๐ฏ\psi _0)`$, which shows that $`๐ฏ`$ is a time reversal operator. Finally, Proposition 2 gives rise to the following spectral representation of the Bateman Hamiltonian
$$\widehat{H}=\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}๐\epsilon E_{\epsilon ,l}|\psi _{\epsilon ,l}\psi _{\epsilon ,l}|=\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}๐\epsilon E_{\epsilon ,l}|\chi _{\epsilon ,l}\chi _{\epsilon ,l}|,$$
(3.15)
with $`E_{\epsilon ,l}`$ defined in (3.3).
## 4 Analyticity, resolvent and resonances
Now, we continue energy eigenfunctions $`\psi _{\epsilon ,l}`$ and $`\chi _{\epsilon ,l}`$ into complex $`\epsilon `$-plane. Note, that $`\epsilon `$-dependence enters $`R_{\epsilon ,l}`$ via the normalization factor $`N_{\epsilon ,l}`$ and the function $`{}_{1}{}^{}F_{1}^{}(a,|l|+1,i\gamma \rho ^2/\mathrm{})`$ ($`a`$ is $`\epsilon `$-dependent, see (3.7)). It is well known (see e.g. ) that confluent hypergeometric function $`{}_{1}{}^{}F_{1}^{}(a,b,z)`$ defines a convergent series for all values of complex parameters $`a`$, $`b`$ and $`z`$ provided $`an`$ and $`bm`$, with $`m`$ and $`n`$ positive integers. Moreover, if $`a=n`$ and $`bm`$, then $`{}_{1}{}^{}F_{1}^{}(a,b,z)`$ is a polynomial of degree $`n`$ in $`z`$. In our case $`b=|l|+1`$ which is never negative and hence $`{}_{1}{}^{}F_{1}^{}(a,|l|+1,i\gamma \rho ^2/\mathrm{})`$ is analytic in $`\epsilon `$. However, it is no longer true for the normalization constant $`N_{\epsilon ,l}`$ given by (3.9). The $`\mathrm{\Gamma }`$-function has simple poles at $`a=n`$, with $`n=0,1,2,\mathrm{}`$, which correspond to
$$\epsilon =\epsilon _{nl}=i\gamma \mathrm{}(|l|+2n+1),$$
(4.1)
on the complex $`\epsilon `$-plane. On the other hand the time-reversed function $`\overline{R_{\epsilon ,l}}`$ has simple poles at $`\epsilon =\overline{\epsilon _{nl}}=\epsilon _{nl}`$.
It is, therefore, natural to introduce two classes of functions that respect these analytical properties of $`\psi _{\epsilon ,l}`$ and $`\chi _{\epsilon ,l}`$. Recall that a smooth function $`f=f(\epsilon )`$ is in the Hardy class from above $`_+^2`$ (from below $`_{}^2`$) if $`f(\epsilon )`$ is a boundary value of an analytic function in the upper, i.e. $`\text{Im}\epsilon 0`$ (lower, i.e. $`\text{Im}\epsilon 0`$) half complex $`\epsilon `$-plane vanishing faster than any power of $`\epsilon `$ at the upper (lower) semi-circle $`|\epsilon |\mathrm{}`$. Define
$$\mathrm{\Phi }_{}:=\left\{\varphi ๐ฎ(_๐ฎ^2)\right|f(\epsilon ):=\chi _{\epsilon ,l}|\varphi _{}^2\},$$
(4.2)
and
$$\mathrm{\Phi }_+:=\left\{\varphi ๐ฎ(_๐ฎ^2)\right|f(\epsilon ):=\psi _{\epsilon ,l}|\varphi _+^2\},$$
(4.3)
where $`๐ฎ(_๐ฎ^2)`$ denotes the Schwartz space , i.e. the space of $`C^{\mathrm{}}(_๐ฎ^2)`$ functions $`f=f(u_1,u_2)`$ vanishing at infinity ($`|๐ฎ|\mathrm{}`$) faster than any polynomial.
It is evident from (3.13) that
$$\mathrm{\Phi }_+=๐ฏ(\mathrm{\Phi }_{}).$$
(4.4)
The main result of this section consists in the following
###### Theorem 1
For any function $`\varphi ^\pm \mathrm{\Phi }_\pm `$ one has
$`\varphi ^+={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}๐ฒ_{nl}^+๐ฒ_{nl}^{}|\varphi ^+,`$ (4.5)
and
$`\varphi ^{}={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}๐ฒ_{nl}^{}๐ฒ_{nl}^+|\varphi ^{}.`$ (4.6)
For the proof see Appendix B. The above theorem implies the following spectral resolutions of the Hamiltonian:
$$\widehat{H}_{}\widehat{H}|_\mathrm{\Phi }_{}=\underset{n=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}E_{nl}^{}|๐ฒ_{nl}^{}๐ฒ_{nl}^+|,$$
(4.7)
and
$$\widehat{H}_+\widehat{H}|_{\mathrm{\Phi }_+}=\underset{n=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}E_{nl}^+|๐ฒ_{nl}^+๐ฒ_{nl}^{}|.$$
(4.8)
In the above formulae $`E_{nl}^\pm `$ is given by (1.8).
The same techniques may be applied for the resolvent operator
$$\widehat{R}(z,\widehat{H})=\frac{1}{\widehat{H}z}.$$
(4.9)
One obtains
$`\widehat{R}_+(z,\widehat{H})`$ $`=`$ $`{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\epsilon }{E_{\epsilon ,l}z}}|\psi _{\epsilon ,l}\psi _{\epsilon ,l}||_{\mathrm{\Phi }_+}`$ (4.10)
$`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{E_{n,l}^+z}}|๐ฒ_{nl}^{}๐ฒ_{nl}^+|,`$
on $`\mathrm{\Phi }_+`$, and
$`\widehat{R}_{}(z,\widehat{H})`$ $`=`$ $`{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\epsilon }{E_{\epsilon ,l}z}}|\chi _{\epsilon ,l}\chi _{\epsilon ,l}||_\mathrm{\Phi }_{}`$ (4.11)
$`=`$ $`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{E_{n,l}^{}z}}|๐ฒ_{nl}^+๐ฒ_{nl}^{}|,`$
on $`\mathrm{\Phi }_{}`$. Hence, $`\widehat{R}_+(z,\widehat{H})`$ has poles at $`z=E_{nl}^+`$, and $`\widehat{R}_{}(z,\widehat{H})`$ has poles at $`z=E_{nl}^{}`$. As usual eigenvectors $`๐ฒ_{nl}^+`$ and $`๐ฒ_{nl}^{}`$ corresponding to poles of the resolvent are interpreted as resonant states. Note, that the Cauchy integral formula implies
$$\widehat{P}_{nl}^+:=|๐ฒ_{nl}^{}๐ฒ_{nl}^+|=\frac{1}{2\pi i}_{\mathrm{\Gamma }_{nl}^+}\widehat{R}_+(z,\widehat{H})๐z,$$
(4.12)
where $`\mathrm{\Gamma }_{nl}^+`$ is a clockwise closed curve that encircles the singularity $`z=E_{nl}^+`$. Similarly,
$$\widehat{P}_{nl}^{}:=|๐ฒ_{nl}^+๐ฒ_{nl}^{}|=\frac{1}{2\pi i}_{\mathrm{\Gamma }_{nl}^{}}\widehat{R}_{}(z,\widehat{H})๐z,$$
(4.13)
where $`\mathrm{\Gamma }_{nl}^{}`$ is an anti-clockwise closed curve that encircles the singularity $`z=E_{nl}^{}`$. One easily shows that
$$\widehat{P}_{nl}^\pm \widehat{P}_{n^{}l^{}}^\pm =\delta _{nn^{}}\delta _{ll^{}}\widehat{P}_{nl}^\pm ,$$
(4.14)
and hence the spectral decompositions of (4.7) and (4.7) may be written as follows:
$$\widehat{H}_\pm =\underset{n=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}E_{nl}^\pm \widehat{P}_{nl}^\pm .$$
(4.15)
Finally, let us note, that restriction of the unitary group $`\widehat{U}(t)=e^{i\widehat{H}t/\mathrm{}}`$ to $`\mathrm{\Phi }_\pm `$ no longer defines a group. It gives rise to two semigroups:
$$\widehat{U}_{}(t):=e^{i\widehat{H}_{}t/\mathrm{}}:\mathrm{\Phi }_{}\mathrm{\Phi }_{},\mathrm{for}t0,$$
(4.16)
and
$$\widehat{U}_+(t):=e^{i\widehat{H}_+t/\mathrm{}}:\mathrm{\Phi }_+\mathrm{\Phi }_+,\mathrm{for}t0.$$
(4.17)
Using (4.15) and the formula for $`E_{nl}`$ one finds:
$`\varphi ^{}(t)=\widehat{U}_{}(t)\varphi ^{}={\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}e^{i\omega lt}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}e^{\gamma (2n+|l|+1)t}\widehat{P}_{nl}^{},`$ (4.18)
for $`t0`$, and
$`\varphi ^+(t)=\widehat{U}_+(t)\varphi ^+={\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}e^{i\omega lt}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}e^{\gamma (2n+|l|+1)t}\widehat{P}_{nl}^+,`$ (4.19)
for $`t0`$. We stress that $`\varphi _t^{}`$ ($`\varphi _t^+`$) does belong to $`L^2(_๐ฎ^2)`$ also for $`t<0`$ ($`t>0`$). However, $`\varphi _t^{}\mathrm{\Phi }_{}`$ ($`\varphi _t^+\mathrm{\Phi }_+`$) only for $`t0`$ ($`t0`$). This way the irreversibility enters the dynamics of the reversed oscillator by restricting it to the dense subspace $`\mathrm{\Phi }_\pm `$ of $`L^2(_๐ฎ^2)`$.
From the mathematical point of view the above construction gives rise to so called rigged Hilbert spaces (or Gelโfand triplets) :
$$\mathrm{\Phi }_{}\mathrm{\Phi }_{}^{},$$
(4.20)
and
$$\mathrm{\Phi }_+\mathrm{\Phi }_+^{},$$
(4.21)
where $`\mathrm{\Phi }_\pm ^{}`$ denote dual spaces, i.e. linear functionals on $`\mathrm{\Phi }_\pm `$. Note, that generalized eigenvectors $`๐ฒ_{nl}^\pm `$ are not elements from $``$. However, they do belong to $`\mathrm{\Phi }_\pm ^{}`$. The first triplet $`(\mathrm{\Phi }_{},,\mathrm{\Phi }_{}^{})`$ is corresponds to the evolution for $`t0`$, whereas the second one $`(\mathrm{\Phi }_+,,\mathrm{\Phi }_+^{})`$ corresponds to the evolution for $`t0`$.
## 5 Batemanโs system in hyperbolic representation
In a recent paper Blasone and Jizba used another representation. They transform Batemanโs Hamiltonian (1.3) into the following form
$$H(y_1,y_2,w_1,w_2)=\frac{1}{2}(w_1^2w_2^2)\gamma (y_1w_2+y_2w_1)+\frac{1}{2}\omega ^2(y_1^2y_2^2),$$
(5.1)
with the new positions
$$y_1=\frac{x+y}{\sqrt{2}},y_2=\frac{xy}{\sqrt{2}},$$
(5.2)
and new canonical momenta
$$w_1=\frac{p_x+p_y}{\sqrt{2}},w_2=\frac{p_xp_y}{\sqrt{2}}.$$
(5.3)
Now, introducing hyperbolic coordinates $`(\varrho ,u)`$:
$$y_1=\varrho \mathrm{cosh}u,y_2=\varrho \mathrm{sinh}u,$$
(5.4)
the canonical quantization leads to the following Hamiltonian defined on the Hilbert space $`=L^2(_+,\varrho d\varrho )L^2(,du)`$:
$$\widehat{H}=\widehat{H}_0+\widehat{H}_{\mathrm{iho}},$$
(5.5)
with
$$\widehat{H}_0=\frac{\mathrm{}^2}{2}\mathrm{}_{\mathrm{\hspace{0.17em}2}}+\frac{\omega ^2}{2}\varrho ^2,$$
(5.6)
and the iho part $`\widehat{H}_{\mathrm{iho}}`$
$$\widehat{H}_{\mathrm{iho}}=i\gamma \mathrm{}\frac{}{u}.$$
(5.7)
In the above formulae $`\mathrm{}_{\mathrm{\hspace{0.17em}2}}`$ denotes the 2D wave operator, that is
$$\mathrm{}_{\mathrm{\hspace{0.17em}2}}=\frac{^2}{y_1^2}\frac{^2}{y_2^2}=\frac{^2}{\varrho ^2}+\frac{1}{\varrho }\frac{}{\varrho }\frac{1}{\varrho ^2}\frac{^2}{u^2}.$$
(5.8)
Clearly, in the $`(\varrho ,u)`$ variables the formula for $`\widehat{H}_{\mathrm{iho}}`$ considerably simplifies and $`\widehat{H}_{\mathrm{iho}}`$ represents the generator of SO(1,1) hyperbolic rotation on the $`(y_1,y_2)`$โplane. In this particular representation $`\widehat{H}_{\mathrm{iho}}`$ defines a self-adjoint operator on $`L^2(,du)`$. The corresponding eigen-problem is immediately solved
$$\widehat{H}_{\mathrm{iho}}\mathrm{\Phi }_\nu =\gamma \mathrm{}\nu \mathrm{\Phi }_\nu ,$$
(5.9)
with $`\mathrm{\Phi }_\nu (u)=e^{i\nu u}/\sqrt{2\pi }`$, and hence it reproduces the continuous spectrum of 2D iho $`\sigma (\widehat{H}_{\mathrm{iho}})=(\mathrm{},\mathrm{})`$. However, there is a crucial difference between elliptic $`(\rho ,\phi )`$ and $`(\varrho ,u)`$ representations. The generalized eigenvectors $`\mathrm{\Phi }_\nu `$ may be analytically continued on the entire complex $`\nu `$โplane. Therefore, the hyperbolic representation does not display the family of resonances corresponding to complex eigenvalues $`\epsilon _{nl}`$ defined in (4.1). Of course one may by hand fix the values of $`\nu `$ to $`\nu =i(2n+|l|+1)`$ but then the corresponding discrete $`\mathrm{\Phi }_{nl}`$ family is neither bi-orthogonal nor bi-complete (cf. Proposition 2).
To show how the complex eigenvalues of Blasone and Jizba appear let us consider $`\widehat{H}_0`$ defined in (5.6). Note, that $`\widehat{H}_0`$ resembles 2D harmonic oscillator given by (2.1). There is, however, crucial difference between $`\widehat{H}_0`$ and $`\widehat{H}_{\mathrm{ho}}`$. The hyperbolic operator โ$`\mathrm{}_{\mathrm{\hspace{0.17em}2}}`$โ, contrary to the elliptic one โ$`\mathrm{}_2`$โ, is not positively defined and hence it allows for negative eigenvalues. It is clear, since in the elliptic $`(\rho ,\phi )`$โrepresentation $`\widehat{H}_0=i\omega \mathrm{}_\phi `$ defines a self-adjoint operator on $`L^2([0,2\pi ),d\phi )`$ with purely discrete spectrum $`\omega \mathrm{}l`$ ($`l`$). Now, the spectral analysis of the Bateman Hamiltonian represented by (5.5) is straightforward:
$$\widehat{H}\psi _{ฯต\nu }=_{ฯต\nu }\psi _{ฯต\nu },$$
(5.10)
with
$$_{ฯต\nu }=ฯต+\gamma \mathrm{}\nu ,$$
(5.11)
and the following factorized form of $`\psi _{ฯต\nu }`$:
$$\psi _{ฯต\nu }(\varrho ,u)=_{ฯต\nu }(\varrho )\mathrm{\Phi }_\nu (u).$$
(5.12)
The radial function $`_{ฯต\nu }`$ solves
$$\widehat{H}_0_{ฯต\nu }=ฯต_{ฯต\nu },$$
(5.13)
and in analogy to (3.6) it is given by
$$_{ฯต\nu }(\varrho )=N_{ฯต\nu }(\sqrt{\omega /\mathrm{}}\varrho )^{i\nu }\mathrm{exp}(\omega \varrho ^2/2\mathrm{})U(b,i\nu +1,\omega \varrho ^2/\mathrm{}),$$
(5.14)
with
$$b=\frac{1}{2}\left(i\nu +1\frac{ฯต}{\mathrm{}\omega }\right).$$
(5.15)
In (5.14) we have used instead of the standard confluent hypergeometric function $`{}_{1}{}^{}F_{1}^{}`$ so called Tricomi function $`U`$ (see e.g. ).<sup>2</sup><sup>2</sup>2Actually, in (and also in ) this function is denoted by $`G`$. We follow the notation of Abramowitz and Stegun . It is defined by
$$U(a,c,z)=\frac{\mathrm{\Gamma }(1c)}{\mathrm{\Gamma }(ac+1)}_1F_1(a,c,z)+\frac{\mathrm{\Gamma }(c1)}{\mathrm{\Gamma }(a)}z_1^{1c}F_1(ac+1,2c,z).$$
(5.16)
A Tricomi function $`U(a,c,z)`$ is an analytical function of its arguments and for $`a=n`$ $`(n=0,1,2,\mathrm{})`$ it defines a polynomial of order $`n`$ in $`z`$:
$$U(n,\alpha +1,z)=(1)^nn!L_n^\alpha (z).$$
(5.17)
Moreover, using the following property of $`U`$ (an analog of (B.5) for $`{}_{1}{}^{}F_{1}^{}`$)
$$U(a,c,z)=z^{1c}U(1+ac,2c,z),$$
(5.18)
one obtains
$$_0^{\mathrm{}}\overline{_{ฯต\nu }(\varrho )}_{ฯต\nu }(\varrho )\varrho ๐\varrho =\frac{\mathrm{}}{2\omega }|N_{ฯต\nu }|^2_0^{\mathrm{}}z^{i\nu }e^zU^2(b,i\nu +1,z)๐z,$$
(5.19)
with $`z=\omega \varrho ^2/\mathrm{}`$. Now for $`b=n`$, $`_{ฯต\nu }`$ belongs to the Hilbert space $`L^2(_+,\varrho d\varrho )`$. It implies
$$ฯต=\mathrm{}\omega (2n+1+i\nu ).$$
(5.20)
and hence it reproduces discrete spectrum โ$`\mathrm{}\omega \times \mathrm{integer}`$โ iff $`i\nu =l=0,\pm 1,\pm 2,\mathrm{}`$. Now, using (5.17), (5.19) and
$$_0^{\mathrm{}}e^zz^\alpha L_n^\alpha (z)L_m^\alpha (z)๐z=\frac{1}{n!}\mathrm{\Gamma }(n+\alpha +1)\delta _{nm},$$
(5.21)
with $`\alpha >1`$, one obtains the following family $`_{nl}L^2(_+,\varrho d\varrho )`$:<sup>3</sup><sup>3</sup>3There is a difference in normalization factor in formulae (37) in . It follows from slightly different definition of $`L_n^\alpha `$.
$$_{nl}(\varrho )=\sqrt{\frac{2\omega /\mathrm{}}{n!\mathrm{\Gamma }(n+l+1)}}(\sqrt{\omega /\mathrm{}}\varrho )^l\mathrm{exp}(\omega \varrho ^2/2\mathrm{})L_n^l(\omega \varrho ^2/\mathrm{}),$$
(5.22)
with $`n=0,1,2,\mathrm{}`$, and $`l=0,1,2,\mathrm{}`$, satisfying
$$_0^{\mathrm{}}\overline{_{nl}(\varrho )}_{n^{}l}(\varrho )\varrho ๐\varrho =\delta _{nn^{}}.$$
(5.23)
We stress that the family $`_{nl}`$ is defined for $`l0`$ only (otherwise it can not be normalized!). Finally, defining
$$\varphi _{nl}(\varrho ,u)=\frac{1}{\sqrt{2\pi }}_{nl}(\varrho )e^{ul},$$
(5.24)
one has
$$\widehat{H}\varphi _{nl}=_{nl}\varphi _{nl},$$
(5.25)
with
$$_{nl}=\mathrm{}\omega (2n+l+1)i\mathrm{}\gamma l.$$
(5.26)
There is, however, crucial difference between families $`๐ฒ_{nl}^\pm (\rho ,\phi )`$ and $`\varphi _{nl}(\varrho ,u)`$. The family $`|๐ฒ_{nl}^\pm `$ corresponds to the poles of $`\psi _{\epsilon ,l}`$ from (3.2). No such correspondence holds for $`|\varphi _{nl}`$ and $`\psi _{ฯต\nu }`$ from (5.12). In particular there is no analog of Theorem 1 for $`|\varphi _{nl}`$. Moreover, $`_{nl}`$ contrary to $`E_{nl}`$ from (1.8) does not fit the formula for complex eigenvalues of Feshbach and Tikochinsky (see detailed discussion in ). It defines simply another family which is however not directly related to the spectral properties of the Bateman Hamiltonian.
## Appendix A
To compute $`N_{\epsilon ,l}`$ in (3.6) let us analyze the quantity $`I_\epsilon =_0^{\mathrm{}}\overline{R_{\epsilon ,l}(\rho )}R_{\epsilon ,l}(\rho )\rho ๐\rho `$. Clearly, this integral diverges ($`I_\epsilon =\delta (0)`$), however, its structure enables one the calculation of $`N_{\epsilon ,l}`$. One has
$$I_\epsilon =\frac{1}{2\gamma }|N_{\epsilon ,l}|^2_0^{\mathrm{}}z_1^{|l|}F_1(a,|l|+1,iz)_1F_1(\overline{a},|l|+1,iz)๐z,$$
(A.1)
where we defined $`z=\gamma \rho ^2`$. Now, the integral in (A.1) belongs to the general class
$$J=_0^{\mathrm{}}e^{\lambda z}z_1^{\mu 1}F_1(\alpha ,\mu ,kz)_1F_1(\alpha ^{},\mu ,k^{}z)๐z,$$
(A.2)
given by the following formula (see Appendix f in ):
$$J=\mathrm{\Gamma }(\mu )\lambda ^{\alpha \alpha ^{}\mu }(\lambda k)^\alpha (\lambda k^{})_2^\alpha ^{}F_1(\alpha ,\alpha ^{},\mu ;\frac{kk^{}}{(\lambda k)(\lambda k^{})}).$$
(A.3)
Using the above formula with $`\lambda =0`$, $`\mu =|l|+1`$, $`\alpha =a`$, $`\alpha ^{}=\overline{a}`$ and $`k=k^{}=i`$ one finds
$$I_\epsilon =\frac{1}{2\gamma }|N_{\epsilon ,l}|^2(i)^a\overline{(i)^a}_2F_1(a,\overline{a},|l|+1;1).$$
(A.4)
Finally, noting that
$$_2F_1(\alpha ,\beta ,\gamma ;1)=\frac{\mathrm{\Gamma }(\gamma )\mathrm{\Gamma }(\gamma \alpha \beta )}{\mathrm{\Gamma }(\gamma \alpha )\mathrm{\Gamma }(\gamma \beta )},$$
one has
$$I_\epsilon =\frac{|l|!}{2\gamma }|N_{\epsilon ,l}|^2(i)^a\overline{(i)^a}\frac{\mathrm{\Gamma }(0)}{\mathrm{\Gamma }(a)\mathrm{\Gamma }(\overline{a})}.$$
(A.5)
Therefore, comparing (A.5) with $`I_\epsilon =\delta (0)`$ one finds
$$N_{\epsilon ,l}=\sqrt{\frac{\gamma }{\pi |l|!}}(i)^a\mathrm{\Gamma }(a),$$
(A.6)
which proves (3.9).
## Appendix B
Due to the Gelโfand-Maurin spectral theorem an arbitrary function $`\varphi ^+\mathrm{\Phi }_+`$ may be decomposed with respect to the basis $`\psi _{\epsilon ,l}`$
$$\varphi ^+=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}_{\mathrm{}}^{\mathrm{}}๐\epsilon \psi _{\epsilon ,l}\psi _{\epsilon ,l}|\varphi ^+.$$
(B.1)
Now, since $`\psi _{\epsilon ,l}|\varphi ^+_+^2`$, we may close the integration contour along the upper semi-circle $`|\epsilon |\mathrm{}`$. Applying the Residue Theorem one obtains
$`\varphi ^+(\rho ,\phi )=2\pi i{\displaystyle \underset{l=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\text{Res}\psi _{\epsilon ,l}(\rho ,\phi )|_{\epsilon =\epsilon _{nl}}\psi _{\epsilon ,l}|\varphi ^+|_{\epsilon =\epsilon _{nl}}.`$ (B.2)
Using the well known formula for the residuum
$$\text{Res}\mathrm{\Gamma }(a)|_{a=n}=\frac{(1)^n}{n!},$$
(B.3)
one obtains
$$\text{Res}\psi _{\epsilon ,l}|_{\epsilon =\epsilon _{nl}}=\frac{i}{\sqrt{i^{2n+|l|+1}}}\sqrt{\frac{1}{2\pi \mathrm{}}}\sqrt{\frac{(n+|l|)!}{n!|l|!}}๐ฒ_{nl}^+.$$
(B.4)
Moreover, the analytical function $`\overline{\psi _{\epsilon ,l}}`$ computed at $`\epsilon =\epsilon _{nl}`$ reads:
$$\overline{\psi _{\epsilon ,l}}|_{\epsilon =\epsilon _{nl}}=i^{n+|l|+1}\sqrt{\frac{\gamma }{\pi |l|!}}(\sqrt{i\gamma /\mathrm{}}\rho )^{|l|}\mathrm{exp}(i\gamma \rho ^2/2\mathrm{}){}_{1}{}^{}F_{1}^{}(n+|l|+1,|l|+1,i\gamma \rho ^2/\mathrm{})\overline{\mathrm{\Phi }_l(\phi )}.$$
Due to the well known relation
$$_1F_1(a,b,z)=e_1^zF_1(ba,b,z),$$
(B.5)
one finds
$$\overline{\psi _{\epsilon ,l}}|_{\epsilon =\epsilon _{nl}}=\sqrt{i^{2n+|l|+1}}\sqrt{\frac{\mathrm{}}{2\pi }}\sqrt{\frac{n!|l|!}{(n+|l|)!}}\overline{๐ฒ_{nl}^{}}.$$
(B.6)
and hence the formula (4.5) follows. In a similar way one shows (4.6).
## Acknowledgments
This work was partially supported by the Polish State Committee for Scientific Research Grant Informatyka i inลผynieria kwantowa No PBZ-Min-008/P03/03.
|
warning/0506/hep-ph0506283.html
|
ar5iv
|
text
|
# Improved measurement of 2โข๐ฝ+๐พ
## I Introduction
CP violation is one of the most important topics in current particle physics research. In the standard model, CP violation arises due to a single complex phase in the Cabibbo-Kobayashi-Maskawa matrix $`V`$ ref:CKM . A major goal of $`B`$ meson physics is to measure the angles and sides of the CKM unitarity triangle. Theoretically clean measurement methods are crucial for obtaining these parameters accurately. The BaBar ref:Babar-sin2b and Belle ref:Belle-sin2b measurements of the parameter $`\mathrm{sin}(2\beta )`$, where $`\beta =\mathrm{arg}\left(V_{cd}V_{cb}^{}/V_{td}V_{tb}^{}\right)`$, confirm the standard model to within the precision of the experiments, and increased precision is expected in the coming years.
Crucial studies of the CKM mechanism and constraints on new physics can be obtained by measuring the CKM angle $`\gamma =\mathrm{arg}\left(V_{ud}V_{ub}^{}/V_{cd}V_{cb}^{}\right)`$. The greatest challenges presented by these measurements is that they require very large data samples and are subject to discrete ambiguities. It is therefore important to use every possible mode and method for measuring $`\gamma `$, and to devise methods that help resolve the ambiguities.
An important class of measurements makes use of decays such as $`BD^{}\pi ^+`$ to measure $`2\beta +\gamma `$. Proposed initially by Dunietz ref:dunietz , the first attempts to measure time-dependent CP asymmetries proportional to $`\mathrm{sin}(2\beta +\gamma )`$ and $`\mathrm{cos}(2\beta +\gamma )`$ have been conducted by BaBar ref:babar-sinbg and Belle ref:belle-sinbg using the modes $`BD^{()}\pi ^+`$ and $`BD^{}\rho ^+`$. While these measurements are currently statistically limited, their precision will become significant as more data are accumulated. At that stage, the greatest difficulty in extracting $`2\beta +\gamma `$ from these results will be the lack of precise knowledge of the ratio between the interfering amplitudes, defined as $`r|A(\overline{B}^0D^{()}h^+)/A(B^0D^{()}h^+)|`$, where $`h^+`$ indicates the light hadron $`\pi ^+`$ or $`\rho ^+`$.
In principle, $`r`$ may be obtained from the difference between the magnitudes of two terms with different time dependences in the decay rate. The relevant terms are $`(1+r^2)`$ and $`(1r^2)\mathrm{cos}\mathrm{\Delta }mt`$, where $`\mathrm{\Delta }m`$ is the $`B^0\overline{B}^0`$ oscillation frequency. However, with $`r๐ช(12\%)`$, extracting it from the $`๐ช(r^2)`$ difference between these $`๐ช(1)`$ terms requires prohibitively large data sets. Thus, the time-dependent measurement has negligible sensitivity to the value of $`r`$, which must therefore be obtained by assuming factorization and SU(3) symmetry to make use of the ratio of branching fractions $`(\overline{B}^0D_s^{()}h^+)/(B^0D^{()}h^+)`$. This approximation ignores the contribution of annihilation diagrams and some SU(3) breaking effects, and is taken to have a theoretical error of roughly 30% ref:babar-sinbg .
In $`BD^{}\rho ^+`$, the single parameter $`r`$ is replaced by a matrix $`\rho _{mn}`$ of ratios between the magnitudes of the $`bu\overline{c}d`$ and $`bc\overline{u}d`$ contributions of the of three different helicity amplitudes contributing to the decay. It has been shown London:2000zi that $`\rho _{mn}`$ may be obtained using only first-order $`๐ช(\rho _{mn})`$ terms. Not having to rely on small second-order terms or external input regarding amplitude ratios, this provides a much improved, theoretically clean measurement of $`2\beta +\gamma `$.
In this paper we generalize and extend that method to other decays that proceed through more than one intermediate state. Examples include $`BD^{}\rho ^+`$, which can interfere with $`BD^{}\rho ^+(1450)`$ and non-resonant $`BD^{}\pi ^+\pi ^0`$; $`BD^{}a_1^+`$, where non-resonant contributions are expected under the $`a_1`$ peak; and $`BD^{}\pi ^+`$, where interference between several excited charmed mesons may be realized in the decays $`D^{}D\pi `$ and $`D^{}D^{}\pi `$, in addition to possible contributions from non-resonant decays.
In all these cases, the interfering contributions have overlapping yet different distributions in relevant analysis variables. The first of these variables is the invariant mass squared $`s`$ of the final state of the resonance. The second variable $`s^{}`$ typically describes an angular distribution that is fully determined by the spin of the resonance. In the case of $`BD^{}a_1^+`$, $`s^{}`$ corresponds to the two variables of the Dalitz plot of the $`a_1`$ decay.
Our method applies equally well to modes with higher excitations, such as $`BD^{}\rho ^+`$, $`BD^{}a_1^+`$, $`BD^{}\rho ^+`$, and $`BD^{}a_1^+`$, in which $`s^{}`$ corresponds to several angular and mass-related variables. In addition to the interference between several resonances and non-resonant contributions, these decays involve several helicity amplitudes, which are treated as different intermediate states in our method.
## II Measuring $`\mathrm{๐}๐ท\mathbf{+}๐ธ`$
Let us consider a decay of the type described above, involving the interference of $`N`$ intermediate states. We denote the final state by $`f`$ ($`\overline{f}`$) if it contains a $`\overline{c}`$ ($`c`$) quark. The four decay amplitudes of interest are
$`A(B^0f)=A(\overline{B}^0\overline{f})`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{N}{}}}A_mg_m(s,s^{})e^{i\mathrm{\Delta }_m},`$
$`A(B^0\overline{f})`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{N}{}}}a_mg_m(s,s^{})e^{i(\delta _m+\gamma )},`$
$`A(\overline{B}^0f)`$ $`=`$ $`{\displaystyle \underset{m=1}{\overset{N}{}}}a_mg_m(s,s^{})e^{i(\delta _m\gamma )},`$ (1)
where $`\mathrm{\Delta }_m`$ ($`\delta _m`$) is the CP-conserving phase and $`A_m`$ $`(a_m)`$ is the magnitudes of the $`bc\overline{u}d`$ ($`bu\overline{c}d`$) decay amplitude proceeding via intermediate state $`m`$, and $`g_m(s,s^{})`$ is a known function of the final state variables $`s`$ and $`s^{}`$ that depends on the nature of the intermediate state $`m`$. For example, for $`f=D^{}\pi ^+\pi ^0`$ and $`m`$ being the index of the $`D^{}\rho ^+`$ intermediate state, $`g_m(s,s^{})=R(s)s^{}`$, where $`s`$ is the square of the $`\pi \pi ^0`$ invariant mass, $`R(s)`$ is a Breit-Wigner function, and $`s^{}`$ is the cosine of the angle between the momenta of the $`B`$ and of one of the pions, calculated in the $`\pi ^+\pi ^0`$ rest frame (the โhelicityโ angle). Vector-vector intermediate states, such as $`D^{}\rho ^+`$, must be further divided into the different helicity amplitude, each of which has a different $`s^{}`$ dependence.
With the above equations, the time-dependent decay rates for $`B^0(t)f`$ and $`B^0(t)\overline{f}`$ become
$`\mathrm{\Gamma }(B^0(t)f)=e^{\mathrm{\Gamma }t}{\displaystyle \underset{m,n}{}}`$ $`[`$ $`I_{mn}+C_{mn}\mathrm{cos}(\mathrm{\Delta }mt)S_{mn}^{}\mathrm{sin}(\mathrm{\Delta }mt)],`$ (2a)
$`\mathrm{\Gamma }(B^0(t)\overline{f})=e^{\mathrm{\Gamma }t}{\displaystyle \underset{m,n}{}}`$ $`[`$ $`I_{mn}C_{mn}\mathrm{cos}(\mathrm{\Delta }mt)S_{mn}^+\mathrm{sin}(\mathrm{\Delta }mt)],`$ (2b)
where for convenience we define the symbols
$`I_{mn}`$ $``$ $`{\displaystyle \frac{1}{2}}\left\{g_mg_n^{}\left(A_mA_ne^{i(\mathrm{\Delta }_n\mathrm{\Delta }_m)}+a_ma_ne^{i(\delta _n\delta _m)}\right)\right\},`$
$`C_{mn}`$ $``$ $`{\displaystyle \frac{1}{2}}\left\{g_mg_n^{}\left(A_mA_ne^{i(\mathrm{\Delta }_n\mathrm{\Delta }_m)}a_ma_ne^{i(\delta _n\delta _m)}\right)\right\},`$
$`S_{mn}^{}`$ $``$ $`\mathrm{Im}\left\{g_mg_n^{}A_na_me^{i(\delta _m\mathrm{\Delta }_n)}e^{i\varphi }\right\},`$
$`S_{mn}^+`$ $``$ $`\mathrm{Im}\left\{g_mg_n^{}A_ma_ne^{i(\delta _n\mathrm{\Delta }_m)}e^{i\varphi }\right\},`$
$`\varphi `$ $``$ $`(2\beta +\gamma ).`$
The decay rates for $`\overline{B}^0`$ decays are obtained from the $`B^0`$ rates by inverting the sign of the $`\mathrm{cos}(\mathrm{\Delta }mt)`$ and $`\mathrm{sin}(\mathrm{\Delta }mt)`$ terms. They double the statistics but do not yield additional information.
Next, we determine the conditions under which all the unknown parameters of Eqs. (2) can be obtained from the measurement, and show that these conditions are satisfied in the typical case of interfering Breit-Wigner resonances and a possible non-resonant contribution.
The three terms of Eq. (2a) are distinguishable based on their different time dependences, thus determining their coefficients. The relative differences between $`I_{mn}`$ and $`C_{mn}`$ are of order $`(a_ma_n)/(A_mA_n)r^210^4`$, which is practically unobservable. As a result, these terms yield the parameters $`A_m`$ and $`\mathrm{\Delta }_m`$, while $`a_m`$ and $`\delta _m`$ are measured from the coefficients of the $`\mathrm{sin}(\mathrm{\Delta }mt)`$ terms, as described later. To study the conditions for obtaining $`A_m`$ and $`\mathrm{\Delta }_m`$, we expand
$`{\displaystyle \underset{m,n}{}}A_mA_n\left\{g_mg_n^{}e^{i(\mathrm{\Delta }_n\mathrm{\Delta }_m)}\right\}`$ $`=`$ $`{\displaystyle \underset{m}{}}|g_m|^2A_m^2`$ (4)
$`+`$ $`2{\displaystyle \underset{m<n}{}}\mathrm{Re}(g_mg_n^{})A_mA_n\mathrm{cos}(\mathrm{\Delta }_n\mathrm{\Delta }_m)`$
$`+`$ $`2{\displaystyle \underset{m<n}{}}\mathrm{Im}(g_mg_n^{})A_mA_n\mathrm{sin}(\mathrm{\Delta }_n\mathrm{\Delta }_m).`$
If $`|g_m|^2`$, $`\mathrm{Re}(g_mg_n^{})`$, and $`\mathrm{Im}(g_mg_n^{})`$ all have different $`s`$ and/or $`s^{}`$ dependences, Eq. (4) yields $`N^2`$ unique observables, which is more than enough to determine the $`2N1`$ unknowns $`A_m`$ and $`\mathrm{\Delta }_m`$ (one of the $`\mathrm{\Delta }_m`$ phases is a global phase and can be chosen arbitrarily) for $`N2`$. This uniqueness condition is satisfied when all the $`g_m`$ are Breit-Wigner functions,
$$g_m(s)=\frac{M_m\mathrm{\Gamma }_m}{sM_m^2+iM_m\mathrm{\Gamma }_m},$$
(5)
even when all contributions have the same $`s^{}`$ dependence. A non-resonant contribution $`g_1=1`$ introduces $`N1`$ degenerate relations:
$$\mathrm{Im}(g_1g_m^{})=|g_m|^2,$$
(6)
where $`g_m(m>1)`$ is a Breit-Wigner function. In this case, the number of observables is reduced to $`N^2(N1)`$. However, a solution still exists for $`N2`$, and this solution is unambiguous when the non-resonant contribution is small enough relative to the resonant contributions. In addition, most practical cases involve resonances with total spins different from 0, and hence $`s^{}`$ dependences that distinguish them from a non-resonant $`s`$-wave contribution. This guarantees a unique solution of Eq. (4) in terms of $`A_m`$ and $`\mathrm{\Delta }_m`$.
We note that these conclusions do not depend on the assumption that the $`a_ma_n`$ terms in $`I_{mn}`$ and $`C_{mn}`$ are negligible. In fact, they apply equally well to the $`a_ma_n\{g_mg_n^{}e^{i(\delta _n\delta _m)}\}`$ terms in Eq. (LABEL:eq:defs).
We now show how the coefficients of the $`\mathrm{sin}(\mathrm{\Delta }mt)`$ terms in Eq. (2) yield the values of the remaining $`2N+1`$ unknowns, namely, $`a_m`$, $`\delta _m`$, and $`\varphi `$. The coefficients are
$`{\displaystyle \underset{mn}{}}S_{mn}^{}`$ $`=`$ $`{\displaystyle \underset{m}{}}A_ma_m|g_m|^2\mathrm{sin}(\varphi \pm \delta _{mm})`$ (7)
$`+`$ $`{\displaystyle \underset{m<n}{}}\mathrm{Im}(g_mg_n^{})[A_ma_n\mathrm{cos}(\varphi \pm \delta _{nm})\pm A_na_m\mathrm{cos}(\varphi \pm \delta _{mn})]`$
$`+`$ $`{\displaystyle \underset{m<n}{}}\mathrm{Re}(g_mg_n^{})[A_ma_n\mathrm{sin}(\varphi \pm \delta _{nm})+A_na_m\mathrm{sin}(\varphi \pm \delta _{mn})]`$
where $`\delta _{nm}\delta _n\mathrm{\Delta }_m`$. If $`|g_m|^2`$, $`\mathrm{Re}(g_mg_n^{})`$, and $`\mathrm{Im}(g_mg_n^{})`$ are all different, Eq. (7) yields $`N^2`$ observables for $`S_{mn}^{}`$ and $`N^2`$ for $`S_{mn}^+`$. It is therefore possible to obtain all the unknowns for $`N2`$.
## III Discrete Ambiguities
In the $`N=1`$ case, only the first line in Eq. (7) is non-vanishing. The measurement of $`\varphi `$ then suffers from an eight-fold ambiguity, due to the invariance of the observable $`\mathrm{sin}(\varphi \pm \delta _{mn})`$ under the three symmetry operations Silva:2002mt :
$$\begin{array}{ccccc}\hfill S_{\pi /2}& & \varphi \delta _{mm}+\pi /2\hfill & ,& \delta _{mm}\varphi \pi /2,\hfill \\ \hfill S_\pi ^{}& & \varphi \delta _{mm}+\pi \hfill & ,& \delta _{mm}\varphi +\pi ,\hfill \\ \hfill S_\pm ^{}& & \varphi \pi \varphi \hfill & ,& \delta _{mm}\delta _{mm}.\hfill \end{array}$$
(8)
In the typical $`N>1`$ case, $`S_{\pi /2}`$ and $`S_\pm ^{}`$ are no longer good symmetries, since they are broken by the $`\mathrm{cos}(\varphi \pm \delta _{mn})`$ terms. Furthermore, these terms are distinguishable from the $`\mathrm{sin}(\varphi \pm \delta _{mn})`$ terms by virtue of the different $`s`$ and/or $`s^{}`$ dependences of $`\mathrm{Re}(g_mg_n^{})`$ and $`|g_m|^2`$ or $`\mathrm{Im}(g_mg_n^{})`$. This further improves the measurement of $`\varphi `$.
## IV Discussion and Summary
In this paper, we have outlined the formalism for measuring $`2\beta +\gamma `$ with neutral $`B`$ meson decays involving interference between several intermediate states. We have shown that, despite involving a more complicated analysis, these decays have distinct advantages over $`BD^{()}\pi ^+`$, once our formalism is applied to their analysis, thus enhancing the overall precision with which $`2\beta +\gamma `$ is known.
First, as already noted for the special case of $`BD^{}V^+`$ decays London:2000zi , our method is sensitive to $`2\beta +\gamma `$ using only first order terms in the ratios $`a_m/A_n`$ between the $`bu\overline{c}d`$ and $`bc\overline{u}d`$ amplitudes. By contrast, in $`BD^{()}\pi ^+`$, or in the analysis of other decay modes that ignores the contribution of multiple intermediate states, one needs to extract $`r=a_1/A_1`$ from $`๐ช(1r^2)`$ terms, or rely on external measurements and incur a large theoretical uncertainty. Since $`r`$ is as small as $`12\%`$, this advantage is realized in our method even when the amplitude of one of the interfering intermediate states is much greater than the others.
Second, the $`BD^{}\pi ^+`$ measurement is subject to an eight-fold ambiguity, while in our method, the ambiguity is only two-fold.
Our method is not completely model-independent, since one has to assume specific forms for the $`g_m`$ functions, such as a Breit-Wigner for the resonances. However, this model dependence is much smaller than the 30% theoretical error estimated for $`r`$. Most resonances are well understood, and their shapes can be studied with the terms of Eqs. (4). In addition, the number of observables in Eq.(7) is greater than the number of unknowns when there are more than two intermediate states. The additional constraints may be used to further reduce the model dependence associated with some $`g_m`$ functions.
We emphasize that these conclusions and the formalism presented here do not depend on a specific final state, but apply whenever enough is known about the $`g_m(s,s^{})`$ functions for a solution to be obtainable, which in practice holds for a majority of the cases.
It is interesting to note some similarities and differences between the method we present here and methods developed for measuring $`\gamma `$ in $`BDK`$. Multi-body final state $`BDK`$ decays (such as $`B^{}DK^{}\pi ^0`$ Aleksan:2002mh , $`B^{}D^{}K^{}`$ Sinha:2004ct , $`B^{}D^{}K^{}`$ Sinha:1997zu , or $`BDK`$ with multi-body $`D`$ decays Giri:2003ty ) have been shown to improve the measurement of $`\gamma `$. This improvement is mostly due to the resolution of ambiguities and the ability to make efficient use of many $`B`$ and $`D`$ modes. As we have shown here, similar advantages are realized by interference between intermediate states in the measurement of $`2\beta +\gamma `$ with multi-body $`BD^{}\pi ^+`$-like modes. But in addition, these measurements benefit mostly from the fact that they do not depend on the very small $`r^2`$ terms. By contrast, $`BDK`$ decays are governed by the amplitude ratio $`r_B=|A(B^{}\overline{D}^0K^{})/A(B^{}D^0K^{})|1020\%`$, which is about an order of magnitude larger than $`r`$. Therefore, the sensitivity advantage brought about by interference between intermediate states is much greater in $`BD^{}\pi ^+`$ than in $`BDK`$.
## V Acknowledgments
We thank Yuval Grossman for useful comments. The work of A.S. was supported by the U.S. Department of Energy under contract DE-FG03-93ER40788. The work of N.S. was supported by the Department of Science and Technology, India.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.